A digital twin framework for civil engineering structures 土木工程结构的数字孪生框架
Matteo Torzoni , Marco Tezzele , Stefano Mariani , Andrea Manzoni , Matteo Torzoni , Marco Tezzele , Stefano Mariani , Andrea Manzoni 、Karen E. Willcox 卡伦-威尔库克斯 a Dipartimento di Ingegneria Civile e Ambientale, Politecnico di Milano, Piazza L. da Vinci 32, Milan, 20133, Italy a 意大利米兰 Politecnico di Milano, Piazza L. da Vinci 32, Milan, 20133,土木与环境工程系 Oden Institute for Computational Engineering and Sciences, University of Texas at Austin, Austin, 78712, TX, United States 美国德克萨斯大学奥斯汀分校奥登计算工程与科学研究所,美国德克萨斯州奥斯汀 78712c MOX, Dipartimento di Matematica, Politecnico di Milano, Piazza L. da Vinci 32, Milan, 20133, Italy c MOX,数学系,米兰理工大学,达芬奇广场 32 号,米兰,20133,意大利
A R T I C L E I N F O A R T I C L E I N F O R M A T I O N
Keywords: 关键词:
Digital twins 数字双胞胎
Predictive maintenance 预测性维护
Bayesian networks 贝叶斯网络
Deep learning 深度学习
Structural health monitoring 结构健康监测
Model order reduction 减少模型顺序
Abstract 摘要
A B S T R A C T The digital twin concept represents an appealing opportunity to advance condition-based and predictive maintenance paradigms for civil engineering systems, thus allowing reduced lifecycle costs, increased system safety, and increased system availability. This work proposes a predictive digital twin approach to the health monitoring, maintenance, and management planning of civil engineering structures. The asset-twin coupled dynamical system is encoded employing a probabilistic graphical model, which allows all relevant sources of uncertainty to be taken into account. In particular, the time-repeating observations-to-decisions flow is modeled using a dynamic Bayesian network. Real-time structural health diagnostics are provided by assimilating sensed data with deep learning models. The digital twin state is continually updated in a sequential Bayesian inference fashion. This is then exploited to inform the optimal planning of maintenance and management actions within a dynamic decision-making framework. A preliminary offline phase involves the population of training datasets through a reduced-order numerical model and the computation of a health-dependent control policy. The strategy is assessed on two synthetic case studies, involving a cantilever beam and a railway bridge, demonstrating the dynamic decision-making capabilities of health-aware digital twins A B S T R A C T 数字孪生概念为推进土木工程系统基于状态的预测性维护范例提供了一个极具吸引力的机会,从而可以降低生命周期成本、提高系统安全性和系统可用性。本研究提出了一种预测性数字孪生方法,用于土木工程结构的健康监测、维护和管理规划。资产-孪生耦合动态系统采用概率图形模型进行编码,可将所有相关的不确定性来源考虑在内。特别是,使用动态贝叶斯网络对从观测到决策的时间重复流进行建模。通过深度学习模型同化感知数据,提供实时结构健康诊断。数字孪生状态以连续贝叶斯推理的方式不断更新。然后在动态决策框架内,利用这些数据为维护和管理行动的优化规划提供信息。初步离线阶段包括通过简化的数字模型生成训练数据集,并计算与健康状况相关的控制策略。该策略在涉及悬臂梁和铁路桥梁的两个合成案例研究中进行了评估,展示了健康感知数字孪生的动态决策能力。
1. Introduction 1.导言
The optimal management of deteriorating structural systems is an important challenge in modern engineering. In particular, the failure or non-optimized maintenance planning of civil structures may entail high safety, economic, and social costs. Within this context, enabling a digital twin (DT) perspective for structural systems that are critical for either safety or operative reasons, is crucial to allow for condition-based or predictive maintenance practices, in place of customarily employed time-based ones. Indeed, having an up-to-date digital replica of the physical asset of interest can yield several benefits spanning its entire lifecycle, including performance and health monitoring, as well as maintenance, inspection, and management planning [1]. 对老化的结构系统进行优化管理是现代工程学面临的一项重要挑战。特别是,民用结构的故障或未优化的维护规划可能会带来高昂的安全、经济和社会成本。在这种情况下,对那些因安全或运行原因而至关重要的结构系统采用数字孪生(DT)视角,对于实现基于状态或预测的维护实践,以取代通常采用的基于时间的维护实践至关重要。事实上,拥有相关物理资产的最新数字副本可以在其整个生命周期内产生多种益处,包括性能和健康监测,以及维护、检查和管理规划[1]。
The DT concept [2-6] has been recently applied to several fields for operational monitoring, control, and decision support, including structural health monitoring (SHM) and predictive maintenance [7,8], additive manufacturing [9], smart cities [10], urban sustainability [11], and railway systems management [12]. It allows for a personalized characterization of a physical asset, in the form of computational models and parameters of interest, that evolves over time and is kept synchronized with its physical counterpart by means of data-collecting devices. Within a civil SHM framework, such a twinning perspective can be enabled by DT 概念[2-6]最近被应用于多个领域的运行监测、控制和决策支持,包括结构健康监测(SHM)和预测性维护[7,8]、增材制造[9]、智能城市[10]、城市可持续性[11]和铁路系统管理[12]。它允许以计算模型和相关参数的形式对物理资产进行个性化描述,并随着时间的推移而不断变化,通过数据采集设备与其物理对应物保持同步。在民用 SHM 框架内,可以通过以下方式实现这种孪生观点
Fig. 1. Predictive digital twin framework for civil engineering structures: graphical abstraction of the end-to-end information flow enabled by the probabilistic graphical model. 图 1.土木工程结构的预测性数字孪生框架:概率图形模型实现的端到端信息流的图形抽象。
the assimilation of data through data-driven structural health diagnostics (from physical to digital), possibly accommodating the quantification and propagation of relevant uncertainties related to, e.g., measurement noise, modeling assumptions, environmental and operational variabilities [13-17]. The resulting updated digital state should then enable prediction of the physical system evolution, as well as inform optimal planning of maintenance and management actions (from digital to physical). 通过数据驱动的结构健康诊断(从物理到数字)对数据进行同化,并可能考虑到与测量噪声、建模假设、环境和运行变异等相关的不确定性的量化和传播[13-17]。由此更新的数字状态应能预测物理系统的演变,并为维护和管理行动的优化规划提供信息(从数字到物理)。
In this work, we propose a DT framework for civil engineering structures. The overall computational strategy is based upon a probabilistic graphical model (PGM) inspired by the foundational model proposed in [18], which provides a general framework to carry out data assimilation, state estimation, prediction, planning, and learning. Formally, such a PGM is a dynamic Bayesian network with the addition of decision nodes, i.e., a dynamic decision network . This is employed to encode the end-to-end information flow, from physical to digital through assimilation and inference, and back to the physical asset in the form of informed control actions. A graphical abstraction of the proposed DT strategy is depicted in Fig. 1. The figure shows a physical-to-digital information flow and a digital-to-physical information flow. These bi-directional information flows repeat indefinitely over time. In particular, we have: 在这项工作中,我们提出了土木工程结构的 DT 框架。整体计算策略基于概率图形模型(PGM),该模型受文献[18]中提出的基础模型的启发,为数据同化、状态估计、预测、规划和学习提供了一个通用框架。从形式上看,这种 PGM 是一个增加了决策节点的动态贝叶斯网络,即动态决策网络 。它用于编码端到端的信息流,通过同化和推理从物理到数字,再以知情控制行动的形式返回到物理资产。图 1 展示了拟议 DT 战略的图形抽象。图中显示了从物理到数字的信息流和从数字到物理的信息流。这些双向信息流随着时间的推移无限重复。具体来说,我们有
From physical to digital. Structural response data are gathered from the physical system and assimilated with deep learning (DL) models, see e.g., [21,22], to estimate the current structural health in terms of presence, location, and severity of structural damage. To solve this inverse problem, we refer to vibration-based SHM techniques, see e.g., [23-26], which exploit the aforementioned collected data, such as displacement or acceleration time histories. This first estimate of the digital state is then employed to estimate an updated digital state, according to control-dependent transition dynamics models describing how the structural health is expected to evolve. 从物理到数字。从物理系统中收集结构响应数据,并与深度学习(DL)模型同化(参见文献[21,22]),以估计当前结构的健康状况,包括结构损坏的存在、位置和严重程度。为解决这一逆向问题,我们参考了基于振动的 SHM 技术,如 [23-26],该技术利用了上述收集的数据,如位移或加速度时间历程。然后,根据描述结构健康状况预期演变方式的控制相关过渡动力学模型,利用对数字状态的首次估计来估计更新的数字状态。
From digital to physical. The updated digital state is exploited to predict the future evolution of the physical system and the associated uncertainty, thereby enabling predictive decision-making about maintenance and management actions feeding back to the physical system. 从数字到物理。利用更新后的数字状态来预测物理系统的未来演变和相关的不确定性,从而对反馈到物理系统的维护和管理行动做出预测性决策。
Offline learning phase. The DT setup considered in this work takes advantage of a preliminary offline learning phase. This phase involves the training of the DL models underlying the structural health identification, and learning the control policy to be applied at each time step of the online phase. The DL models are trained in a supervised fashion, with labeled data pertaining to specific damage conditions generated by exploiting physics-based numerical models. To efficiently assemble a training dataset representative of potential damage and operational conditions the structure might undergo during its lifetime, we exploit a reduced-order modeling strategy for parametrized systems relying on the reduced basis method [27]. The health-dependent 离线学习阶段。本工作中考虑的 DT 设置利用了初步离线学习阶段。该阶段包括对结构健康识别所依据的 DL 模型进行训练,以及学习在线阶段每个时间步骤所要应用的控制策略。DL 模型采用监督方式进行训练,利用基于物理的数值模型生成的与特定损坏条件相关的标记数据。为了有效地收集代表结构在其寿命期内可能发生的潜在损坏和运行条件的训练数据集,我们采用了一种参数化系统的降阶建模策略,该策略依赖于降阶基方法[27]。与健康相关的
Fig. 2. Dynamic decision network encoding the asset-twin coupled dynamical system. Circle nodes denote random variables, square nodes denote actions, and diamond nodes denote the objective function. Bold outlines denote observed quantities, while thin outlines denote estimated quantities. Directed solid edges represent the variables' dependencies encoded via conditional probability distributions, while directed dashed edges represent the variables' dependencies encoded via deterministic functions. 图 2.编码资产-双胞胎耦合动态系统的动态决策网络。圆形节点表示随机变量,方形节点表示行动,菱形节点表示目标函数。粗体轮廓表示观测量,细体轮廓表示估计量。有向实线表示通过条件概率分布编码的变量依赖关系,有向虚线表示通过确定性函数编码的变量依赖关系。
control policy is also computed offline, by maximizing the expected future rewards for the planning problem induced by the PGM. 控制策略也是离线计算的,方法是使 PGM 诱导的规划问题的预期未来回报最大化。
The elements of novelty that characterize this work are the following: (i) the adaptation of the PGM digital twinning framework to the health monitoring, maintenance, and management planning of civil engineering structures; (ii) the assimilation of vibration response data is carried out by exploiting DL models, which allow automated selection and extraction of optimized damagesensitive features and real-time assessment of the structural state. This work shows how to incorporate in the DT framework high-dimensional multivariate time series describing the sensor measurements, while tracking the associated uncertainties. The proposed computational framework is made available in the public repository digital-twin-SHM [28]. The code implements the PGM framework as a dynamic decision network. This enables us to easily specify the graph topology from a few time slices, and then unroll it for any number of time steps in the future. 这项工作的新颖之处在于以下几点:(i) PGM 数字孪生框架适用于土木工程结构的健康监测、维护和管理规划;(ii) 利用 DL 模型对振动响应数据进行同化,从而自动选择和提取优化的损伤敏感特征,并对结构状态进行实时评估。这项工作展示了如何将描述传感器测量的高维多变量时间序列纳入 DT 框架,同时跟踪相关的不确定性。提出的计算框架可在公共存储库 digital-twin-SHM [28] 中获取。代码以动态决策网络的形式实现了 PGM 框架。这使我们能够轻松地从几个时间片中指定图拓扑结构,然后在未来任意数量的时间步骤中展开它。
The remainder of this paper is organized as follows. In Section 2, we describe the proposed DT framework. In Section 3, the computational procedure is assessed on two test cases, respectively related to a cantilever beam and a railway bridge. Conclusions and future developments are drawn in Section 4. 本文的其余部分安排如下。在第 2 节中,我们介绍了所提出的 DT 框架。第 3 节中,我们在两个测试案例中对计算程序进行了评估,这两个案例分别与悬臂梁和铁路桥梁有关。第 4 节是结论和未来发展。
2. Predictive digital twins using physics-based models and machine learning 2.利用物理模型和机器学习预测数字双胞胎
In this section, we describe the methodology characterizing our DT framework in terms of the PGM encoding the assettwin coupled dynamical system in Section 2.1; the population of training datasets exploiting physics-based numerical models in Section 2.2; and the DL models underlying the structural health identification in Section 2.3. 在本节中,我们将在第 2.1 节中介绍 DT 框架的方法论特征,包括编码资产双耦合动力系统的 PGM;第 2.2 节中介绍利用基于物理的数值模型的训练数据集群;以及第 2.3 节中介绍结构健康识别所依据的 DL 模型。
2.1. Probabilistic graphical model for predictive digital twins 2.1.预测数字双胞胎的概率图模型
The digital twin assimilates vibration recordings shaped as multivariate time series , consisting of time series made of sensor measurements equally spaced in time, for instance in terms of accelerations or displacements. The vector comprises the parameters representing the operational, damage, and (possibly) environmental conditions. Each recording refers to a time interval , within which measurements are recorded with a sampling rate . For the problem settings we consider, the interval is short enough for the operational, environmental, and damage conditions to be considered time-invariant, yet long enough not to compromise the identification of the structural behavior. 数字孪生系统将振动记录同化为多变量时间序列 ,由时间间隔相等的 传感器测量值(例如加速度或位移)组成的 时间序列。矢量 包括代表运行、损坏和(可能)环境条件的参数。每次记录指的是一个时间间隔 ,在该时间间隔内以采样率 记录测量值。对于我们所考虑的问题设置, 时间间隔足够短,可将运行、环境和损坏条件视为时间不变,但又足够长,不会影响结构行为的识别。
The PGM that defines the elements comprising the asset-twin coupled dynamical system, and mathematically describes the relevant interactions via observed data and control inputs, is the dynamic decision network sketched in Fig. 2. Circle nodes in the graph denote random variables at discrete times, square nodes denote actions, and diamond nodes denote the objective function. Bold outlines denote observed quantities, while thin outlines denote estimated quantities. The directed acyclic structure of the PGM encodes the assumed conditional dependencies. Edges in the graph represent dependencies between random variables. Solid edges represent the variables' dependencies encoded via conditional probability distributions, while dashed edges represent the variables' dependencies encoded via deterministic functions. 图 2 所示的动态决策网络定义了资产-孪生兄弟耦合动态系统的组成要素,并通过观测数据和控制输入对相关的相互作用进行了数学描述。图中的圆形节点表示离散时间的随机变量,方形节点表示行动,菱形节点表示目标函数。粗体轮廓表示观察到的数量,细体轮廓表示估计的数量。PGM 的有向无循环结构编码了假定的条件依赖关系。图中的边代表随机变量之间的依赖关系。实线表示通过条件概率分布编码的变量依赖关系,虚线表示通过确定性函数编码的变量依赖关系。
We consider a non-dimensional time discretization, and denote discrete time steps by . The physical time duration between successive time steps may vary depending on the application, and is governed by the update frequency of the DT via data assimilation, so that the DT is updated once per time step. Thanks to the modeled conditional dependencies between random variables, the graph topology is specified from the first two time steps, and can then be unrolled for any number of time steps. 我们考虑非一维时间离散化,用 表示离散时间步长。连续时间步之间的物理时间长度可能因应用而异,并受数据同化 DT 更新频率的制约,因此 DT 每个时间步更新一次。由于随机变量之间的条件依赖关系模型化,图拓扑结构从头两个时间步开始指定,然后可以在任意数量的时间步中展开。
The physical state , with denoting the realization of the random variable at time , encapsulates the variability in the health state of the asset, which is usually only partially observable. The probability distribution encoding the relative likelihood that , for any possible , is denoted with . The digital state is characterized by those parameters employed to capture the variability of the physical asset by means of the computational models comprising the DT. In our framework, the digital state is given as a vector of length two, describing the presence/location and magnitude of damage in the asset. The physical-todigital information flow is governed by the observed data , which are assimilated by the DT to update the digital state. The assimilation is carried out using the DL models described in Section 2.3, providing a first estimate of the digital state . This estimation is then used in a Bayesian inference formulation, together with the prior belief from the previous time step, to estimate an updated digital state according to a control-dependent transition dynamics model describing how the digital state is expected to evolve. The updated digital state can thus be exploited to compute quantities of interest , such as modal quantities or other response features, through the computational models comprising the DT. For instance, quantities of interest can be useful to perform posterior predictive checks on the tracking capabilities of the DT to assess how it matches the reality, by comparing sensor measurements with the corresponding posterior estimates. However, we point out that this capability is not exploited in the present work, and that the node is kept in the graph in agreement with the foundational model proposed in [18]. Nevertheless, the updated digital state is eventually exploited to inform the digital-to-physical information flow, in the form of control inputs; in Fig. 2, and denote the belief about what action to take and the control input effectively enacted on the asset, respectively. At each time step, is estimated according to a health-dependent control policy, that maps the belief over the digital state onto the control actions feeding back to the physical asset. Finally, the reward quantifies the performance of the asset for the time step and can be equivalently perceived as a negative cost to be maximized. 物理状态 ,其中 表示随机变量 在时间 的实现,它包含了资产健康状态的变化,通常只能部分观测到。对于任何可能的 ,编码 的相对可能性的概率分布用 表示。数字状态 的特征是通过构成 DT 的计算模型来捕捉物理资产变化的参数。在我们的框架中,数字状态是一个长度为 2 的向量,描述了资产中损坏的存在/位置和程度。物理-数字信息流受观测数据 的支配,DT 通过同化这些数据来更新数字状态。同化使用第 2.3 节所述的 DL 模型进行,提供数字状态的初步估计 。然后,根据描述数字状态预期演变方式的控制相关过渡动力学模型,将这一估计值与上一时间步的先验信 念 一起用于贝叶斯推理公式,以估计出更新的数字状态 。更新后的数字状态可用于计算 ,如模态量或其他响应特征。例如,通过比较传感器测量值和相应的后验估计值,相关数量可用于对 DT 的跟踪能力进行后验预测检查,以评估其与实际情况的匹配程度。 不过,我们要指出的是,本研究并没有利用这一功能,而是根据 [18] 中提出的基础模型,在图中保留了 节点。尽管如此,更新后的数字状态 最终还是会以控制输入的形式为数字到物理信息流提供信息;在图 2 中, 和 分别表示对资产采取何种行动的信念和有效实施的控制输入。在每个时间步骤中, 都会根据健康控制策略进行估算,该策略将数字状态的信念映射到反馈给物理资产的控制行动上。最后,奖励 量化了资产在该时间步长内的性能,可等同于需要最大化的负成本。
The key assumptions behind our PGM are that the physical state is only observable indirectly via the sensed structural response, and the physical and digital states evolve according to a Markovian process. This implies that the conditional probabilities associated with the random variables at one time step depend only on the random variables at the previous time step, and are independent of all past states. The resulting graph topology encodes a conditional independence structure that allows us to conveniently cast the asset tracking within a sequential Bayesian inference framework. Indeed, by exploiting conditional independence and Bayes rule, the joint distributions over variables can be factorized up to the current time step , as follows: 我们的 PGM 背后的关键假设是,物理状态只能通过感应到的结构响应间接观测到,物理状态和数字状态按照马尔可夫过程演化。这意味着与某一时间步的随机变量相关的条件概率仅取决于前一时间步的随机变量,而与过去的所有状态无关。由此产生的图拓扑结构编码了一种条件独立性结构,使我们能够方便地将资产跟踪置于顺序贝叶斯推理框架内。事实上,利用条件独立性和贝叶斯规则,变量的联合分布可以因式分解到当前时间步 ,如下所示:
with factors: 因素:
and factorize the belief about the digital state , conditioned on the digital state at the previous time step , the last enacted control input , and the data assimilation outcome . In our PGM, the spaces of the digital states and control inputs are discrete, thus the relevant causal relationships are modeled by means of conditional probability tables (CPTs). In particular, plays the role of a predictor forward in time, parametrized by means of a control-dependent CPT describing how the digital state is expected to evolve. Such a CPT should embody any a priori knowledge that the DT designer has with respect to the asset and the relevant operational conditions. can be estimated offline from historical data, see e.g., [29,30], or learned online from experience. On the other hand, updates the digital state estimate to account for data assimilation. This is encoded by means of a CPT mapping the estimate provided by the DL models, onto a belief about . Such a CPT is a confusion matrix measuring the offline (expected) performance of the DL models in correctly identifying the digital state among all the possible outcomes of and respectively encapsulate the evaluation of the computational models comprising the DT to estimate quantities of interest, and the computation of the reward function quantifying the performance of the asset. Finally, the control factor is encoded by means of a health-dependent control policy computed as described in the following. Since the spaces of the unobserved variables are discrete, we can propagate and update the relative belief exactly with a single pass of the sum-product message-passing algorithm [19]. 和 对数字状态的信念进行因式分解 ,条件是前一时间步骤的数字状态 、最后颁布的控制输入 和数据同化结果 。在我们的 PGM 中,数字状态和控制输入的空间是离散的,因此相关的因果关系是通过条件概率表(CPT)来建模的。特别是, 在时间上扮演着预测者的角色,通过与控制相关的 CPT 参数来描述数字状态的预期演变过程。这种 CPT 应包含 DT 设计者对资产和相关运行条件的任何先验知识。 可以根据历史数据离线估算(参见 [29,30] 等),也可以根据经验在线学习。另一方面, 更新数字状态估计,以考虑数据同化。这是通过将 DL 模型提供的估计值 与 的信念映射的 CPT 来编码的。这样的 CPT 是一个混淆矩阵,用于衡量 DL 模型在所有可能结果中正确识别数字状态的离线(预期)性能。 和 分别封装了由 DT 组成的计算模型的评估,以估算相关数量,以及量化资产性能的奖励函数的计算。最后,控制因素 是通过依赖健康的控制策略 进行编码的,计算方法如下所述。 由于未观测变量的空间是离散的,我们只需通过一次和积传递信息算法 [19],就能准确地传播和更新相对信念。
The control policy is computed offline under the simplifying assumption of sufficient sensing capability to provide an accurate estimate of the structural health, allowing us to decouple the sensing and control problems. This involves solving the planning problem induced by the expected evolution of the structural health, maximizing the expected reward over the planning horizon. Considering an infinite planning horizon, this can be stated as the optimization problem: 控制策略 是离线计算的,其简化假设是有足够的传感能力来提供对结构健康状况的准确估计,从而使我们能够将传感和控制问题分离开来。这就需要解决由结构健康状况的预期变化引起的规划问题,使规划期内的预期收益最大化。考虑到规划期限为无限期,可以将其视为优化问题:
Fig. 3. Dynamic decision network employed to predict the future evolution of the digital state and the associated uncertainty. Circle nodes denote random variables, and diamond nodes denote the objective function. Directed solid edges represent the variables' dependencies encoded via conditional probability distributions, while directed dashed edges represent the dependencies encoded via deterministic functions 图 3.用于预测未来数字状态演变和相关不确定性的动态决策网络。圆圈节点表示随机变量,菱形节点表示目标函数。有向实线表示通过条件概率分布编码的变量依赖关系,有向虚线表示通过确定性函数编码的变量依赖关系。
where is the discount factor. Here, this is solved using the dynamic-programming value iteration algorithm [31]. The reward function to be optimized is chosen as: 其中 是折扣系数。这里采用动态编程值迭代算法 [31]。需要优化的奖励函数选为
Herein, and quantify the rewards relative to control inputs and health state, respectively, and is a weighting factor, useful to tune the trade-off between risk-averse and risk-seeking behavior. After learning is selected as the best point estimate of . 在这里, 和 分别量化了相对于控制投入和健康状态的奖励, 是一个加权系数,用于调整规避风险行为和寻求风险行为之间的权衡。经过学习, 被选为 的最佳估计点。
Starting from the updated digital state at the current time step , future prediction is achieved by unrolling until a prediction time the portion of PGM relative to , and (see Fig. 3). All other nodes are removed from the prediction graph, as neither data assimilation nor actions are performed on the asset while forecasting its evolution. The factorization in Eq. (1) can be extended over the prediction horizon as: 从当前时间步骤 的更新数字状态 开始,通过展开至预测时间 PGM 相对于 和 的部分,实现未来预测(见图 3)。所有其他节点都从预测图中删除,因为在预测资产变化时,既不会对资产进行数据同化,也不会对资产采取任何行动。公式 (1) 中的因式分解可以在预测范围内扩展为
The algorithmic description of the online phase of the proposed digital twinning framework is reported in Algorithm 1. The operations repeat each time new observational data are provided. Note that the considered PGM digital twinning framework is general, and can easily be adapted to deal with physical assets other than civil engineering structures by reorganizing the topology of the graph, if necessary. 拟议数字孪生框架在线阶段的算法说明见算法 1。每次提供新的观测数据时,都会重复这些操作。需要注意的是,所考虑的 PGM 数字孪生框架是通用的,如有必要,可以通过重新组织图的拓扑结构,很容易地适应于处理土木工程结构以外的其他有形资产。
Algorithm 1 Online phase - algorithmic description
Input: observational data \(O_{t}=o_{t}\)
assimilate \(o_{t}\) with the DL models to provide \(D_{t}^{\mathrm{NN}}=d_{t}^{\mathrm{NN}}\). \(\quad \triangleright\left(O_{t}\right) \rightarrow\left(D_{t}^{\mathrm{NN}}\right)\)
infer \(D_{t}\) and \(U_{t}\) by updating \(d_{t-1}\), given \(u_{t-1}^{A}, d_{t}^{\mathrm{NN}}\), and the CPTs encoding \(\phi_{t}^{\text {history }}, \phi_{t}^{\mathrm{NN}}\) and \(\phi_{t}^{\text {control }} \triangleright\left(D_{t-1}, D_{t}^{\mathrm{NN}}, U_{t-1}^{A},\right) \rightarrow\left(D_{t}, U_{t}\right)\)
infer the future evolution of \(D_{t}\) and \(U_{t}\), given the updated \(d_{t}\), and the CPTs encoding \(\phi_{t}^{\text {history }}\) and \(\phi_{t}^{\text {control }} . \quad \triangleright\left(D_{t_{c}}\right) \rightarrow\left(D_{t_{p}}, U_{t_{p}}\right)\)
select \(U_{t}^{A}=u_{t}^{A}\) as the best point estimate of \(U_{t}=u_{t} . \quad \triangleright\left(U_{t}\right) \rightarrow\left(U_{t}^{A}\right)\)
return control input to be enacted \(u_{t}^{A}\), expected evolution of \(D_{t}\) and \(U_{t}\).
2.2. Numerical models for simulation-based damage identification 2.2.基于模拟的损伤识别数值模型
As anticipated in the previous section, the assimilation of structural response data to identify the structural state is carried out through DL models. A simulation-based strategy is exploited to train the DL models on the basis of vibration responses. The training data are numerically generated by simulating physics-based models so that the effect of damage on the structural response can be systematically reproduced [32]. In particular, the structure to be monitored is modeled as a linear-elastic continuum, discretized in space through finite elements. Its dynamic response to the applied loadings, under the assumption of linearized kinematics, is described by the following semi-discretized form of the elasto-dynamic problem: 如上一节所述,通过 DL 模型对结构响应数据进行同化以识别结构状态。在振动响应的基础上,采用基于模拟的策略来训练 DL 模型。训练数据是通过模拟物理模型数值生成的,因此可以系统地再现损伤对结构响应的影响[32]。具体而言,需要监测的结构被模拟为线性弹性连续体,并通过有限元在空间中离散化。在线性化运动学假设下,其对外加载荷的动态响应由以下弹性动力问题的半离散形式描述:
which is referred to as the full-order model (FOM). Here denotes time; are the vectors of nodal displacements, velocities and accelerations, respectively; is the number of degrees of freedom (dofs); is the mass matrix; is the damping matrix, assembled according to the Rayleigh's model; is the stiffness matrix; is the vector of nodal forces induced by the external loadings; and and are the initial conditions (at ), in terms of nodal displacements and velocities, respectively. The mass matrix is not a function of because the mass properties of the structure are unaffected by the employed damage description or by the operational conditions. The solution of Problem (11) is advanced in time using the Newmark integration scheme (constant average acceleration method) [33], to provide and , for , with being the vector of nodal displacements at time . 称为全阶模型(FOM)。 表示时间; 分别是节点位移、速度和加速度矢量; 是自由度(dofs)数; 是质量矩阵; 是阻尼矩阵,根据瑞利模型组装; 是刚度矩阵; 是由外部载荷引起的节点力矢量; 和 分别是节点位移和速度的初始条件(在 处)。质量矩阵 不是 的函数,因为结构的质量特性不受采用的损伤描述或运行条件的影响。利用纽马克积分法(恒定平均加速度法)[33]对问题 (11) 的解进行时间推进,以提供 和 , ,其中 是时间 的节点位移矢量。
With reference to civil structures, we focus on the early detection of damage patterns characterized by a small evolution rate, whose prompt identification can reduce lifecycle costs and increase the safety and availability of the structure. In this context, a localized reduction of the material stiffness stands as the simplest damage mechanism resulting from a time scale separation between damage growth and damage assessment, see e.g., [34-36]. Here, local stiffness reduction is obtained by parametrizing the stiffness matrix via two variables and , included in the parameter vector , respectively describing the location and magnitude of the applied stiffness reduction, similarly to [37-39]. In particular, labels the specific damage region, among a set of predefined damage locations, where identifies the damage-free baseline. The parameter describes the magnitude of the stiffness reduction taking place within the predesignated region associated with . 在民用结构方面,我们的重点是早期检测以小演化率为特征的损伤模式,对其进行及时识别可以降低生命周期成本,提高结构的安全性和可用性。在这种情况下,材料刚度的局部降低是最简单的损坏机制,其产生于损坏增长与损坏评估之间的时间尺度分离,参见 [34-36]。在这里,局部刚度减小是通过参数向量 中的两个变量 和 对刚度矩阵进行参数化而得到的,这两个变量分别描述了应用刚度减小的位置和幅度,与 [37-39] 类似。其中, 在一组预定义的 损伤位置中标示出特定的损伤区域, 标示出无损伤基线。参数 描述了与 相关的预定区域内发生的刚度降低幅度。
As increases, the computational cost associated with the solution of the FOM for any sampled also grows, and the generation of synthetic datasets becomes prohibitive. To address this challenge, a projection-based reduced-order model (ROM) is exploited in place of the FOM to speed up the offline dataset population phase, similarly to [38,39]. The ROM is obtained by a proper orthogonal decomposition (POD)-Galerkin reduced basis method [27,40-42]. This reduced-order modeling strategy is chosen because POD has been investigated and validated in the context of structural dynamics [43,44] and structural analysis [45,46], its appealing offline-online decoupling, and the availability of efficient criteria for the selection of POD basis functions. It is worth noting that alternative reduced-order modeling approaches can also be employed to alleviate the computational burden during the offline dataset generation. For instance, one could use spectral POD [47-49], or Grassmannian diffusion maps [50], as viable alternatives to the reduced basis method. 随着 的增加,为任何采样 求解 FOM 所需的计算成本也在增加,合成数据集的生成变得令人望而却步。为了应对这一挑战,与文献[38,39]类似,我们利用基于投影的降阶模型(ROM)来替代 FOM,从而加快离线数据集的生成。ROM 是通过适当正交分解 (POD) -Galerkin 降阶基方法获得的 [27,40-42]。之所以选择这种降阶建模策略,是因为 POD 已在结构动力学[43,44] 和结构分析[45,46] 中得到研究和验证,它的离线-在线解耦很有吸引力,而且有高效的 POD 基函数选择标准。值得注意的是,在离线数据集生成过程中,也可以采用其他降阶建模方法来减轻计算负担。例如,可以使用光谱 POD [47-49] 或格拉斯曼扩散图 [50],作为还原基方法的可行替代方法。
The ROM approximation to the solution of Problem (11) is obtained by linearly combining POD basis functions , as , where is the basis matrix collecting the POD basis functions and is the vector of unknown POD coefficients. By enforcing the orthogonality between the residual and the subspace spanned by the first POD modes through a Galerkin projection, the following -dimensional semi-discretized form is obtained: 通过线性组合 POD 基函数 ,得到问题 (11) 解的 ROM 近似值,如 ,其中 是收集 POD 基函数的基矩阵, 是未知 POD 系数向量。通过伽勒金投影法加强残差与第一个 POD 模式所跨子空间之间的正交性,可得到以下 -dimensional 半具体化形式:
The solution of this reduced-order system is advanced in time using the same strategy employed for the FOM model, and then projected onto the original FOM space as . Here, reduced matrices , and , and the reduced vector play the same role as their high-fidelity counterparts, yet with dimension instead of , according to the following relationships: 采用与 FOM 模型相同的策略,在时间上推进这一简化系统的求解,然后投影到原 FOM 空间,即 。根据以下关系,简化矩阵 和 以及简化矢量 与高保真矢量扮演相同的角色,但维度为 而非 :
The basis matrix is obtained by POD, exploiting the so-called method of snapshots as follows. First, a snapshot matrix is assembled from solution snapshots, computed by integrating in time the FOM solution for different values of parameters . The computation of an optimal reduced basis is then carried out by factorizing through a singular value decomposition. We use a standard energy-based criterion to set the order of the approximation. For further details see, e.g., . 基矩阵 由 POD 获得,利用所谓的快照法如下。 首先,通过对不同参数值的 FOM 解进行时间积分,计算出 解的快照矩阵 。然后,通过奇异值分解对 进行因式分解,计算出最佳还原基。我们使用基于能量的标准来设定近似的阶数 。更多详情,请参阅 等。
To populate the training dataset , the parametric space of vector is taken as uniformly distributed, and then sampled via the Latin hypercube rule. The number of samples is equal to the number of instances collected in as: 为了填充训练数据集 ,向量 的参数空间被视为均匀分布,然后通过拉丁超立方规则进行采样。样本数等于 中收集的实例数 :
where the vibration recordings associated with the -th sampling of , with , are labeled by the corresponding values of and , and are obtained as follows. With reference to displacement recordings, nodal values in are first collected as by solving Problem (12). The relevant vibration recordings are then obtained as: 其中,与 -th 采样相关的振动记录 与 ,与 ,由 和 的相应值标注,并按如下方式获得。关于位移记录,首先通过求解问题(12)收集 中的节点值作为 。然后得到相关的振动记录 如下:
where is a Boolean matrix whose -th entry is equal to 1 only if the -th sensor output coincides with the -th dof. In order to mimic the measurement noise, each vibration recording in is corrupted by adding an independent, identically distributed Gaussian noise, whose statistical properties depend on the target accuracy of the sensors. In the following, the index will be dropped for ease of notation, unless necessary. 其中, 是一个布尔矩阵,只有当 -th 传感器输出与 -th dof 重合时, -th 项才等于 1。为了模拟测量噪声, 中的每个振动记录都会被添加一个独立、同分布的高斯噪声所破坏,该噪声的统计特性取决于传感器的目标精度。在下文中,为便于记述,除非必要, 。
2.3. Data assimilation via artificial neural networks 2.3.通过人工神经网络进行数据同化
The factor in our PGM encodes the assimilation of observed data through the DL models underlying the identification of the structural health. In this section, we describe the adopted DL models, the aspects related to their training, and how they are used to assimilate observational data to detect, locate, and quantify the presence of structural damage. 在我们的 PGM 中, 因子通过作为结构健康识别基础的 DL 模型对观测数据进行同化。在本节中,我们将介绍所采用的 DL 模型、与其训练有关的方面,以及如何使用这些模型来同化观测数据,以检测、定位和量化结构损伤的存在。
Every time new observational data are acquired, they are first processed with a classification model to address damage detection/localization. Classification involves the prediction of an output class to categorize a given input. Here, the classes are those described through the parameter. Whenever a damage is identified in the -th region, , the observational data are further processed with regression models , one for each damageable region, to quantify the associated amount of damage . 每次获取新的观测数据 时,都会首先使用分类模型 进行处理,以解决损坏检测/定位问题。分类包括预测输出类别,对给定输入进行分类。这里的类别是指通过 参数描述的类别。每当在 -th 区域( )发现损坏时,就会使用回归模型 进一步处理观测数据 ,每个损坏区域一个回归模型,以量化相关的损坏量 。
The aforementioned classification and regression tasks are addressed by means of DL models. The use of DL models for SHM purposes has the advantage of automating the feature engineering stage characterizing the pattern recognition paradigm for SHM [35,51]. Indeed, a DL model is trained to select and extract optimized damage-sensitive features from raw sensor recordings through an end-to-end learning process. Moreover, since the DL model is learned offline, the structural state can be next assessed in real-time regardless of considering continuous or discrete variables, which would be difficult to achieve with other optimization techniques, such as nonlinear programming, stochastic optimization, and metaheuristic methods. 上述分类和回归任务可通过 DL 模型来完成。将 DL 模型用于 SHM 目的具有将特征工程阶段自动化的优势,这也是 SHM 模式识别范例的特点 [35,51]。事实上,通过端到端的学习过程,DL 模型可以从原始传感器记录中选择和提取优化的损伤敏感特征。此外,由于 DL 模型是离线学习的,因此无论考虑连续变量还是离散变量,都可以对结构状态进行下一步实时评估,而这是其他优化技术(如非线性编程、随机优化和元搜索方法)难以实现的。
The model addresses the multi-class classification task underlying the damage detection/localization problem, namely . The target label categorizes one of the predefined damage scenarios described through parameter . In particular, is a one-hot encoding , with entries equal to 1 if the target class is and 0 otherwise, with . This is needed because DL models cannot operate on nominal data directly. They require all input variables and output variables to be numeric. The one-hot encoding converts the nominal feature described by the parameter into a multidimensional binary vector. The number of dimensions corresponds to the number of categories, and each category gets its dimension. Each category is encoded by mapping it to a vector in which the entry corresponding to the category's dimension is 1 , and the rest are 0 . 该模型 解决了损害检测/定位问题的多类分类任务,即 。目标标签 对通过参数 描述的 预定义损坏情况之一进行分类。具体而言, 是一个单次编码 ,如果目标类别 是 ,则条目 等于 1,否则为 0, 。这是因为 DL 模型不能直接对名义数据进行操作。它们要求所有输入变量和输出变量都是数值型的。单次编码将 参数描述的标称特征转换为多维二进制向量。维数与类别数相对应,每个类别都有自己的维数。每个类别的编码方式是将其映射到一个向量中,其中与类别维度相对应的条目为 1,其余为 0。
The estimated counterpart of is obtained as . By employing a Softmax activation function for the output layer of , the entries of are interpreted as the confidence levels by which is assigned to the -th damage class, with . In particular, the Softmax activation function converts the real-valued vector , provided by the output layer of , into a discrete probability distribution as: 的估计对应值为 。通过对 的输出层使用 Softmax 激活函数, 的条目被解释为置信度 ,根据置信度, 被分配到 -th damage class, 。具体而言,Softmax 激活函数将 的输出层提供的实值向量 转换为离散概率分布:
When is exploited for prediction, the most likely class is selected as the one that best categorizes the processed measurements U. 当利用 进行预测时,最有可能的类别将被选择为最能将处理后的测量结果 U 分类的类别。
The model addresses the regression task underlying the damage quantification problem, namely , with . The estimated counterpart of is obtained as . Hence, the regression models, one for each damageable region, map the vibration recordings associated with the -th damage region, onto the estimated magnitude of the stiffness reduction taking place within the relative damage region. Since all models are learned following the same procedure, the index will be dropped in the following for ease of notation. 模型解决了损害量化问题的基础回归任务,即 ,与 。 的对应估计值为 。因此,每个可损坏区域都有一个回归模型,将与 -th 损坏区域相关的振动记录 映射到相对损坏区域内发生的刚度降低的估计幅度上。由于所有 模型的学习过程相同,为便于记述,下文中将去掉 这一索引。
Since the space of digital states in the PGM is discrete, the outcomes of and are accommodated within the PGM by discretizing the range in which the damage level can take values in uniform intervals, thus resulting in possible states. The same reasoning is followed to compute the confusion matrix encoding the factor. In particular, measures the offline performance of and in assimilating noisy FOM data to classify the digital state, among the possible outcomes of . 由于 PGM 中的数字状态空间是离散的, 和 的结果在 PGM 中是可以容纳的,方法是将损害程度 取值的范围离散化,在 的统一区间内取值,从而产生 的可能状态。同样的推理也适用于计算编码 因子的混淆矩阵。其中, 衡量的是 和 在同化噪声 FOM 数据对数字状态进行分类时,在 的 可能结果中的离线性能。
The models and are trained separately. The datasets dedicated to the training of and are derived from dataset in Eq. (14) as follows. The dataset used to learn is obtained from Eq. (14), as 模型 和 是分别训练的。在公式(14)中,用于训练 和 的数据集来自数据集 ,如下所示。用于学习 的数据集由公式 (14) 得出,即
The dataset used to learn is derived from Eq. (14), as 用于学习 的数据集由公式 (14) 得出,即
where is the number of training instances in , all characterized by a structural damage within the same predefined region. 其中 是 中的训练实例数,所有训练实例的特征都是在同一预定义区域内的结构损坏。
The set of weights and biases parametrizing is denoted as . This is optimized minimizing the probabilistic categorical cross-entropy between the predicted and target class labels over : 对 进行参数化的权重和偏差集表示为 。 上预测类别标签和目标类别标签之间的概率分类交叉熵 ,从而达到最优化:
which provides a measure of the distance between the discrete probability distribution describing , and its estimated counterpart . 它提供了描述 的离散概率分布与其估计的对应 之间距离的度量。
The set of weights and biases parametrizing is learned through the minimization of the following mean squared error loss function: 参数 的权重和偏差集是通过最小化以下均方误差损失函数来学习的:
Fig. 4. L-shaped cantilever beam: details of synthetic recordings related to displacements , loading condition, and predefined damage regions . 图 4.L 形悬臂梁:与位移 、加载条件和预定义损伤区域 有关的合成记录细节。
which provides a measure of the distance between the target magnitude of the stiffness reduction , and its approximated counterpart . 它提供了刚度降低的目标幅度 与它的近似值 之间距离的度量。
The algorithmic description of the procedures and computations characterizing the preliminary offline phase of the proposed digital twinning framework is reported in Algorithm 2. The implementation details of the deep learning models are reported in the Appendix. 算法 2 报告了拟议数字孪生框架初步离线阶段的程序和计算的算法描述。深度学习模型的实现细节见附录。
Algorithm 2 Preliminary offline phase - algorithmic description
Input: parametrization of the operational and damage conditions
PGM implementing the prediction graph
set up the physics-based numerical model of the structure to be monitored.
assemble the snapshot matrix of the structural response via FOM analyses.
compute the POD basis functions via singular value decomposition of the snapshots matrix.
use the ROM to populate the training dataset \(\mathcal{D}\) with vibration recordings at sensor location.
use the recordings and labels in \(\mathcal{D}\) to derive \(\mathcal{D}_{\mathrm{CL}}\) and \(\mathcal{D}_{\mathrm{RG}}\).
train the classification model \(\mathrm{NN}_{\mathrm{CL}}\) on \(\mathcal{D}_{\mathrm{CL}}\) and the regression models \(\mathrm{NN}_{\mathrm{RG}}\) on \(\mathcal{D}_{\mathrm{RG}}\).
test the generalization capabilities of \(\mathrm{NN}_{\mathrm{CL}}\) and \(\mathrm{NN}_{\mathrm{RG}}\) on noisy FOM data.
compute the confusion matrix encoding the \(\phi_{t}^{\mathrm{NN}}\) factor.
compute the control policy \(\pi\left(D_{t}\right)\) by solving the planning problem induced by the PGM.
return trained DL models, \(\phi_{t}^{\mathrm{NN}}\) factor, control policy \(\pi\left(D_{t}\right)\).
3. Numerical experiments 3.数值实验
This section demonstrates the proposed methodology for two test cases: an L-shaped cantilever beam and a railway bridge. 本节针对两个测试案例(L 型悬臂梁和铁路桥)演示了所提出的方法。
The FOM and ROM in Problem (11) and Problem (12) are implemented in the Matlab environment, using the redbKIT library [53]. The PGM framework for predictive digital twins is implemented in Python, using the pgmpy library [54]. All computations have been carried out on a PC featuring an AMD Ryzen 5950X CPU @ and 128 GB RAM. The NN architectures are implemented through the Tensorflow-based Keras API [55], and trained on a single Nvidia GeForce RTX 3080 GPU card. 问题(11)和问题(12)中的 FOM 和 ROM 是在 Matlab 环境中使用 redbKIT 库[53]实现的。预测数字双胞胎的 PGM 框架在 Python 中使用 pgmpy 库[54]实现。所有计算均在一台配备 AMD Ryzen 5950X CPU @ 和 128 GB 内存的电脑上进行。NN 架构通过基于 Tensorflow 的 Keras API [55]实现,并在一块 Nvidia GeForce RTX 3080 GPU 卡上进行训练。
3.1. L-shaped cantilever beam 3.1.L 型悬臂梁
The first test case deals with the L-shaped cantilever beam depicted in Fig. 4. The structure is made of two arms, each one having a length of , a width of , and a height of . The assumed mechanical properties are those of concrete: Young's modulus , Poisson's ratio , density . The structure is excited by a distributed vertical load , acting on an area of close to its tip. The load varies in time according to , with and , respectively being the load amplitude and frequency. Following the setup described in Section 2, these parameters have a uniform distribution within their respective ranges. 第一个测试案例是图 4 所示的 L 形悬臂梁。该结构由两个臂组成,每个臂的长度为 ,宽度为 ,高度为 。假定力学性能为混凝土的力学性能:杨氏模量 ,泊松比 ,密度 。该结构受到分布式垂直荷载 的激励,荷载作用于 靠近顶端的区域。荷载随时间变化,根据 , 和 ,分别为荷载振幅和频率。按照第 2 节所述设置,这些参数在各自范围内均匀分布。
Fig. 5. L-shaped cantilever beam - Confusion matrix measuring the offline performance of the DL models in correctly categorizing the digital state. Results are reported in terms of classification accuracy, measuring how observational data are classified with respect to the ground truth digital state. Digital states are ordered first for damage location and then for damage level. 图 5.L 形悬臂梁 - 衡量 DL 模型在正确分类数字状态方面离线性能的混淆矩阵。结果以分类准确率的形式进行报告,衡量观测数据如何根据基本真实数字状态进行分类。数字状态首先根据损坏位置排序,然后根据损坏程度排序。
3.1.1. Dataset assembly 3.1.1.数据集组装
Synthetic displacement time histories are obtained in relation to dofs along the bottom surface of the structure, to mimic the monitoring system depicted in Fig. 4. Each recording is provided for a time interval with an acquisition frequency . Recordings are corrupted with an additive Gaussian noise yielding a signal-to-noise ratio of 100 . 合成位移时间历程 与沿结构底面的 dofs 有关,以模拟图 4 所示的监测系统。每次记录的时间间隔为 ,采集频率为 。记录会受到加性高斯噪声的干扰,信噪比为 100。
In addition to the damage-free baseline condition, damage is simulated by considering possible damage classes, each referring to a reduction of the material stiffness within a subdomain , with , as depicted in Fig. 4. The stiffness reduction can occur with a magnitude , and is held constant within the considered time interval. 如图 4 所示,除了无损伤基线条件外,还通过考虑 可能的损伤类别来模拟损伤,每个类别指的是子域 内材料刚度的降低, 。刚度降低的幅度 ,并在考虑的时间间隔内保持不变。
The FOM is obtained with a finite element discretization using linear tetrahedral elements and resulting in dofs. The basis matrix is obtained from a snapshot matrix , assembled through 400 evaluations of the FOM, at varying values of the input parameters sampled via Latin hypercube rule. By prescribing a tolerance on the fraction of energy content to be disregarded in the approximation, the order of the ROM approximation turns out to be . FOM 是使用线性四面体元素进行有限元离散化得到的,结果是 dofs。基矩阵 由快照矩阵 获得,该快照矩阵通过对 FOM 的 400 次求值组合而成,输入参数的不同值 通过拉丁超立方规则采样。通过对近似中要忽略的能量含量部分设定容差 ,ROM 近似的阶数为 。
The dataset is built with instances collected using the ROM. This is then employed to train and , as described in the previous section. In the absence of experimental data, the testing phase of and of is carried out through noise-corrupted FOM solutions. In particular, the asset is monitored by processing batches of noisy observations, relative to the same damage location and damage magnitude , yet featuring varying operational conditions set by and . As the health of the asset evolves over time, the DT assimilates a batch of noisy observations at each time step, to dynamically estimate the variation in the structural health parameters underlying the digital state. 数据集 是利用 ROM 收集的 实例建立的。然后,如上一节所述,利用该数据集对 和 进行训练。在没有实验数据的情况下, 和 的测试阶段通过噪声干扰 FOM 解决方案进行。具体而言,通过处理成批的 噪声观测数据对资产进行监控,这些观测数据相对于相同的损坏位置 和损坏程度 ,但具有 和 设定的不同运行条件。当资产的健康状况随时间发生变化时,DT 会在每个时间步吸收一批噪声观测数据 ,以动态估计数字状态下结构健康参数的变化。
3.1.2. Digital twin framework 3.1.2.数字孪生框架
The two structural health parameters within the digital state are . In order to accommodate the outcome of the DL models within the PGM and to compute the CPT encoding the factor, the range in which the damage level can take values is discretized in intervals , thus resulting in possible digital states. The number of intervals and the width of each interval are chosen arbitrarily, and there are no restrictions in this respect. The resulting digital states are then sorted to follow the lexicographic order. 数字状态下的两个结构健康参数是 。为了在 PGM 中容纳 DL 模型的结果,并计算编码 因子的 CPT,损害程度 的取值范围被离散化为 个区间 ,从而产生 种可能的数字状态。 间隔的数量和每个间隔的宽度可任意选择,在这方面没有任何限制。然后按照词典顺序对得到的数字状态进行排序。
The confusion matrix reported in Fig. 5 measures the offline performance of and in assimilating noisy FOM data to classify the digital state, among the possible outcomes of . The (unknown) ground truth digital state is detected by the DL models with an overall classification accuracy of . Moreover, it can be argued from the confusion matrix that most of the misclassifications are due to the damage scenarios related to a stiffness reduction within or within . This is a quite expected outcome since measurements closer to the clamped side are only marginally affected by the presence of damage close to the free end of the beam, thus yielding a smaller sensitivity of sensor recordings to damage. This confusion matrix then serves as the CPT encoding the factor. 图 5 中的混淆矩阵衡量了 和 在同化噪声 FOM 数据以对数字状态进行分类方面的离线性能,其中包括 的可能结果。DL 模型检测到了(未知的)地面数字状态,其总体分类准确率为 。此外,从混淆矩阵中可以看出,大部分错误分类是由于 或 内的刚度降低造成的。这是意料之中的结果,因为靠近夹紧侧的测量结果只受到靠近梁自由端损坏的轻微影响,因此传感器记录对损坏的敏感度较低。该混淆矩阵可作为 因子的 CPT 编码。
For the present case, we consider four possible control inputs, each provided with a CPT modeling the transition probability from to after taking the action , and collectively encoding the factor. These internal models of how structural health is expected to evolve do not reflect the prescribed ground truth evolution, which is unknown to the DT. The considered control inputs are the following: 在本例中,我们考虑了四种可能的控制输入,每种输入都有一个 CPT,模拟了在采取 操作后从 到 的过渡概率 ,并共同编码了 因子。这些结构健康状况预期演变的内部模型并不反映规定的地面实况演变,因为地面实况演变对 DT 来说是未知的。考虑的控制输入如下
Do nothing (DN) action. There is no maintenance action planned in this case and the physical state will evolve according to a stochastic deterioration process. 不采取任何行动(DN)。在这种情况下,没有计划进行维护,物理状态将按照随机劣化过程发展。
Minor imperfect maintenance (MI) action. A maintenance action is performed and the asset may be restored from its current condition to a healthier damage state. This can be traced back to, e.g., patching and sealing cracked surfaces, rectifying and replacing expansion joints, or tightening/replacing loose/missing knot bolts for steel members. 轻微不完善维护 (MI) 行动。执行维护行动后,资产可能会从当前状态恢复到较健康的损坏状态。这可以追溯到诸如修补和密封开裂的表面、纠正和更换伸缩缝或拧紧/更换钢构件松动/缺失的连接螺栓等。
Major imperfect maintenance (MA) action. A maintenance action is performed and the asset may be restored from its current condition to a healthier damage state, with a higher probability of improvements than in the previous case. This can be traced back to, e.g., repairing heavily damaged slabs, piers, and steel members, and retrofitting compromised structural elements. 主要不完善维护 (MA) 行动。执行维护行动后,资产可能会从目前的状态恢复到较健康的损坏状态,改善的概率高于前一种情况。这可以追溯到修复严重损坏的楼板、桥墩和钢构件,以及改造受损的结构部件等。
Perfect maintenance (PM) action. A maintenance action is performed and the asset is restored from its current condition to the damage-free state. This can be traced back to the replacement of excessively compromised structural elements. 完美维护 (PM) 行动。执行维护行动后,资产将从当前状态恢复到无损坏状态。这可以追溯到更换过度损坏的结构元件。
3.1.3. Results: two available actions 3.1.3.结果:两个可用行动
We first illustrate the DT capabilities to assimilate observational data and track the structural health evolution, by restricting the available actions to DN and PM. The (unknown) ground truth evolution of structural health varies depending on the most recently applied control input, which can be either DN or PM. In the absence of maintenance, the physical state evolves following a deterioration process. We prescribe a (simulated) stochastic degradation process that monotonically deteriorates the structural health. The degradation process features a probability of damage inception equal to 0.5 . Damage may develop in any of the predefined regions with , and then propagate with increments sampled from a Gaussian probability density function (pdf) centered at and featuring a standard deviation equal to (negative increments are rounded to zero). The effect of a PM action is simulated by restoring the physical state to its undamaged configuration. At each time step during the operation, new observational data are simulated according to the (unknown) ground truth structural health and the most recently enacted control input. The DT assimilates the data and estimates the digital state, eventually suggesting the next control input to enact. Note that the prescribed trajectory of the structural health parameters is arbitrarily chosen to fully display the capabilities of the DT. Nevertheless, the DT would be equally capable of tracking the structural health evolution also considering either more or less aggressive degradation processes. 我们首先通过将可用操作限制为 DN 和 PM 来说明 DT 吸收观测数据和跟踪结构健康演变的能力。结构健康状况的(未知)基本真实演变取决于最近应用的控制输入,控制输入可以是 DN 或 PM。在没有维护的情况下,物理状态会随着恶化过程而变化。我们预设了一个(模拟的)随机退化过程,使结构健康状况单调恶化。退化过程的特点是损坏开始的概率 等于 0.5。损坏可能在 的任何预定区域发生,然后以 的增量传播,增量取自以 为中心的高斯概率密度函数 (pdf),标准偏差等于 (负增量四舍五入为零)。通过将物理状态恢复到未损坏的配置来模拟 PM 作用的效果。在运行过程中的每个时间步长,都会根据(未知的)地面真实结构健康状况和最新的控制输入模拟新的观测数据。DT 对数据进行同化,并对数字状态进行估计,最终提出下一个要执行的控制输入。请注意,为了充分展示 DT 的能力,结构健康参数的规定轨迹是任意选择的。尽管如此,DT 同样能够跟踪结构健康状况的变化,并考虑到更强或更弱的退化过程。
The state transition model encoding is conditioned on the most recently issued control input. The transition probability from to associated with the DN action assumes that damage may start in any subdomain , with , with probability 0.05 , and then grow to the next interval with the same probability. The transition model assumed for the PM action instead maps the belief to a belief associated with a damage-free condition, independently of the current condition. The corresponding CPTs are transition matrices, where the diagonal entries represent the probability of staying in the same state. The lower-left and upper-right triangles are associated with the probabilities of the system of deteriorating and improving its condition, respectively. Therefore, the DN transition matrix is a lower-left triangular matrix, with the highest probability assigned to remaining in the same state, consistent with what is expected for the deterioration of civil structures. The transition to the next interval is the second most likely transition, while improvements have a zero probability. Once the structure has reached the last interval, it remains in this condition with a probability equal to 1 . In contrast, the PM transition matrix is an upper-right triangular matrix with probabilities equal to 1 in the first row. 编码 的状态转换模型以最近发布的控制输入为条件。与 DN 操作相关的 从 到 的过渡概率假定,损坏可能从任何子域 开始,其中 的概率为 0.05,然后以相同的概率增长到下一个 区间。而 PM 动作所假设的转换模型则是将 的信念映射到与无损坏条件相关的信念 ,与当前条件无关。相应的 CPT 是过渡矩阵,其中对角线项表示保持相同状态的概率。左下角和右上角的三角形分别代表系统状态恶化和改善的概率。因此,DN 过渡矩阵是一个左下角三角形矩阵,保持相同状态的概率最高,这与土木工程结构恶化的预期一致。过渡到下一个 区间的可能性次之,而改进的可能性为零。一旦结构到达最后一个 区间,则保持这种状态的概率等于 1。相比之下, PM 过渡矩阵是一个右上三角矩阵,第一行的概率等于 1。
At each time step, the DT selects a control input to be enacted on the asset. Taking a DN action yields a positive reward, but also gives the chance of worsening the asset's structural health. On the other hand, the PM action responds to the degrading structural health, yet yields a negative reward. The computation of the costs associated with the health state and control inputs encapsulates the evaluation of the factor quantifying the performance of the asset. In particular, the two reward functions in Eq. (9) are defined as: 在每个时间步长内,DT 都会选择对资产实施控制输入 。采取 DN 操作会产生正回报,但也有可能导致资产结构健康状况恶化。另一方面,PM 操作会对不断恶化的结构健康状况做出响应,但会产生负回报。与健康状态和控制输入相关的成本计算包含了对量化资产性能的 因子的评估。具体而言,公式 (9) 中的两个奖励函数定义如下
where targets the cost assigned to each control input and measures the cost associated with the structural health state. These non-dimensional rewards represent indicative values the decision-maker is charged due to the condition of the structure. Although these values are not based on real data, actual values are not usually hard to find. State agencies and companies provide lists with services and costs [56]. The three cases in distinguish between the absence of damage, the presence of damage within the harm closed to the clamped side, and the presence of damage far from the clamp, respectively. Note how these penalize the progressive deterioration of the structural health as a function of can resemble a variety of aspects, like reduction in the level of service due to deterioration, working accidents, structural reliability, and structural failure probability [56]. 其中 代表分配给每个控制输入的成本, 代表与结构健康状态相关的成本。这些非维度奖励代表了决策者因结构状况而被收取的指示值。虽然这些数值并非基于真实数据,但实际数值通常并不难找到。国家机构和公司会提供服务和成本清单 [56]。 中的三种情况分别区分了无损坏、在与夹紧侧封闭的伤害范围内有损坏和远离夹紧侧有损坏。请注意,这些对结构健康状况逐渐恶化的惩罚是 的函数,可以类似于各种方面,如恶化导致的服务水平下降、工作事故、结构可靠性和结构失效概率[56]。
During the offline phase, we solve the planning problem induced by the PGM to compute the control policy , which maps the digital state belief to actions and encodes the control factor . The optimization of is carried out as described in Section 2.1, assuming a discount factor and a weighting factor . The computed control policy recommends that the asset operates until when and , respectively if and if , at which point it should be repaired. 在离线阶段,我们通过解决 PGM 诱导的规划问题来计算控制策略 ,该策略将数字状态信念映射为行动,并对控制因子 进行编码。 的优化过程如第 2.1 节所述,假定有一个贴现因子 和一个权重因子 。计算出的控制策略建议资产运行到 和 时,如果 和 ,则应分别修复。
Fig. 6 depicts a simulated online phase of the DT up to time step . Results are reported in terms of the (unknown) ground truth digital state, and the corresponding DT estimate after assimilating the observational data. The graphs report the evolution of the digital state only for the damaged regions, nevertheless, damage can potentially affect all predefined damageable regions. The DT proves capable of accurately tracking the digital state evolution with relatively low uncertainty. The corresponding estimation of the control inputs is reported in the bottom part of the figure, demonstrating that the DT is able to promptly suggest the PM action within one time step of when the (unknown) ground truth structural health demands it. 图 6 描述了直至时间步长 的 DT 模拟在线阶段。结果以(未知)地面实况数字状态和同化观测数据后的相应 DT 估计值来报告。图中仅报告了受损区域的数字状态变化,但受损可能会影响所有 预定义的可受损区域。事实证明,DT 能够以相对较低的不确定性准确跟踪数字状态的演变。图中下部显示了对控制输入的相应估算,表明 DT 能够在(未知的)地面真实结构健康状况需要时,在一个时间步长内及时建议 PM 操作。
Fig. 6. L-shaped cantilever beam - Online phase of the digital twin framework with two possible actions: DN (do nothing), and PM (perfect maintenance). Probabilistic and best point estimates of: (top) digital state evolution against the ground truth digital state; (bottom) control inputs informed by the digital twin, against the optimal control input under ground truth. In the top panels the background color corresponds to . In the bottom panel it corresponds to . 图 6.L 型悬臂梁 - 数字孪生框架的在线阶段,有两种可能的操作:DN(什么都不做)和 PM(完美维护)。概率和最佳点估算:(上图)数字状态演变与地面数字状态的对比;(下图)数字孪生提供的控制输入与地面数字状态下的最佳控制输入的对比。在上图中,背景颜色对应于 。下图中的背景颜色与 对应。
Fig. 7. L-shaped cantilever beam - Digital twin future predictions with two possible actions: DN (do nothing), and PM (perfect maintenance). The starting time is . In the top panel the probability relates to the amount of damage in . In the bottom panel it corresponds to . 图 7.L 型悬臂梁--数字孪生未来预测,有两种可能的操作:DN(什么都不做)和 PM(完美维护)。起始时间为 。顶部面板中的概率 与 中的损坏量有关。下图中的概率与 对应。
Fig. 7 depicts the predicted evolution of the digital state and of the corresponding informed control inputs, starting from . The prediction horizon is extended over 20 time steps in the future so that . The DT prediction engine informs about the expected future degradation of the structural health, allowing to plan future interventions. 图 7 描述了从 开始的数字状态和相应的知情控制输入的预测变化。预测范围扩展到未来 20 个时间步长,因此 。DT 预测引擎会告知未来结构健康状况的预期恶化情况,以便规划未来的干预措施。
3.1.4. Results: four available actions 3.1.4.结果:四项可用行动
We now consider all four possible control inputs. We prescribe a stochastic degradation process with a probability of damage inception equal to 0.5 . Damage may develop in any of the predefined regions with damage level sampled from a uniform distribution , and then propagate as in the previous case. This more aggressive degradation process is used to spot in a few time steps the effectiveness of the decision-making capabilities of the DT. The effect of the MI and MA actions on the asset is simulated according with stochastic repair processes, for which the structural health is forced to improve. The effect of a MI action is simulated with decrements sampled from a Gaussian pdf centered at and featuring a standard deviation equal to , 现在我们考虑所有四种可能的控制输入。我们预设了一个随机降解过程,损坏开始的概率 等于 0.5。损坏可能发生在任何一个预定义区域,损坏程度从均匀分布 中采样,然后像前一种情况一样传播。这种更激进的退化过程可用于在几个时间步骤内发现 DT 决策能力的有效性。根据随机修复过程模拟 MI 和 MA 操作对资产的影响,迫使结构健康状况得到改善。MI 操作的效果是通过 递减率模拟的,该 递减率取自以 为中心、标准偏差等于 的高斯 pdf、
Fig. 8. L-shaped cantilever beam - Online phase of the digital twin framework with four possible actions: DN (do nothing), PM (perfect maintenance), MI (minor imperfect maintenance), and MA (major imperfect maintenance). Probabilistic and best point estimates of: (top) digital state evolution against the ground truth digital state; (bottom) control inputs informed by the digital twin, against the optimal control input under ground truth. In the top panels the background color corresponds to . In the bottom panel it corresponds to . 图 8.L 型悬臂梁--数字孪生框架的在线阶段,有四种可能的操作:DN(什么都不做)、PM(完美维护)、MI(轻微不完美维护)和 MA(重大不完美维护)。概率和最佳点估算:(上图)数字状态演变与基本真实数字状态的对比;(下图)数字孪生提供的控制输入与基本真实下的最优控制输入的对比。在上图中,背景颜色对应于 。下图中的背景颜色与 对应。
while the effect of a MA action is modeled with decrements sampled from a Gaussian pdf centered at and featuring a standard deviation equal to . In both cases, the damage-free condition is assumed to be recovered if the resulting structural state features . 而 MA 作用的影响是通过 递减率建模的,该 递减率取自以 为中心、标准偏差等于 的高斯 pdf。在这两种情况下,如果得到的结构状态特征为 ,则假定无损伤状态得到恢复。
The transition model associated with the MI action assumes no improvement in the structural health with probability 0.1 , improvement of one interval with probability 0.75 , and improvement of two intervals with probability 0.15 . The resulting CPT is an upper-right triangular transition matrix, as deterioration from any state upon a repair action is assumed to have zero probability. The highest probability is assigned to improvements of one interval, followed by improvements of two intervals. There is also a lower probability of remaining in the same deteriorated state, which reflects a failed maintenance. Similarly, the MA action assumes no improvement with probability 0.05 , improvement of one interval with probability 0.3 , improvement of two intervals with probability 0.4 , and improvement of three intervals with probability 0.25 . In this case, the highest probability is assigned to improvements of two intervals, followed by improvements of one intervals, three intervals, and finally, the lowest probability is associated with the possibility of a failed maintenance. 与 MI 操作相关的过渡模型 假设结构健康状况没有改善的概率为 0.1,改善一个 间的概率为 0.75,改善两个 间的概率为 0.15。由此产生的 CPT 是一个右上三角转换矩阵,因为假定采取修复措施后从任何状态恶化的概率为零。改进一个 间隔的概率最高,其次是改进两个 间隔。保持相同恶化状态的概率也较低,这反映了维修失败。同样,MA 操作假设无改进的概率为 0.05,改进一个 间隔的概率为 0.3,改进两个 间隔的概率为 0.4,改进三个 间隔的概率为 0.25。在这种情况下,两个 间隔的改进概率最高,其次是一个 间隔的改进,再次是三个 间隔的改进,最后是与维护失败的可能性相关的最低概率。
The two reward functions in Eq. (9) are chosen as: 公式 (9) 中的两个奖励函数选为
We assume a discount factor , and a weighting factor . The resulting control policy recommends that the asset should operate until when , after which: if , a MI action should be performed when , and a PM action should be performed when ; while, if , the MI and MA actions should be performed, respectively when and when , and a PM action should be performed when . 我们假设贴现率为 ,加权系数为 。由此得出的控制策略 建议资产运行到 时,之后:如果 ,则应在 时执行 MI 操作,在 时执行 PM 操作;如果 ,则应分别在 和 时执行 MI 和 MA 操作,在 时执行 PM 操作。
Fig. 9. L-shaped cantilever beam - Digital twin future predictions with four possible actions: DN (do nothing), PM (perfect maintenance), MI (minor imperfect maintenance), and MA (major imperfect maintenance). The starting time is . In the top panel the probability relates to the amount of damage in . In the bottom panel it corresponds to . 图 9.L 型悬臂梁--数字孪生未来预测,四种可能的操作:DN(什么都不做)、PM(完美维护)、MI(轻微不完美维护)和 MA(重大不完美维护)。起始时间为 。顶部面板中的概率 与 中的损坏量有关。下图中的概率与 对应。
Fig. 8 depicts a simulated online phase of the DT up to . The DT accurately tracks the digital state evolution and timely suggests the appropriate control inputs most of the time. In particular, the DT proposes the optimal control input, except for the time steps and featuring a sub-optimal action. In both cases, a MI action is proposed in place of a DN, because the DT estimates a instead of a related to a stiffness reduction within . This is in line with what was observed in the confusion matrix of Fig. 5, due to the limited sensitivity of recordings to damage scenarios affecting the terminal region of the beam. This peculiar type of misclassification turns out to be the most pathological in the confusion matrix and is therefore capable of potentially spoiling the assimilation of observational data. Nevertheless, the DT reverts to correctly tracking the structural health of the asset within one time step. 图 8 描述了 DT 的模拟在线阶段,最高可达 。DT 准确地跟踪了数字状态的演变,并在大多数情况下及时提出了适当的控制输入。特别是,除了 和 这两个时间步骤采用了次优操作外,DT 都提出了最优控制输入。在这两种情况下,由于 DT 估计的是 ,而不是与 内刚度降低有关的 ,因此提出了 MI 操作,而不是 DN。这与图 5 的混淆矩阵中观察到的情况一致,这是因为记录对影响梁末端区域的损坏情况的灵敏度有限。在混淆矩阵中,这种特殊类型的错误分类最为严重,因此有可能破坏观测数据的同化。尽管如此,DT 仍能在一个时间步长内正确跟踪资产的结构健康状况。
Fig. 9 depicts the predicted evolution of the digital state and control inputs, from and over 20 time steps in the future. The DT prediction correctly suggests taking with high probability a MA action, followed by two MI actions, and accordingly predicts the corresponding evolution of the structural health. Comparing the DT prediction with what is effectively experienced during the online phase (see Fig. 8), note how the DT prediction closely resembles the actual evolution of the digital state and control inputs. This is a remarkable result in terms of DT prediction capabilities, since the DT is not aware of the future values of the structural health parameters, and the relative transition models do not match their real (stochastic) evolution. 图 9 描述了从 到未来 20 个时间步的数字状态和控制输入的预测变化。DT 预测正确地建议采取高概率的 MA 操作,随后是两个 MI 操作,并相应地预测了结构健康状况的相应变化。将 DT 预测与在线阶段的实际情况进行比较(见图 8),可以发现 DT 预测与数字状态和控制输入的实际演变非常相似。就 DT 预测能力而言,这是一个了不起的结果,因为 DT 并不知道结构健康参数的未来值,而且相对过渡模型与它们的实际(随机)演变并不匹配。
3.2. Railway bridge 3.2.铁路桥
The second case study concerns the railway bridge depicted in Fig. 10. It is an integral concrete portal frame bridge located along the Bothnia line in the Swedish suburbs of Hörnefors. It features a span of , a free height of , and a width of (edge beams excluded). The thickness of the structural elements is for the deck, for the frame walls, and for the wing walls. The bridge is founded on two plates connected by stay beams and supported by pile groups. The concrete is of class C35/45, whose mechanical properties are: . The superstructure consists of a single track with sleepers spaced apart, resting on a ballast layer deep, wide and featuring a density . The geometrical and mechanical modeling data have been adapted from former research activities on the relevant soil-structure interaction, see . 第二个案例研究涉及图 10 所示的铁路桥。这是一座整体式混凝土门式框架桥,位于瑞典赫内福斯郊区的 Bothnia 线沿线。它的跨度为 ,自由高度为 ,宽度为 (不包括边梁)。桥面结构件的厚度为 ,框架墙的厚度为 ,翼墙的厚度为 。桥梁的基础是两块板,由留置梁连接,并由桩组支撑。混凝土的等级为 C35/45,其机械性能为: .上部结构由单轨组成,枕木间距为 ,铺设在 深、 宽的道碴层上,密度为 。几何和机械建模数据取自以前有关土壤与结构相互作用的研究活动,见 。
The bridge is subjected to the transit of Gröna Tåget trains type, at a speed . Only trains composed of two wagons are considered, thus characterized by 8 axles, each one carrying a mass ton. The corresponding load model is described in [38], and consists of 25 equivalent distributed forces transmitted by the sleepers to the deck through the ballast layer with a slope 4:1, according with Eurocode 1 [59]. 该桥承受 Gröna Tåget 型列车的通过,速度为 。只考虑由两节车厢组成的列车,因此有 8 根车轴,每根车轴的质量为 吨。相应的荷载模型见 [38],包括 25 个由枕木通过坡度为 4:1 的道碴层传递到桥面的等效分布力,符合 Eurocode 1 [59]。
3.2.1. Dataset assembly 3.2.1.数据集组装
Synthetic displacement time histories are obtained from sensors deployed as depicted in Fig. 11. Each recording is provided for a time interval with an acquisition frequency . This setting allows to record train passages at the lowest speed of , and properly catches the structural response at the maximum speed of . Recordings are corrupted with an additive Gaussian noise yielding a signal-to-noise ratio of 120 . 合成位移时间历程 由 传感器获取,如图 11 所示。每次记录的时间间隔为 ,采集频率为 。这样的设置可以在 的最低速度下记录列车通过情况,并在 的最高速度下正确捕捉结构响应。记录被加性高斯噪声干扰,信噪比为 120。
In addition to the undamaged condition, the presence of damage in the structure is accounted for using a localized stiffness reduction that can take place within predefined subdomains , with , as depicted in Fig. 11. The stiffness reduction can occur with a magnitude , and is kept fixed while a train travels across the bridge. 如图 11 所示,除了未损坏状态外,结构中存在的损坏可通过在 预定义的子域 内进行局部刚度降低来考虑,子域为 。刚度减小的幅度为 ,并在火车通过桥梁时保持固定。
Fig. 10. Hörnefors railway bridge. 图 10.赫内福斯铁路桥。
Fig. 11. Railway bridge: details of synthetic recordings related to displacements , and predefined damage regions . 图 11.铁路桥:与位移相关的合成记录详情 ,以及预定义损坏区域 。
The FOM features dofs, resulting from a finite element discretization with an element size of and a reduced size of for the deck, to enable a smooth propagation of the traveling load. The presence of the ballast layer is accounted for through an increased density for the deck and for the edge beams. The embankments are accounted for through distributed springs, modeled as a Robin mixed boundary condition (with elastic coefficient ) applied on the surfaces facing the ground. The structural dissipation is modeled by means of a Rayleigh's damping matrix, assembled to account for a damping ratio on the first two structural modes. FOM 的特点是 dofs,这是有限元离散化的结果,其元素尺寸为 ,甲板的尺寸减小到 ,从而使行驶荷载得以平稳传播。压载层的存在是通过增加桥面和边梁的密度来考虑的。路堤通过分布式弹簧进行计算,在面向地面的表面采用罗宾混合边界条件(弹性系数为 )建模。结构耗散通过雷利阻尼矩阵建模,该矩阵用于计算 前两个结构模态的阻尼比。
The ROM is obtained from a snapshot matrix S, assembled through 400 evaluations of the FOM for different values of parameters . By setting the error tolerance to POD modes are to be considered. ROM 由一个快照矩阵 S 得出,该快照矩阵是通过对不同参数值的 FOM 进行 400 次评估得出的 。通过将误差容限设置为 ,可以考虑 POD 模式。
The training dataset is built with instances collected using the ROM. Also in this case, the testing phase of and of is carried out considering noisy FOM solutions. The monitoring of the asset is then simulated by assimilating noisy observations at each time step. As the structural health of the bridge evolves over time, the DT estimates the variation in the structural health parameters every time a train travels across the bridge. 训练数据集 是利用 ROM 收集的 实例建立的。在这种情况下, 和 的测试阶段也要考虑到 FOM 的噪声解决方案。然后,通过在每个时间步吸收 的噪声观测数据,模拟对资产的监测。由于桥梁的结构健康状况会随着时间的推移而发生变化,因此每次列车通过桥梁时,DT 都会估算出结构健康参数的变化情况。
3.2.2. Digital twin framework 3.2.2.数字孪生框架
As in the previous case, the two structural health parameters within the digital state are . The range in which the damage level can take values is discretized in intervals. The resulting possible digital states are sorted first for damage location and then for damage level. 与前一种情况一样,数字状态下的两个结构健康参数是 。损坏程度 的取值范围按 间隔离散。由此产生的 可能的数字状态首先按损坏位置排序,然后按损坏程度排序。
The confusion matrix measuring the offline performance of and of in correctly categorizing the digital state is reported in Fig. 12. The ground truth digital state is detected with an overall classification accuracy of . In this case, the majority of misclassifications are due to confusing adjacent digital states relative to the same damage location, thus yielding a tridiagonal band matrix. 衡量 和 在正确分类数字状态方面离线性能的混淆矩阵见图 12。地面数字状态检测的总体分类准确率为 。在这种情况下,大部分错误分类都是由于混淆了相对于同一损坏位置的相邻数字状态,从而产生了一个三对角带状矩阵。
For the present case, we consider the following three possible control inputs: 在本例中,我们考虑了以下三种可能的控制输入:
Do nothing (DN) action. There is no maintenance action planned in this case and the physical state will evolve according to a stochastic deterioration process. 不采取任何行动(DN)。在这种情况下,没有计划进行维护,物理状态将按照随机劣化过程发展。
Fig. 12. Railway bridge - Confusion matrix measuring the offline performance of the DL models in correctly categorizing the digital state. Results are reported in terms of classification accuracy, measuring how observational data are classified with respect to the ground truth digital state. Digital states are ordered first for damage location and then for damage level. 图 12.铁路桥梁 - 衡量 DL 模型在正确分类数字状态方面离线性能的混淆矩阵。结果以分类准确率的形式报告,衡量了观测数据相对于基本真实数字状态的分类情况。数字状态首先根据损坏位置排序,然后根据损坏程度排序。
Perfect maintenance (PM) action. A maintenance action is performed and the asset is restored from its current condition to the damage-free state. 完美维护 (PM) 行动。执行维护操作后,资产将从当前状态恢复到无损坏状态。
Restrict operational conditions (RE) action. The operational conditions of the bridge are restricted by allowing only lightweight trains, carrying less than 18 ton per axle, to travel across the bridge. Such a restriction results in a lower deterioration rate, but also yields a lower revenue generated by the infrastructure. 限制运行条件(RE)行动。限制大桥的运营条件,只允许每轴载重量低于 18 吨的轻型列车通过大桥。这种限制降低了老化率,但也降低了基础设施的收益。
In the cases where the most recently issued control input is either DN or RE, the physical state undergoes a degradation process that monotonically deteriorates the structural health. When operational conditions are not restricted, we prescribe a stochastic degradation process featuring a probability of damage inception equal to 0.5 . Damage may develop in any of the predefined regions with damage level sampled from a uniform distribution , and then propagate with increments sampled from a Gaussian pdf centered at and featuring a standard deviation equal to (negative increments are rounded to zero). When the operations are restricted and only lightweight trains are allowed to travel across the bridge, we instead assume a probability of damage inception equal to 0.25 . In this eventuality, damage may develop with damage level sampled from a uniform distribution , and then propagate with increments sampled from a Gaussian pdf centered at and featuring a standard deviation equal to . The resulting trajectory of the structural health parameters is intended to represent periods of gradual degradation in the structural health, as well as sudden changes due to discrete damage events. Also in this case, the effect of a PM action is simulated by restoring the physical state to its undamaged configuration 在最近发布的控制输入为 DN 或 RE 的情况下,物理状态会经历一个退化过程,使结构健康状况单调恶化。在运行条件不受限制的情况下,我们规定了一个随机退化过程,其特点是损坏开始的概率 等于 0.5。损坏可能发生在任何一个预定区域,其损坏程度从均匀分布 中采样,然后以 增量传播,其增量从以 为中心的高斯分布中采样,标准偏差等于 (负增量四舍五入为零)。当运营受到限制,只允许轻型列车通过桥梁时,我们假设损坏发生的概率等于 0.25。在这种情况下,损坏可能会以从均匀分布 中采样的损坏程度发展,然后以从以 为中心、标准偏差等于 的高斯 pdf 中采样的 增量传播。由此产生的结构健康参数轨迹旨在表示结构健康逐渐退化的时期,以及由于离散损坏事件引起的突变。在这种情况下,还可以通过将物理状态恢复到未损坏时的配置来模拟 PM 作用的效果
The transition model associated with the DN action assumes that damage may start in any subdomain , with , with probability 0.1 , and then grow to the next interval with the same probability. For the transition model associated with the RE action, this probability is assumed to decrease to 0.03 . The CPTs associated with the DN and RE actions are therefore lower-left triangular transition matrices. The highest probability is assigned to remaining in the same state, followed by the transition to the next interval, with zero probability of improvements. The transition model assumed for the PM action instead maps the belief to a belief associated with a damage-free condition, independently of the current condition. The CPT associated with the PM action is therefore an upper-right triangular transition matrix with probabilities equal to 1 in the first row. 与 DN 操作相关的过渡模型 假设损坏可能从任何子域 开始, ,概率为 0.1,然后以相同的概率增长到下一个 区间。因此,与 DN 和 RE 行动相关的 CPT 是左下三角过渡矩阵。保持相同状态的概率最高,其次是过渡到下一个 区间,改善的概率为零。PM 行动的过渡模型是将 的信念映射到与无损伤状态相关的信念 ,与当前状态无关。因此,与 PM 行动相关的 CPT 是一个右上三角过渡矩阵,第一行的概率等于 1。
In this case, the two reward functions in Eq. (9) are chosen as: 在这种情况下,公式 (9) 中的两个奖励函数选为
where the last contribution in penalizes excessively compromised structural states with a significantly negative reward. 其中 中的最后一个贡献以显著的负奖励来惩罚过度受损的结构状态。
3.2.3. Results 3.2.3.结果
During the offline phase, we solve the planning problem in Eq. (8) by assuming a discount factor , and a weighting factor . The resulting control policy recommends that the asset operates in ordinary conditions until when , after which point it should fall back to the more conservative RE regime in order to minimize further degradation. Once reached , the bridge should be finally repaired. 在离线阶段,我们通过假设贴现因子 和权重因子 来解决公式(8)中的规划问题。由此得出的控制策略 建议资产在普通条件下运行,直到 ,之后应回到更保守的 RE 状态,以尽量减少进一步的退化。一旦达到 ,则应最终修复桥梁。
Fig. 13. Railway bridge - Online phase of the digital twin framework with three possible actions: DN (do nothing), PM (perfect maintenance), and RE (restrict operational conditions). Probabilistic and best point estimates of: (top) digital state evolution against the ground truth digital state; (bottom) control inputs informed by the digital twin, against the optimal control input under ground truth. In the top panels the background color corresponds to . In the bottom panel it corresponds to . 图 13.铁路桥梁 - 数字孪生框架的在线阶段,有三种可能的操作:DN(什么都不做)、PM(完美维护)和 RE(限制运行条件)。概率和最佳点估算:(上图)数字状态演变与数字状态基本事实的对比;(下图)数字孪生提供的控制输入与基本事实下的最佳控制输入的对比。在上图中,背景颜色对应于 。下图中的背景颜色与 对应。
Fig. 14. Railway bridge - Digital twin future predictions with three possible actions: DN (do nothing), PM (perfect maintenance), and RE (restrict operational conditions). The starting time is . In the top panel the probability relates to the amount of damage in . In the bottom panel it corresponds to . 图 14.铁路桥--数字孪生未来预测与三种可能的行动:DN(什么都不做)、PM(完美维护)和 RE(限制运行条件)。起始时间为 。顶部面板中的概率 与 中的损坏量有关。下图中的概率与 对应。
Fig. 13 reports a sample simulation of the DT online phase up to time step . The DT correctly tracks the digital state with relatively low uncertainty. Damage initially develops within , and the DT follows its evolution with a limited delay of at most two time steps, with respect to the ground truth, due to the need of updating the relative prior belief from the previous time steps. The RE action is suggested as soon as the DT estimates a , after which point the DT keeps on tracking the structural health parameters evolving with a lower deterioration rate. A PM action is finally suggested due to an excessively compromised structural state. A similar behavior can be observed for the following damage scenario affecting . 图 13 报告了 DT 在线阶段到时间步长 的模拟样本。DT 以相对较低的不确定性正确跟踪了数字状态。损坏最初发生在 内,由于需要更新前一时间步长的相对先验信念,DT 在跟踪其演变过程时,相对于地面实况最多延迟两个时间步长。一旦 DT 估计出 ,就会建议采取 RE 措施,之后 DT 会继续跟踪结构健康参数,并以较低的劣化率演化。最后,由于结构状态过度恶化,建议采取 PM 操作。在以下影响 的损坏情况中也可以观察到类似的行为。
Fig. 14 reports the predicted evolution of the digital state and control inputs, from and over 20 time steps in the future. The DT predicts the expected degradation of the structural health according to the transition model associated with the DN action, before predicting to take a RE action with relatively high probability after a few time steps. The DT prediction is close to what is effectively experienced online (see Fig. 13). However, besides having the estimated digital state two time steps behind the ground 图 14 报告了数字状态和控制输入的预测变化,从 到未来 20 个时间步骤。DT 根据与 DN 操作相关的过渡模型预测结构健康状况的预期恶化,然后在几个时间步骤后以相对较高的概率预测采取 RE 操作。DT 预测与在线实际体验非常接近(见图 13)。不过,除了估计的数字状态比地面状态晚两个时间步之外
truth value, the prediction is also too optimistic in terms of deterioration rate, which suggests the use of a more refined transition model. 在真值方面,预测的恶化率也过于乐观,这就需要使用一个更加完善的过渡模型。
4. Conclusions 4.结论
In this work we have proposed a predictive digital twin approach to the health monitoring, maintenance, and management planning of civil structures, to advance condition-based and predictive maintenance practices. The presented strategy relies upon a probabilistic graphical model inspired by [18]. This framework is used to encode the asset-twin coupled dynamical system, the relevant end-to-end information flow via observational data (physical to digital) and control inputs (digital to physical), and its evolution over time, all with quantified uncertainty. The assimilation of observational data is carried out with deep learning models, leveraging the capabilities of convolutional layers to automatically select and extract damage-sensitive features from raw vibration recordings. The structural health parameters comprising the digital state are used to capture the variability of the physical asset. They are continually updated in a sequential Bayesian inference fashion, according to control-dependent transition dynamics models describing how the structural health is expected to evolve. The updated digital state is eventually exploited to predict the future evolution of the physical system and the associated uncertainty. This enables predictive decision-making about maintenance and management actions. 在这项工作中,我们提出了一种预测性数字孪生方法,用于土建结构的健康监测、维护和管理规划,以推进基于状态的预测性维护实践。所提出的策略依赖于受文献 [18] 启发而开发的概率图形模型。该框架用于编码资产-双耦合动态系统、通过观测数据(物理到数字)和控制输入(数字到物理)进行的相关端到端信息流及其随时间的演变,所有这些都具有量化的不确定性。观测数据的同化采用深度学习模型,利用卷积层的功能,从原始振动记录中自动选择和提取损伤敏感特征。构成数字状态的结构健康参数用于捕捉物理资产的可变性。这些参数根据描述结构健康状况预期演变方式的控制相关过渡动力学模型,以贝叶斯推理的方式进行连续更新。更新后的数字状态最终被用来预测物理系统的未来演变和相关的不确定性。这样就能对维护和管理行动做出预测性决策。
The computational procedure takes advantage of a preliminary offline phase which involves: (i) using physics-based numerical models and reduced order modeling, to overcome the lack of experimental data for civil applications under varying damage and operational conditions while populating the datasets for training the deep learning models; (ii) learning the health-dependent control policy to be applied at each time step of the online phase, to map the belief over the digital state onto actions feeding back to the physical asset. 计算程序利用了初步离线阶段,包括:(i) 使用基于物理的数值模型和降序建模,以克服在不同损坏和运行条件下民用应用实验数据的缺乏,同时填充用于训练深度学习模型的数据集;(ii) 学习在线阶段每个时间步应用的健康控制策略,将对数字状态的信念映射到反馈到物理资产的行动上。
The proposed strategy has been assessed against the simulated monitoring of an L-shaped cantilever beam and a railway bridge. In the absence of experimental data, the tests have been carried out considering high-fidelity simulation data, corrupted with an additive Gaussian noise. The obtained results have proved the digital twin capabilities of accurately tracking the digital state evolution under varying operational conditions, with relatively low uncertainty. The framework is also able to promptly suggest the appropriate control input, within at most two time steps of when the (unknown) ground truth structural health demands it. 针对 L 型悬臂梁和铁路桥梁的模拟监测,对所提出的策略进行了评估。在缺乏实验数据的情况下,测试采用了高保真模拟数据,并加入了高斯噪声。获得的结果证明,数字孪生具有在不同运行条件下准确跟踪数字状态演变的能力,而且不确定性相对较低。当(未知)地面真实结构健康状况需要时,该框架还能在最多两个时间步骤内及时建议适当的控制输入。
Although the capabilities of health-aware digital twins are showcased in the specific context of monitoring the structural integrity of civil structures to advance predictive maintenance practices, the applicability of the presented framework is general. Indeed, the proposed framework can be adapted for various types of structures and engineering systems by adjusting the components within the dynamic Bayesian network to align with the specific characteristics of the problem at hand. The solution to the inverse problem (if any) can be estimated by assimilating available observational data using methods other than deep neural networks, for instance through Markov chain Monte Carlo sampling algorithms. Similarly, the state transition models are closely tied to the employed parametrization of the digital state and the availability of historical data. The same applies to the available control inputs, which are likely to vary for different structures, such as those in mechanical or aerospace systems, and the method chosen for solving the associated planning problem. Additionally, the graph topology can be easily reorganized to adapt to situations where observational data are not acquired after issuing a control input, or when control inputs are issued with a different frequency than that governing the digital twin update. 虽然健康感知数字孪生的功能是在监测土木工程结构完整性以推进预测性维护实践的特定背景下展示的,但所提出的框架具有普遍适用性。事实上,通过调整动态贝叶斯网络中的组件,使其符合当前问题的具体特征,所提出的框架可适用于各种类型的结构和工程系统。逆问题的解决方案(如果有的话)可以通过使用深度神经网络以外的方法(例如马尔科夫链蒙特卡罗采样算法)吸收可用的观测数据来估算。同样,状态转换模型也与所采用的数字状态参数化和历史数据的可用性密切相关。这同样适用于可用的控制输入,不同的结构(如机械或航空航天系统中的结构)可能会有不同的控制输入,以及解决相关规划问题所选择的方法。此外,图形拓扑结构可以轻松重组,以适应在发出控制输入后未获取观测数据的情况,或发出控制输入的频率与数字孪生更新频率不同的情况。
Future research lines will investigate the ability of the digital twin to update the transition dynamics models by learning from previous data. As suggested by the railway bridge case study, this will allow for a more accurate prediction of the expected evolution of the digital state, thus enabling predictive decision-making better tailored to the monitored asset. Another aspect of interest concerns solving the planning problem induced by the probabilistic model using reinforcement learning algorithms, capable of taking into account a finite planning horizon representing the design lifetime of the asset. 未来的研究方向将研究数字孪生系统通过学习以前的数据来更新过渡动态模型的能力。正如铁路桥梁案例研究中提出的那样,这将允许对数字状态的预期演变进行更准确的预测,从而使预测性决策更适合受监控的资产。另一个值得关注的方面是利用强化学习算法解决概率模型引起的规划问题,该算法能够考虑到代表资产设计寿命的有限规划范围。
Code availability 代码可用性
The implementation code used for the experiments presented in Section 3 is available in the public repository digitaltwin-SHM [28]. The code implements the proposed digital twin framework and can be used to generate the graphs of digital state estimation and prediction reported in this paper. The DL models trained according to the implementation details reported in the Appendix are also made available in the same repository. 第 3 节中介绍的实验所使用的实现代码可从公共存储库 digitaltwin-SHM [28] 中获取。该代码实现了所提出的数字孪生框架,可用于生成本文所报告的数字状态估计和预测图。根据附录中报告的实现细节训练的数字孪生模型也可在同一资源库中获取。
Declaration of competing interest 利益冲突声明
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper. 作者声明,他们没有任何可能会影响本文所报告工作的已知经济利益或个人关系。
Data availability 数据可用性
The observational data used to run the experiments presented in Section 3 are available in the public repository digitaltwin-SHM [28]. The Matlab library for finite element simulation and reduced-order modeling of partial differential equations employed to generate these data is available in the public repository Redbkit [53]. 用于运行第 3 节中介绍的实验的观测数据可在公共资源库 digitaltwin-SHM [28] 中获取。用于生成这些数据的有限元模拟和偏微分方程降阶建模的 Matlab 库可从公共资源库 Redbkit [53] 中获取。
Acknowledgments 致谢
Matteo Torzoni acknowledges the financial support from the Politecnico di Milano, Italy through the interdisciplinary Ph.D. Grant "Physics-informed deep learning for structural health monitoring". Marco Tezzele and Karen E. Willcox acknowledge support from the "NASA University Leadership Initiative" under Cooperative Agreement 80NSSC21M0071. Andrea Manzoni acknowledges the project "Dipartimento di Eccellenza 2023-2027", funded by MUR, Italy, and the project "FAIR (Future Artificial Intelligence Research)", funded by the NextGenerationEU program within the PNRR-PE-AI scheme (M4C2, Investment 1.3, Line on Artificial Intelligence). Matteo Torzoni 感谢意大利米兰理工大学通过跨学科博士论文 "结构健康监测的物理信息深度学习 "提供的资金支持。Marco Tezzele 和 Karen E. Willcox 感谢 "美国国家航空航天局大学领导力计划 "根据合作协议 80NSSC21M0071 提供的支持。Andrea Manzoni 感谢由意大利 MUR 资助的 "Dipartimento di Eccellenza 2023-2027 "项目,以及由 PNRR-PE-AI 计划内的 NextGenerationEU 项目(M4C2,投资 1.3,人工智能项目)资助的 "FAIR(未来人工智能研究)"项目。
Appendix. Implementation details 附录。实施细节
In this Appendix, we discuss the implementation details of the DL models described in Section 2.3. The architectures, as well as the relevant hyperparameters and training options, have been chosen through a preliminary study, aimed at minimizing and , while retaining the generalization capabilities of and of , with . Since all models share the same architecture, the index will be dropped in the following for ease of notation. 在本附录中,我们将讨论第 2.3 节中描述的 DL 模型的实现细节。这些架构以及相关的超参数和训练选项都是通过初步研究选择的,目的是尽量减少 和 ,同时保留 和 的泛化能力,以及 的泛化能力。由于所有 模型都具有相同的架构,为便于记述,下文将去掉 这一索引。
In the present work, and are set as 12-layers DL models, whose architecture is outlined in Table A.1a and in Table A.2a, respectively. and feature a damage-sensitive feature extractor required to be insensitive to transformations in the input not related to damage. This is implemented through the composition of three one-dimensional (1D) convolutional units. Convolutional layers naturally embed good relational inductive biases such as locality and translation equivariance [52,60], which prove highly effective to detect time correlations within and across time series. The resulting sparse connectivity and parameter sharing also make them computationally efficient. Each convolutional unit consists of a convolutional layer, followed by a Tanh activation function, max pooling, and dropout. The extracted features are expected to be sensitive to the presence of damage, but insensitive to measurement noise and operational variability. The extracted features are then reshaped through a flatten layer and run through a stack of three fully-connected layers: the first two are Tanh-activated, while the output layer of is Softmax-activated, and the output layer of has no activation function. 在本研究中, 和 被设定为 12 层 DL 模型,其结构分别如表 A.1a 和表 A.2a 所示。 和 具有损伤敏感特征提取器,要求它对输入中与损伤无关的变换不敏感。这是由三个一维(1D)卷积单元组成的。卷积层天然地嵌入了良好的关系归纳偏差,如局部性和平移等差数 [52,60],这对检测时间序列内和时间序列间的时间相关性非常有效。由此产生的稀疏连通性和参数共享也使其具有很高的计算效率。每个卷积单元由一个卷积层、一个 Tanh 激活函数、最大池化和剔除组成。提取的特征预计会对是否存在损坏敏感,但对测量噪声和操作变异不敏感。然后,提取的特征通过扁平层进行重塑,并通过三个全连接层的堆叠运行:前两个层为 Tanh 激活层,而 的输出层为 Softmax 激活层, 的输出层没有激活函数。
Table A. 1 表 A. 1
- (a) employed architecture, and (b) selected hyperparameters and training options. - (a) 采用的架构,以及 (b) 选定的超参数和训练选项。
(a)
(b)
Layer 层数
Output shape 输出形状
Activation 激活
Input 输入
0 - Input 0 - 输入
None 无
-
Convolution kernel size: 卷积核大小
1 - Conv1D
Tanh
0
Dropout rate: 辍学率:
2 - MaxPooling1D
None 无
1
Weight initializer: 重量初始化器:
Xavier 泽维尔
3 - Dropout 3 - 中途退学
None 无
2
regularization rate: 正规化率:
4 - Conv1D
Tanh
3
Optimizer: 优化器:
Adam 亚当
5 - MaxPooling1D
None 无
4
Batch size: 批量大小:
6 - Dropout 6 - 中途退学
None 无
5
Initial learning rate: 初始学习率:
7 - Conv1D
Tanh
6
Allowed epochs: 允许的年代:
250
8 - MaxPooling1D
None 无
7
Learning schedule: 学习时间表:
cosine decay 余弦衰减
9 - Dropout 9 - 中途退学
None 无
8
Weight decay: 重量衰减:
0.05
10 - Flatten 10 - 压平
None 无
9
Early stop patience: 早期停止的耐心:
15 epochs 15 个纪元
11 - Dense 11 - 密集
Tanh
10
Train-val split: 火车与汽车分离:
12 - Dense 12 - 密集
Tanh
11
13 - Dense 13 - 密集
Softmax 软磁
12
Table A. 2 表 A. 2
- (a) employed architecture, and (b) selected hyperparameters and training options. - (a) 采用的架构,以及 (b) 选定的超参数和训练选项。
(a)
(b)
Layer 层数
Output shape 输出形状
Activation 激活
Input 输入
0 - Input 0 - 输入
None 无
-
Convolution kernel size: 卷积核大小
1 - Conv1D
Tanh
0
Dropout rate: 辍学率:
2 - MaxPooling1D
None 无
1
Weight initializer: 重量初始化器:
Xavier 泽维尔
3 - Dropout 3 - 中途退学
None 无
2
regularization rate: 正规化率:
4 - Conv1D
Tanh
3
Optimizer: 优化器:
Adam 亚当
5 - MaxPooling1D
None 无
4
Batch size: 批量大小:
6 - Dropout 6 - 中途退学
None 无
5
Initial learning rate: 初始学习率:
7 - Conv1D
Tanh
6
Allowed epochs: 允许的年代:
250
8 - MaxPooling1D
None 无
7
Learning schedule: 学习时间表:
cosine decay 余弦衰减
9 - Dropout 9 - 中途退学
None 无
8
Weight decay: 重量衰减:
0.05
10 - Flatten 10 - 压平
None 无
9
Early stop patience: 提前停止的耐心:
15 epochs 15 个纪元
11 - Dense 11 - 密集
Tanh
10
Train-val split: 火车与汽车分离:
12 - Dense 12 - 密集
Tanh
11
13 - Dense 13 - 密集
None 无
12
Using the Xavier's weight initialization [61], and are trained by minimizing the following loss functions, respectively: 利用 Xavier 权重初始化方法 [61], 和 分别通过最小化以下损失函数进行训练:
where and denote the regularization rate over the relative model parameters and . The loss functions and are minimized using the first-order stochastic gradient descent optimizer Adam [62], for a maximum of 250 allowed epochs. The corresponding learning rates and are initially set to , and decreased for of the allowed training steps using a cosine decay schedule with weight decay equal to 0.05 . The optimization is carried out considering an splitting ratio of the dataset for training and validation purposes, with of the data randomly taken and set aside to monitor the learning process. We use an early stopping strategy to interrupt learning, whenever the loss function value attained on the validation set does not decrease for a prescribed number of patience epochs in a row. The hyperparameters and training options for and for are reported in Table A. and in Table A. , respectively. 其中 和 表示 相对于模型参数 和 的正则化率。损失函数 和 使用一阶随机梯度下降优化器 Adam [62]最小化,最多允许 250 个历时。相应的学习率 和 初始设置为 ,并在 的允许训练步骤中使用权重衰减等于 0.05 的余弦衰减计划来降低学习率。在进行优化时,我们考虑到了用于训练和验证的数据集的分割比例为 ,并随机抽取 的数据用于监控学习过程。只要验证集上的损失函数值在规定数量的耐心历时内连续不下降,我们就会使用提前停止策略来中断学习。 和 的超参数和训练选项分别见表 A. 和表 A. 。
References 参考资料
[1] A. Thelen, X. Zhang, O. Fink, Y. Lu, S. Ghosh, B.D. Youn, M.D. Todd, S. Mahadevan, C. Hu, Z. Hu, A comprehensive review of digital twin - part 1: modeling and twinning enabling technologies, Struct. Multidiscip. Optim. 65 (2022) 354, http://dx.doi.org/10.1007/s00158-022-03425-4. [1] A. Thelen, X. Zhang, O. Fink, Y. Lu, S. Ghosh, B.D. Youn, M.D. Todd, S. Mahadevan, C. Hu, Z. Hu, A comprehensive review of digital twin - part 1: modeling and twinning enabling technologies, Struct.Multidiscip.Optim.65 (2022) 354, http://dx.doi.org/10.1007/s00158-022-03425-4.
[2] E.J. Tuegel, A.R. Ingraffea, T.G. Eason, S.M. Spottswood, Reengineering aircraft structural life prediction using a digital twin, Int. J. Aerosp. Eng. 2011 (2011) 154798, http://dx.doi.org/10.1155/2011/154798. [2] E.J. Tuegel, A.R. Ingraffea, T.G. Eason, S.M. Spottswood, Reengineering aircraft structural life prediction using a digital twin, Int. J. Aerosp.J. Aerosp.2011 (2011) 154798, http://dx.doi.org/10.1155/2011/154798.
[3] E. Tuegel, The airframe digital twin: Some challenges to realization, in: 53rd AIAA/ASME/ASCE/AHS/ASC Structures, Structural Dynamics and Materials Conference, 2012, http://dx.doi.org/10.2514/6.2012-1812. [3] E. Tuegel, The airframe digital twin: Some challenges to realization, in:53rd AIAA/ASME/ASCE/AHS/ASC Structures, Structural Dynamics and Materials Conference, 2012, http://dx.doi.org/10.2514/6.2012-1812.
[4] M. Grieves, J. Vickers, Digital twin: Mitigating unpredictable, undesirable emergent behavior in complex systems, in: Transdisciplinary Perspectives on Complex Systems: New Findings and Approaches, Springer, 2017, pp. 85-113, http://dx.doi.org/10.1007/978-3-319-38756-7_4. [4] M. Grieves, J. Vickers, Digital twin: Mitigating unpredictable, undesirable emergent behavior in complex systems, in:Transdisciplinary Perspectives on Complex Systems:新发现与新方法》,施普林格出版社,2017 年,第 85-113 页,http://dx.doi.org/10.1007/978-3-319-38756-7_4。
[5] A. Rasheed, O. San, T. Kvamsdal, Digital twin: Values, challenges and enablers from a modeling perspective, IEEE Access 8 (2020) 21980-22012, http://dx.doi.org/10.1109/ACCESS.2020.2970143.
[6] S.A. Niederer, M.S. Sacks, M. Girolami, K. Willcox, Scaling digital twins from the artisanal to the industrial, Nat. Comput. Sci. 1 (5) (2021) 313-320, http://dx.doi.org/10.1038/s43588-021-00072-5. [6] S.A. Niederer, M.S. Sacks, M. Girolami, K. Willcox, Scaling digital twins from the artisanal to the industrial, Nat.Comput.1 (5) (2021) 313-320, http://dx.doi.org/10.1038/s43588-021-00072-5.
[7] Y. Ye, Q. Yang, F. Yang, Y. Huo, S. Meng, Digital twin for the structural health management of reusable spacecraft: a case study, Eng. Fract. Mech. 234 (2020) 107076, http://dx.doi.org/10.1016/j.engfracmech.2020.107076. [7] Y. Ye, Q. Yang, F. Yang, Y. Huo, S. Meng, Digital twin for the structural health management of reusable spacecraft: a case study, Eng.Fract.Mech.234 (2020) 107076, http://dx.doi.org/10.1016/j.engfracmech.2020.107076.
[8] M. Podskarbi, D.J. Knezevic, Digital twin for operations - present applications and future digital thread, in: OTC Offshore Technology Conference, Houston, Texas, 2020, http://dx.doi.org/10.4043/30553-MS. [8] M. Podskarbi, D.J. Knezevic, Digital twin for operations - Present applications and future digital thread, in:OTC Offshore Technology Conference, Houston, Texas, 2020, http://dx.doi.org/10.4043/30553-MS.
[9] A. Phua, C. Davies, G. Delaney, A digital twin hierarchy for metal additive manufacturing, Comput. Ind. 140 (2022) 103667, http://dx.doi.org/10.1016/ j.compind.2022.103667. [9] A. Phua, C. Davies, G. Delaney, A digital twin hierarchy for metal additive manufacturing, Comput.Ind.140 (2022) 103667, http://dx.doi.org/10.1016/ j.compind.2022.103667.
[10] M. Jans-Singh, K. Leeming, R. Choudhary, M. Girolami, Digital twin of an urban-integrated hydroponic farm, Data-Cent. Eng. 1 (2020) 20, http: dce.2020.21. [10] M. Jans-Singh、K. Leeming、R. Choudhary、M. Girolami,城市综合水培农场的数字孪生,Data-Cent.Eng.1 (2020) 20, http: dce.2020.21.
[11] A. Tzachor, S. Sabri, C.E. Richards, A. Rajabifard, M. Acuto, Potential and limitations of digital twins to achieve the sustainable development goals, Nat. Sustain. 5 (10) (2022) 822-829, http://dx.doi.org/10.1038/s41893-022-00923-7. [11] A. Tzachor, S. Sabri, C.E. Richards, A. Rajabifard, M. Acuto, Potential and limitations of digital twins to achieve the sustainable development goals, Nat.Sustain.5 (10) (2022) 822-829, http://dx.doi.org/10.1038/s41893-022-00923-7.
[12] G. Arcieri, C. Hoelzl, O. Schwery, D. Straub, K.G. Papakonstantinou, E. Chatzi, Bridging POMDPs and Bayesian decision making for robust maintenance planning under model uncertainty: An application to railway systems, Reliab. Eng. Syst. Saf. (2023) 109496, http://dx.doi.org/10.1016/j.ress.2023.109496. [12] G. Arcieri, C. Hoelzl, O. Schwery, D. Straub, K.G. Papakonstantinou, E. Chatzi, Bridging POMDPs and Bayesian decision making for robust maintenance planning under model uncertainty:An application to railway systems, Reliab.Eng.Syst.Saf.(2023) 109496, http://dx.doi.org/10.1016/j.ress.2023.109496.
[13] M.G. Kapteyn, D.J. Knezevic, D. Huynh, M. Tran, K.E. Willcox, Data-driven physics-based digital twins via a library of component-based reduced-order models, Internat. J. Numer. Methods Engrg. 123 (13) (2020) 2986-3003, http://dx.doi.org/10.1002/nme.6423. [13] M.G. Kapteyn, D.J. Knezevic, D. Huynh, M. Tran, K.E. Willcox, Data-driven physics-based digital twins via a library of component-based reduced-order models, Internat.J. Numer.Methods Engrg.123 (13) (2020) 2986-3003, http://dx.doi.org/10.1002/nme.6423.
[14] B. Moya, A. Badías, I. Alfaro, F. Chinesta, E. Cueto, Digital twins that learn and correct themselves, Internat. J. Numer. Methods Engrg. 123 (13) (2022) 3034-3044, http://dx.doi.org/10.1002/nme.6535. [14] B. Moya, A. Badías, I. Alfaro, F. Chinesta, E. Cueto, Digital twins that learn and correct themselves, Internat.J. Numer.Methods Engrg.123 (13) (2022) 3034-3044, http://dx.doi.org/10.1002/nme.6535.
[15] W. Sisson, P. Karve, S. Mahadevan, Digital twin approach for component health-informed rotorcraft flight parameter optimization, AIAA J. 60 (3) (2022) 1923-1936, http://dx.doi.org/10.2514/1.J060770.
[16] B. Moya, I. Alfaro, D. Gonzalez, F. Chinesta, E. Cueto, Physically sound, self-learning digital twins for sloshing fluids, PLoS One 15 (6) (2020) e0234569, http://dx.doi.org/10.1371/journal.pone.0234569.
[17] K. Cherifi, V. Mehrmann, P. Schulze, Simulations in a digital twin of an electrical machine, 2022, http://dx.doi.org/10.48550/arXiv.2207.02171, arXiv preprint arXiv:2207.02171.
[18] M.G. Kapteyn, J.V. Pretorius, K.E. Willcox, A probabilistic graphical model foundation for enabling predictive digital twins at scale, Nat. Comput. Sci. 1 (5) (2021) 337-347, http://dx.doi.org/10.1038/s43588-021-00069-0. [18] M.G. Kapteyn, J.V. Pretorius, K.E. Willcox, A probabilistic graphical model foundation for enabling predictive digital twins at scale, Nat.Comput.1 (5) (2021) 337-347, http://dx.doi.org/10.1038/s43588-021-00069-0.
[19] D. Koller, N. Friedman, Probabilistic Graphical Models: Principles and Techniques, MIT Press, Cambridge, MA, 2009. [19] D. Koller, N. Friedman, Probabilistic Graphical Models:Principles and Techniques, MIT Press, Cambridge, MA, 2009.
[20] S.J. Russell, Artificial Intelligence: A Modern Approach, Pearson Education, London, UK, 2020. [20] S.J. Russell,Artificial Intelligence:A Modern Approach, Pearson Education, London, UK, 2020.
[21] M. Torzoni, A. Manzoni, S. Mariani, Structural health monitoring of civil structures: a diagnostic framework powered by deep metric learning, Comput. Struct. 271 (2022) 106858, http://dx.doi.org/10.1016/j.compstruc.2022.106858. [21] M. Torzoni, A. Manzoni, S. Mariani, Structural health monitoring of civil structures: a diagnostic framework powered by deep metric learning, Comput.Struct.271 (2022) 106858, http://dx.doi.org/10.1016/j.compstruc.2022.106858.
[22] V. Giglioni, I. Venanzi, V. Poggioni, A. Milani, F. Ubertini, Autoencoders for unsupervised real-time bridge health assessment, Comput.-Aided Civ. Infrastruct. Eng. 38 (2023) 959-974, http://dx.doi.org/10.1111/mice.12943. [22] V. Giglioni、I. Venanzi、V. Poggioni、A. Milani、F. Ubertini,Autoencoders for unsupervised real-time bridge health assessment,Comput.-Aided Civ. Infrastruct.Infrastruct.Eng.38 (2023) 959-974, http://dx.doi.org/10.1111/mice.12943.
[23] M. Torzoni, A. Manzoni, S. Mariani, A multi-fidelity surrogate model for structural health monitoring exploiting model order reduction and artificial neural networks, Mech. Syst. Signal Process. 197 (2023) 110376, http://dx.doi.org/10.1016/j.ymssp.2023.110376. [23] M. Torzoni, A. Manzoni, S. Mariani, A multi-fidelity surrogate model for structural health monitoring exploiting model order reduction and artificial neural networks, Mech.Syst.Signal Process.197 (2023) 110376, http://dx.doi.org/10.1016/j.ymssp.2023.110376.
[24] M. Torzoni, A. Manzoni, S. Mariani, A deep neural network, multi-fidelity surrogate model approach for Bayesian model updating in SHM, in: European Workshop on Structural Health Monitoring, Springer International Publishing, 2023, pp. 1076-1086, http://dx.doi.org/10.1007/978-3-031-07258-1_108. [24] M. Torzoni, A. Manzoni, S. Mariani, A deep neural network, multi-fidelity surrogate model approach for Bayesian model updating in SHM, in:European Workshop on Structural Health Monitoring, Springer International Publishing, 2023, pp.
[25] L. Rosafalco, M. Torzoni, A. Manzoni, S. Mariani, A. Corigliano, A self-adaptive hybrid model/data-driven approach to SHM based on model order reduction and deep learning, in: Structural Health Monitoring Based on Data Science Techniques, Springer International Publishing, 2022, pp. 165-184, http://dx.doi.org/10.1007/978-3-030-81716-9. [25] L. Rosafalco, M. Torzoni, A. Manzoni, S. Mariani, A. Corigliano, A self-adaptive hybrid model/data-driven approach to SHM based on model order reduction and deep learning, in:基于数据科学技术的结构健康监测》,施普林格国际出版公司,2022 年,第 165-184 页,http://dx.doi.org/10.1007/978-3-030-81716-9。
[26] S. Eftekhar Azam, S. Mariani, Online damage detection in structural systems via dynamic inverse analysis: A recursive Bayesian approach, Eng. Struct. 159 (2018) 28-45, http://dx.doi.org/10.1016/j.engstruct.2017.12.031. [26] S. Eftekhar Azam、S. Mariani,《通过动态逆分析在线检测结构系统中的损坏》:A recursive Bayesian approach, Eng.Struct.159 (2018) 28-45, http://dx.doi.org/10.1016/j.engstruct.2017.12.031.
[27] A. Quarteroni, A. Manzoni, F. Negri, Reduced Basis Methods for Partial Differential Equations: An Introduction, Vol. 92, Springer, 2015, http://dx.doi, org . [27] A. Quarteroni, A. Manzoni, F. Negri, Reduced Basis Methods for Partial Differential Equations:An Introduction, Vol. 92, Springer, 2015, http://dx.doi, org .
[29] G. Morcous, Performance prediction of bridge deck systems using Markov chains, J. Perform. Constr. Facil. 20 (2) (2006) 146-155, http://dx.doi.org/10. 1061/(ASCE)0887-3828(2006)20:2(146). [29] G. Morcous, Performance prediction of bridge deck systems using Markov chains, J. Perform.Constr.Facil.20 (2) (2006) 146-155, http://dx.doi.org/10.1061/(ASCE)0887-3828(2006)20:2(146).
[31] R.S. Sutton, A.G. Barto, Reinforcement Learning: An Introduction, MIT Press, Cambridge, MA, 2018. [31] R.S.萨顿、A.G.巴托:《强化学习》(Reinforcement Learning:导论》,麻省理工学院出版社,马萨诸塞州剑桥市,2018 年。
[32] T. Taddei, J.D. Penn, M. Yano, A.T. Patera, Simulation-based classification; a model-order-reduction approach for structural health monitoring, Arch. Comput. Methods Eng. 25 (2018) 23-45, http://dx.doi.org/10.1007/s11831-016-9185-0. [32] T. Taddei, J.D. Penn, M. Yano, A.T. Patera, Simulation-based classification; a model-order-reduction approach for structural health monitoring, Arch.Comput.Methods Eng. 25 (2018) 23-45, http://dx.doi.org/10.1007/s11831-016-9185-0.
[33] N.M. Newmark, A method of computation for structural dynamics, J. Eng. Mech. 85 (3) (1959) 67-94, http://dx.doi.org/10.1061/JMCEA3.0000098. [33] N.M. Newmark, A method of computation for structural dynamics, J. Eng. Mech.Mech.85 (3) (1959) 67-94, http://dx.doi.org/10.1061/JMCEA3.0000098.
[34] A. Teughels, J. Maeck, G. De Roeck, Damage assessment by FE model updating using damage functions, Comput. Struct. 80 (25) (2002) 1869-1879, http://dx.doi.org/10.1016/S0045-7949(02)00217-1. [34] A. Teughels, J. Maeck, G. De Roeck, Damage assessment by FE model updating using damage functions, Comput.80 (25) (2002) 1869-1879, .80 (25) (2002) 1869-1879, http://dx.doi.org/10.1016/S0045-7949(02)00217-1.
[35] C. Farrar, K. Worden, Structural Health Monitoring: A Machine Learning Perspective, John Wiley & Sons, 2013, http://dx.doi.org/10.1002/9781118443118. [35] C. Farrar, K. Worden, Structural Health Monitoring:A Machine Learning Perspective, John Wiley & Sons, 2013, http://dx.doi.org/10.1002/9781118443118.
[36] A. Pandey, M. Biswas, Damage detection in structures using changes in flexibility, J. Sound Vib. 169 (1) (1994) 3-17, http://dx.doi.org/10.1006/jsvi. 1994.1002 . [36] A. Pandey, M. Biswas, Damage detection in structures using changes in flexibility, J. Sound Vib.169 (1) (1994) 3-17, http://dx.doi.org/10.1006/jsvi.1994.1002 .
[37] L. Rosafalco, A. Manzoni, S. Mariani, A. Corigliano, Fully convolutional networks for structural health monitoring through multivariate time series classification, Adv. Model. Simul. Eng. Sci. 7 (2020) 38, http://dx.doi.org/10.1186/s40323-020-00174-1. [37] L. Rosafalco, A. Manzoni, S. Mariani, A. Corigliano, Fully convolutional networks for structural health monitoring through multivariate time series classification, Adv. Model.Simul.Simul.7 (2020) 38, http://dx.doi.org/10.1186/s40323-020-00174-1.
[38] L. Rosafalco, M. Torzoni, A. Manzoni, S. Mariani, A. Corigliano, Online structural health monitoring by model order reduction and deep learning algorithms, Comput. Struct. 255 (2021) 106604, http://dx.doi.org/10.1016/j.compstruc.2021.106604. [38] L. Rosafalco、M. Torzoni、A. Manzoni、S. Mariani、A. Corigliano,《通过模型阶次减少和深度学习算法进行在线结构健康监测》,Comput.Struct.255 (2021) 106604, http://dx.doi.org/10.1016/j.compstruc.2021.106604.
[39] M. Torzoni, L. Rosafalco, A. Manzoni, S. Mariani, A. Corigliano, SHM under varying environmental conditions: an approach based on model order reduction and deep learning, Comput. Struct. 266 (2022) 106790, http://dx.doi.org/10.1016/j.compstruc.2022.106790. [39] M. Torzoni, L. Rosafalco, A. Manzoni, S. Mariani, A. Corigliano, SHM under varying environmental conditions: an approach based on model order reduction and deep learning, Comput.Struct.266 (2022) 106790, http://dx.doi.org/10.1016/j.compstruc.2022.106790.
[40] F. Chinesta, A. Huerta, G. Rozza, K. Willcox, Model reduction methods, in: E. Stein, R. de Borst, T.J.R. Hughes (Eds.), Encyclopedia of Computational Mechanics, second ed., John Wiley & Sons, 2017, pp. 1-36. [40] F. Chinesta, A. Huerta, G. Rozza, K. Willcox, Model reduction methods, in:E. Stein, R. de Borst, T.J.R. Hughes (Eds.), Encyclopedia of Computational Mechanics, second ed., John Wiley & Sons, 2017, pp.
[41] P. Benner, S. Grivet-Talocia, A. Quarteroni, G. Rozza, W.H.A. Schilders, L.M. Silveira (Eds.), Volume 1: System- and Data-Driven Methods and Algorithms, De Gruyter, Berlin, Boston, 2021, http://dx.doi.org/10.1515/9783110498967.
[42] G. Rozza, G. Stabile, F. Ballarin, in: G. Rozza, G. Stabile, F. Ballarin (Eds.), Advanced Reduced Order Methods and Applications in Computational Fluid Dynamics, Society for Industrial and Applied Mathematics, Philadelphia, PA, 2022, http://dx.doi.org/10.1137/1.9781611977257. [42] G. Rozza, G. Stabile, F. Ballarin, in:G. Rozza, G. Stabile, F. Ballarin (Eds.), Advanced Reduced Order Methods and Applications in Computational Fluid Dynamics, Society for Industrial and Applied Mathematics, Philadelphia, PA, 2022, http://dx.doi.org/10.1137/1.9781611977257.
[43] G. Kerschen, J.C. Golinval, Physical interpretation of the proper orthogonal modes using the singular value decomposition, J. Sound Vib. 249 (5) (2002) 849-865, http://dx.doi.org/10.1006/jsvi.2001.3930. [43] G. Kerschen, J.C. Golinval, Physical interpretation of the proper orthogonal modes using the singular value decomposition, J. Sound Vib.249 (5) (2002) 849-865, http://dx.doi.org/10.1006/jsvi.2001.3930.
[44] G. Kerschen, J.C. Golinval, A. Vakakis, L.A. Bergman, The method of proper orthogonal decomposition for dynamical characterization and order reduction of mechanical systems: An overview, Nonlinear Dynam. 41 (1) (2005) 147-169, http://dx.doi.org/10.1007/s11071-005-2803-2. [44] G. Kerschen, J.C. Golinval, A. Vakakis, L.A. Bergman, The method of proper orthogonal decomposition for dynamical characterization and order reduction of mechanical systems. An overview, Nonlinear Dynam:An overview, Nonlinear Dynam.41 (1) (2005) 147-169, http://dx.doi.org/10.1007/s11071-005-2803-2.
[45] M. Guo, J.S. Hesthaven, Reduced order modeling for nonlinear structural analysis using Gaussian process regression, Comput. Methods Appl. Mech. Engrg. 341 (2018) 807-826, http://dx.doi.org/10.1016/j.cma.2018.07.017. [45] M. Guo, J.S. Hesthaven, Reduced order modeling for nonlinear structural analysis using Gaussian process regression, Comput.Methods Appl.Engrg.341 (2018) 807-826, http://dx.doi.org/10.1016/j.cma.2018.07.017.
[46] M. Tezzele, L. Fabris, M. Sidari, M. Sicchiero, G. Rozza, A multi-fidelity approach coupling parameter space reduction and non-intrusive POD with application to structural optimization of passenger ship hulls, Internat. J. Numer. Methods Engrg. 124 (5) (2023) 1193-1210, http://dx.doi.org/10.1002/ nme. 7159 . [46] M. Tezzele, L. Fabris, M. Sidari, M. Sicchiero, G. Rozza, A multi-fidelity approach coupling parameter space reduction and non-intrusive POD with application to structural optimization of passenger ship hulls, Internat.J. Numer.Methods Engrg.124 (5) (2023) 1193-1210, http://dx.doi.org/10.1002/ nme.7159 .
[47] M. Sieber, C.O. Paschereit, K. Oberleithner, Spectral proper orthogonal decomposition, J. Fluid Mech. 792 (2016) 798-828, http://dx.doi.org/10.1017/ jfm.2016.103. [47] M. Sieber, C.O. Paschereit, K. Oberleithner, Spectral proper orthogonal decomposition, J. Fluid Mech.792 (2016) 798-828, http://dx.doi.org/10.1017/ jfm.2016.103.
[48] A. Towne, O.T. Schmidt, T. Colonius, Spectral proper orthogonal decomposition and its relationship to dynamic mode decomposition and resolvent analysis, J. Fluid Mech. 847 (2018) 821-867, http://dx.doi.org/10.1017/jfm.2018.283. [48] A. Towne, O.T. Schmidt, T. Colonius, Spectral proper orthogonal decomposition and its relationship to dynamic mode decomposition and resolvent analysis, J. Fluid Mech.847 (2018) 821-867, http://dx.doi.org/10.1017/jfm.2018.283.
[49] A. Lario, R. Maulik, O.T. Schmidt, G. Rozza, G. Mengaldo, Neural-network learning of SPOD latent dynamics, J. Comput. Phys. 468 (2022) 111475, http://dx.doi.org/10.1016/j.jcp.2022.111475. [49] A. Lario、R. Maulik、O.T. Schmidt、G. Rozza、G. Mengaldo,SPOD 潜在动力学的神经网络学习,J. Comput.468 (2022) 111475, http://dx.doi.org/10.1016/j.jcp.2022.111475.
[50] K.R. dos Santos, D.G. Giovanis, K. Kontolati, D. Loukrezis, M.D. Shields, Grassmannian diffusion maps based surrogate modeling via geometric harmonics, Internat. J. Numer. Methods Engrg. 123 (15) (2022) 3507-3529, http://dx.doi.org/10.1002/nme.6977. [50] K.R. dos Santos, D.G. Giovanis, K. Kontolati, D. Loukrezis, M.D. Shields, Grassmannian diffusion maps based surrogate modeling via geometric harmonics, Internat.J. Numer.Methods Engrg.123 (15) (2022) 3507-3529, http://dx.doi.org/10.1002/nme.6977.
[51] C.M. Bishop, Pattern Recognition and Machine Learning, in: Information Science and Statistics, Springer, New York, NY, 2006. [51] C.M. Bishop, Pattern Recognition and Machine Learning, in:信息科学与统计学》,施普林格出版社,纽约州纽约市,2006 年。
[52] I. Goodfellow, Y. Bengio, A. Courville, Deep Learning, MIT Press, Cambridge, MA, 2016.
[54] A. Ankan, A. Panda, pgmpy: Probabilistic graphical models using python, in: Proceedings of the 14th Python in Science Conference (SCIPY 2015), Citeseer, 2015. [54] A. Ankan, A. Panda, pgmpy:Probabilistic graphical models using python, in:Proceedings of the 14th Python in Science Conference (SCIPY 2015), Citeseer, 2015.
[56] K. Papakonstantinou, M. Shinozuka, Planning structural inspection and maintenance policies via dynamic programming and Markov processes. Part II: POMDP implementation, Reliab. Eng. Syst. Saf. 130 (2014) 214-224, http://dx.doi.org/10.1016/j.ress.2014.04.006. [56] K. Papakonstantinou, M. Shinozuka, Planning structural inspection and maintenance policies via dynamic programming and Markov processes.Part II: POMDP implementation, Reliab.Eng.Syst.Saf.130 (2014) 214-224, http://dx.doi.org/10.1016/j.ress.2014.04.006.
[57] M. Ülker-Kaustell, Some Aspects of the Dynamic Soil-Structure Interaction of a Portal Frame Railway Bridge (Ph.D. thesis), KTH Royal Institute of Technology, 2009. [57] M. Ülker-Kaustell,Some Aspects of the Dynamic Soil-Structure Interaction of a Portal Frame Railway Bridge(博士论文),KTH Royal Institute of Technology,2009.
[58] T. Arvidsson, J. Li, Dynamic Analysis of a Portal Frame Railway Bridge Using Frequency Dependent Soil Structure Interaction (Ph.D. thesis), KTH Royal Institute of Technology, 2011. [58] T. Arvidsson,J. Li,《利用频率相关土体结构相互作用对门式刚架铁路桥进行动态分析》(博士论文),KTH 皇家理工学院,2011 年。
[59] European Committee for Standardization, Part 2: Traffic loads on bridges, in: EN 1991-2 Eurocode 1: Actions on Structures, 2003, pp. 66-74. [59] 欧洲标准化委员会,《第 2 部分:桥梁上的交通荷载》,EN 1991-2 Eurocode 1: Actions on Structures,2003 年,第 66-74 页:EN 1991-2 Eurocode 1: Actions on Structures, 2003, pp.
[60] P. Battaglia, J.B.C. Hamrick, V. Bapst, A. Sanchez, V. Zambaldi, M. Malinowski, A. Tacchetti, D. Raposo, A. Santoro, R. Faulkner, C. Gulcehre, F. Song, A. Ballard, J. Gilmer, G.E. Dahl, A. Vaswani, K. Allen, C. Nash, V.J. Langston, C. Dyer, N. Heess, D. Wierstra, P. Kohli, M. Botvinick, O. Vinyals, Y. Li, R. Pascanu, Relational inductive biases, deep learning, and graph networks, 2018, http://dx.doi.org/10.48550/arXiv.1806.01261, arXiv preprint arXiv:1806.01261. [60] P. Battaglia, J.B.C. Hamrick, V. Bapst, A. Sanchez, V. Zambaldi, M. Malinowski, A. Tacchetti, D. Raposo, A. Santoro, R. Faulkner, C. Gulcehre, F. Song, A. Ballard, J. Gilmer, G.E. Dahl, A. Vaswani, K. Allen, C. Nash, V.J. Langston, C. Dyer, N. Heess, D. Wierstra, P. Kohli, M. Botvinick, O. Vinyals, Y. Li, R. Wierstra.Nash, V.J. Langston, C. Dyer, N. Heess, D. Wierstra, P. Kohli, M. Botvinick, O. Vinyals, Y. Li, R. Pascanu, Relational inductive biases, deep learning, and graph networks, 2018, http://dx.doi.org/10.48550/arXiv.1806.01261, arXiv preprint arXiv:1806.01261.
[61] X. Glorot, Y. Bengio, Understanding the difficulty of training deep feedforward neural networks, J. Mach. Learn. Res. 9 (2010) 249-256. [61] X. Glorot, Y. Bengio, Understanding the difficulty of training deep feedforward neural networks, J. Mach.Learn.9 (2010) 249-256.
[62] D. Kingma, J. Ba, Adam: A method for stochastic optimization, in: 3rd International Conference on Learning Representations, Vol. 3, 2015, pp. 1-13. [62] D. Kingma, J. Ba, Adam: A method for stochastic optimization, in:3rd International Conference on Learning Representations, Vol. 3, 2015, pp.