[Paper Review] A digital twin framework for civil engineering structures

Matteo Torzoni, Marco Tezzele+3 more|arXiv (Cornell University)|Aug 2, 2023

Integrated Circuits and Semiconductor Failure AnalysisEngineering55 references8 citations

TL;DR

This paper proposes a predictive digital twin framework for civil structures that fuses a probabilistic graphical model with deep learning for real-time health assessment and maintenance planning, demonstrated on a cantilever beam and a railway bridge.

ABSTRACT

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.

Motivation & Objective

• Motivate condition-based and predictive maintenance for civil structures to reduce lifecycle costs and improve safety and availability.
• Introduce a digital twin (DT) framework that couples physical assets with a probabilistic graphical model to handle uncertainty.
• Develop an offline training phase using reduced-order models and learning a health-dependent maintenance policy.
• Demonstrate the framework through two synthetic case studies (cantilever beam and railway bridge) to illustrate dynamic decision-making for health-aware DTs.

Proposed method

• Encode asset-twin dynamics as a dynamic decision network, a dynamic Bayesian network with decision nodes.
• Use deep learning models to assimilate vibration data and estimate the digital state for damage detection and quantification.
• Employ a reduced-order model (POD-Galerkin) to generate training data for damage identification.
• Train DL models offline for damage detection (classification) and damage quantification (regression) using simulated data from physics-based models.
• Infer digital and plan actions using a sequential Bayesian framework and a health-dependent control policy via dynamic programming.
• Provide an online phase algorithm that assimilates data, updates states, forecasts evolution, and selects actions.

Experimental results

Research questions

• RQ1 How can a probabilistic graphical model be used to couple physical asset dynamics with digital twin states for civil structures?
• RQ2 Can deep learning-based data assimilation reliably identify damage and quantify stability within the digital twin framework?
• RQ3 How effective is the offline-trained health-dependent policy for maintenance planning in a dynamic decision process?
• RQ4 What is the workflow of the online phase for updating the digital twin and choosing maintenance actions in real time?
• RQ5 Do the proposed methods generalize to different civil structures beyond the tested case studies?

Key findings

• A probabilistic graphical model enables end-to-end data assimilation, state estimation, prediction, planning, and learning within a digital twin for civil structures.
• Deep learning models are used to detect damage and quantify its severity from vibration data, integrated via a CPT-based Bayesian update for the digital state.
• A reduced-order model (POD-Galerkin) accelerates dataset generation for training the DL models, enabling efficient offline learning.
• The framework supports online updating and health-aware decision making through a dynamic decision network and a learned health-dependent policy.
• Two synthetic case studies (cantilever beam and railway bridge) demonstrate the framework’s ability to perform dynamic, health-aware maintenance planning.