MS17.6 - Structural Health Monitoring

This study is driven by the ever-growing enthusiasm from industry and academia towards the development of a robust, lean, and autonomous structural health monitoring system. Owing to their lightweight, reduced carbon footprint and new design possibilities composite materials are ubiquitous in various industrial applications, including large-scale safety-critical applications in aerospace, automotive, renewable, marine and construction. However, their multilayered, anisotropic properties not only lead to complex failure modes but also makes it a challenge to detect and monitor damage initiation and growth. Non-intrusive inspection methods are vital to the continuous monitoring regime and to move towards a predictive maintenance regime for these structures. This relies heavily on the capabilities of the onboard monitoring system and its efficacy of signal acquisition and processing on the edge, to extract essential signal features. Such signal features, when mapped to parameterized damage metrics, can characterize damages, and help to realize an automated framework for assessment of structural integrity and maintenance intervention points. This paper presents a systematic ultrasonic guided wave-based active inspection approach integrated with suitable data conditioning and processing operations to identify fundamental wave modes in the recorded signal. An experimental setup consisting of a signal generation/reception system was used to actuate a carbon fiber reinforced composite panel with user-defined guided wave signals over a range of carrier frequencies. A sparse array of sensors attached to this panel was used to record the responses to actuation. The recorded signals were conditioned to extract structural acoustic response in the ultrasonic frequency band and were investigated with techniques to extract frequency components and wave packets in time-frequency domain. The final objective was to filter individual structural wave modes from which experimentally measured dispersion curves could be extracted autonomously and consistently for active ultrasonic interrogation signals introduced by the transducer network. The individual wave velocities and directional properties were compared with a semi-analytical finite element model. Authors highlight their vision of an integrated cyber-physical system with the ability to make intelligent projections and extrapolations upon exploring scenarios that the physical model is yet to encounter. Quantifying the uncertainty associated with these predictions is crucial to realize an autonomous structural health monitoring system and a research area with immense potential.

Structural health monitoring of bridges aims to provide an assessment of the condition of the structure, using collected structural response. The efficacy of the methods usually is constrained by the number and spatial distribution of sensors. Developing methods to map the response from known to unknown locations has been a challenging yet interesting area of study in recent years. In this research, we propose and study a novel framework to estimate and reconstruct the dynamic response of bridges at the connection level from the vibration response at global locations. The bridge is considered as a dynamic system in which vehicle excitations are the input and responses at sensor locations are the outputs. The proposed method studies an output-only problem and the input is considered unknown. The response at two types of output locations of the bridge, one at a global and one at the connection detail level are used to learn the dynamic relationship between the time signals via convolutional neural networks. This model-free framework is validated through a finite element simulation to reconstruct the strain response at the gusset plate of a truss bridge from the responses at other nodes along the bridge.

Structural health monitoring relies on direct measurements from the structure for a variety of necessary investigations including operational modal analysis, life-cycle assessment, damage detection, and model updating. Although, because of complications and inherent inaccuracies in this direct measurement, engineers have been seeking to facilitate the data collection using machine learning-based approaches for a more robust strategy. In this work, a novel transfer learning framework is proposed to enable spatiotemporal strain estimation from acceleration measurement of the bridge for unmeasured or faulty locations and times. This task is executed by extraction of time-dependant and location-dependant contents of collected vibration signals and reconstruction of strain signal merely by providing the desired time and location. The framework has been verified on a simulation case study and showed high accuracy signal reconstruction metrics.

SHM is vital in quantitatively identifying engineered critical structural damage due to its potential economic and security interests. Convolutional Neural Network (CNN) is a popular method used for SHM on damage localization and classification. However, traditional CNN methods have limitations in predicting performance uncertainty and only provide point evaluations without indicating their accuracy. To address this issue, this paper introduces a PCNN framework, which combines a traditional CNN with a probabilistic layer to generate overall confidence intervals (CIs) for prediction results, as well as conditional probability distributions (CPDs) and likelihood for each prediction result. The PCNN method provides a manner to quantify the prediction uncertainty of neural networks and determine the confidence of each prediction. The paper also recommends using Leaky ReLU as the activation function, which retains negative value information. The effectiveness of the PCNN method is illustrated through case studies of carbon fiber-reinforced polymer beams with different layups. The results show that PCNN is effective in giving damage location prediction for CIs, CPDs and likelihood.