As such, careful engineering and considerable domain knowledge are required to extract damage-sensitive features from the raw data which are then fed into a suitable ML model. Conventional machine learning techniques are however limited in their ability to process the large amounts of measured sensor data in their raw form. Machine learning in SHM aims at building models or representations for mapping input patterns in measured sensor data to output targets for damage assessment at different levels, Rytter. This motivates the use of a data-driven approach for SHM where-in damage assessment is dealt with, at least at a lower level, primarily as being a type of statistical pattern recognition problem thus circumventing some of the major challenges associated with the physics-based approach.ĭuring the last few decades, machine learning (ML) techniques have been extensively employed by researchers both for vibration based and ultrasonic guided wave based damage detection.
Owing to the advances in information and sensing technologies in recent times, it has now become feasible to monitor a large number of parameters in-situ in large/complex real-world structures on either a continuous or sporadic basis. As the underlying system complexity increases, such an approach becomes much less dependable. This limits the application of such methods to the health monitoring of rather simple structures with pre-defined boundary conditions and well-controlled environments. However, difficulty in the modelling of complex real-world structures, considerations of multiple sensing modalities, material and/or geometric non-linearity, and uncertainty in material properties, boundary conditions, environmental/operational variations are some of the factors that make this exclusive reliance on the system physics rather impractical in case of complex real-world structures. Physics-based methods, in some form or the other, rely primarily on the physical laws governing the structural behavior in order to extract meaningful information about the damage and its evolution from the measured sensor data.
An optimal approach for large/complex real-world structures must be data-driven with the system-physics/domain knowledge embedded in some form. The big picture of structural health monitoring (SHM) indicating that as the complexity of the structure increases, exclusive reliance on the system physics for analyzing the measured sensor data is reduced, and an approach that is primarily data-driven is adopted. Lastly, as the final attribute of an optimal SHM approach, a sensing paradigm for non-contact full-field measurements for damage diagnosis is presented. Other ways of incorporating domain knowledge into the machine learning pipeline are then presented through case-studies on various aspects of NDI/SHM (visual inspection, impact diagnosis). Forward and inverse problems involving partial differential equations are solved and comparisons reveal a clear superiority of physics-informed approach over one that is purely datadriven vis-à-vis overfitting/generalization. As a step towards the goal of automated damage detection (mathematically an inverse problem), preliminary results are presented from dynamic modelling of beam structures using physics-informed artificial neural networks. Physics-informed learning, which involves the integration of domain knowledge into the learning process, is presented here as a potential remedy to this challenge. Most of the supervised machine-learning/deep-learning techniques, when trained using this inherently limited data, lack robustness and generalizability. However, the lack of sensor data corresponding to different damage scenarios continues to remain a challenge. The availability of this historical data has engendered a lot of interest in a data-driven approach as a natural and more viable option for realizing the goal of SHM in such structures. With the advances in information and sensing technology (sensors and sensor networks), it has become feasible to monitor large/diverse number of parameters in complex real-world structures either continuously or intermittently by employing large in-situ (wireless) sensor networks. A physics-based approach to structural health monitoring (SHM) has practical shortcomings which restrict its suitability to simple structures under well controlled environments.