Gaussian process regression for active sensing probabilistic structural health monitoring: experimental assessment across multiple damage and loading scenarios

Abstract

In the near future, Structural Health Monitoring (SHM) technologies will be capable of overcoming the drawbacks in the current maintenance and life-cycle management paradigms, namely, cost, increased downtime, less-than-optimal safety management paradigm and the limited applicability of fully-autonomous operations. One of the most challenging tasks is structural damage quantification. Existing quantification techniques face accuracy and/or robustness issues when it comes to varying operating and environmental conditions. In addition, the damage/no-damage paradigm of current industrial frameworks does not offer much information to maintainers on the ground for proper decision-making. In this study, a novel structural damage quantification framework is proposed based on the widely-used Damage Indices (DIs) and Gaussian Process Regression Models (GPRMs) in order to overcome the aforementioned shortcomings. The proposed framework takes a simple approach to the damage quantification problem by using DI values for training, and provides confidence bounds on the estimated states. In addition, the proposed method is shown to quantify multiple structural states simultaneously from incoming DI test points. This framework is applied to three test cases: a Carbon Fibre-Reinforced Plastic (CFRP) coupon with attached weights as simulated damage, an aluminum coupon with a notch and an aluminum coupon with attached weights as simulated damage under varying loading states. The novel state prediction method presented herein is applied to single-state quantification in the first two test cases, as well as the third one assuming the loading state is known. Finally, the proposed method is applied to the third test case assuming neither the damage size nor the load is known in order to predict both simultaneously from incoming DI test points. In applying this framework, two forms of GPRMs (standard and variational heteroscedastic) are used in order to critically assess their performance with respect to the three test cases.

Explainable Machine Learning Framework for Guided Waves Signal Reconstruction and Structural Health Monitoring Under Varying Operating and Environmental States

“…Even in constant and controlled environments, propagations of damage and vibrations of sensor locations can lead to fault detections. Under this circumstance, Gaussian Process Regression Models (GPRM) have been developed to integrate with time series models to account for the uncertainties [8,9]. Yet these approaches still require domain expertise which can be restricted for online monitoring purpose.…”

Active sensing ultrasonic guided wave-based damage diagnosis via stochastic stationary time-series models

“…SHM methods may be broadly classified as local (or “hotspot”) or global. 2,3 For the local monitoring of a structure, a wide variety of methods are available based on ultrasound, 3–5 eddy currents, 6 acoustic emissions, 7 and thermal field principles. 8 On the other hand, the vibration-based family of methods is classified within the global monitoring category and uses random excitation/response signals combined with statistical model building, and statistical decision-making to diagnose the current health state of the structure/system.…”

“…On the contrary, some of the local active sensing SHM approaches, such as ultrasonic guided wave-based methods, are highly sensitive to local effects and can detect minor structural changes. 3,5,14 Such waves can easily be generated/collected by piezoelectric transducers in the form of an applied strain/voltage. 15,16 Damage/health indices/indicators (D/HI) are widely used metrics for performing damage detection.…”

“…Recently, steps have been taken toward formulating probabilistic DIs using Gaussian mixture models, Gaussian process regression, and other probabilistic or statistical tools. 5,17,32,33 For the case of Gaussian mixture models, instead of comparing the individual healthy and damaged state DIs, their corresponding probability distributions are compared via the Kullback-Liebler divergence or appropriate modifications. 32 However, such probabilistic DI-based formulations may show suboptimal performance if the reflections part of the guided wave signals is included in the analysis.…”

Digital Twin-Based Damage Quantification on Composite Structures

“…Machine learning techniques are utilized to alleviate extensive numerical campaigns by offering great interpolation opportunities via surrogate models [11]. A common approach to acquire training data for the surrogate models is by directly receiving experimental measurements from the subject physical asset while operating [12]. On the other hand, leveraging on numerical models to produce the relevant training data is an alternative; numerical models are used to generate the training dataset of the surrogate models that would be utilized to perform SHM strategies [13,14].…”