Sim-to-real localization: Environment resilient deep ensemble learning for guided wave damage localization

Abstract

Guided ultrasonic wave localization systems use spatially distributed sensor arrays and wave propagation models to detect and locate damage across a structure. Environmental and operational conditions, such as temperature or stress variations, introduce uncertainty into guided wave data and reduce the effectiveness of these localization systems. These uncertainties cause the models used by each localization algorithm to fail to match with reality. This paper addresses this challenge with an ensemble deep neural network that is trained solely with simulated data. Relative to delay-and-sum and matched field processing strategies, this approach is demonstrated to be more robust to temperature variations in experimental data. As a result, this approach demonstrates superior accuracy with small numbers of sensors and greater resilience to spatially nonhomogeneous temperature variations over time.

A data-driven hybrid approach to generate synthetic data for unavailable damage scenarios in welded rails for ultrasonic guided wave monitoring

“…The neural network was trained solely with simulated data, and the analysis was extended to experimental data with temperature variations. 18 This approach proved to be robust to uncertainty and showed a competitive performance to traditional localisation methods. Deep convolutional neural networks were also employed by Zhang et al, 19 Wu 20 and Rautela and Gopalakrishnan 21 in plate-like structures for damage detection, though the techniques were applied to 2D time-frequency wavelet transform image data instead of 1D time series signals.…”

“…16 The main building block in most of these approaches is a convolutional layer that applies a filter to the training data to extract the underlying features. In the studies by Khurjekar and Harley, 17,18 a deep convolutional neural network-based framework for damage localisation in the presence of uncertainty was proposed and applied to a 1 m square plate. Damage location was modelled as a multi-modal probability distribution, which made it possible to identify multiple damage locations in the plate.…”