CyberSHM: A cyberphysical continuous monitoring technology for safety-critical structures

Continuous monitoring is crucial for ensuring the proper functioning and longevity of operational structures in safety-critical applications. To address, we are exploring the use of smart, self-sensing structures that combine edge computing, physics-informed and machine learning-enabled monitoring techniques. Of particular interest are thin-walled laminated composite structures with high-density cores and discontinuities that are especially challenging due to their complex waveguide behaviour. This talk will focus on the use of acousto-ultrasonic signals to monitor and interrogate such thin-walled structures for hidden barely visible damages.

The dispersion and wave scattering associated with these structures make traditional time-of-arrival (ToA) techniques ineffective for holistic damage identification. A singular focus on ToA results in under exploitation of several signal features needed for a multiclass representation of damages. Data-driven machine learning-based approaches are being explored which can map complex set of signal features to acoustic source characteristics. But rather than working as a black-box model for damage identification, these must complement the physics-based model predictions to incorporate physical plausibility and thus establish a robust grey-box predictor. Physics-based dispersion characteristics is modelled with a semi-analytical approach which allows for interlaminar damage features to be incorporated into the model. The data-driven component focusses on training a high-dimensional Bayesian surrogate model which maps complex signal features in the time-frequency domain to the damage parameters such as location, type and severity. Inverse identification is performed with a Bayesian approach which quantifies and incorporates measurement and model-form uncertainty into robust predictions of structural damage metrics and the associated confidence bounds.

The stated aim of continuous monitoring presents several challenges ranging from reducing the footprint of signal acquisition/processing hardware to combining cloud computing with edge computing to be deployed for conditioning and transmission of signals for real-time decision making. We conceptualise this as a Cyberphysical Structural Health Monitoring or a CyberSHM system which is an automated monitoring framework integrated with the internet and working collaboratively with human end-users. The study uses carbon-fibre composite panels with stiffeners as a test bench, subjecting them to impact and fatigue loading and monitored with a CyberSHM system, thus realising a generalized automated approach for online monitoring of thin-walled structures highlighting its effectiveness, challenges and a futuristic vision of this technology.

Speaker: Dr. Abhishek Kundu

Biography: Dr Abhishek Kundu is a Senior Lecturer at the Computational Mechanics & Engineering AI research group at the Cardiff School of Engineering, Cardiff University, UK and an elected member of the Royal Aeronautical Society. His research interests span the fields of structural health monitoring (SHM), stochastic structural dynamics, uncertainty quantification, machine learning and Bayesian identification. His main contribution lies in efficient computational techniques for the study of stochastic structural dynamics systems and control and data-driven approaches for SHM. He completed his PhD from Swansea University as Zienkiewicz scholar in 2014. Dr Kundu has authored more than 50 scientific publications and was awarded the best paper at the European Workshop on Structural Health Monitoring (EWSHM 2018). Amongst his main research engagements, he has been the recipient of Royal Academy of Engineering’s Industrial Fellowship with Airbus and currently serves as the principal investigator in the EPSRC funded project on CyberSHM.