Supervisors: Emmanouil Kakouris, Rachel Edwards and Peter Brommer

Summary:

Monitoring of damage degradation in materials is vital as any cracks will reduce local stiffness, accelerating the ageing process of the physical assets. In-situ ultrasonic monitoring data is valuable for giving the inspectors information on the level of what may be wrong with the structural systems. However, purely data-driven representations do not always give inspectors enough information and the ability to predict how the system will behave in the future. This can be achieved by utilising some knowledge of the physics that underpins the system. This project will use the fusion of experimental data and physical models for failure prediction of materials, by taking measurements and developing models that are hybrid in nature.

Background:

Early detection of cracks allows mitigation measures to be taken to prevent damage and possible failure of a structural component. Ultrasonic non-destructive evaluation is one of the most efficient methods for detecting flaws and defects in materials, since it does not damage the object being tested. It offers the potential for both detection and sizing of cracking (Xiang and Edwards 2020). Other evaluation techniques may offer improved sensitivity in certain situations, but ultrasound is often the best choice.

In this project, a research software tool will be developed, able to acquire data from a network of ultrasound sensors and infer a response at other positions (unmeasured locations) along the structural system. The proposed model will, initially, use a data-driven approach to understand the underlying physics of elastic wave propagation in a material, and then will be A.I. trained to predict future degradation of the system.

The model will utilise a Bayesian filtering logic to tackle uncertainty in the system (Tatsis et al., 2022). This means that the physics-based model will predict the damage state of the material at the next point and time and then experimental measurements will be used to correct the predicted values. This will also be enhanced by utilising reduced order model representations (Triantafyllou and Kakouris, 2020) that are computationally efficient and rapid enough to be able be merged with data as such short time rates.

Links to HetSys Training

The project will develop rapid and accurate hybrid numerical models by blending experimental data with physics-based models to predict structural damage in materials at future stages. The HetSys training program will equip the student with skills in continuum mechanics (PX912), solid mechanics, including computational damage modelling of materials (PX920), and computer programming skills for the delivery of the research software tool (PX913). It will also provide training in stochastic modelling techniques by developing a Bayesian methodology (PX914) to tackle uncertainties of the structural system and experimental observations. To enhance the robustness of the hybrid model, the project will develop reduced order model representations (PX914) that are not computationally taxing and are able to provide rapid simulation results as data is attained. In addition, we will provide training in the latest advances in experimental damage detection and evaluation of materials.

The project brings together significant levels of inter-disciplinary expertise within the supervision team (Dr Kakouris – Civil Engineering: computational damage modelling of materials, multiscale modelling techniques, Dr Edwards – Physics: ultrasonic non-destructive damage evaluation of materials, and Dr Brommer – Mechanical Engineering: Bayesian modelling, uncertainty quantification). Thus, the project provides a great opportunity for the student to enhance their interdisciplinary cooperation with both academia and industry for future cooperation in industrial and publicly funded projects.

The development of hybrid numerical models through this project aligns with the emerging scientific field of the synergy of monitoring data and physics laws with great potentials in future.

References

Tatsis et al. (2022), “A hierarchical output-only Bayesian approach for online vibration-based crack detection using parametric reduced-order models”, Mechanical Systems and Signal Processing, 167, 108558 https://doi.org/10.1016/j.ymssp.2021.108558.

Triantafyllou and Kakouris, (2020), "A generalised phase‐field multi‐scale finite element method for brittle fracture", International Journal for Numerical Methods in Engineering, 121, pp. 1915-1945, https://doi.org/10.1002/nme.6293.

Xiang and Edwards (2020), “Multiple Wavemode Scanning for Near and Far-Side Defect Characterisation”, Journal of Nondestructive Evaluation, 39, 9, https://doi.org/10.1007/s10921-019-0651-0.