Hybrid Machine Learning and Agent-Based Metaheuristic

This framework combines autonomous agents and a decision tree to search for the optimal modal basis and vibration measurements prediction.

Embedded structural health monitoring (SHM) systems are complex and costly.

Mostly, the monitoring task can only be executed using a limited set of sen-

sors due to the inaccessibility of measurement locations. On the other hand,

hybrid agent-based metaheuristic approaches are an emergent tool in modelling

complex systems. Therefore, this work proposes a new hybrid machine learning

(ML) model and agent-based metaheuristic framework for predicting (virtual

sensing) vibration responses at unmeasured locations. This framework combines

autonomous agents and a decision tree (ML model) to search for the optimal

modal basis and vibration measurements to predict responses accurately at un-

measured locations. As a case study, virtual sensing of vibration responses on the

main deck of a catamaran, using a sparse set of accelerometers, was performed.

Those measurements were obtained during a sea trial. Statistical analysis was

performed using the ANOVA test to evaluate the robustness and accuracy of the

proposed methodology. In addition, three metrics in time and frequency domains

were also used: Root Mean Square Error (RMSE), Time-Response Assurance Cri-

terion (TRAC) and Frequency-Response Assurance Criterion (FRAC). Results

showed significant accuracy in the prediction task. In summary, this framework

is a viable solution for online structural health monitoring (SHM) applications

without requiring a high computational cost..

Preprint submitted to Structural and Multidisciplinary Optimization journal