A sequential bayesian operational mode classification for SHM under discrete operational variability

Structural Health Monitoring (SHM) evaluates a structure’s condition by comparing its current state to a baseline representing a healthy state. However, many structures operate under multiple regimes, requiring multiple discrete baselines. This operational variability challenges conventional mitigation approaches, highlighting the need for methods that can automatically account for discrete variations in operation. This study leverages the concept of operational modes to address multiple discrete operating regimes as classes which are subsequently determined via a robust Bayesian classifier capable of handling variations, noise, and transition periods between modes. Specifically, a Bayesian Multinomial Logistic Regression model integrated with a Hidden Markov Model (BMLR-HMM) is proposed, where the probabilistic approach ensures robustness to noise, supports informed decision-making for each classification, and exploits the temporal structure in the evolution of the modes. The model was first tested on a simple. simulated dataset, and then using the vibrational responses from gearboxes on a Crew Transfer Vessel during operation. Results demonstrate high accuracy in mode classification, with the probabilistic nature of the model enabling uncertainty quantification, forming the basis for informed decision-making when classifications are less confident. This approach contributes to the development of robust classifiers for operational analysis in dynamic environments, with applications in SHM and other domains requiring mode identification for performance or integrity assessment.