MODELING AND DAMAGE DETECTION IN SYSTEMS WITH MULTIPLE OPERATIONAL REGIMES THROUGH A BAYESIAN MULTI-MODEL TIME SERIES APPROACH

Vibration-based Structural Health Monitoring (SHM) has been extensively studied for decades; however, developing robust algorithms is challenging when the underlying system operates under multiple operational regimes. The dynamic characteristics of these systems are observed as multiple discrete regimes, each with a significant effect on the vibration response features, which likewise appear as multiple discrete regimes in the feature space. Hence, the development of practical SHM algorithms requires appropriate modeling of the system across operational regimes. To address this challenge, this study proposes a Bayesian Multi-model approach for damage detection. The approach utilizes time-series models for each operating regime, with the Bayesian framework constructing probability distribution models based on available data. During the inspection phase, a time-series model is fitted to newly acquired data and compared against established reference models. The closest match provides an indication of the current operating regime, while deviations serve as an damage indicator. To demonstrate this approach, a Finite Element (FE) model simulating an output shaft on a ship is used, subjected to three distinct axial load states to represent different operational regimes. Results show that the Multi-model approach can detect propagating damage at an early stage, precisely accounting for operational variability. Therefore, the proposed Multi-model effectively improves the reliability of damage detection in systems with changing operational conditions.