The development of a damage model for the use in machine learning driven SHM and comparison with conventional SHM Methods

Approaches to damage detection can be categorised into two main approaches: model-driven methods and data-driven methods. Model driven methods pose the risk of the model departing from real physical meaning and are generally computationally expensive. Data driven methods per contra are limited by the experimental data available for all likely damage scenarios, and therefore can be impractical and costly. This paper presents the development of a damage model using finite elements (FEs) for the use in machine learning driven structural health monitoring (SHM). This method maintains a model that has physical meaning thereby removing the need for numerous experimental damage scenarios as a validated FE model can be used to simulate a plethora of likely damage scenarios. Two case studies are presented; a cantilever beam and a representative three-story building structure, for which the novel method is compared to both data-driven and model-driven methods.