Assessment criteria for optimal sensor placement for a structural health monitoring system

Machine learning algorithms have been extensively used to implement structural health monitoring (SHM) systems to detect the occurrence of damage within a structure. To obtain the most effective data for SHM decision making, it is desirable to perform sensor placement optimisation (SPO), with a particular focus on damage identification. However, comparatively little attention has been paid to systematic assessment criteria appropriate to the design of a sensor system for SHM. This paper focusses on studying the evaluation criteria at different stages of a sensor-system design process, ranging from the measurement of linear associations to the detailed evaluation of the overall probability of correct classification. The effects of the investigated criteria are demonstrated using a physics-based model with uncertain parameters related to material proprieties. Predictions of the dynamic response of the structure in different states of interest are used to derive features.