Comparative Analysis of Bioinspired and Domain Adaptation Approaches for Structural Health Monitoring Under Varying Environmental Conditions

Comparative Analysis of Bioinspired and Domain Adaptation Approaches for Structural Health Monitoring Under Varying Environmental Conditions Structural Health Monitoring (SHM) is of great importance across various engineering fields, offering crucial insights into the integrity and performance of structures. With the advent of advanced sensor technologies and data analytics, there has been a shift towards data-driven methods, such as machine learning and pattern recognition. However, the training and test data should not include variations in operational and environmental conditions (EOCs), which present a complex challenge in identifying structural damage. Indeed, the EOCs influence the dynamic properties and extracted features, and their effects can be similar to the damage-induced variations, limiting their detection. Several methods have been developed to account for these effects in SHM. Among these, bio-inspired approaches have emerged, aiming to mimic natural mechanisms for damage identification, even in the presence of multiple operational and environmental conditions, explicitly modelling them. Other methods seek to remove the impact of EOCs via feature harmonisation. This study aims to compare these approaches, focusing on the bio-inspired Negative Selection Algorithm (NSA), chosen for its ability of self/non-self discrimination under varying EOCs, and on the Normal Condition Alignment (NCA), a Statistic Alignment (SA) approach which harmonises the multiple undamaged condition data into a shared domain, where novelty detection can be performed. The study focuses on the numerical simulation of a population of bi-dimensional shear-frames under multiple sources of heterogeneity and damage conditions. The comparison is conducted by exploiting the numerical case study of bi-dimensional shear-frames under multiple environmental and damage conditions, allowing for a controlled analysis of the effectiveness of each approach under varying EOCs. The results will provide insights into the strengths and limitations of both techniques, contributing to the ongoing development of more robust SHM methodologies