Evaluation of Varying Noise Levels' Effects on Damage Detection in Structural Health Monitoring

The public opinion's growing awareness of ensuring enough safety level of existing structures was sadly promoted by recent iconic collapses, for instance, the Morandi viaduct collapse in Genoa in 2018 among others. Therefore, the current historical period represents an important catalyst for an unprecedented renewal period in the entire civil engineering sector, especially devoted to research and development of smart structural health monitoring (SHM) systems and artificial intelligence (AI) data-driven supported solutions. The monitoring of structural health can be formally pursued at five different investigation levels. The foremost represents the screening damage detection task which is of crucial importance to optimize time and economic resources to further deeper investigation level for understanding the nature of actual structural damage. Indeed, an effective damage identification procedure permits to promptly intervene with targeted maintenance and safety restoration actions. In this study, the authors analyzed the effects of different noise levels on a recently proposed AI-assisted data-driven damage detection procedure based on the analysis of output-only vibrational response data collected from accelerometer sensors deployed on civil structures. Specifically, three reasonable signal-to-noise ratio (SNR) levels, viz. 20dB, 40dB, and 60 dB, have been calibrated in order to simulate existing low-cost Internet of Things (IoT) sensors currently available on the market for smart monitoring solutions, i.e. focusing on micro-electro-mechanical system (MEMS) accelerometer sensors. The current preliminary numerical results showed that the chosen AI-based damage detection procedure presents a fairly good noise immunity performance, with 97% of accuracy for 1D convolutional neural network (1D-CNN) model considering three noise levels.