ABSTRACT
Fatigue assessment of bridge infrastructure entails in-situ instrumentation of structural components. Consequently, the quality of collected data can be compromised considerably owing to various noise sources. Previously, the authors proposed a novel Artificial Neural Network (ANN) architecture for strain signal estimation from field acceleration measurements. This study aims to examine the impact of acquired acceleration data quality on strain signal estimation using the proposed ANN. Data collected from a real-scale case study is utilized to analyze noise levels in the input acceleration signal, measured strain signal, and subsequently the predicted strain signal to show if and how the quality of data has been affected by the ANN model. This is an effort to further understand the so-called black box neural networks specifically in the context of uncertainty propagation. Understanding the impact of data quality on the ANN’s performance is crucial for enhancing the accuracy and trustworthiness of structural health monitoring in real-world applications.
