Application of Intelligent Low-Cost Accelerometers for Bridge Monitoring With a Deep Learning Approach

Abstract

Despite the crucial role of structural health monitoring (SHM) in ensuring the integrity and safety of essential infrastructure, its adoption is often limited by the high costs of traditional sensors. This study introduces an innovative approach for creating intelligent, high-performing low-cost accelerometers using a deep learning framework rooted in long short–term memory (LSTM) neural networks. Initially, commercial sensors are temporarily installed alongside low-cost accelerometers on a bridge to facilitate the training process. Once the training is complete, the commercial sensors are removed, leaving the calibrated low-cost accelerometers permanently in place to perform continuous SHM tasks. In a case study, a bridge was equipped with an array of six low-cost and six commercial sensors. The efficacy of this innovative approach is corroborated through a comparative analysis of mode shapes and eigenfrequencies derived from both the low-cost and commercial sensors, as well as intelligent low-cost accelerometers.