Integration of Multi-Gaussian fitting and LSTM neural networks for health monitoring of an automotive suspension component

For mechanical structures under harsh operational conditions, a structural health monitoring method is a promising tool to reduce safety risk and maintenance costs. With the rapid development of advanced sensing and data analysis techniques, a massive research effort has been performed to monitor structural health from vibration signals. However, practical engineering components usually have nonlinear dynamic characteristics and high volume of measured data. Therefore, applying a health monitoring system to an actual project is far from an easy work. Based on the recent field studies, a novel method is proposed to achieve a trade-off between prediction accuracy and computation efficiency. In this method, we integrate a multi-Gaussian fitting feature extraction method and an LSTM-based damage identification method to develop a health monitoring system with available vibration signals. Through the proposed method named GaPSD/LSTM, frequency-domain features from sequential vibration signals are firstly extracted by multi-Gaussian fitting of power spectral density (PSD) curve. Then, a long short-term memory (LSTM) neural network is used to predict partial damage level. The proposed GaPSD/LSTM is validated by automotive suspension durability tests. The experimental results prove that the proposed method can significantly reduce computation time in the condition of achieving great prediction accuracy, compared with several state-of-the-art baseline methods.