CWT-LSTM for damping identification in spur pair Gear System

LSTM networks have been successfully applied to predict damping in numerous studies. However, this technique often struggles with the non-linearities in gear systems. To overcome these limits, we propose a novel approach that combines Continuous Wavelet Transform (CWT) with Long Short-Term Memory (LSTM) networks. The novelty of this work lies in the use of the CWT to extract time-frequency characteristics. These characteristics are used as inputs to reduce noise, enhancing the robustness of damping estimation. So, this paper employs the continuous wavelet transform to clean time series data and improving overall data quality. The core challenge of this paper is to optimize the data prediction results, and for this purpose, a continuous wavelet transform technique is applied to the input data before the application of the LSTM method. Numerical results demonstrate the effectiveness of the CWT-LSTM approach in identifying damping, outperforming traditional LSTM method in terms of both accuracy and robustness. The proposed technique uses the CWT coefficients of position and velocity as inputs, which are scaled before feeding it into the LSTM model to found damping model as output. This approach has the potential to significantly improve the design of gear systems in engineering applications.