An adaptive self-generated WaveletNet method for intelligent diagnosis of mechanical equipment bearing system

Abstract The advancement of industrial automation and intelligent systems has led to a growing need for enhanced stability and reliability in mechanical equipment. This research introduces a new deep learning model called Self-Generated WaveletNet, specifically designed for adaptive fault diagnosis under varying operational conditions. The model is developed based on the Continuous Wavelet Transform (CWT) theory, enabling the creation of a composite wavelet kernel by integrating multiple wavelets through linear superposition. This approach allows the model to dynamically adjust the weight of each wavelet based on its suitability for the signal being analyzed. The study includes two comparative experiments: the first evaluates the adaptability of individual wavelet networks for fault diagnosis across diverse operating conditions, highlighting their specific strengths and limitations; the second demonstrates that the Self-Generated WaveletNet, with its adaptive nature, outperforms single wavelet networks in analyzing different operational signals. The research also validates the effectiveness of the Self-Generated WaveletNet in real-world industrial settings by showcasing its efficiency and accuracy in fault diagnosis using the aero-engine bearing dataset from Harbin Institute of Technology (HIT). In conclusion, the Self-Generated WaveletNet overcomes the limitations of manual wavelet basis selection in traditional wavelet networks by autonomously adjusting the weights of wavelet functions. This innovation offers a precise and efficient solution for fault diagnosis, significantly enhancing the capabilities of industrial automation and intelligent systems.