An fault diagnosis method based on self-organizing map neural networks for the artillery automatic machine

To address the diagnostic challenges associated with rotating chamber bushings, curved sliding plates, and cam mechanisms in artillery automatic systems, this paper presents a fault diagnosis method leveraging Self-Organizing Map (SOM) neural networks. Ensemble Empirical Mode Decomposition (EEMD) is employed to decompose multisensor signals under both fault and normal conditions, followed by the use of Kullback-Leibler (K-L) divergence to extract informative Intrinsic Mode Functions (IMFs). Feature vectors constructed from four characteristic parameters (peak, waveform, margin, and kurtosis indices) serve as inputs to a SOM neural network for clustering-based classification. Simulation experiments conducted on six operating states using 54 datasets (36 for training, 18 for testing) demonstrate the effectiveness of the proposed method. The SOM network achieves full cluster separation after 1,000 training iterations, and testing yields a fault identification accuracy of 94.44%, validating the method's robustness and diagnostic precision across multiple failure modes in artillery automatic mechanisms.