Vibration signal of high speed EMU air compressor

 The timely and accurate diagnosis of severe faults in the high-speed train air compressor is crucial due to the potential for significant safety issues. In response to this problem, this paper proposes a high-speed train air compressor fault diagnosis method based on an improved complete ensemble empirical mode decomposition adaptive noise (ICEEMDAN) and t-Distributed Stochastic Neighbor Embedding (t-SNE) algorithm. Firstly, the joint denoising of ICEEMDAN and wavelet thresholding is employed to address the issue of unrealistic components in traditional CEEMDAN signal decomposition. This approach ensures denoising while preserving the integrity of the original signal. Finally, vibration and pressure signals from the air compressor are subjected to feature extraction, constructing high-dimensional feature vectors. The t-SNE manifold learning algorithm is applied for secondary feature extraction, creating an MPGA-SVM (Multi-Objective Genetic Algorithm-Support Vector Machine) fault diagnosis model. Experimental results demonstrate that the ICEEMDAN and wavelet thresholding denoising method improves the signal-to-noise ratio by 6.5% and reduces the mean square error by 16.1% compared to the Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (CEEMDAN) and wavelet thresholding denoising method. t-SNE, in comparison to ISOMAP and LLE, produces a minimum intra-class distance of 1.28 and a maximum inter-class distance of 43.1. The accuracy of the MPGA-SVM fault diagnosis model reaches 98.33%, confirming its effectiveness and reliability. These studies provide important theoretical support for the improvement and application of air compressor fault diagnosis methods.