Accurate Identification and Location Technology of Fault Types based on Multi-Dimensional Feature Fusion Analysis Algorithm

With increasing complexity in industrial equipment operations, conventional single-feature diagnostic methods struggle to detect anomalies promptly under dynamic conditions. This study introduces an intelligent fault identification framework using Multi-Dimensional Feature Fusion Analysis. First, vibration, temperature, and pressure signals undergo wavelet threshold denoising with rigrsure-optimized thresholds to extract transient features. Next, a Gaussian Mixture Model (GMM) clusters these features through expectation-maximization iterations, resolving nonlinear correlations while mitigating overfitting risks. Finally, a Convolutional Neural Network (CNN) with spatial attention mechanisms fuses multi-source features hierarchically, enabling simultaneous fault classification and heatmap-based localization. Experimental validation on NASA's CMAPSS dataset demonstrates significant improvements: the proposed method achieves 90.2% average accuracy across five fault types, outperforming Random Forest by 12.3%. CNN-based heatmaps localize fault origins with 92% spatial resolution accuracy, proving 15.6% superior to traditional spectral analysis in handling concurrent failure modes. These results confirm GMM-guided feature fusion effectively addresses generalization limitations in complex industrial scenarios.