基于SDP图像转换和融合自注意力机制U-net神经网络的电机故障诊断

打开文本图片集

中图分类号：TB9 文献标志码：A 文章编号：1674-5124（2026）02-0162-13

Abstract： To addressthe challenges of dificult fault feature extraction in motor systems and the smal-sample problem where actual samples cannot meet model requirements， this study develops an efficient and accurate motor fault diagnosis method. The proposed approach integrates parameter-optimized Symmetrized Dot Patern （SDP） and the semantic segmentation model U-net.First， SDP is employed to analyze the three-phase vibration signals of the motor， transforming fault features into SDP images and enhancing feature signals through symmetry-based color inversion. The generated images undergo data generalization and augmentation before being fed into the U-net network for upsampling and downsampling processing，followed by fault classification via a multi-layer perceptron. Tests conducted on a normal motor， broken-bar motor，single-phase open-circuit motor， and rotor misalignment motor under three load conditions （0 HP，1 HP，and 2 HP）on an experimental platform show that the improved U-net achieves a fault diagnosis accuracy exceeding 98% Experimental results verifythat this method efectivelysolves the problems offeature extractionand small samples in motor fault diagnosis， demonstrating superior performance and reliable diagnosticcapabilities across various motor fault scenarios.

Keywords： motor fault diagnosis;symmetry point mode; difusion model; U-net neural network; self-attntion mechanism; multi-level perceptrons

0引言

电机作为现代工业体系中不可或缺的关键设备，正经历着大规模集成化与智能化的变革。（剩余15403字）