基于多模态特征融合的电机故障检测方法研究

Research on a Motor Fault Detection Method Based on Multi-modal Feature Fusion

摘要:

针对电机故障特征深度挖掘与精准识别需求,提出了一种基于BiLSTM-Attention的故障检测方法,通过构建多模态特征融合数据集,有效克服了单一模态特征数据集特征有限的不足；实验结果表明,在该融合数据集上训练的模型,准确率、精确率、召回率和F1值分别达到98.64%、97.74%、98.85%和0.9904；其中,准确率较单一模态特征数据集下训练模型提升5.07%；此外,为增强模型在复杂工业现场强干扰环境下的运行稳定性,在训练中引入标准差为0.02的高斯噪声进行数据增强,模型准确率仅下降0.61%,体现了良好的抗干扰能力与鲁棒性；实验结果证明,该方法对于提升电机故障检测性能具有一定的理论价值和实际应用前景。

Abstract:

To address the demand for deep mining and accurate identification of motor fault features, this paper proposes a BiLSTM-Attention-based fault detection method; By constructing a multi-modal feature fusion dataset, this method effectively overcomes the limitation of insufficient feature representation in single-modal feature datasets;. The experimental results show that the model trained on the proposed fusion dataset achieves an accuracy of 98.64%, a precision of 97.74%, a recall of 98.85%, and an F1-score of 0.9904, respectively; Specifically, its accuracy is 5.07% higher than that of the model trained on the single-modal feature dataset; In addition, to enhance the operational stability of the model in the high-interference environment of complex industrial fields, Gaussian noise with a standard deviation of 0.02 is introduced for data augmentation during model training; Under this interference condition, the model accuracy only decreases by 0.61%, which demonstrates its excellent anti-interference capability and robustness; The experimental results verify that the proposed method has favorable theoretical value and practical application prospects for improving the performance of motor fault detection.