基于机器学习算法的智能诊断方法在轴承故障诊断中应用广泛,但是大多数诊断模型均基于某一种机器学习算法.不同的机器学习算法具有不同的特点和适用范围,基于单一算法的智能诊断模型提取故障信息的能力存在一定的局限性,容易导致误诊和漏报.针对该问题,提出一种轴承故障混合智能诊断模型.该模型的详细技术路线如下:首先,提取对轴承故障敏感的时域和频域特征来构造特征集;其次,利用MLP神经网络、多分类SVM和随机森林三种智能分类器对轴承故障进行初步诊断;最后,利用D-S证据理论进行决策级融合,以综合利用多种机器学习算法的优点和适应性,并输出最终的判决结果.实验表明该模型具有内部纠正机制,较之单分类器诊断模型准确率明显提升,更加适用于轴承的故障诊断.
Intelligent diagnosis methods based on machine learning algorithms are widely used in bearing fault diagnosis,but most diagnosis models are based on a certain machine learning algorithm.Different machine learning algorithms have different characteristics and application scopes,and the ability of an intelligent diagnosis model based on a single algorithm to extract fault information has certain limitations,which can easily lead to misdiagnosis and omission.To address this problem,a hybrid intelligent diagnosis model for bearing faults is proposed.The detailed technical route of the model is as follows:Firstly,the time domain and frequency domain features sensitive to bearing faults are extracted to construct feature sets;secondly,three intelligent classifiers,MLP neural network,multi-classification SVM and random forest,are used to preliminarily diagnose the bearing faults;finally,D-S evidence theory is used for decision-level fusion,so as to comprehensively utilize the advantages and adaptability of various machine learning algorithms and output the final judgment result.Experiments show that the model has an internal correction mechanism,the accuracy of the model is significantly improved compared with the single classifier diagnosis model,and it is more suitable for bearing fault diagnosis.