基于集成学习概率神经网络的电机轴承故障诊断

Motor Bearing Fault Diagnosis Method Based on Integrated Learning Probabilistic Neural Network

电机作为一种重要的驱动设备,在工业生产中得到了广泛的应用。滚动轴承是电机最重要的部件之一,一旦发生故障,将严重影响生产过程。因此,对电机轴承故障进行诊断,对保证安全、正常生产具有重要意义。本文采用概率神经网络(PNN)实现了电机轴承的故障分类,针对作为概率神经网络最重要参数之一的平滑因子σ需要通过经验或不断尝试的方式人为设定的问题,提出了一种基于正弦余弦优化算法(SCA)的自适应概率神经网络(SPNN);针对同一个训练集和测试集,PNN会得到固定的识别结果,从而在一定程度上降低PNN泛化能力的问题,建立了利用SPNN作为弱分类器的基于AdaBoost的集成学习模型(ASPNN),采用输出概率线性组合的方式得到强分类器的输出结果;将ASPNN模型应用于电机轴承故障诊断,仿真结果表明,与PNN、SPNN和MSVM相比,本文方法在电机轴承故障诊断方面具有更好的性能。

As an important driving equipment,motor has been widely applied in industrial production.Rolling bearing is one of the most important parts of the motor.Once a fault occurs,the production process will be seriously affected.So,the diagnosis of motor bearing fault has great significance for ensuring the safe and normal production.In this paper,the probabilistic neural network(PNN)is introduced to achieve the fault classification of motor bearing.In order to deal with the shortcoming that the smoothed factor,one of the most important parameters of PNN,which usually needs to be artificially set from experience or repeated attempt,we propose an adaptive probabilistic neural network(SPNN)based on sine cosine algorithm(SCA).It is noted that PNN will get the fixed recognition results for the same training set and test set,which may reduce the generalization ability of PNN to some extent.Hence,this paper proposes an AdaBoost-based integrated learning model(ASPNN)using SPNN as a weak classifier.Moreover,the output results of the strong classifier are obtained by means of the linear combination of the output probability of SPNN.Finally,the proposed ASPNN model is applied to the fault diagnosis of motor bearing.Simulation results show that the proposed method has better performance in motor bearing fault diagnosis than PNN,SPNN and multiclassification support vector machine(MSVM).