基于GAPSO-SVM的多级齿轮箱故障诊断新方法

A new fault diagnosis method for multistage gearbox based on GAPSO-SVM

多级齿轮箱是机械传动的重要部件,针对运行过程中的状态识别问题,研究并提出一种基于振动信号的小波包分解能量谱特征提取和支持向量机(support vector machine,SVM)的智能评估新方法。用小波包分解算法对振动信号进行分解,提取时频信号的能量谱构建多级齿轮箱状态特征集,训练SVM模型。针对SVM的惩罚因子C和高斯核参数g选择困难的问题,结合遗传算法(genetic algorithm,GA)和粒子群算法(particle swarm optimization,PSO)的基因粒子群算法(genetic algorithm-particle swarm optimization,GAPSO)优化SVM参数。GAPSO同时具有GA全局搜索的性能和PSO快速收敛特点。将优化后的SVM算法应用于多级齿轮箱故障诊断,结果表明,GAPSO-SVM模型故障识别精度为98.55%,高于基本的SVM、PSO-SVM和BP神经网络,而且泛化能力强,该方法更适合多级齿轮箱故障诊断。

Multistage gearbox is an important part in mechanical transmission.Aiming at the problem of state recognition during operation,a new intelligent evaluation method based on the wavelet packet decomposition energy spectrum feature extraction and support vector machine(SVM)is researched and proposed.The wavelet packet decomposition algorithm is used to decompose the vibration signal,with the energy spectrum of the time-frequency signal extracted to construct multistage gearbox state feature set.In view of the difficulty in selecting the factor C and Gaussian kernel parameter g of SVM,genetic algorithm-particle swarm optimization(GAPSO),which combines genetic algorithm(GA)and particle swarm optimization(PSO),optimizes SVM parameters.GAPSO has both the global search performance of GA and the fast convergence of PSO.The optimized SVM algorithm is applied to fault diagnosis of multi-stage gearbox.The results show that the fault identification accuracy of GAPSO-SVM model is 98.55%.The accuracy is higher than that by the basic SVM,PSO-SVM and BP neural network,and has strong generalization ability.This method is more suitable for fault diagnosis of multistage gearbox.