基于改进粒子群优化SVM的轴承故障识别研究

曹进华

摘  要：为了提高轴承故障严重程度识别的准确率，本文提出基于改进粒子群算法优化SVM的轴承故障识别方法。针对粒子群算法易陷入局部最优的不足，引入Levy飞行方式改进粒子群算法的寻优过程。在运算过程中，该方法通过粒子群的进化程度，将粒子种群动态的划分为较优子群和较差子群;较差子群以PSO算法为指导进行全局搜索，较优子群中引入Levy飞行方式，粒子围绕最优个体进行精细化的寻优过程;两个子群通过种群之间个体的重组和全局最优个体的更新实现信息交换。通过实验数据分析的结果表明：基于LPSO优化SVM参数提高了轴承故障识别的准确率，效果优于其他几种方法。

关键词：粒子群算法;支持向量机;故障识别;滚动轴承

中图分类号：TH133;TP181     文献标识码：A 文章编号：2096-4706（2019）12-0148-04

Abstract：In order to improve the recognition accuracy of bearing fault severity identification. In view of the problem，a bearing fault recognition based on improvement PSO algorithm optimized SVM is proposed. Due to the demerits of PSO optimization algorithm，such as easily relapsing into local optimum，introducing Levy flight strategy to improve PSO algorithm. In the process of computation，the method divides the dynamics of particle population into better subgroups and worse subgroups by the evolutionary degree of particle swarm. The worse subgroups are searched globally under the guidance of PSO algorithm. Levy flight mode is introduced into the better subgroups，and the particles are refined around the optimal individuals. The information exchange between the two sub-populations is realized by the reorganization of individuals and the updating of the globally optimal individuals. The results of experimental data analysis show that optimization of SVM parameters based on LPSO improves the accuracy of bearing fault identification，and the effect is better than other methods.

Keywords：PSO;SVM;fault recognition;rolling bearing

0  引  言

長期以来，滚动轴承故障模式识别一直是故障诊断领域的焦点问题。通过建立有效模型，及时检测滚动轴承的状态信息，准确识别轴承故障可以避免“维修不足”和“过剩维修”所带来的经济损失，降低维修成本，提高机械设备正常运转可靠性。近年来，因为支持向量机（SVM）方法的小样本学习能力和泛化能力突出，在故障模式识别领域得到广泛应用。张超等人[1]将支持向量机用于完成齿轮箱的故障诊断。Moura等[2]通过支持向量回归机预测失效和可靠性的问题。翟永杰等[3]通过分级聚类的支持向量机实现汽轮机故障诊断。目前，SVM的研究重点在于提高其分类性能，其关注焦点集中于优化惩罚参数和核参数。……