基于S-PSO分类算法的故障诊断方法

Fault diagnosis method based on S-PSO classification algorithm

将监控数据的已知状态作为先验类别标签,构造出新的有监督的粒子群优化(S-PSO)分类算法,并对设备进行故障诊断。为提高故障诊断的准确率,降低随机性对分类算法的影响,提出了新的基于动态邻域的自适应探测更新(ADU-DN)的干预更新策略来拓展粒子搜索整个解空间的能力,引导粒子自适应地跳出局部最优区域,确保获得全局最优解;同时设计出基于最小类内距离、最大类间距离和训练样本最大分类精度的适应度函数,使得输出的最优类别中心兼顾了这3个因素,增强了分类算法在故障诊断中的通用性和容错性,提高了测试样本的分类精度。S-PSO分类算法有效克服了聚类算法只考虑数据间相似性特征、不考虑数据蕴含的物理意义以及不能很好指导样本分类的缺陷。对GE90发动机孔探图像纹理特征分类进行了对比研究,研究数据表明:S-PSO分类算法表现出了较强的鲁棒性,在故障诊断中的分类精度高于支持向量机(SVM)和常用神经网络模型。

On the basis of taking the known states of monitoring data as prior class labels,a novel supervised particle swarm optimization（S-PSO）classification algorithm is constructed and used in fault diagnosis.In order to improve the accuracy of fault diagnosis and reduce the influence of randomness on the classification algorithm,a novel intervention updating strategy named an adaptive detecting updating based on dynamic neighborhood（ADU-DN）is proposed to expand the particles＇ ability of searching the entire solution space and guide the particles＇ adaptively jumping out of the local optimal region,ensuring to obtain the global optimal solution.Meanwhile,a fitness function based on minimum distance of intra-class,maximum distance of inter-class and maximum classification precision of train samples is designed to make these three factors constraint the output optimal class centers,enhance classification algorithm＇s generality and fault-tolerant ability in fault diagnosis and increase the classification precision of test samples.The S-PSO classification algorithm overcomes the defects of the clustering algorithms that only consider the similarity features of data but not the physical meanings,and do not guide the classification of samples well.A comparison study on the GE90 engine borescope image texture feature classification is carried out and the research data show that S-PSO classification algorithm has a strong robustness and the classification precision is higher than the support vector machine（SVM）and the common neural network model.