论文部分内容阅读
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。
The known state of the monitoring data is used as a priori class label to construct a new supervised S-PSO classification algorithm and to diagnose the equipment. In order to improve the accuracy of fault diagnosis and reduce the impact of randomness on the classification algorithm, a new adaptive update based on dynamic neighborhood update (ADU-DN) is proposed to expand the particle search ability of the entire solution space and guide Particles adaptively jump out of the local optimal region to ensure that the global optimal solution is obtained. At the same time, a fitness function based on the minimum intraclass distance, the maximum interclass distance, and the maximum classification accuracy of the training samples is designed so that the optimal category center of the output is balanced These three factors enhance the commonality and fault tolerance of the classification algorithm in fault diagnosis and improve the classification accuracy of test samples. The S-PSO classification algorithm effectively overcomes the defects that the clustering algorithm only considers the similarity between the data, does not consider the physical meaning of the data and can not guide the sample classification well. The comparative research on image texture feature classification of GE90 engine borehole drilling is carried out. The research data show that the S-PSO classification algorithm shows strong robustness, and the classification accuracy in fault diagnosis is higher than that of support vector machine (SVM) Neural Network Model.