相关向量机模型及其在T^2控制图异常识别中的应用

On Fault Identification of Mean Shifts in T^2 Control Chart Using Relevance Vector Machines

研究了关联向量机(Relevance Vector Machine)的基本原理及其训练方法,针对T2控制图的诊断问题,在假设过程协方差矩阵保持不变的前提下,建立了一种基于RVM的T2控制图均值偏移模型,该模型单独对过程各变量进行均值偏移诊断,并能指出均值偏移的方向。以一个三维过程为例,根据不同的均值偏移模型,产生SVM训练样本和测试样本。仿真分析结果表明,文章构建的模型对不同均值偏移模式下的测试数据的分类正确率均超过70%,大部分情况下超过80%。

The theory of Relevance Vector Machine（RVM） and its training method are studied.Based on a brief of introduction of T2 control chart,a RVM-based T2 control chart model is proposed under the assumption of constant variance-covariance matrix.In the model,the process variables are analyzed independently with RVM and it can give the directions of means shifts.A simulation with an three dimensional example.It shows in the simulations that the proposed model can give comparatively accurate diagnosis results.In the testing process,the correct classification ratios of different means shifts combinations are all over 70% and a large proportion of them are bigger than 80%.