基于PCA和SVM的控制图失控模式智能识别方法

控制图是在线质量控制的重要工具,而利用控制图进行异常过程模式识别却是个困难问题。该文在分析现有控制图识别技术在实际应用中存在缺陷的基础上,提出了一种基于主元分析(PCA)和支持向量机(SVM)的控制图失控模式识别方法。首先,将控制图作为信息图用于趋势模式数据集提取;然后,通过对数据集的高维特征进行线性组合并向低维空间投影的方法,降低了分类器的输入维数,提高了各维特征的敏感性;最后,为了克服神经网络方法速度慢和泛化能力弱的缺陷,利用SVM小样本学习能力,有针对性地设计SVM多分类器进行模式识别。用一个含有6种趋势的20维特征仿真数据集对该方法进行检验,通过主元分析后,数据集的特征被降到了3维并保留了88%的分类信息。进一步的识别结果表明,该方法相对现有的BP、SLFM识别方法达到更高的识别率和识别速度,适合质量控制图在线实时识别。

Abnormal Pattern Recognition Method for Control Chart Based on Principal Component Analysis and Support Vector Machine

Control chart is one of important tools for on-line quality control,but it is difficult to recognize which pattern exists on the chart.A new method based on combined principal component analysis(PCA)with support vector machine(SVM)techniques was presented after limitation of control chart-recognizer using in practice was analyzed.Firstly,universal control chart was to extract data sets of trend pattern.Secondly,through linearly associating the trend data sets in higher dimensional feature space and projecting them to lower dimensional feature space,the number of recogniter input dimension could be reduced and their sensitivity to pattern could be increased.Finally,in order to overcome the shortcoming of BP algorithm and improve the recognition speed and quality,a SVM-recognizer was designed as a classifier to lower dimensional feature data.20 dimensional simulated data sets,including six patterns,were used to test the method.Through PCA,the data sets have been reduced to 3 dimensions and contain 88% classification-messages.Compared with BP and SLFM methods,this method can obtain quicker result and higher recognition rate,so it is more suitable for on-line real time quality control.