基于神经网络—模糊集的多元过程质量异常检测及分类

Quality variation detection and classification in multivariate process based on neural network and fuzzy set technology

传统的多元控制图用于多元质量过程监控难以给出异常相关的进一步信息。而对于与过程异常关联的质量特性或其组合的检测及其偏移量的测定或定性,对于异常原因的快速诊断、纠正措施的及时制定意义重大,从而减少过程异常导致的不合格产品量。本文提出一种利用神经网络和模糊集技术对多元过程质量异常进行检测及分类的方法,神经网络模块利用其模式识别功能对过程偏移信号作出解释,确定引发异常的质量特性或其组合;模糊分类模块利用其模糊聚类功能对神经网络的输出信号加以分类,确定异常质量特性或其组合的偏移程度。并以ARL为评价指标,与多元T2控制图做了比较。

Traditional multivariate control charts for monitoring multivariate quality control process cannot provide any variation related information when variation occurs in the process. However the immediate detection of the possible quality characteristics or their combination and measurement of their magnitude of shifts can facilitate quick diagnose and corrective action before many nonconforming units are manufactured. In this paper, a neural fuzzy model for detecting mean shifts and classifying their magnitude in multivariate process is proposed. The neural network （NN） model is trained to interpret and confirm the possible quality characteristics or their combination. The fuzzy classifying module takes the outputs of NN and classifies them into various decision intervals so that confirm the shifts magnitude. Based on ARL index, the performance of the model is compared with the multivariate T^2 control chart.