基于混合模型与改进多分类马田系统的控制图模式识别

Control Chart Pattern Recognition Based on Hybrid Model and Improved Multi-classification Mahalanobis-Taguchi System

为提高生产过程中产品质量的智能监控水平,提出基于时间序列混合模型及改进多分类马田系统的控制图模式识别算法。选用时间序列混合模型对控制图实时数据进行特征提取;改进马田系统的阈值计算方法并制定多类判别准则,将表征的特征向量代入改进多分类马田系统分类器中进行特征约减及模式识别。最后,将该识别算法应用于控制图公开数据集及生产案例中,以验证算法的有效性,并与其他算法对比了分析,结果表明,基于时间序列混合模型及改进多分类马田系统算法能简化识别系统,识别精度高,是一种更为有效的控制图模式识别方法。

In order to improve intelligent monitoring level of product quality during the production, time series hybrid model and improved multi-classification Mahalanobis-Taguchi system were proposed to construct control chart pattern recognition algorithm. Firstly, time series hybrid model was applied to extract features for the control chart real-time data. Secondly, threshold calculation method was improved and multi-classification’s discriminant was defined. Improved multi-classification Mahalanobis-Taguchi system was applied to reduce the dimension of features and recognize control chart patterns. Finally, to verify the validity of the algorithm, the public control chart dataset and manufacture cases were tested and the results were compared with other algorithms. Results indicate that an algorithm that is based on time series hybrid model and improved multi-classification Mahalanobis-Taguchi system may yield several successes including higher accuracy and simplify recognition system. Therefore, it is an effective method of control chart pattern recognition.