基于信息熵和超限学习机的制造过程控制图模式识别模型

Pattern-recognition model of manufacturing-process control chart based on information entropy and extreme learning machine

为提高复杂生产过程中质量监控的效率,文中将信息熵特征提取方法与超限学习机(Extreme Learning Machine, ELM)分类算法相结合,提出了一种基于信息熵的控制图模式识别ELM模型。首先,根据控制图模式的波动情况,采用奇异谱分解提取不同模式的信息熵特征;其次,将提取的过程信号信息熵特征作为ELM模型的输入,训练控制图模式识别ELM模型;最后,通过蒙特卡洛方法模拟生成不同控制图模式信号,采用仿真试验对比分析不同识别模型的监控效率,并进一步利用实际生产数据验证所提ELM识别模型的有效性。仿真试验和实际数据分析表明,所提识别模型具有训练成本低、识别精度高的特点,能够用于实际生产制造过程的实时质量控制。

In this article,in order to improve efficiency of quality monitoring in complex the production process,efforts are made to propose the Extreme Learning Machine model based on information entropy,which is used for pattern recognition of control chart,in combination with the information-entropy feature-extraction method with the classification algorithm of extreme learning machine.Firstly,according to the shift patterns of control chart,the information-entropy features of different patterns are extracted by means of singular spectrum decomposition.Secondly,the extracted information-entropy features of process signal are used as the input of the ELM model,so as to train the ELM model for pattern recognition of control chart.Finally,different pattern signals of control chart are generated with the help of Monte Carlo simulation,and then the monitoring efficiency of different recognition models is compared and analyzed by means of the simulation experiments;thus,it is verified that the ELM recognition model is effective according to the data on production.The results obtained from the simulation experiments and the data analysis show that this recognition model with the characteristics of low training cost and high accuracy in recognition can be used for realtime quality control in the actual manufacturing process.