面向过程输出质量与输入参数的集成诊断方法

为了诊断导致加工过程质量失控的具体输入参数, 提出了一种面向多输入多输出制造过程的输出质量特征和输入参数集成诊断方法. 克服了传统质量诊断方法仅仅能够诊断输出质量的不足, 能够同时对引起输出质量失控的具体质量特征和对该质量特征有影响作用的加工过程输入参数进行诊断. 通过建立残差T2控制图对过程的输出质量进行监控, 当发现过程失控时, 利用BN-MYT分解法对T2统计量进行分解, 找出导致过程失控的输出变量; 通过该输出变量与输入参数所对应的神经网络模型, 读取神经元权值和阀值, 带入灵敏度计算公式求解出灵敏度矩阵; 比较灵敏度大小, 找出对失控变量影响最大的输入变量. 该方法克服了传统诊断方法仅能对输出质量特性进行诊断的不足, 实现了同时对输出质量及引起输出质量变异的输入参数的诊断.

Integrated Diagnostic Approach Oriented to Process Output Quality Characteristics and Input Parameters

An integrated quality diagnosis method is proposed to detect the input parameters. It can diagnose both the output quality and input quality in a multiple-input-multiple-output (MIMO) manufacturing process. This integrated diagnosis method overcomes the deficiencies of traditional quality control and diagnosis method that can only diagnosis the output quality of manufacturing process. It can detect the input parameters of the manufacturing process and provide sensitivities analysis results for adjustment of input parameter. The quality out of control situation can be firstly detected by the establishment of residual error T2 control chart. Then, the origin output quality parameters that arouse the process quality anomaly can be found out by BN-MTY approach. It integrated the Bayesian network and MYT theory to estimate the origin output quality parameters through the decomposition of residual error of T2 control chart. Neural network and sensitivity analysis are used in the integrated network to get the weight and threshold value of nerve cell in the forecasting network. They are applied to calculate the sensitivities of input parameters to the root output quality by sensitivities computational formula. Sensitivities represent the importance of the input parameters to the output quality failure. This integrated quality diagnosis method can both diagnose the output quality characteristics and the input parameters.