Key Process Quality Control Based On Probabilistic Neural Network Ensemble

Quality control of key processes in product manufacturing is an important measure to ensure the quality of the final product.The statistical process control(SPC)methods commonly used in process quality control are difficult to adapt to the increasing complexity and automation of manufacturing.Combining machine learning methods in the field of artificial intelligence with SPC and pattern recognition of SPC control charts to achieve intelligent control of process quality has become a new trend.Machine learning methods represented by BP neural network,decision tree,support vector machine,and neural network integration are all studied in the field of manufacturing process quality intelligent control.This thesis introduces the integration of probabilistic neural networks into this field,and combines the specific requirements of the quality control of key processes of reducer assembly.1)The pattern recognition method of control charts using probabilistic neural network integration as a pattern classification tool is studied.First,the control chart pattern data set is created by the MontCarlo simulation method,and the pattern characteristics are determined.Different training sample sets were generated by the Bagging method,and the parameters of the probabilistic neural network(PNN)were randomly selected in a reasonable range to obtain the individual PNNs.Finally,the majority of voting algorithms are used to integrate the individual PNN outputs,and through the improved binary particle swarm optimization(BPSO)algorithm,the optimal PNN ensemble for control pattern recognition is achieved by the individual combination optimization with the goal of optimizing the pattern classification accuracy.2)A quality control model for manufacturing process based on PNN ensemble was studied and established.The model consists of sliding window sampling,data preprocessing,pattern feature extraction,and pattern classification.Among them,aiming at the problem that the normal mode and the mixed mode are confusing in pattern classification,the PNN integration and the decision tree bi-level classification mechanism are proposed.The former layer is roughly classified by PNN integration to identify the abnormal mode.The next layer is subdivided by a decision tree,correcting the normal and mixed modes of confusion.Contrast experiments with simulation data to verify the effectiveness of the model.3)Based on the established process quality control model,according to the actual requirements of the gear meshing backlash control in the assembly of the reducer,a key process quality control prototype system was designed and developed according to the software engineering ideas and processes.The core functions of the system are realized through the interoperability between.NET and MATLAB.Online data acquisition and real-time control pattern recognition are supported.Preliminary application verification is performed in the laboratory.