Process Adjustment And Monitoring For The Integrated SPC And APC System Based On Neural Network

The integrated SPC/APC system as a process control approach to reduce variation in manufacturing industry is becoming a new research hotspot. SPC, as an important tool to detect and eliminate process abnormal disturbance, is used to monitor the process output which adopts control chart technique. However, the process output are auto-correlated due to the complexity in modern manufacturing environment which break the assumption that the process data are independent for the traditional SPC techniques and increase the chance of false alarms. APC, which is based on process feedback compensation in order to keep the process output as close as to some desired target is an effective tool to remove self-correlation. In the framework of the integrated SPC/APC system, when APC is used to adjust and tune the process, it reduces the impact on the system caused by abnormal factors at the same time .That will affect the effectiveness of SPC and delay detecting and eliminating the process abnormal disturbance. These shortcomings need remedy from two aspects to approve the performance of the integrated SPC/APC. More suitable controller to adjust the process should be designed and SPC monitoring technique should be improved.The artificial neural network is a highly complex nonlinear dynamic system with good learning, associative memory, mapping and generalization abilities which is consisted of a large number of simple process neural units that connected with each other. So it is widely used in the fields such as system identification, time series forecasting, pattern recognition and so on. This thesis introduces artificial neural network into the integrated SPC/APC system. The study on process adjustment and monitoring in this thesis are as following: (1) For auto-correlated process, a time series forecasting approach based on radial basis neural network (RBF) has been presented to upgrade the performance of MMSE controller. Based on above process adjustment, a self-adaptive BP neural network controller has been designed which had be trained initially through collecting discrete data and then proceed online. Those works above are a part of the new APC. Simulation results show that compared with traditional APC strategies, the APC strategy based on artificial neural network has high accuracy and fast data-processing features. (2) To enchance the process monitoring for the integrated SPC and APC, when considering the different process model and different process abnormal disturbance, BP neural network models for the process monitoring have been given which combining the process adjustment variance and output variance to construct a new characteristic variance as the input of the designed BP neural network so as to building a classification model with good ability to monitor the process while compared with the traditional control chart techniques. Abnormal factors which existing both in the process adjustment variance and the output have been taken into account comprehensively. Such approach has more comprehensive, accurate and timely detection of process abnormal factors. (3) This thesis takes MSE and ARL respectively as performance index to evaluate process adjustment and monitoring through comparative study by simulation and analysis. The results show that the integrated SPC and APC based on artificial neural network can effectively improve the performance of integrated SPC and APC system.