Multistage Manufacturing Quality Control And Diagnosis Based On Gearbox Casing

Production processes for quality control and diagnosis to improve product quality has very important significance. The quality of predictive control and diagnosis is control technology that based on the current processing information for predicting the next time the state of system. As SPC traditional control is belong to lag control, this method can work only when the process is in abnormal state. So we can say that it is a kind of "appear problems, analyze problems, solve the problem" method and is difficult to satisfy the demands of the rate of zero waste. The qualities of predictive control what is a "forecast to be problems, solve the problem" process can control advanced. The key to predictive control technology is the establishment of forecasting model, on the one hand, it determines the system of control precision; On the other hand, it is difficult to establish a precise prediction model because of the complexity of processing, time-varying and nonlinear. The main work in this paper is divided into three parts: (1) Principal component analysis on the factors of the manufacturing quality of the Gearbox casing; (2) Cluster analysis based on the quality diagnosis of the Gearbox casing; (3) Association rules based on the manufacturing quality control of the Gearbox casing. In this paper, a more simple and practical method of quality control and diagnosis is made.Specifically include the following elements:1. Introduced the research background, the significance of the research topic, the concept of quality control, results and research directions on the quality control and diagnostic techniques at home and abroad.2. Data pre-processing and smoothing the original data and the sub-processing, to reduce the impact of noise, that is created the conditions of clustering and correlation analysis the next principal component.3. Through principal component analysis, we can use the new six main components to replace the 24 steps of the deviation size in the processing; six principal components that retain 98.074 percent quality information expressed by the original 24 steps of the deviation size in the processing and successfully reduce the data dimension from 24 to 6-dimensional. Then we can classify the gearbox shell by PCA, determine the quality grade of each types according to the integrated scores of various types of principal component and judge which step occurred the quality question according to various types of each principal component of its own size so as to repair and maintain the equipment.4. Sample data of vertical and horizontal clustering are carried out in this paper. Vertical cluster that is using traditional two-step clustering method to classify the seizure of 300 gearbox shell. Through analysis and comparison, we found that the distinction between various types is most obvious when clustering is six, so clustering six is the more reasonable results. Horizontal clustering that is using hierarchical clustering method to cluster 24 deviation size of processing size in the work piece to be seized. Deviation of size 24 will be divided into three categories, and extracted the representatives of each type variables, namely the step 5, 18, 25. Finally, on behalf of extracted three variables made the discriminate analysis, the result indicates that three typical variables can greatly reduce computation and ensure the accuracy of discriminator well. Through the above analysis, we can use 3-bit processing instead of 24 original deviation size of processing size in the work piece to be seized, namely the step 5,18,25. It can greatly reduce the detection of the amount of labor workers and working hours.5. This chapter puts forward the concept of step association; Association rules in manufacturing systems are often used for fault diagnosis. However, purely analyzing the relationship between the steps to control the process quality has not yet been found. Firstly we can made pre-processing to the old data according to the original data and the actual production, combined with the association rules mining requirements; secondly we make use of association rules to get the results of processing requirements after work spaces through the control of a number of front steps control. We can also adjust the control previously to the front step of processing dimension and avoid a bad result of processing. Then the second part of this chapter compares with the correct rate in testing data sets using C5.0, C & RT and CHAID Decision Tree Algorithm, ultimately choose C5.0 algorithm to build decision tree according to consider various factors. Finally the pruned decision tree greatly reduce the computational complexity and almost don't reduce the ratio of good judgment at the same time. Then decision tree can be used to generate the rule set to control and judge the quality of processing the gearbox casing.6. At the end of our article, we summarized our work and prospect for the future research. In this paper, based on the SPC data of a gearbox plant and modern data mining techniques, present a suitable quality control and diagnosis method. It is beneficial to improve the quality of gearbox shell.