Research On Prediction And Diagnosis Method Of Processing Quality Abnormalities

Quality is more important in today due to intensive competition in market. To win the market, the enterprise must assure its product quality and can predict quality of products, thus can minimize waste in production, achieving high quality at low cost. In the processing procedure, affecting factors do not act as one, it is possibly for them to act together, thus exists combined effect. In the course of machining, the procedure parameters are not always the same, they change because they are affected by the machined parts, e.g., machining tools and fixtures and other things related with machining will wear, even damage. The interrelationship between the process quality and the processing factors makes the product quality is more difficult to control. If we are able to find the association, it is much easier to control the product quality, to achieve quality improvement, winning more extensive market.Due to multiple quality factors and their complex interaction in multi-process production, it is difficult to build an exact mathematical model, and the created model not intuitive. Data mining is a knowledge-disvovery technique, in which clustering method is the one in common use. The Clustering method can classify a given sample into several clusters according to some principles under non-supervised and supervised conditions, and the result has a better interpretation. Acounting for the enomous and dynamic process quality data, the influence uncertainty of the 5M1E(man, machine, material, method, measurement, environment) on the process quality, we made use of the data mining method and dynamic statistical process control technique to study the process quality.It was intented for the paper to study the process quality trends and diagnosis method. The quality control chart is a significant content in the statistical process control. Abnormalities of quality characteristics data can be judged by control chart, and also that of the process, but which one is the abnormal quality factor can not be located.In this paper, pattern recognition methods can be adopted to recognize quality control charts. Clustering methods in the data mining was used to cluster the identified control chart patterns. Statistical process control techniques were used to analyze quality characteristics data and diagnosis for the failure. Quality abnormalities were defined and procedures for the quality prediction, princeples of judging abnormalities, shortcomings and advantages of control charts for judging abnormalities, processes and cautions for clustering method to judge abnormalities were given in the paper. In the studies, abnormal prediction models were used to study the patterns of quality control charts. The relationships between quality characteristics and quality affecting factors can be obtained. In the paper, the influence of quality factor changing on the variation of quality characteristics was analyzed and the relationships between the abnormalities of quality characteristics and failure metadata were analyzed. And conceptual-driven diagnosis principles and processes were stated. To use concepts to perfoming diagnosis, several theories and methods were proposed in the paper. For this, it is suggested that data mining techniques were applied. Statistical process control indexes, control chart pattern recognition methods, selection of clustering methods and its measure criteria, validation of its validity were determined in the paper. By custering, the relationships between quality characteristics and quality affecting factors were gotten.Before predicting, control charts were used to analyze the quality data. Using the pattern generated judged the process abnormalities. By clustering of ordered swatch, quality data patterns abnormal and normal were clustered, getting clusters of the combination of quality affecting factors.This method in fact is to cluster for the quality affecting data variables. As clustering criteria in the paper, Fisher judging algorithm is a typical one used for the clustering of ordered swatch. When clustering, Euclian distance is used as measure criteria. By analyzing the dissimilarities between interclusters and similarties between intraclusters, we can know the differences between quality affecting factors. Hurst index can be used as a self-similarity measure of quality sequence data, by which the changing of quality combined factors over time can be derived. In this paper, by analyzing differences between quality affecting factors and Hurst indexes, the influence of quality affecting factoring on qulity characteristics was achieved.By analyzing the phenomena of quality abnormalities, we gave the prediction method and model of quality abnormalities. To achieve the quality prediction, we must specify quality affecting fators and analyze the influence of their changing on quality characteristics. By control chart to divide the quality data, using clustering method to cluster the patterns in the control charts, with the pattern's similarity as the clustering criteria, anylyzing and comparing the quality affecting factors generating the same pattern quality data, we can find the associationships between control chart patterns and quality affecting factors, then predict the quality according to the current state of quality affecting factors.To achieve quality control, quality improvement, making quality stable, it is necessary to decrease the fluctuation of quality characteristics, i.e., the deviation changing range between machined values and the required. The key of this is to find variation sources and determine the influence of them on quality characteristics. It is a key factor for the evaluation of clustering quality. In this paper, Pearson's correlation coefficient is introduced as measure criterion of relational clustering. By it, clustering can be performed automatically without specifying the number of clusters. The significance of differences between the variance and mean is used as a measure to cluster data formed by the same factors. Quality diagnosis can be carried out by the method.Before performing quality diagnosis, we must analyzied quality characteristics data and failure metadata. Conceptual clustering techniques in the data mining technique were used to cluster the quality data with the correlation of control chart patterns as the clustering criteria. By comparing the quality characteristics data with the failure metadata, we can find the relationships between. Then according to quality characteristics data, it can be known what the failure is. At last with the gear box housing machining as a example, optimum combinations of quality affecting factors were obtained by statistical process analysis. Hurst indexes of the combinations of quality affecting factors over time were computed. Logistic regression models were used and the affecting degree of quality affecting factors on quality characteristics data was known. Then clustering method was used to cluster the new obtained data sample identified by control chart pattern with Pearson correlation coefficient as the measure criteria. The extracted concept of conditional propability by Baysian method was used to derive the corresponding relations between quality data and failure metadata. By the example, we have validated the proposed method.