Research On Diagnosis Methods Of Process Quality Under Incomplete Information

The quality of the product is related to the survival and development of the enterprise,which is an important manifestation of its market competitiveness.It is also the result of quality factor influence or error accumulation in its design,manufacture and assembly.In particular,under the context of the ever-increasingly diversified product specifications,the more and more complex production process,as well as the gradual predominance of the multi-variety and small batch,how to diagnose,control and prevent quality problems in time has become one of the focus of quality control field.In complex manufacturing environments,there is often a certain degree of incompleteness and uncertainty in the mastery of process state information due to the variety of processes and the complexity of quality issues.The applicability of statistical-based quality control methods has been severely limited,because its statistical reliability conditions are harsh and often unable to be fully satisfied.How to choose the quality diagnosis and control method reasonably according to the characteristics of information,dealing with the incomplete information correctly and obtaining the relative real-time and accurate diagnosis results,has become an important issue of process quality diagnosis in complex manufacturing environment.This thesis takes the process quality diagnosis with incomplete information in complex manufacturing environment as the research object.Based on the status of quality diagnostic technology and incomplete information processing theory,this paper conducts researches on several typical situations of incomplete information.The main contents are as following:In Chapter 1,scientific backgrounds and significance of research are comprehensively elaborated firstly.Then,on the basis of summarizing the present domestic and overseas research in diagnosis technology about process quality problems,the main problems of the diagnosis of process quality are put forward according to the status quo and characteristics of information acquisition in complex manufacturing environment.Finally,the content framework and structure of this thesis are presented.Chapter 2 focuses on the identification,diagnosis and control process of quality problems,and introduces quality issues,quality problems,errors and other related concepts to analyze the information needed in the process quality diagnosis.Then the ExGQM method is proposed to discuss the completeness requirement of process quality information,and the incomplete understanding of information is analyzed based on the existing research.Finally,according to the incomplete information research,several typical incompleteness of quality information,such as attribute value absence,information redundancy and inconsistency,information uncertainty and sample shortage,are summarized.On the basis of incomplete information analysis,it is essential to combine appropriate diagnosis methods to realize the diagnosis and control of quality problems,so as to effectively prevent further impact on product quality.To this end,Chapter 3 puts forward the processing strategy under the condition of the incomplete information,and points out the direction for selecting the proper follow-up quality diagnosis method.According to all kinds of incomplete information processing strategies,the method selection based on knowledge/rule is carried out from the perspectives of the advantages and applicability of the method,and corresponding diagnostic strategies are formulated.Finally,the key problem of quality diagnosis with incomplete information is pointed out in combination with the demand of quality diagnosis in complex manufacturing trend.In Chapter 4,as for the problem of how to obtain implicit knowledge from the lack of attributes information,redundancy,duplication,and inconsistency,according to the information filling and attribute reduction strategy,the research is carried out.Firstly,aiming at the limitation of Roustida algorithm in process quality information completions,this chapter improves the algorithm and gives the corresponding algorithm,which expands the scope of application in complex engineering practice,and realizes the integrity and completion of incomplete information.Then,based on the relatively complete information,attribute reduction and rule extraction are carried out by using genetic algorithm and generalized diagnosis reasoning,so as to find the simplest way to express the relationship between condition attributes and decision attributes.Finally,an example of the diagnosis for grinding quality problem of bearing ring is given to verify the feasibility and effectiveness of the research.By taking advantage of the Bayesian Networks to integrate experience-based qualitative judgment and quantitative assessment of statistical probability,Chapter 5 intends to establish the initial Bayesian Network structure by pre-causal assumptions about relevant influencing factors and quality issues of multi-source information fusion.However,the numerous nodes and fuzzy structure in the initial network may increase the difficulties for structural learning and conditional probability reasoning,so the initial structure is optimized by the K2 algorithm based on the score/search.Meanwhile,the influence of random factors in the production process is incorporated into the reasoning model by the Leaky Noisy-OR node model,so that the probability reasoning is closer to the actual while simplifying the estimated conditional probability.Finally,the feasibility and effectiveness of the proposed model and its optimization method are verified by the diagnosis of surface morphology quality of channel grinding.Compared with the naive Bayesian Network in the light of uncertain or incomplete information,this study has obvious advantages in reducing the structural complexity and the number of conditional probability estimates.In Chapter 6,a diagnostic method for small sample problems is proposed.First,the case library including the accumulated well-known mistakes in the past is established by making full use of case-based reasoning;then,according to the index strategy of feature classification,matching algorithm and similarity calculation,a similar cases from the case library is retrieved;at last,the exclusive solution will be obtained by consultation and modification of the former solution.In case of the limitation of formal representation for process quality problems,this chapter combines the extension theory with case-based reasoning,and puts forward the knowledge expression structure based on matter-element model.The compatibility of the quality problem feature terms in the multilevel index structure is realized by the extension operator.Then,the vibration problem of 608-2RS deep groove ball bearings is solved by combining the hierarchical case organization based on domain knowledge and the nearest neighbor retrieval strategy,so as to proving the feasibility and effectiveness of the method.Finally,the advantages of this method are verified by comparison with traditional case-based reasoning in terms of retrieval mode,retrieval time complexity and scope of application.The Chapter 7 is the research conclusion of this thesis.And some research limitations and prospects for further research are suggested.