Multivariate Empirical Bayes Modeling And Quality Monitoring For Autobody Measuring Data

Auto-body manufacturing quality affects the quality of cars ultimately, thus control ofit is highly valued by the car industry overseas and at home.Auto-body is usually gauged by CMM. Only one or two cars will be selected forinspection, but the number of gauging points amounted to hundreds, which leads to lowaccuracy and long circle of feedback. And other factors, like fluctuating batch quality ofStamping Parts, fixture abrasion, batches rotation, decrease the stability of measurementdata and increase the difficulty of quality assessment.In contrast to typical statistical theory, bayes statistic can improve estimatingaccuracy of the sample on the ground of measurement data. It has been applied in theresearch field in small sample measurement data of auto-body effectively, but there stillexists some problems in the following. On one hand, multivariate empirical bayesmodeling is complicated, which enhance the difficulty of making full use of multivariatehistorical information of auto-body data. On the other hand, when stability of data is low,the result of prior distribution doesn’t be in accordance with that of current sampleinformation, which affects the estimating accuracy heavily.In order to solve the problems above, this thesis proposes multivariate empiricalbayes model with dynamic intervention, specifically aiming at the key problems ofmultivariate empirical bayes modeling, data stability measuring, dynamic intervention,dynamic intervention efficiency and estimating accuracy and the development andapplication of data analysis software which improves the estimating accuracy of smallsample measurement data of auto-body. The main points of this thesis are illustrated in the following:1. Multivariate empirical bayes modeling of auto-body quality. To begin with,analyze measured method, arrangement of measurement point and measured data, thenconclude statistical characteristics of auto-body measured data. What’s more, fuse priordistribution with measurement data and estimate quality indexes, such as, mean values,standard derivation values, etc through the combination of multivariate empirical bayesmodeling with relevant information of measurement points. Finally, research therelationship between estimating accuracy and model parameters and analyze the variationrule of data stability decrease and estimating accuracy of the model.2. Dynamic intervention of monitoring for data stability and MEB. In the first place,construct bayes control charts which are suitable for small sample measurement data;analyze and set suitable control limits based on alarm sensitivity of control charts, thusrealize real time monitoring of data stability. Secondly, directing at the situation when theprocess stability is low, recognize the current derivation modes type on the groundanalysis of them. On top of that, correct multivariate empirical bayes modelingautomatically based on the results of modeling analysis.3. Modeling dynamic intervention efficiency and estimating accuracy of qualityevaluation. On one hand, analyze the efficiency of dynamic intervention algorithm,including accuracy rate of deviation modes recognition, starting time estimating errors,propose data grouping algorithms based on foregoing results of analysis. On the otherhand, using massive real data of auto-body small samples as analysis objects; discuss theestimating accuracy and feasibility of MEB after fusing of dynamic interventionalgorithm.4. Software development and engineering application. Based on the above discussion,development auto-body data analysis software; realize functions of grouping ofmeasurement data, estimate of quality indexes,(monitoring for data stability and deviationmodes recognition. Apply the software in quality evaluation and monitoring of frames ofBIW, frames of lift-gate and check the validity of algorithm in this thesis.The results show that the multivariate empirical bayes model with dynamicintervention algorithm propose in this thesis can improve the estimating accuracy ofauto-body small sample data, whose results can provide effective preference for process monitoring and alarm. This method not only can it be applied to auto-body measurementdata processing, but also it can be applied to other manufacturing fields, such as highspeed trains, aeronautics and astronautics etc.