Study On Some Key Issues In Product Quality Control Of Injection Molding Process

Injection molding is an important technology for plastics processing. High-quality product is oneof major concerns during process control of injection molding process. As a typical small-volume,multiple-product batch process, injection molding is characterized as a multi-variable, multi-phase,nonlinear, non-stationary and periodic process. In addition, injection molding is prone to be sufferedfrom various disturbances and uncertainties, resulting fluctuated final product quality. In control offinal product quality for injection molding process, there remain a lot of problems which have notbeen solved thoroughly, for example,(1) how to make full advantages of the intrinsic informationfrom process data of injection molding;(2) how to improve the accuracy of soft-sensing models forthe final product quality;(3) how to develop reasonable and feasible quality control and optimizationschemes. Aiming at those issues, the main researches conducted in this thesis and the achievedcontributions are summarized as below.1. A comprehensive overview of quality control methods for injection molding process ispresented. First, the three-level-structure, open-loop and closed-loop quality control schemes areintroduced; then various closed-loop quality control methods are discussed in three groups, i.e.,process parameter optimization, model-based optimization and model-free optimization; finally,several perspectives are put forward for the future development of quality control of injection moldingprocess.2. Manifold based dimension reduction and feature analysis methods are studied and comparedfor injection molding process. Laplacian Eigenmap (LE) is selected for data clustering ofhigh-dimensional process variables. In addition, the low-dimensional manipulate variables can bestudied in Riemannian space; therefore, it is possible to apply natural gradient to the model-freeoptimization based quality control scheme for injection molding process.3. A product quality prediction method based on operating mode recognition has been proposedand verified for injection molding process. During offline modeling, LE is used to reduce datadimension and mine the distribution features of multiple operating modes, and a Mean Shift basedclustering algorithm is used to obtain the underlying operating patterns. Meanwhile, an onlineoperating mode recognition algorithm is proposed by making use of the principle of Mean Shift. Afterthat, a PLS-LSSVM modeling method, where PSO was used for parameter determination, is used todeveloped soft-sensing model of produce quality. 4. A novel quality control scheme using natural gradient based model-free optimization has beenproposed and verified for injection molding process. From the angles of Riemannian manifold andstochastic optimization, process measurements on the operational variables and product qualityvariables are directly used to search iteratively the optimal process settings for the required qualityspecifications. Meanwhile, an approximation method is derived for the calculation of the naturalgradient, and a multivariate iterative sensitivity matrix based on Riemannian geodesic distance isproposed to obtain a novel adaptive stepping strategy.