Research For Recognizable Virtual Measurement With Injection Molding Big Data Application

One of the core values of Industry 4.0 is to integrate people’s needs into the manufacturing of enhanced products,systems and services to enable a higher level of product customization.Low yields represent lost capacity and wasted capital,so improving product yields is a key factor in the profitability of the injection molding industry.In order to achieve this goal,it is necessary to be able to predict and control the quality of injection molded products during the production process.Most of the methods to predict and control the molding quality of injection molded products rely on a large amount of historical process parameter data when the product fails to maintain a stable production,but sometimes it is not easy to identify and analyze most of the data,so it is impossible to establish an accurate prediction model,which makes it impossible to effectively predict the molding quality of products.Usually,the products are tested by sampling,and this method can only obtain the quality of the sampled products,while the quality of other injection molded products without sampling is still unknown,and increasing the sampling rate is not a feasible solution.And the use of the traditional sampling and testing method is bound to produce a certain measurement delay,which will produce a considerable risk of defects or nonconforming products.Therefore,there is a need for a method that can assess the molding quality of injection molded products online and in real-time prediction.And the identifiable performance evaluation(RPE)method has multiple identification functions with dual identification characteristics,which can analyze the defect factors with different characteristics for two-by-two identification.Therefore,this study mainly uses the recognizable performance evaluation method to analyze the interactive dynamic effects of process parameters on injection molding problems in the injection molding manufacturing process,and combines the use of virtual metrology technology and other related research methods to achieve real-time,online comprehensive inspection of injection molded products,so as to achieve predictive control of injection molded product quality.In this research topic,the following work is focused on.(1)In this paper,based on the recognizable performance evaluation(RPE)method virtual metrology(VM)method,APC is used in its data pre-processing,and a two-stage algorithm is proposed in combination with the concept of digital twin,a class of neural network is used as its base prediction model,and the concept of product quality state transfer is proposed.The identifiable performance evaluation method and Pearson correlation coefficient are also applied in the data identification processing of the prediction model,where the identifiable performance evaluation method is the key to link different algorithms(neural network-like,Pearson correlation coefficient,etc.)of this study in tandem,making it more accurate in predicting and assessing the molding quality of products.(2)In this study,the experimental study of "Optimizing the molding quality of industrial components based on the reverse warpage characteristics of glass fibers" successfully verified the reliability of applying the virtual measurement two-stage algorithm for optimal control of product molding quality by combining the concept of digital twin.(3)In this study,the experimental research on "virtual measurement control of molding quality of recycled material products(plastic bottle caps)based on large column extension data and experimental results" successfully verified the feasibility of using the concept of virtual measurement,combined with APC and Pearson correlation coefficient,to propose a quality monitoring method to predict and control the molding quality of recycled material products.The feasibility of using the concept of virtual measurement and combining APC and Pearson correlation coefficient to propose a quality monitoring method for predicting and controlling the molding quality of recycled materials and reflecting the concept of product quality transfer through the change of product quality.(4)In this study,the experimental research of "developing deep learning network(RPE+ANN)for predictive control of bumper molding quality based on machine data" successfully verified the effectiveness of a new deep learning network proposed in this study based on identifiable performance evaluation method and artificial neural network for predictive control of molding quality of injection molded products.The effectiveness of a new deep learning network based on identifiable performance assessment method and artificial neural network for predictive control of molded product quality is successfully verified,and the authenticity of combining the unique RPE method with dual identification characteristics and PCC to improve the accuracy of the prediction model is confirmed.(5)The identifiable performance evaluation(RPE)method proposed in this study has multiple identification functions,so that it can perform dual identification characteristics,i.e.,two-by-two identification between injection molding defect factors with different characteristics(look large,look small,look small,and unitless),instead of only identifying defect factors with a single characteristic,which further improves the reliability and accuracy of the virtual prediction model of this study.