Predictive Quality Control of Rubber Injection Molding

Abstract This study presents a predictive quality control (PQC) approach for rubber injection molding (IM), aiming to reduce molding related defects of rubber products via facilitating machine learning (ML) based quality modelling. Predictive quality of cyclical processes like injection molding is a fast-emerging field of research with a predominant focus on thermoplastic. Integrating predictive quality models into control strategies offers potential savings by reducing waste. However, the complex interactions between measured process variables and setting parameters place additional demands on the controllability of these models. While previous research (Olbrich et al. in 15th Fall Rubber Colloquium. KHK, Hannover. 2024) was focused on modelling molding related quality properties with process data, the aim of this work was to integrate ML models into closed-loop control approaches to control process deviations. The integrated model serves as a quality observer, which substitutes quality measurements. If the predicted quality deviates from the target value, the controller adjusts cyclically the setpoints of the IM process to countermeasure deviations. As part of this research a demonstrator was developed and tested to predict the weight of an elastomer test plate as quality property. Due to the long cycle times caused by vulcanization, the opportunities for experimental trials are considerably restricted. For this, the controller was designed and tuned virtually, facilitating the predictive quality models. The proposed predictive quality approach was afterwards successful validated on differed defined disturbance scenarios, like material change overs between rubber batches of different viscosities. The proposed approach therefore demonstrates significant potential for reducing waste while enhancing quality assurance, in rubber IM.