Prediction of Hotspots in Injection Moulding by Using Simulation, In-Mould Sensors, and Machine Learning

Abstract

Injection moulding is an industrial process for the mass production of plastic components, with many parameters affecting the quality of this process. Hotspot regions in the component occur due to non-optimised process variables or limitations in the cooling system and can lead to warpage or shrinkage. Hotspots should be minimised to avoid part defects and achieve the required dimensional tolerances for precision components. This work outlines a machine-learning-based approach for predicting the maximum hotspot temperature in an injection moulded component using process simulation and in-mould sensor data. The hotspots were identified through software simulation, and then their locations and temperatures were confirmed through an actual experiment using in-mould thermocouples. Two different machine learning approaches, artificial neural network (ANN) and support vector regression (SVR), were developed using the extracted data from the sensors and a design of experiment (DOE) method. The performance of linear and Gaussian kernels was compared for the SVR method. The Gaussian SVR resulted in superior performance compared to the linear kernel. The Gaussian SVR was then compared to the ANN prediction method, where ANN showed a slightly better prediction performance. This study has two primary outcomes. First, we show the simulation results can be used to identify critical areas of the part for real-time monitoring. Secondly, embedding sensors in these locations and applying the collected data for a machine learning analysis, provides a good indication of potential quality issues such as warpage and shrinkage post-production. The use of ANN indicates an accurate prediction performance, facilitating rapid optimisation of the process for the minimisation of hotspots.