Product quality prediction for injection molding based on temperature field infrared thermography and convolutional neural network

Quality prediction of injection molded products holds significant importance in terms of economic benefits. Researches have shown that the quality is closely related to product temperature field distribution while quantitatively predicting product quality through temperature field analysis remains a challenge. In this paper, a quality prediction method was proposed, combining infrared thermography and conventional neural network (CNN) model. An online system based on infrared camera has been constructed for temperature field measurement of injection molded products. Mass, tensile strength, and warpage deformation were selected as quality indexes of products, and corresponding injection molding experiments were conducted under different process parameters and thermal images of products were collected, to establish a dataset for training and testing of the proposed model. The collected images were preprocessed using image segmentation and data augmentation to improve the training performance of the model. A CNN network model was built to map the complex nonlinear relationship between temperature field and quality of the product. Optimization was performed on hyperparameters to improve model 's predictive performance which was evaluated by five-fold cross-validation, while the extraction and visualization of temperature features were achieved. Results show that the average relative errors of mass, tensile strength, and warpage deformation through optimization are 3.48 %, 3.60 %, and 8.00 %, respectively. Besides, the temperature in the boundary region of product has a significant impact on its quality. This method contributes to the development of quality prediction and monitoring for injection molded products.