Hybrid learning integration of iterative weighted least squares and backpropagation neural networks for advanced manufacturing analysis

Traditional statistical models present limitations in capturing the complexity of advanced manufacturing processes. To address this challenge, the integration of the iteratively weighted least squares (IWLS) method in the training of a backpropagation neural network (BPNN) model is proposed. This combination allows for the application of generalized linear model (GLM) statistical inference, forming a hybrid learning approach that enhances both predictive accuracy and model interpretability. The hybrid approach provides robust statistical support, identifying critical factors in manufacturing processes and offering a comprehensive tool for model evaluation and an in-depth understanding of process dynamics. The applicability of the method is demonstrated through the analysis of two manufacturing processes and identifying their key factors. The results of the deviance analysis and hypothesis testing highlight metal temperature and solidification time as crucial factors in the die casting process and input voltage as a crucial factor in the TIG welding process. This method underscores its practical value by improving efficiency and competitiveness in the manufacturing industry, providing detailed and accurate evaluations of process factors.