A deep hybrid framework for quality prediction in complex manufacturing processes assisted by polynomial feature engineering

Accurate quality prediction of manufactured workpieces is difficult for intelligent transformation of equipment due to nonlinearity and multi-source coupling of implicit relationships during the machining process. To address this challenge, a novel quality prediction framework for manufactured workpieces based on polynomial feature derivation and hybrid neural networks is developed. In the developed method, second-order polynomials are adopted to obtain implicit relationships of workpieces embedded within multidimensional measured data. Random forest is utilized to assess the significance of these derived features, enabling the filtering of redundant information. Building upon this foundation, a hybrid neural network assisted by convolutional neural networks and long short-term memory neural networks is designed for the quality prediction of workpieces. Competitive experimental results demonstrate the effectiveness of the proposed method in workpiece quality prediction, offering a novel perspective for intelligent workpiece management during the intelligent transformation of equipment.