Multi-Source Data-Driven Quality Prediction for Machining Processes Using Hybrid Machine Learning and Deep Learning

Abstract

Research was conducted on the application of machine learning and deep learning technologies to machining quality prediction for parts in the context of industrial intelligence. Two practical projects were investigated: the first was the prediction of the flatness of engine cylinder blocks in combustion-chamber face milling on a horizontal machining center; the second was the prediction of irregular-shaped defects in circular holes during the reaming of cylinder-head guide holes. In the first project, features were extracted from both the time and frequency domains and combined with process and tool information, and machine learning algorithms such as Random Forest, XGBoost, LightGBM, and Support Vector Regression were used for regression prediction. In the second project, a one-dimensional convolutional neural network and LSTM were employed to learn directly from multi-channel vibration signals, thereby achieving classification prediction of whether irregular-shaped defects would occur in circular holes after machining. Experimental results show that the proposed methods perform well in reducing the frequency of manual sampling inspection and lowering the output of defective products. The machine learning and deep learning models can effectively capture the correlation patterns between machining quality and multi-source sensing data, achieving accurate quality prediction and defect classification in this machining scenario. This further verifies the rationality of the model design. Sensor-data-based machine learning/deep learning methods can significantly improve the automation and accuracy of machining quality monitoring and provide strong support for quality assurance in intelligent manufacturing.