Quality Prediction in Intelligent Manufacturing Processes Based on Ensemble Learning

In the intelligent manufacturing process, large amounts of data are generated from various sensors (e.g., temperature, pressure, vibration), forming high-dimensional, multimodal datasets. This data complexity makes it challenging for traditional single models to adequately capture the intricate relationships within. Additionally, characteristics such as nonlinearity and dynamism among data, data imbalance issues, and real-time requirements impose new demands on predictive models. Ensemble learning, which combines the strengths of multiple base models, emerges as a promising approach to address these challenges effectively. This paper focuses on quality prediction in the TFT-LCD (thin-film transistor liquid crystal display) production process, utilizing datasets collected during production. Starting with data pre-processing, we removed duplicates and highly correlated series, followed by feature selection, where feature sets for prediction were determined through importance ranking. Finally, a Stacking ensemble learning model was constructed, with model tuning and training applied, followed by quality prediction using the trained model. The prediction results indicate that the Stacking ensemble learning algorithm selected in this study outperforms individual base learners in terms of the mean squared error (MSE) metric, improving by 10% over the best-performing single model.