Research on Industrial Equipment Fault Prediction Based on Model Stacking and Data Augmentation

Industrial equipment fault prediction is crucial for ensuring production safety and efficiency, but the characteristics of high-dimensional data, strong noise, and sample imbalance caused by complex operating environments pose challenges to prediction. Existing research still has shortcomings in model selection, performance improvement, and feature understanding. This study aims to evaluate the performance of various machine learning and deep learning models in equipment fault prediction, explore the effects of model fusion and data augmentation, and identify key features. The main work includes: comparing the performance of single models such as K-nearest Neighbor Classification (KNN), Decision Tree (DT), Gradient Boosting Decision Tree (GBDT), Random Forest (RF), Convolutional Neural Network (CNN), Deep Neural Network (DNN), Long Short Term Memory (LSTM), and eXtreme Gradient Boosting (XGboost); applying Mix Up data augmentation to alleviate sample imbalance; and adopting the Stacking strategy to fuse base models. Experimental results show that the Stacking fusion model effectively improves overall prediction performance, especially in identifying minority class faults; Mix Up has some help in improving the ability to identify minority classes, but the effect is limited. This study provides an effective technical solution for industrial equipment fault prediction and reveals directions for further optimization.