Dynamic Weight Optimization for Health Index Construction in Mass Spectrometer Ion Sources

This paper presents a deep learning-based method for assessing the health status of industrial equipment. The proposed Real-time Adaptive Dynamic Weighted Fusion (RADWF) model integrates Convolutional Neural Networks (CNN) and Long Short-Term Memory (LSTM) networks for extracting spatiotemporal features. Additionally, a dynamic weight optimization mechanism is employed to improve the accuracy and robustness of the Health Index (HI) predictions and fault early warning. The paper details the CNN and LSTM components of the feature extraction module and elaborates on the data preprocessing steps, emphasizing Ideal Value Normalization's role in ensuring data consistency. The RADWF model is also compared against traditional machine learning models and other dynamic weight optimization techniques, demonstrating superior performance across multiple metrics. Experimental results confirm that RADWF not only captures degradation trends effectively but also suppresses short-term fluctuations, ensuring stable predictions. These findings provide theoretical support and practical insights for intelligent health management in industrial and aerospace equipment.