Application of Online Semi-Supervised Learning Embedded With Chaotic Dynamics in Equipment Health Prognostics

The development of artificial intelligence (AI) methods with high generalization and robustness for industrial equipment prognostics remains a significant challenge. Traditional lifespan prediction models often struggle with complex, dynamic tasks that require real-time performance, primarily due to limited data availability and the inherent constraints of supervised learning approaches. These models are typically overly reliant on labeled data, which is difficult to obtain in real-world industrial settings, and they suffer from poor dynamic adaptability when dealing with varying operating conditions. To address these limitations, we propose a novel data augmentation method inspired by chaos theory, designed specifically for real-time feature extraction from industrial equipment. In addition, we introduce a semi-supervised learning framework that integrates Kent mapping with model predictive control (MPC), enabling continuous, real-time correction of predictions. This approach is capable of learning from sparse labeled data while maintaining high accuracy in forecasting the remaining useful life (RUL) of critical industrial components, such as lithium-ion batteries, turbine engines, and bearings. By overcoming the traditional model limitations—such as excessive dependence on labeled data and poor adaptability to dynamic environments—our method demonstrates strong predictive capabilities and holds significant promise for real-world industrial applications, offering improved reliability and efficiency in equipment health monitoring and maintenance planning.