predicting the deterioration trend of rolling element bearings through an adaptive relevance vector machine utilizing limited historical data

Rotating equipment, akin to human systems, necessitates diligent care and monitoring to ensure optimal functioning. given that nearly half of rotating equipment failures can be attributed to bearing malfunctions, it is crucial to develop effective predictive solutions. one promising approach is developing a model capable of forecasting bearing deterioration once it enters the degradation stage. with the rise of artificial intelligence, numerous studies have sought to estimate bearing lifespan or detect deterioration. however, these methods often rely on continuous data collection, which is frequently unavailable in industrial settings. this paper introduces a relevance vector machine (rvm) model that effectively provides predictions utilizing limited historical data while also offering results with a defined confidence level. to validate this model, run-to-failure tests are conducted in the laboratory, complemented by vibration analysis of two electro-fans in an industrial environment. the model is developed through three stages: identifying the optimal health indicators marking the onset of degradation, determining the best indicators for describing the deterioration trend, and configuring the rvm through hyperparameter optimization. the model's robustness is further evaluated against data reduction and measurement intervals, demonstrating superior predictive capabilities with accuracies exceeding 92.4% in laboratory data and over 91.1% in industrial data.