Classification and Recognition of Roller Bearing Damage in Lift Installations Using Supervised Machine Learning and Vibration Analysis

The resilience of rotating components, specifically traction sheaves and diverter pulleys in lift installations, is of paramount importance. However, these components frequently undergo fatigue failure due to their exposure to intense cyclic and dynamic loading conditions. Traditional methods for estimating bearing life, which show insufficiency in adapting to the dynamic operating conditions of lifts (such as variable load, speed, and direction), often fail to anticipate these breakdowns. An experimental laboratory rig comprising a rotating disk-shaft assembly with intentionally damaged components emulating real-world scenarios has been designed to address this challenge. Vibration data, representative of actual operational conditions, were systematically captured using accelerometers. This data was then leveraged to extract salient vibration features, which served as inputs to train artificial neural network (ANN) models within a supervised machine learning framework. The trained models have shown the capacity to identify and categorise damage patterns, thereby enabling a comprehensive understanding of fatigue failure mechanisms in these systems. The findings from this research demonstrate the potential for developing robust and efficient condition monitoring methodologies, which could significantly enhance both the longevity and safety of lift installations.