Development of Advanced Techniques For Gear Wear Monitoring and Prediction

Abstract

Gears are widely used in industrial machinery, and gear failure is a main cause of machine failure. Since gear wear is often the initial stage of gear failure, its monitoring and prediction are key to minimising machine downtime, maintenance costs, and safety risks. However, existing gear wear monitoring and prediction techniques face some ongoing technical challenges, including providing direct wear information and wear assessment and prediction in a cost-effective and efficient manner. To tackle the challenges, this research aims to develop a set of advanced techniques for gear wear monitoring and prediction. The four objectives of the research and their corresponding methodologies and outcomes are summarised as follow.

(a) To develop a method to obtain direct and comprehensive wear information without disassembling the gearbox. This objective was realised by combining surface replication with image analysis, allowing easy acquisition of high-resolution mould images showing wear evolution on a tooth flank. (b) To investigate the relationship between the features of worn gear surfaces and those of wear debris. To further understand the role of wear debris analysis in wear assessment, a study on various features of macropits and wear particles in the same fatigue process was conducted and provided new insights into gear pitting and its monitoring. (c) To develop an automated system for gear wear assessment. Deep learning models were developed to identify wear mechanisms and severities using gear mould images and wear debris images. High classification accuracies were achieved, and comparisons between the two image sources were made. (d) To develop a gear wear prediction model using direct wear information. A deep generative model was developed and trained on time series of gear mould images. Tests showed that the model using the state-of-the-art AI technology can generate realistic and accurate predictions.

Overall, this research addressed the main limitations of existing methods and provided a direct and evidence-based tool for monitoring and predicting gear wear. Its specific contributions include a new moulding-imaging method for monitoring gear wear evolution, a detailed comparison between worn gear surfaces and wear debris in a wear process, and AI and image-based gear wear assessment and prediction models for the first time. The techniques could be performed during regular inspections of machines and used with online methods for increased robustness.