Research on image processing-based oil particle contaminant detection methods and devices

Abstract

With the continuous advancement of automation, regular condition monitoring and fault diagnosis are necessary to ensure the long-term operational stability of mechanical equipment. Lubricating oil, often referred to as the “lifeblood” of mechanical equipment, serves various functions such as energy transfer, anti-wear, system lubrication, corrosion prevention, rust prevention, and cooling. Therefore, detecting contaminants in the lubricating oil of mechanical equipment, especially particulate contaminants, is a prerequisite for ensuring the normal operation of the equipment. This paper designs and constructs an oil particle contaminant target detection algorithm based on Faster RCNN-Contrastive Language-Image Pretraining (CLIP), which combines a deep learning model with traditional image processing for extracting information about oil particle contaminants. This algorithm can extract information on oil particle pollutants’ location, type, quantity, size, and shape. On a smaller dataset, the MAP score of Faster RCNN-CLIP is 68.43%, the F1 score is 62.35%, while the corresponding metrics of Faster RCNN are 24.57% and 20.23%, respectively, and YOLOv5 are 40.79% and 31.22%. These results indicate that Faster RCNN-CLIP is a model more suitable for oil particle contaminant detection, especially under limited data resources.An oil particle contaminant detection device was also developed to break free from the laboratory constraints. This device integrates the oil above particle contaminant extraction model and mainly consists of an oil delivery unit, collection unit, control unit, display unit, and backend. The device is portable and efficient in detection and has IoT capabilities, enabling real-time remote detection of oil particle contaminants. This new method brings a more flexible and efficient solution to real-time fault diagnosis for oil condition monitoring.