Abstract
To solve the problems of low accuracy and the significant impact of human factors when manually adjusting the oil feed rate (FR) and base number (BN) of the cylinder oil in a marine slow-speed diesel engine, a new method for optimising cylinder lubrication adjustments based on image deep learning is proposed. This method involves inspecting the wear and carbon deposition on the inner surface of the cylinder liner and the piston ring surface through the scavenging port. Images captured during an inspection are normalised, clustered and segmented, and features are extracted to create a comprehensive cylinder lubrication image database. Then, based on the HU invariant moment and HSV colour space, the images are fused and similarity matched, and image search software is developed. This approach addresses the randomness and significant errors that are often associated with the manual adjustment of the cylinder oil FR. The software is implemented to manage the cylinder lubrication on a very large crude carrier. The results demonstrate that the proposed optimisation method can achieve optimal control targets: the residual oil iron content is found to be below 25 ppm, the residual oil BN is above 30 mg KOH/g, and the cylinder liner wear rate is below 0.1 mm/1,000 h, thereby reducing the risk of excessive wear on the cylinder liner.