Study On Quantitative Analysis Method For Qualitative Ferrography Of Equipment Wear Condition

Wear is the main failure mode of large-scale mechanical equipment.How to accurately obtain the wear condition information of the equipment,determine its wear degree,position and cause,find and eliminate it timely in the early stage of the fault,is of great significance to realize the predictive maintenance of large equipment and ensure its reliable operation.Ferrography analysis is an important technical means in equipment wear condition monitoring,but there are still some problems,such as the majority of qualitative conclusions of diagnosis results,strong dependence on the professional level of operators,and low level of intelligence.For this reason,this thesis explores the quantitative analysis method for qualitative ferrography of equipment wear condition,in order to improve the objectivity and intelligence level of equipment wear condition discrimination.Based on the principle of magnetic balance,a new type of wear particle concentration detection device is designed,its principle is analyzed and the relevant mathematical model is established.Based on the COMSOL finite element magnetic field simulation technology,the structural parameters are simulated and optimized,and the main signal processing and control circuits are designed.The linearity and reproducibility of the wear particle concentration detection device are tested by the oil samples collected on site.the results show that it can effectively detect the ferromagnetic wear particle content in the lubricating oil and analyze the trend of the wear particle concentration of the equipment.it can be used as a pre-diagnosis instrument for precision ferrography analysis.Explore the deposition law of wear particles in the ferrogram making process of the rotary ferrography analysis system,study the minimum view to distinguish the wear condition of the equipment,and plan the acquisition area scheme of the ferrographic image.Combined with artificial intelligence and image processing technology,a model named WP-Mask R-CNN is constructed which is suitable for instance segmentation of ferrographic image,and the wear particle image data set is made to train and test the model.Transfer learning and fine-tuning techniques are used to train the model,and the test set ferrographic image is used to verify the performance of the model.The results show that the model has a good effect on instance segmentation of ferrographic image.The quantitative analysis method for qualitative ferrography is explored.Based on the concept of parameters describing the geometric characteristics of individual wear particles in ferrographic images,combined with the wear debris group theory,the dimensionless characteristic parameters of wear particles are constructed.This thesis discusses the wear law and mechanism of mechanical equipment in the whole life cycle,analyzes the relationship among equipment wear failure mode,wear mechanism and wear particle characteristics,and constructs a fuzzy neural network model to distinguish the wear condition of equipment.Combined with the analysis of wear particle concentration,the model is trained and verified by oil samples collected on site.the results show that the proposed quantitative analysis method for qualitative ferrography can effectively improve the objectivity of ferrography diagnosis.Based on B/S architecture,The system software for quantitative analysis method for qualitative ferrography of equipment wear condition is designed and developed.Combined with the work content and flow of wear condition monitoring of large mechanical equipment,the front-end interactive interface and back-end database of the software are designed to realize the efficient management of monitoring equipment information,oil sample analysis data,ferrographic image,wear particle concentration trend analysis,ferrography precision diagnosis results and other data information.The networking,automation and intelligence level of ferrography analysis is improved.This thesis has 65 figures,14 tables and 99 references.