Research On Cutting State Monitoring Technology Based On Machine Learning

In 2015,the State Council officially released the "Made in China 2025 Plan",proposing to make a transition from production manufacturing to service-oriented manufacturing,and finally realize intelligent manufacturing.Intelligent cutting database is an important way to realize intelligent manufacturing in the field of numerical control machining.It has a great promotion to the network,intelligentization,integration and greening of production.Its intelligent function can not be realized without advanced technologies such as Internet of Things,sensors technology and data mining and other advanced technologies.Based on the problems of simple structure,single function and low level of intelligence in the current cutting database,combined with the problems existing in the tool condition monitoring,machine vision,multi-sensor fusion and data mining,this paper proposes the establishment of a tool failure discriminant model based on machine vision,tool wear and processing quality prediction model based on the multi-sensor fusion,and a slender shaft dimension error prediction model based on machine learning.The intelligent cutting database function module design and core algorithm development have been researched to achieve its online monitoring capabilities.The main research results are as follows:(1)Based on the problem of tool failure in the cutting process of difficult-to-machine materials,a failure-monitoring system of machine vision was set up,and the surface texture images were acquired.The histogram equalization was used to pretreat of the images after gray-scale processing,the gray level co-occurrence matrix was calculated and the related features were extracted.Analyzed the feature quantity and proposed two discriminating conditions of tool breakage.(2)Based on the problems of tool wear and surface quality monitoring during the cutting process,a multi-sensor fusion monitoring system was set up to acquire the cutting force,vibration and surface texture images.The cutting signal was processed in time domain,frequency domain and wavelet analysis,and the principal components analysis method was used to reduce dimension.The genetic algorithm,the particle swarm optimization and the grid search algorithm were respectively used to optimize the support vector machine algorithm.The intelligent prediction models of the tool wear and surface quality based on support vector machine were established.(3)Based on the problem of tungsten alloy cutting in slender bar,the influencing factors of the dimension error of the slender shaft parts were studied,and the dimensional error finite element model based on the elastic deformation of the process system was established.The tool wear,chip shape,surface roughness and residual stress in tungsten alloy cutting were studied,and the tool and cutting parameters were recommended.Dimensional error model in tungsten alloy slender bar prediction has a serious limitation.The dimension error intelligent prediction model of the tungsten alloy slender bar is established based on machine learning,and the prediction accuracy is high.