Research On Intelligent Monitoring Method For Vibration States And Tool Wear In Machining

Intelligent manufacturing technique can efficiently help to improve the quality and processing efficiency in machining process.Intelligent manufacturing is the newest development of the machining technology and also the development direction pointed out by the “Made in China 2025” Program.Intelligent monitoring technology is the important to improve the machining automation.Tool vibration and wear state both have significant influence on the reliability and safety of the machining process and also on the quality of the products.The monitoring of the tool vibration and wear is one of the most popular research topics in machining state monitoring.Aiming at solving these two monitoring tasks,this paper works to apply the intelligent technology to the machining state monitoring to improve the automation degree of the machining process.The main research work and results are listed as follow:(1)Section 2 proposes a new intelligent monitoring method for cutting vibration state.Firstly,based on the relationship between the acceleration signal and the chatter phenomenon in the time-frequency domain,an new theory named the energy aggregation process is proposed to explain the chatter process.Then based on this theory,an revised Hilbert-Huang transform method is proposed to characterize the cutting vibration signal.Thirdly,using the extracted feature space,the Gaussian mixture model is adopted for unsupervised clustering modeling.Finally,based on the clustering boundary,an automatic threshold calculation method is proposed.Experiments prove that compared with the wavelet method,the proposed method is faster by 74 ms in advance.The threshold can be adaptively adjusted.The method can effectively protect the workpiece from vibration damage.Complexity analysis proves this method able to realize real-time monitoring.(2)Section 3 proposes an automatic feature construction method for cutting signal.Aiming at solving the strong experience-dependence problems in vibration state monitoring,an automatic feature construction method based on feature learning theory is proposed.Using the deep network model with special learning algorithm,the proposed method can automatically mine the intrinsic relationship between the vibration signal and the vibration state to realize automatic analysis of the cutting signal.The extracted features are obtained under strict mathematical derivation and are the best representation in the mathematical sense.Based on the automaticly extracted features,a voting strategy is proposed to stabilize the monitoring results,avoiding the oscillation caused by the fluctuation of the signal uncertainty.Experiments show that the proposed detection method is faster and more accurate.(3)Section 4 designs a new tool wear detection device and algorithm.Because the current measuring device is complex,the tool needs to be disassembled from the machine and tested on the dedicated device.This process have negative effect on the process quality.This paper designs a compact and flexible high-quality optical measurement platform to realize in-suit tool wear measurement.Telecentric lens is adopted to improve the imaging quality of wear zone.Based on the high qualified tool wear image,an automatic calculation method of the wear value is designed to realize the automatic measurement of the tool flank wear.Titanium alloy turning tool wear test verifies the effectiveness of the measurement platform and algorithm.(4)Section 5 proposes a new co-supervised ensemble learning method for on-line tool wear monitoring.Aiming at improving the accuracy and adaptability of the current monitoring method,a new hybrid tracking method is proposed to couple the direct measurement method and the signal based indirect prediction method to track the transformation of the tool wear.A compromise is made between the measurement frequency and the monitoring accuracy.The direct measurement component provide accurate tool wear state information.The indirect prediction component controls the frequency of the direct measurement.A novel co-validated ensemble learning method is proposed,which makes the indirect component able to examine whether its prediction is effective.When the prediction is found to fail,the direct measurement will be called to check the tool wear state,and the indirect prediction model will get calibrated using the measured tool wear sample.As a local monitoring model,the method does not have the traditional training phase and testing phase.The performance can be controlled by model parameters rather than the prepared wear data samples.Experiments prove that the proposed method can achieve good monitoring performance(R = 0.9920,RMSE = 3.6413um)with very few direct measurement(22.2%).In conclusion,this paper designs a new algorithm based on the machine learning theory for the tool vibration monitoring problem in the cutting process.To solve the problem of tool wear monitoring,a new measuring device and a new monitoring algorithm are designed.This paper realizes the intelligent real-time online monitoring of the tool vibration and wear.The detection accuracy and speed are both better than the existing method.Experiments prove the method can effectively protect the workpiece and improve product quality.