Research On Mapping Relationship Between Eigenvalues And VB Values Of Tool Wear Monitoring System

Machining is an important part of the manufacturing industry, in which the tool is thedirect performer. Tool wear has not only a direct impact on the demand plans of tool, costaccounting but also closely related to the processing quality and accuracy of the workpiece.Therefore, the research on prediction technology of tool wear is in a very importantsignificance. In practical monitoring process to tool wear, only when getting multipleinformation about the object monitored, it can have more accurate predictions. It uses manysensors to extract a plurality of monitoring signal characteristics in order to get accuratepredictions. It combines characteristics on tool wear to establish a tool wear prediction modelbased on independent component analysis support vector regression theory.This paper selects acoustic emission signals and current signals as monitor signals, twotypes of which are less affected by cutting conditions, so it has many features like highmonitoring accuracy, strong anti-interference ability, high sensitivity, real-time onlinemonitoring and it’s easy to use, et al. Acoustic emission signals and current signals are closelyrelated to tool wear condition, which are used for feature extraction from these two signals. Itcan obtain the feature vectors reflecting the value of different tool VB. It collects acousticemission signals and current signals under the different values of tool VB through turningexperiments. Analysis of these signals indicate that through analyzing acoustic emissionsignal and extracting features by combining singular value decomposition with wavelet packettransform and analyzing current signal and extracting features by empirical modedecomposition, coupled with the parameters of cutting (spindle speed, cutting depth, feed rate)as secondary feature, they are as the initial feature vector of tool wear state commonly.Independent component analysis has been developed in recent years, which originatesfrom mixed-signal separation problem solving. Independent component analysis selects thedirection of independent data in the sample space for data dimensionality reduction. Theoriginal feature vectors reflecting tool wear are analyzed through independent componentanalysis to get more accurate feature vectors reflecting tool wear. The data analysis shows that the first or second-order correlation between each component of the signal, but also has theability to explore and remove high-order-related information. So the output component isstatistically independent from each other, as well as non-Gaussian distribution. Feature vectorfrom independent component analysis can act as input of SVR. Because that support vectorregression machine adopts structural risk minimization principle, which can establish afunction model based on a small sample, and it also can deal with trained samples which havemore fitting and compleity. Therefore, the paper uses support vector regression model in theparticle swarm optimization to predict the amount of tool wear. Support vector regressionmachine uses structural risk minimization principle to establish a function model based on asmall sample, considering the training sample’ fit and complexity. So this paper uses supportvector regression model in the particle swarm optimization to predict the amount of tool wear.The results show that support vector regression optimized by the particle swarm is betterthan the neural network in terms of precision and speed. System based on support vectorregression has good stability,fast recognition speed, which can make more accuratepredictions on tool wear.