The Research Of Tool Condition Monitoring Based On Ellipsoid ARTMAP And Mutilsensor Fusion

In modern manufacturing, the identification of machining condition is importantto reduce processing time and improve the product quality. Cutting tool is animportant component for machining system; it will inevitably become blunt becauseof wear along with the cutting process, particularly in the machining of somerelatively high hardness material. The change will directly affects the workpiecesurface finish and production efficiency, therefore the real-time detection of toolstatus is essencial for actual cutting process.This paper studies the application of Ellipsoid Adaptive Resonance theory (EAM)in the monitoring of tool condition, and achieves the condition monitoring of toolcondition based on the relationship established between force, vibration signalcharacters and tool status. Firstly, the principle of EAM and its modeling method aredescribed, and the influence of some parameters on the classification of EAM is alsoanalyzed. To further verify the effectiveness of EAM in the monitoring of tool status,the TC4titanium milling experiment is designed, and by collecting the force signaland vibration signal during the machining process, some features in time domain,frequency domain and time-frequency domain are extracted. Using the fast correlationbased filtering (FCBF) feature selection method to reduce the feature dimension, andthus the classification of tool status in the milling process of titanium is researched.Multi-sensor fusion technology, which combined with force and vibration in this test,can take advantage of the complementarity between multiple sensor information,thereby avoiding the limitations of using a single sensor and improving themonitoring reliability effectively. In this paper, the comparision between EAM andseveral common methods, Fuzzy Adaptive Resonance (FAM), K-means (KNN),Naive Bayes (NB), Classification and Regression tree (CART) have confirmed theeffectiveness and accuracy of EAM in the monitoring of tool wear status. Furthermore,the samples are selected to testify the incremental learning ability of EAM for thesamples with the same and different classes as the training samples. The result hasshown that EAM has good incremental learning ability.Experimental results and analysis have shown that EAM can be used in themonitoring of tool wear status with high classification accuracy, and EAM has the incremental learning ability, which has made EAM can be used in the online real-timemonitoring. As we all know, online monitoring is essential for the realization of theapplication of tool wear monitoring system in real industry.