Development of four in-process surface recognition systems to predict surface roughness in end milling

Abstract

Surface roughness is one of the important factors in tribology and in evaluating the quality of machining operations. To realize full automation and achieve zero defect production, an effective technique is needed for on-line, real-time monitoring of surface roughness during machining. An in-process surface recognition system (ISRS), was developed for predicting real-time surface roughness, Ra, in end-milling operations. The parameters are spindle speed, feed rate, depth of cut, and the cutting, vibration between tool and workpiece. The cutting vibration is measured by an accelerometer and a proximity sensor. The analyses of the data and the ISRS building model are carried out using multiple regression analysis and the neural fuzzy system. In the statistical model, surface roughness is predicted by a multiple regression equation. For the fuzzy models, the fuzzy rules base is built by a one pass operation making use of successful training data. Surface roughness is predicted by a fuzzifier, a fuzzy inference engine, a fuzzy rules base, and a defuzzifier. Experimental results show that in the statistical model, feed rate is the most significant independent variable to predict the surface roughness, Ra. Vibration data contributes to increase R Square and improve prediction ability with a 91% accuracy rate. In the neural fuzzy model, the fuzzy rules base can be generated automatically within 4 seconds by the training data, and Ra can be predicted with a 96% accuracy rate. Based on the multiple regression equation or fuzzy rules base, the ISRS can predict surface roughness within 0.5 second during end-milling. Therefore, ISRS has potential for use in real-time operations.