Abstract

Today, machining processes are among the most common and important industrial operations. Considering the wide applications of the machining processes, the importance
of having an accurate and reliable monitoring system is clear. Therefore, several researches and studies have been performed to respond to this demand. Despite considerable efforts in this field, the achievement is not satisfactory and still the request for having an accurate, reliable,
automatic, inexpensive and robust monitoring algorithm is remained without a good answer. This study tries to design monitoring algorithms with aforementioned specifications.
The monitoring algorithms predict the amount of tool flank wear in the face milling process. They are designed based on the pattern recognition concept. The algorithms analyze
signals in four steps: preprocessing, feature extraction, feature selection and classification. Descriptors, wavelet transform and S-Transform are applied for feature extraction. Principal component analysis (PCA) and independent component analysis (ICA) perform the feature selection step. Neural network (NN), which is an artificial intelligence method, classifies the data and makes the algorithms intelligent. By combining these methods, five
intelligent algorithms are developed. The results show that the most accurate algorithm between these five algorithms is the combination of S-Transform, ICA and NN. Results also
confirm the good performance of S-Transform for feature extraction comparing with the wavelet transform or descriptors. Applying the best designed algorithm, the effect of sensor fusion on the accuracy of algorithm and the ability of monitoring algorithm for working in different operating conditions are studied as well. It is also shown that the accuracy of the best designed algorithm for indicating the tool status, sharp or dull, is better than the accuracy of predicting the value of the tool wear. Applying S-Transform for machining monitoring and designing five intelligent, practical, inexpensive and accurate algorithms for tool wear prediction can be considered as the key outcomes of this thesis.