This study investigated a methodology for an on-line condition monitoring of tool wear during milling. Based on a comprehensive literature survey, a novel method called 'feature filtered fuzzy clustering' is proposed and developed. Different from the existing fuzzy clustering techniques used for the condition monitoring of machine faults and tool wear, this method can make a realistic identification and classification of tool wear under various cutting conditions by the classification of the features on which the effects of cutting conditions were removed. To realise this method, the relationships between the cutting conditions and the features under three pre-defined wear states corresponding to three clustering centres (initial, normal, severe) were established by experiments, which were undertaken under the conditions defined by experimental design, and non-linear multiple regression analysis. During experiments, a sensor fusion strategy was applied in order to get information from different aspects of the milling process. A mathematical model for fuzzy clustering based on the conception of distance has been verified by experiments using inserts both with artificially created flank wear and accelerated natural flank wear during milling in a CNC tool. In order to obtain appropriate features, the effectiveness of applying different physical parameters, i.e. cutting forces, power consumption of the spindle motor, AE RMS and AE pseudo ring-down count, for monitoring of tool wear has been investigated. Employing data fusion approach and its effect on the classification results has been investigated. Also the feasibility of applying Fourier and Walsh transforms to cutting force signals during monitoring of tool wear has been investigated.
|Date of Award
- Nottingham Trent University