Automatic supervision of blanking tool wear using pattern recognition analysis

Abstract Most of the blanking operations are now done on automatic high speed presses. The faster operation, closer dimensional and form tolerances, higher precision demand automatic supervision of blanking tool conditions in order to avoid producing a large volume of defective parts unnoticed. In this paper, an architecture of an automatic supervisory system for monitoring blanking punch wear under various die wear conditions is proposed. The system employs an autoregressive (AR) time-series model to predict the on-line captured peak blanking force. The AR model coefficients are updated by a modified least mean square (MLMS) adaptive filter so as to minimize the prediction error. An optimum number of AR model coefficients are selected to form the pattern vector which is classified by a least mean squared error (LMSE) classifier. Classification of the punch wear states is accomplished by a linear discriminant function (LDF). The performance of the system was evaluated through a series of blanking experiments. Experimental results indicated a high success rate for recognizing blanking punch wear under various die wear conditions.