An empirical multi-sensor estimation of tool wear

Abstract Automation of metal cutting machinery requires continuous estimation of tool wear. Variations in the type of machining process, materials or tools make a reliable estimation of the tool state by a single sensor signal difficult. A multi-sensor system has been implemented for cutting process monitoring in a lathe. Once tool life intervals are selected, a study of optimal descriptors capable of characterising sensor signals is carried out. Data dispersion inherent to a noisy signal suggests strict quantifier selection over a wide initial set. Pattern recognition procedures such as distance functions, neural networks and information entropy-based procedures offer empirical methods which deal with non-homogeneous data with length flexibility capabilities. An experimental example shows multiple parameter tool wear estimation in a multisensor environment. Good estimation of wear is obtained through the sensor system implanted in the machine.