Rmili W., Serra R., Ouahabi A., Gontier C.

The main objective of this paper is to develop a signal processing technique using vibratory signals in order to provide a robust tool wear monitoring system able to increase machining performance. During machining, frictions generated at high temperature between the cutting tool and the work-piece are the principal source of tool damage. This phenomenon decreases the quality of machining. For this reason, it is necessary to develop a monitoring system in order to detect and evaluate tool wear during the cutting process. Our work was based on the analysis of the vibratory signatures, generated in turning process. This analysis permuted to reveal the influence of a cutting tool wear on the produced vibrations. In the inversed case, we looked for determining which parameters could be extracted from the vibratory signals to describe the machining state. These parameters formed the input vectors of an on-line monitoring system, based on the neural networks. Cutting tests were performed in turning operation under specified cutting conditions without applying cutting fluid and a coated tungsten carbide insert was used. Sensor data corresponding to different tool wear levels (from the first use up to destruction) were acquired using piezoelectric accelerometers. Classical processing schemes in the time and frequency domain were first applied in order to obtain general signal characteristics. Furthermore, the acquired signals were studied using a time-frequency analysis. Classification of the tool wear data was obtained by neural networks structures which presented an efficient tool less affected by the noise and more credible to develop a model process from the actual system parameters. Keywords: Tool wear, turning, vibration, time-frequency, neural networks.