Tool wear monitoring in turning operations using ultrasonic waves and artificial neural networks

Ultrasound waves propagating at 10MHz nominal frequency were used to monitor the gradual wear of carbide inserts on-line during turning operations. A dual-mode ultrasound transducer was placed behind the cutting inserts in a modified tool holder assembly. Ultrasound waves were pulsed at a rate of 10KHz inside the cutting inserts towards the flank and nose surfaces of the tool in a pulse-echo mode. The reflected ultrasound waves, received by the same transducer, were digitized at 100MHz sampling rate and processed using discrete wavelet transform. The powers of coefficients, yielded from the frequency decomposition of the ultrasound signals, were used to search for an optimum artificial neural network architecture that correlates between the reflected ultrasound wave processing parameters and the true state of wear on the cutting inserts. Machine vision was used to measure the gradual wear dimensions at the flank face of the tools during interrupted intervals of cutting.; An analytical model that describes the theoretical behavior of ultrasound waves inside the cutting inserts and the effect of tool wear on the reflected waves was derived. The theoretical relation between the gradual wear level and the change in the ultrasound waves was used to verify the soundness of the empirical results.; The results indicate that the powers of coefficients extracted from the fourth level of decomposition of the ultrasound signal using discrete wavelet transforms yield optimum correlation to the true state of wear on the cutting tools using a three-layer Multi-Layer Perceptron artificial neural network. The best correlation coefficient was found to be 95.9% with an estimated error less than 0.003 inches.