Acoustic Emission based Tool Condition Monitoring System in Drilling

were filtered through digital band pass filter to avoid the affects of low frequency vibration. The filtered AE signals were analyzed in time-domain and time-frequency domain to extract features which are sensitive to drill wear. Root means square (RMS) value which is also a representative parameter for total AE energy of the signal has shown increasing trend with increasing drill wear. In time-frequency domain, wavelet packet transform has been applied to the AE signals, and RMS values of the wavelet coefficients in selected frequency bands are considered as the monitoring features which also show similar increasing trends. The relationships among the features and wear values are found to be non-linear. Artificial neural networks (ANN) are efficient tools to map such non-linear relationships if effectively trained through experimental data. An ANN model trained through back propagation learning algorithm has been developed here to correlate the extracted features to tool wear at different cutting conditions. Experimental results show that drill wear prediction of ANN model based on wavelet packet features is more accurate compared to that based on time domain features.