Abstract:
Intelligent machining centers have become important part of manufacturing systems because of
increased demand of the productivity. Tool Condition Monitoring is an integral part of these
systems. Acoustic emission from machining process is an important indicator of tool health.
Acoustic emission for a metal cutting process can be divided into two categories,structure-born
acoustic emission and airborne acoustic emission. Structure-borne acoustic emission needs high
processing power whereas, background noise is a great challenge in case of airborne acoustic
emission. Reducing the background noise may help in developing a low-cost system. Four
different machine algorithms, have been used as adaptive filters in order to reduce the
background noise. These algorithms include feedforward neural network trained with
Levenberg-Marquardt algorithm, self-organizing maps, K-mean clustering algorithm and Particle
Swarm Optimization (PSO). Acoustic signals from four different machines in background are
acquired and are introduced to a machining signal at different RPMs and feed-rates at a constant
depth of cut. The four machines are 3-axis milling machine, 4-axis mini-milling machine, a
variable speed DC motor and a grinding machine. These background noise signals are filtered
through the proposed algorithms. Backpropagation neural network shows the better performance
for the filtering while the other algorithms work only for dominant noise. The average accuracy
of the backpropagation neural network is found to be 75.82%. The filtered signal is reconstructed
using Auto Regressive Moving Average (ARMA) technique. An average increase of 71.3% in
SNR is found before and after signal reconstruction. ARMA shows a promising results for signal
reconstruction for machining process.
