MACHINE TOOL FAILURE ANALYSIS USING IMAGE AND SIGNAL PROCESSING TECHNIQUES

Abstract

Effective Tool condition monitoring provides a powerful way to lower cost and to improve product quality in manufacturing systems. Tool condition refers to the distinction between a sharp, semi-dull, and dull tool. Modern tool condition monitoring involves estimating and detecting the condition of cutting tools by processing sensory information using signal analysis techniques. This thesis presents work that investigates the correlation between tool wear and quantities characterising texture of machined surfaces and sound generated during the machining process. Machined surfaces are imaged and processed using image analysis techniques including image segmentation, edge detection and run length statistical techniques. Computing the area under the peaks of column projections and calculating the run length statistical parameters are effective in directly monitoring the deterioration of machined surfaces and therefore indirectly monitoring the condition of the cutting tool. Sound produced during the machining process is recorded and analysed using power spectral density and wavelet decomposition methods. Sound energy and energy of the wavelet decomposition coefficients are good features to distinguish tool condition when monitoring the tool condition by processing sound generated during machining process. The simulation results also show that, tool condition monitoring can be effectively achieved even in noisy environments by canceling the noise using the wavelet shrinkage methods discussed in this thesis. Features of the machined surfaces and sound are combined to estimate the condition of the cutting tool using the radial basis function neural network and sensor fusion techniques in this thesis. The results clearly indicate that tool condition monitoring (the ability to distinguish between sharp, semi-dull, or dull tools) can be successfully achieved by analysing machined surfaces and sound produced during the machining process using image processing, signal processing and sensor fusion techniques. This thesis therefore paves the way towards real time, low cost and reliable cutting tool failure diagnosis.