Surface roughness prediction in turning using three artificial intelligence techniques; A comparative study

In today's high competitive industry, it is imperative to achieve high level production rates of high quality products. Manufacturers with efficient process predictabilities are better equipped to gain a competitive edge in the market. Surface finish is one of the most important measures for determining the quality of products in machining. This work presents a comparative study utilizing several Artificial intelligence (AI) techniques to predict surface roughness in turning. Extensive experimental work was conducted to obtain vibration signals during the turning process of steel rods at different speeds, feeds and depth of cut (DOC) conditions. The vibration signals are then analyzed using several spectral and statistical techniques to extract features which are then used as inputs to the AI algorithms. Surface roughness measures are divided into three classes; smooth, medium and rough surface finish. Three AI methods are compared for their effectiveness in predicting the surface roughness class from the vibration signals. The methods used are canonical genetic algorithm, particle swarm optimization (PSO) and differential evolution (DE) algorithms. Six different feature sets from vibration data were used and preliminary results show the effectiveness of the combination of particular feature sets and the differential evolution algorithm in predicting classes.