Predicting Surface Roughness in Grinding Using Neural Networks

Abstract

High rates of manufactured items have been machined by grinding at some stage of their production process, or have been processed by machines whose precision is a direct result of abrasive operations. However, even being the grinding process the most used in industry for obtaining high level of surface quality, it remains as one of the most difficult and least understood processes (Wang et al., 2005). That maybe has origin in the mistaken faith the process is extremely complex to be understood due to the large number of cutting edges and irregular geometry, high cutting speed, and very small depth of cut which varies from grain to grain. In addition, according to (Haussi & Diniz, 2003), grinding is the process indicated when the workpiece demands good surface, dimensional and geometrical quality. Thus, the grinding process is usually one of the last steps in the machining operations chain. When the workpiece reaches this point, it has high aggregated value, which makes a possible rejection very expensive. Monitoring of machining processes is mandatory for their optimization and control. Acoustic emission (AE) has become an increasingly popular monitoring technique. The sensors are inexpensive, easy to mount, and analog signal processing is comparatively simple, but good techniques for extracting reliable process information from the signals are still lacking (Hundt et al., 1997; Aguiar et al., 2002). Electrical power signals have also been largely used in grinding researches. The signal can be monitored either by the electric current of the electric motor or by the product between voltage and current signals, which gives the electrical power consumed by the electric motor. Thus, an estimate of the cutting force can be easily obtained if a model of the electric motor is available (Aguiar et al., 2002). Some researchers have shown the acoustic emission and the cutting power signals combined can provide significant results for monitoring the grinding process phenomena (Aguiar et al., 2002; Dotto et al., 2006; Kwak & Ha, 2004; Aguiar et al., 2006). Neural network has attracted a special interest in grinding research owing to its functions of learning, interpolation, pattern recognition, and pattern classification. Various examples of applications into the production engineering field have been reported (Wang et al., 2005; Dotto et al., 2006; Kwak & Ha, 2004; Aguiar et al., 2006; Wang et al., 2001). According to (Wang et al., 2005), surface roughness is one of the most important factors in assessing and determining the quality of a part. In practical, predicting and controlling the O pe n A cc es s D at ab as e w w w .ite ch on lin e. co m