Grind-Hardening Modeling with the Use of Neural Networks

Abstract

This paper describes the modeling procedure and results of a non-conventional process, called grind-hardening. The main idea of the grind-hardening process is that the heat dissipation in the cutting area is used for the heat treatment of the workpiece. Grind hardening is a complex manufacturing process governed by a multiplicity of parameters.

In order to satisfy the need for industrial exploitation of the process, it must first be thoroughly investigated and optimized. This goal can be achieved by efficient modeling. For this purpose, artificial intelligence methods were used, namely Neural Networks. This advanced simulation method is highly efficient in the case when relationships among parameters are non-linear, which is the case in grind-hardening.

The case studied in this paper is a double-face grind-hardening process. The part in question is a punched disk simultaneously ground and hardened on both sides. Quantitative and qualitative parameters are used to describe the process. The qualitative parameters are modeled using vector representation. Experimental data taken from this process are used to train the network.

After the network training stage has been completed, the network is then used to determine the impact of the process parameters on the working result, namely the surface hardness on both sides of the part. The network results, concerning the level of accuracy of its predictions for different combinations of process parameters, have been obtained and evaluated as satisfactory.