Modeling tool wear in end-milling using enhanced GMDH learning networks

This paper presents an enhanced approach to predictive modeling for determining tool-wear in end-milling operations based on enhanced-group method of data handling (e-GMDH). Using milling input parameters (speed, feed, and depth-of-cut) and response (tool wear), the data for the model is partitioned into training and testing datasets, and the training dataset is used to realize a predictive model that is a function of the input parameters and the coefficients determined. In our approach, we first present a methodology for modeling, and then develop predictive model(s) of the problem being solved in the form of second-order equations based on the input data and coefficients realized. This approach leads to some generalization because it becomes possible to predict not only the test data obtained during experimentation, but other test data outside the experimental results can also be used. Moreover, this approach makes it easy to present the realized solution in a form that can be further optimized for the input parameters using some optimization techniques. The results realized using our e-GMDH method are promising, and the comparative study presented shows that the e-GMDH outperforms polynomial neural network (PNN); moreover, it is more flexible than the conventional GMDH, which tends to produce nonlinear solutions even for simple problems. In the investigation, the extended particle swarm optimization (PSO) technique was applied to obtain the optimal parameters. Consequently, the modeling approach is extremely useful in realizing a computer-aided process-planning system in an advanced manufacturing environment.