engleski

Analytical Study of Different Approaches to Determine Optimal Cutting Force Model

Determination of optimal machining parameters is an engineering task with aim to reduce the production cost and achieve desired product quality. Such exercise can be tackled on many different ways. The goal of this work is to present some of the possible approaches and to benchmark them among each other. These principles are analyzed: response surface methodology (RSM), evolutionary algorithms (GA & GP), support vector regression (SVR) and artificial neural networks (ANN). All methods implement completely different data handling philosophies with the same goal, to build the model which is able to predict cutting force in satisfying manner. Those aspects are chosen to be evaluated and compared: average percentage deviation of all data, ability to find generalized model and minimize the risk of over fitting and at least the runtime of each single model determination. Average percentage deviation is one of the best indicators of the quality of model. The ability to find generalized model is good indicator of the flexibility of model, and shows how model deals with unknown data. The runtime is important in a real time environment or in scenarios where conditions change frequently. Cutting force data used in this benchmark comes from experimental research of longitudinal turning process.

Cutting Force; Modelling; Response Surface Methodology; Genetic algorithms; Genetic Programming; Support Vector Regression; Artificial Neural Networks

nije evidentirano