Abstract: Fuzzy system is adopted to model milling force on the basis of the characteristics of milling force and research purpose. Milling test of titanium alloy is designed and carried out on a numerical control milling machine, and milling force is measured by the dynamometer. From the test, training data and test data are obtained. After analyzing the advantages and disadvantages of basic particle swarm optimization (PSO) algorithm, gradient descent algorithm is incorporated into basic PSO algorithm to constitute an improved PSO algorithm. Using training data, fuzzy system is trained by gradient descent algorithm, basic PSO algorithm and improved PSO algorithm respectively. Results show that the improved PSO algorithm has better convergence than the other two algorithms. Two kinds of experience formula of milling force are obtained through regression analysis, regression functions are index form and linear full factor polynomial form. The models obtained by each method are tested by testing data respectively, the prediction effect of the fuzzy system trained by improved PSO algorithm is better than other methods. The prediction results show that the fuzzy system trained by improved PSO algorithm is reliable to model milling force.

Key words: Fuzzy system, Milling force, Particle swarm algorithm, Regression analysis