In order to improve the surface machining quality of slow tool servo (STS) turning in complex surfaces, the optimal method of tool path generation (TPG) was studied. Taking into consideration the problem of large discrete errors and interpolation errors in TPG, equal height discretization method and interpolation algorithm for non-uniform nodes were proposd and the acceleration continuous condition was introduced. Simulation results showed that equal height discretization method could reduce the discrete error by more than 70%. The interpolation errors could be reduced to two orders of magnitude by transforming segment cubic Hermite interpolation into segment cubic spline interpolation. Finally, the processing experiments were performed. The results showed that the form error PV value for the workpiece of the toric surface obtained by equal height discretization method and non-uniformity processing and segment cubic spline interpolation reached 0.002mm. The PV value of the sinusoidal array surface was about 0.009mm, and its surface roughness value was 0.064μm. The results proved this method can effectively reduce the discrete errors and interpolation errors, as well as improve the surface machining quality.

Surface roughness is an important index to evaluate the quality of a machined surface. In order to accurately predict the surface roughness for slow tool servo (STS) turning, taking toric surface as an example, response surface methodology (RSM) was used to perform the process test. The second-order response surface prediction model was established and the variance analysis and reliability test were carried out. The results showed that the average prediction error was 7.6%. In order to obtain the best process parameters, standard particle swarm optimization (PSO) was used. The results showed that the global optimization ability of standard PSO was poor. In order to solve the problem, compression factor was introduced and particle swarm optimization with compression factor (WCF-PSO) was constructed, which enhanced the convergence of PSO effectively. WCF-PSO was used to optimize the process parameters and the results obtained were R t =0.87mm, a f =0.01mm/r, a p =0.05mm, Δθ=8.70°, with a corresponding surface roughness of Ra=0.0486μm. The results of the verification test showed that the actual value was Ra=0.0520μm, and the error was only 7.0%, indicating that WCF-PSO had a better optimization effect.

To investigate the effect of slow tool servo turning process parameters on surface roughness, we established a high precision surface roughness prediction model. A guide to the selection of turning process parameters was compiled, and a turning test was conducted based on a response surface method (RSM) central composite design. ANOVA explores the influence law of process parameters on surface roughness. A RSM BP neural network model, and MEA-BP surface roughness model were established and the prediction performance of the three models was evaluated. The results show that the significant process parameters affecting surface roughness are tool radius, discrete angle, feed rate, and cutting depth in descending order; and the prediction errors of RSM, BP, and MEA-BP are 11.41%, 19.67%, and 5.54%. This suggests that the MEA-BP model has the highest prediction accuracy with the same test data, RSM is second, whilst the single BP model struggles to capture multiple data characteristics and its prediction accuracy is poor. In addition, MEA can effectively solve the BP model falling into local optimum and improve the model prediction accuracy. Process parameters Surface roughnessResponse surface method BP neural network Mind evolutionary algorithm