Feed Rate Optimization For Three-axis Rough Milling Based On Spindle Power Model Of Machining Process

In the milling process,process personnel often select a relatively conservative feed rate when planning G codes.The strategy of selecting feed rate based on the experience of process personnel or process manuals can guarantee the processing quality and efficiency,but it cannot exert the best performance of the process system.In the current three-axis milling roughing feed rate optimization research and implementation,there are still a problem that the process data generated from the machining process is not effectively applied to modeling and process optimization.The data of the processing process is a large data volume,derived in daily processing scenarios and related to process system characteristics.The daily processing scenarios also put forward higher requirements for the difficulty and cost of implementation of the method.In view of the above problems,this paper proposes an optimization method for the feed rate of three-axis milling roughing based on the spindle power model of the machining process.This method uses the spindle power collected in the CNC system as the model output to reduce the cost and difficulty of data collection.The tool-work piece-cutting model is used to extract process parameters and a processing data fusion strategy based on instruction domain analysis is proposed to programmatically generate the data set required for modeling and prediction.To model the processing process of a specific process system with the neural network suitable for big data modeling.Finally optimize the feed rate of the processing tasks in the specific process system based on the trained spindle power model.The specific work is as follows:1.Generated the data set required for modeling and prediction based on the big data of the processing process.Finished the calculation of the process parameters of the tool point with the tool-work piece-cutting model which constructed by Z-map and scanning envelope method.The calculation accuracy is that when the tool point spacing is 1mm and the Z-mapgrid spacing is 0.5mm,the error is within 3%.A search method by the relationship between the different surfaces was proposed to reduce the single-point calculation time complexity from O(N)to O(1)for the modeling of irregular blanks.Further,a data fusion method was proposed based on the method of instruction domain analysis,which realized the alignment and data fusion of the spindle power and the process parameters in the machining process and generated the corresponding data set.2.The model of spindle power neural network based on the specific process system by the processing process data was trained.The model used the training set and the test set by the method proposed in 1,and used the feed rate,spindle speed,cut width and cut depth as inputs to establish the spindle power neural network prediction model based on a specific process system.Different from the feed rate,spindle speed,cut width,and cut depth,other factors that have an impact on the spindle power became the characteristic of the model.Aiming at the problem of long training time caused by excessive data volume,the application of small batch stochastic gradient descent and the decay strategy of learning rate shortened the training time.Through experimental verification,the average error of the model prediction was 4.91%.3.Based on the improved MOEA / D,three-axis milling roughing feed rate optimization was finished.The process parameter extraction algorithm in 1 and the spindle power prediction model of the specific system trained in 2 were used to optimize the processing tasks in the specific process system.The goals of the optimization included processing efficiency and load balancing.The classic MOEA/D was chosen as the framework of optimization.The parent selection strategy,the Chebyshev aggregation strategy,the neighborhood update strategy and the neighborhood number determination strategy were improved to improve the applicability and optimization effect of the algorithm.It was verified by experiments that the optimization efficiency and load balancing can be improved.The above-mentioned data generation method makes the acquisition of input data required for modeling and prediction sufficiently convenient.The use of spindle power instead of milling force as the output of the model reduces the cost and difficulty of response data collection.The application of BP neural network is also suitable for the modeling of large amounts of data in the processing Scenarios.In addition,the modeling and optimization for specific process systems fully consider the impact of process systems on machining process modeling and optimization.In summary,according to the characteristics of processing data,the whole process of applying processing data to modeling and optimization is realized.