Research On Evaluation Method Of Data-driven Thermal Error Prediction Model For CNC Machine Tools

In the contemporary industry,high-end precision CNC machine tools have become the foundation,and the demand for improving the accuracy of machine tools continues to rise.A large number of studies have proved that thermal error is the key factor causing the decline of processing accuracy.The thermal error prediction model is usually used to improve the machining accuracy by inputting the compensation amount in the opposite direction to the thermal error.Due to the complicated working conditions of CNC machine tools and the constant changes of ambient temperature,the related data-driven modeling methods are difficult to establish accurate prediction models for various working conditions,and each model has its own applicable application scenarios and features.So the pros and cons of various thermal error prediction models can not be evaluated comprehensively.Therefore,a set of evaluation index system and evaluation method of thermal error prediction model is established,and the best model is selected,which has practical guiding significance for subsequent thermal error compensation.In this paper,the representative thermal error prediction model proposed in recent years is used as the evaluation object,and the key technologies of constructing evaluation index system and evaluation method are studied in depth.The main research contents are as follows:(1)Based on the characteristics of the thermal error prediction model,the characteristics of the quasi-static thermal error prediction model and the dynamic thermal error prediction model are analyzed.The modeling methods are compared.Combining with the principles that should be followed in the construction of evaluation index system,the evaluation indexes are extracted from the three aspects of goodness of fit,prediction accuracy and robustness,and the implementation methods and physical meanings of these evaluation indexes are studied.(2)In view of the disadvantage that the weights of indexes in the evaluation process are easy to appear impersonality and insensitive to single-index anomalies,a weighting method based on variable weight improved difference coefficient CRITIC is proposed,which can obtain more reasonable weight distribution of indexes.Analyze the limitations of the grey relational analysis method and the VIKOR method in the comprehensive evaluation,and propose to integrate the two to achieve complementary advantages.The G-VIKOR comprehensive evaluation method based on two-way KL distance adjustment is designed and implemented,and the models are sorted according to the comprehensive utility value,and the best model scheme is selected.(3)For the specific scenes where the internal and external influence factors of the machine tool change greatly,there are different requirements for the performance of the model.Therefore,the bias of the criterion layer is generated.he bias information of the criterion layer is rationalized and quantified by using intuitionistic normal fuzzy numbers.Considering that the traditional TODIM method does not take the reference points of the bias of the criterion layer into account,an improved TODIM evaluation square based on intuitionistic normal fuzzy numbers is proposed.According to the dominance degree of each scheme relative to the reference point,the most suitable model scheme for a particular scenario is selected.The feasibility and validity of the evaluation methods studied and implemented in this paper are verified by applying this two evaluation methods proposed in this paper to the real monitoring temperature field data set of ZK5540 A heavy-duty CNC machine tool.