Research On Electrochemical Machining Prediction Model Based On Improved BP Neural Network

Electrolytic micro-machining is one of the key technologies in the field of j aero-engine parts manufacturing.It has the advantages of high production efficiency,good surface quality,no tool loss,and is suitable for processing high-hardness heat-resistant alloys.In order to solve the problems of many times of actual processing,high production cost,difficulty in selecting processing parameters,etc.,based on the theory of artificial neural network,an improved momentum-adaptive learning BP algorithm is used to construct BP neural network prediction model.According to the actual electrolytic processing,the influence of electrolytic processing parameters on the test results is analyzed.The basic BP neural network algorithm is improved,and the BP neural network prediction model based on momentum-adaptive learning is established.The prediction accuracy of electrolytic machining test results is high.It lays a certain theoretical foundation for deeper research,experimentation and argumentation of electrolytic processing prediction models.The main research work of this thesis is as follows:Firstly,the theoretical basis of electrolytic machining is studied and the parameters of the prediction model are selected.Under the condition of specific constant temperature electrolyte,the current value,electrolytic voltage,pulse power frequency,pulse power supply duty ratio,initial machining gap and electrode feed speed of electrolytic processing have great influence on the test results,and the above parameters are used as the input layer of the prediction model,and the processing time and the upper and lower apertures of the microholes are used as the output layer of the prediction model.Secondly,the theoretical basis of BP neural network is studied and the basic BP algorithm is improved.Aiming at the shortcomings of the basic BP algorithm,such as slow convergence speed and easy to fall into local minimum,many improved schemes are given.After screening and comparison,gain the improved momentum-adaptive learning BP algorithm which the network learning efficiency is high,the convergence speed is fast,the training times are small,and it is easy to reach the global minimum.Finally,50 sets of orthogonal experiments of electrolytic micro-machining are designed.The training set and test set of neural network prediction model are obtained through experimental data.The initial parameters of the neural network and the improvement scheme of the algorithm were determined.The MATLAB programming software is used to establish the improved BP neural network prediction model.Estimation the established prediction model by contrast method and cross-validation method,and the BP neural network prediction model is verified by electrolytic processing test.The prediction results are basically consistent with the actual processing results,and the validity of the momentum-adaptive learning BP neural network electrolytic machining prediction model is verified.