Research On The Prediction Method Of Grinding Surface Roughness Considering Time-varying Characteristics

As a typical hard and brittle material,silicon carbide(Si C)ceramic is widely used in aerospace,electronic communication and other fields due to its excellent physical properties such as high hardness and high wear resistance.However,the Si C material must be processed to become a part with a certain shape and dimensional accuracy in order to exert its excellent physical properties.In recent years,the continuous development of the ceramic industry has made people have stricter requirements for the surface processing quality of Si C ceramics.The low processing efficiency and the difficulty of measuring the surface quality have become the primary factors restricting the further promotion and application of Si C ceramics.With the promotion of the Si C in various fields,real-time prediction of the surface quality during the grinding process of Si C,and the development of a set of practical prediction methods attach great importance to improving the surface quality and promoting the further application of Si C.Aiming at the disadvantages of the current surface roughness prediction models,this paper proposes a Si C surface roughness prediction method that takes into account the time-varying characteristics of the grinding process.On the basis of the force signal and vibration signal collected during the grinding process,the real-time and advanced prediction of surface roughness is achieved with the help of traditional machine learning methods-BP network and LSTM network.The main contents of this paper are as follows:(1)Firstly,the significance of considering the time-varying characteristics of the grinding process is proposed,and then researched the influence of the state of the grinding wheel at different stages in the continuous machining process on the surface roughness of the workpiece.Finally,the grinding force and vibration signal are used to predict grinding surface roughness,and the experiment platform for this is developed.(2)Developing the scheme of signal fusion.In this paper,considering the timevarying characteristics of the grinding process,force signals and vibration signals are selected as the monitoring signals of the grinding surface roughness prediction model.Then a set of 186 features are extracted in the time domain,frequency domain,and time-frequency domain of the signal and those factors that fluctuate greatly with the grinding process are eliminated by using the monotonicity analysis.Finally,five features that can reflect the grinding process are obtained based on the principal component analysis method.(3)On the basis of the BP neural network,a real-time prediction model of grinding surface roughness is proposed.The prediction performance of the network when a single force signal,a vibration signal and the fused signal are used as input signals are researched,and the mean square error MSE values are 0.416,0.8 and 0.0021 respectively.The results can not only verify the effectiveness of the signal fusion scheme in this paper,but ashow that the BP network model proposed in this paper can achieve high-precision real-time prediction of grinding surface roughness.(4)Based on the long short-term memory network(LSTM),an advanced prediction model of grinding surface roughness is proposed.The research results show that the fusion signal can be used as input to achieve one-step advance and multi-step high-precision prediction of the grinding surface roughness.In addition,considering the difficulty of data acquisition and the efficiency of model training,the relationship between the amount of input model data and prediction accuracy are explored.The research results not only greatly reduce the difficulty of data acquisition and improve the training efficiency of the model,but the critical points of prediction accuracy shown in the results are also instructive for the identification of grinding wheel and machine tool status.