Study On Intelligent Prediction Method Of Roll Grinding Surface Roughness Based On Polymorphic Signal Fusion

Roller is a direct-acting component that causes continuous plastic deformation of metal.It is widely used in the field of sheet rolling and supports the development of key industries such as aviation,aerospace,automobile,and military industries.Due to the effect of high temperature and high pressure,the surface of the roll is seriously worn and needs to be repaired by grinding regularly,and the roughness is the core index to measure the surface quality of the roll,which is closely related to the wear resistance,corrosion resistance,fatigue resistance and other properties of the roll,and directly affects the sheet quality.Traditional roll grinding needs to measure the surface roughness after processing.Since the roll is a large and long-shaft part,the measurement workload is large and the efficiency is insufficient.In response to this problem,this paper researches on the intelligent prediction method of roll grinding surface roughness based on multi-state signal fusion,and realizes intelligent and efficient prediction of roughness.The related work is of great significance to the online monitoring of roll grinding quality and the improvement of processing efficiency.The main contents of this article are as follows:Firstly,the main reasons that affect the roughness formation in the process of roll grinding are analyzed.On this basis,the current signal,vibration signal and acoustic emission signal are selected as the describing signals of surface roughness,and a complete signal acquisition system is built based on a roll grinder.The orthogonal test was further carried out to analyze the influence of different process parameters on the surface roughness.Based on the results,a mixed full factorial experiments combining two factors with four levels and two factors with three levels was designed.Each describing signal and the surface roughness at different positions of the roll after grinding,and preprocesses the signal and roughness.The wavelet packet technology is used to de-noise the original signal,and eigenvalue extraction in time domain and frequency domain is performed on the denosing signal,and the principal component analysis method is used to fuse the dimensionality of many features and constructing different input features.Surface roughness prediction is studied by using random forest algorithm and multilayer perceptron network in traditional machine learning algorithm,the results show that the traditional machine learning method can obtain a certain prediction effect(accuracy rate is about 78%),different signals have different capabilities to describe roughness,and the multi-signal fusion scheme has more advantages than the single-signal scheme.In order to further improve the prediction effect of roll surface roughness,further research based on deep learning will be carried out.In view of the original signal dimension is too large,grinding roughness formed in the process of temporal characteristics and different signal has the characteristics of different characterization ability,a convolutional long short-term memory network combined with self-attention mechanism for feature fusion is constructed.The neural network is used to predict the surface roughness of roll grinding.The results show that deep learning has obvious advantages over traditional machine learning,and the prediction accuracy exceeds 83%.Finally,in view of the problem of model performance degradation over a long time span,transfer learning is applied to improve the prediction performance of the model to a higher level under a small sample data set,which effectively expands the applicability of the proposed method for intelligent prediction of surface roughness.