| In the pelletizing production line of mines,the fan system is one of the most important large machinery,which is the main structure of heat discharge and recovery.Once the fan system breaks down,it will stop the operation of the entire production line,and even cause serious accidents.At present,the fault diagnosis of some fans in pellet production lines in China is still based on manual detection,which not only requires workers to have rich experience,but also wastes manpower and material resources.The foresight of fault diagnosis and the accurate judgment of fault type are the most important.In order to ensure the pelletizing production line can run for a long time and stably,it is necessary to study the online fault diagnosis method of fans in pelletizing production line.In this thesis,the main extraction fan is taken as an example,aiming at the problems of slow model training,large amount of training data and low fault diagnosis accuracy in fault diagnosis,two models are proposed.Different models are selected for different operation periods of the main extraction fan to achieve the optimal effect.The main work of this thesis is as follows:(1)Aiming at the shortcomings of the fault diagnosis model,which requires slow training speed and large amount of running data,a model based on improved social network search algorithm(ISNS)combined with multi-classification support vector data description(M-SVDD)is proposed.Firstly,the optimal individual guidance strategy and T distribution are introduced to optimize the social network search algorithm(SNS),so as to avoid the two problems of slow search speed and easy to fall into local optimal.Then,the improved ISNS algorithm is used to optimize the two important parameters in M-SVDD: the penalty factor and the parameters of Gaussian kernel function,so as to improve the accuracy of fault diagnosis.The M-SVDD model establishes its own high-dimensional hyperspheres through 29 main parameters in the time domain and frequency domain signals of normal and abnormal data.By judging whether the new data is in a certain kind of hypersphere,the fault type can be quickly determined,and the accuracy can be improved without affecting the speed.The experimental results show that the M-SVDD model optimized by the improved algorithm can meet the needs of rapid fault diagnosis in the pellet main extraction fan fault diagnosis,especially in the case of less early operating data,and can also achieve a high accuracy.(2)In order to overcome the low fault diagnosis accuracy of the neural network in the late operation period of the main extraction fan,an improved empirical wavelet transform combined with residual network(Res Net)model is proposed.A new method is proposed to divide the spectrum boundary of the empirical wavelet transform.Firstly,the left minimum,the maximum and the right minimum of the initial boundary are obtained by the local maximum method.Then,a new fitting curve is obtained by cubic spline interpolation,and the partition boundary is established by using the local minimum of the new curve.Secondly,the largest component of the square envelope kurtosis value after the improved empirical wavelet transform is extracted to reconstruct the signal.Finally,the reconstructed new signal is input into the residual network,so as to achieve vibration signal noise reduction and more efficient feature extraction.The comparison experiment shows that this method is very effective in improving the accuracy of fault diagnosis.The experimental results show that compared with the original signal direct input residual network,the pre-processed signal input residual network improves the speed and accuracy of fault diagnosis by 12%,and can reach a higher fault diagnosis accuracy faster. |
