Research On Fault Early Warning Of Primary Air Fan Based On Convolution Neural Network

With the opening of China’s thermal power "large unit" era,the stable operation of fans is more important for power plants.Early fault warning of fans can reduce the number of unexpected downtime of fans and bring huge economic benefits to thermal power plants.However,the existing equipment fault monitoring and early warning technology of power plants is limited,so it is difficult to send alarm to fans in time.In this paper,convolution neural network is constructed to fully mine the effective information in the operation data of primary air fan,and predict the vibration of primary air fan bearing so as to realize the early warning of fan fault.To achieve the goal,the following works have been carried out in this thesis.Study the internal structure of the primary air fan,summarize the common faults of the fan and the causes of the faults.Aiming at the bearing vibration signals which can show most faults,the convolution neural network bearing vibration prediction model is constructed based on the historical data of normal state of fan.In the selection of model input,the feature selection method based on k-nearest neighbor mutual information is adopted.Finally,based on the operation data of the primary air fan,the vibration prediction model of the fan bearing is established and the simulation test is carried out.The test results show that the method has high accuracy.The internal structure of CNN is studied,and the hyperparameters that have a greater impact on the performance of CNN are summarized.Aiming at the problem of initial hyperparameter selection,a convolutional neural network structure algorithm optimized based on genetic algorithm is proposed,which adaptively optimizes the hyperparameters in CNN,and solves the uncertainty problem of manual selection.Experiments show that the optimized CNN has higher accuracy in bearing vibration prediction.The difference between the predicted value of the model and the true value implies a wealth of fault information.In order to be able to accurately judge the abnormal changes of weak signals under the premise that false alarms and missed alarms are as small as possible.This paper proposes to use the sliding window method to analyze the model prediction residuals,so as to realize the failure warning of the primary air fan.If the mean or standard deviation of the residuals in the window exceeds the set warning threshold.Finally,a simulation study is carried out with a fan failure of a power plant as an example.The results verify that the proposed method can detect the early failure of the fan in time and realize.accurate failure warning.