Research On Fault Diagnosis Method And Stall Prediction Of A Centrifugal Fan

As an important part of the mechanical equipment, fans have been widely used in lots of departments of the national economy. In power plant, it is the power source of the smoke system, and its running states are directly related to the security and economic operation of the power plant. Therefore, it is of great significance to research the fault diagnosis and ensure the safe and reliable operation of the fan. In this paper, G4-73No8D centrifugal fan is chosen as the study object. The fault diagnosis and the prediction of the rotating stall of the fan are studied based on the experimental research. The main research contents and results are as follows:(1) The simulation research the build of sample library of fan faults. The simulation of typical failure of the mechanical vibration and the rotating stall of the fan are realized based on the G4-73No8D centrifugal fan test bench. A high speed data acquisition system is established for data acquisition, and the fault sample library is built. The types of the mechanical vibration faults include the imbalance, misalignment, bearing’s looseness and rub-impact. The operating states include faults of different severity, occurred in different parts. The simulation of the rotating stall failure is the gradual process of the fan working from the normal operation to the strong stall station under the different working condition of the diverter opening angle.(2)The intelligent fault diagnosis model of the fan based on wavelet packet and the signal complexity is researched. The sample entropy features, the symbolic dynamics entropy features, the wavelet packet energy features and the wavelet packet singular value features of the vibration signal are extracted, and different features of each feature vector are analyzed. The characteristic is found that different fault conditions have certain distinction, which provides the basis for intelligent fault diagnosis. Researching on the intelligent diagnosis methods, we know that the neural network is easy to fall into the local minimum and the convergence speed is low, the parameters of the SVM model are not easy to choose. Therefore, the BP neural network, improved by increasing the momentum term and the adaptive adjustment vector optimization, and the SVM, improved by the particle swarm optimization, are used to be fault classifier, classify the fault features and then the intelligent fault diagnosis model established. Result shows that the intelligent fault diagnosis model based on the complexity analysis and the wavelet packet analysis can make an accurate diagnosis for the fan faults and the computational efficiency is high. (3) The fault diagnosis model combined SDP analysis and image matching is established. Symmetrized Dot Pattern (SDP) transforms the vibration signal into SDP graphics within polar coordinates through the corresponding calculation formula; it can fully describe the characteristics of the signal. When all types of the vibration signals of the fan are transformed with the SDP, we can get SDP figure of vibration signals in different operating states. Establish a known fault SDP template map and match the unknown fault SDP chart with fault template image. The matching results are just the fault classification results. The study shows that selecting a suitable template has a great impact on matching results. The matching accuracy rate of failure is low by using a single template, and the multi-fault template increases the number of additional computation and extends the computing time. Clustering fault template by the means of the clustering analysis can ensure the accuracy of matching and will not increase the additional computation.(4)Establish the rotating stall prediction model of the fan based on the phase space transformation and the support of vector regression. The stall prediction model of the fan is trained by the progressive signal from the normal state to the stall state so as to realize the real-time prediction of rotating stall. Extracting the signal features by the means of phase space transformation and finding the nature of the law hidden in the one-dimensional time sequences provide more substantial information for regression research. Improved SVR regression model is used for real-time prediction of rotating stall and wavelet transform method is used for detecting the starting point of stall. Multi-step prediction is studied for a longer prediction time. The result shows that rotating stall prediction model used in this paper can predict the starting point of stall five steps ahead and meet the time requirements of stall warning.