Study On The Vibration Fault Diagnosis Of Centrifugal Fan

As modern power generators'single capacity is growing larger, the units'safe operation is more and more important. Diagnosis technology is necessary for rotating machinery vibration fault, in which failure feature extraction and classification are the key problems. Le value, approximate entropy and Hurst index are used in two-phase flow recognition which is common parameter in the vibration acceleration time domain used to describe the vibration signal of fault characteristic. Meanwhile, some theory applied in brain waves and electrocardiogram denoising is also used to signal pretreatment, and with improved k-means algorithm, centrifugal fan vibration fault test is carried out.Auto regressive and EMD method are used to analyze misalignment fault, through which relationships between fan loads and fault frequency and relationships among the misalignment fault, fan loads and the fan speed are obtained.Based on gauss moments, the algorithm named FASTICA is used for blind source separation to get separated signals, through the original vibration analysis, the fault can be confirmed. After assigning the independent component zero except the fault component to get a new signal matrix, the denoising signal can be obtained with separable matrix inverse matrix and signal matrix multiplication. According to the distance criterion, denoising effect with other several methods was compared.The vibration signal is pretreated by EMD theory and reconstructed. After extracting the IMF1 to IMF8, and normalizing its energy characteristics, the feature vectors can be obtained and put into the improved k-means algorithm for classification.The original idea about k-means algorithm is introduced. Through the result about 4 cluster centers are distributed in two separate spaces. The shortcoming about the algorithm that is easy to fall into local optimum is displayed. Cluster center mobile criterion is made use of in the algorithm optimization, and combined with three time-domain signal characteristic parameters, the fault is classified.