Stall Detection And Mechanical Failure Diagnosis Of Centrifugal Fan Based On The SDP Analysis

As a power source of the smoke system in the power plant, the running state of centrifugal fan has the direct bearing on the safe and economic operation of power plant. However, with modern power plant units becoming large-scale, and systematic automated and the poor working environment, it is prone to faults for centrifugal fan. With faults, fan can not operate normally. Even more, the faults may cause the entire unit downtime, resulting in economic losses and security risks. Therefore, it has great significance to perform real-time status tracking evaluation and accurate diagnosis of fan for the security and economic about power plant units. Fan faults mainly contain two aspects: fluid unsteady flow fault and mechanical fault. Based on the pressure signals acquired by rotating stall simulation experiments and the vibration signals acquired by mechanical faults simulation experiments, stall detection and mechanical faults diagnosis methods are researched.Stall detection method based on SDP analysis and image matching was proposed. The SDP technique was used to analyze the pressure signal, and the SDP pattern templates under normal and stall operation state were established. Finally, the image matching method was used to identify the similarity between the templates and the real-time SDP pattern for performing stall detection. In order to strengthen the individual features of operation state and weaken the common features of interference factors, two optimization approaches were put forward:(a) the wavelet noise reduction was utilized to denoise the pressure signal;(b) the principal component analysis was utilized to extract the feature of the SDP pattern. It is shown that the method optimized by the wavelet noise reduction, could detect the starting point of rotating stall within 0.3656 s, which cannot meet the timeliness requirement. Meanwhile, the wavelet noise reduction method cannot well meet the need to detect stall of online fan. The method optimized by the principal component analysis, could detect the starting point of rotating stall within 0.0656 s, 0.3 s faster than the method optimized by wavelet noise reduction. Meanwhile, it can meet the need to detect the rotating stall of online centrifugal fan.Stall detection method based on the analyses of SDP and pattern features was proposed. The wavelet transform was used to analyze the pressure signals, and the SDP pattern features under different operation states were analyzed based on the time-domain characteristics of the pressure signal and the SDP technique. Finally, the comprehensive autocorrelation coefficient about the variances of radius and angles was deemed as criterion to perform real-time stall detection. It is shown that stall detection method based on the analyses of SDP and pattern features could detect the starting point of rotating stall within 0.125 s, which can meet the timeliness requirement. Meanwhile, it can meet the need to detect the rotating stall of online centrifugal fan.Fault diagnostic method based on the SDP analysis and improved BP neural network was proposed. The SDP technique was utilized to reconstruct the vibration signals of 13 kinds of running states. Then, the features of the SDP pattern of each running state were extracted and the fault eigenvectors based on the multiple feature fusion were constructed. Finally, the sample set of the eigenvectors was trained and tested by the improved BP neural network to diagnose the mechanical failure of the fan. The diagnostic accuracy rate is 100%.