Automatic Fault Diagnosis in Wind Turbine Applications

Abstract

A wind turbine may have any—out of hundreds—types of bearing on its generator or other components. The bearing type is often unknown until a possible defect occurs or until a diagnostic engineer performs a diagnosis. We identify faults on different wind turbine components by matching their fault frequencies with frequency harmonics and sidebands found in the spectrum analysis. It is crucial to know the fault frequencies, such as ball pass inner and outer race defect frequencies, before the diagnosis. Lack of expertise and experience in identifying bearing type and frequency may result in an incorrect diagnosis. Moreover, not knowing the bearing frequency leads to the inability to track bearing fault development continuously. In this paper, we propose automatic diagnosis, a novel method to identify prominent peaks above carpet level in power spectrum analysis, and then label them for potential faults. Prominent peaks are noticeable spectral peaks above the noise floor, and we calculate the noise floor using median filtering. Once we extract the peaks, we implement priority labeling. The fittest bearing frequency is intelligently estimated using approximate peak matching and envelope of bandpass segment. We use the estimated frequency to identify, quantify, and track early and late bearing faults without knowing the bearing type. Using data from wind turbines monitored by Brüel & Kjær Vibro, we show that automatic diagnosis accurately detects faults, such as ball pass outer race fault, ball spin frequency fault, generator imbalance, and gear fault. We expect automatic diagnosis to reduce manual identification of fault frequencies. Implementing automatic diagnosis could significantly reduce the cost of simultaneously monitoring the condition of hundreds, or even thousands, of wind turbines.