A unified fault diagnostic technique for wind turbine drivetrain using adaptive time space vibration analysis and machine learning models

Abstract

Owing to the cost effective and less carbon intensive electricity generation techniques makes wind energy is a prime contributor in the renewable energy sector. But the reliability of the wind turbines depends on the maintenance strategies. The unscheduled downtime of the wind turbines not only increases the Operation and Maintenance (OandM) cost it also affects the reliability and availability of wind energy generation. Premature failure in the rotating components is the prime cause for the unscheduled lengthier downtime. Condition monitoring is an elemental tool that monitors the turbine dynamics continuously, and detect the faults at the incipient state by which the progression of failure can be inferred and thereby severe damages can be avoided by planning appropriate maintenance strategy at right time. In this research work, a fault diagnostic technique based on adaptive time-space processing by vibration signal processing with machine learning model proposed to classify the typical and frequent tribological faults such as scuffing, polishing, fretting corrosion, overheating and assembly damage in wind turbine drivetrain. In wind turbines, the variable environmental factors cause variation in the loads that influences the vibration test, the corresponding vibration spectra exhibits dominant wideband blurs at higher frequency so that the peaks may appear wider than normal and amplitude levels reduced that influences the performance of fault diagnostics considerably. Variable speed wind turbines add additional complexity in terms of nonlinear operations. Low speed rotor shafts in wind turbine drive imposecritical challenge in vibration monitoring. At low speeds the vibration amplitude is very less and diminishing and also susceptible to environmental noise. Multiple and time varying transmission paths attenuate the vibration signal in planetary epicyclic gears. Nonlinear load distribution in wind planetary gears induce sideband modulations in vibration spectra even at normal conditions. Extraction of fault signatu