| Doubly-fed wind turbines have an advantage over direct-drive models in terms of price and occupy about 85% of the market.However,there are still some problems in practical applications,such as a high proportion of downtime due to gearbox failures,Insufficient mining of operational monitoring data,and the manual extraction of feature quantities based on shallow learning diagnosis methods.In this paper,a combination of theoretical analysis and experimental verification is used.First,the2.5MW FZCR2500 gearbox is predicted based on the time-domain characteristics of gear failures.Then,shallow learning and deep learning are used to diagnose the wind power gearbox.The SDAE-based fault diagnosis method for the root cracks of the wind power gearbox is determined,and it is verified by a scaled-down prototype of the wind power gearbox built in the laboratory.The main research contents are as follows:(1)Explains the research background and significance of gearbox fault diagnosis,and confirms the use of big data processing methods and technologies to carry out early fault diagnosis research on wind turbine gearboxes;Furthermore,the research status of traditional fault diagnosis methods for wind power gearboxes and wind power gearbox fault diagnosis methods based on deep learning are reviewed.In view of the insufficient information mining of operation monitoring data and the need for manual extraction of feature quantities based on shallow learning diagnosis methods,wind power Gearbox Research on Early Fault Diagnosis of Gearbox Based on Deep Learning;Sorted out the general idea and technical route of the paper.(2)The basic structure,working principle and structural parameters of the2.5MW FZCR2500 double-fed wind power gearbox are introduced.The failure mode and the weakest link of the gear transmission system are determined according to the fault tree analysis;the gears and the weakest links in the gearbox are considered and calculated.The main characteristic frequencies of the bearings,the 2.5MW FZCR2500 wind turbine gearbox characteristic frequency database and the bearing characteristic frequency database have been summarized and established.(3)The shortcomings of the BP algorithm are summarized through the overview of shallow machine learning,and the advantages of deep learning are summarized according to the overview of deep learning neural networks;in order to select a suitable deep learning model for effective early fault diagnosis of wind power gearboxes,and try to avoid the problems existing in the existing models,compare the current main deep learning models,and select the method of early diagnosis of wind power gearbox faults by stacking noise reduction autoencoders with excellent comprehensive performance.(4)In order to monitor the real-time operating status of the wind power gearbox,make a preliminary assessment of its operating status,and correctly diagnose its early fault status,the laboratory has built a 2.5MW FZCR2500 double-fed wind power gearbox scale prototype test system;According to the failure modes and weak links of the fault tree analysis in Chapter 2,the test system faults are implanted to simulate typical failure modes,and the fault status of the wind power gearbox is preliminarily diagnosed through data collection and fault feature extraction;The layer model further diagnoses its fault.(5)The principle and method flow of the stacked noise reduction autoencoder are introduced.6 types of failure data collected through the wind power gearbox scale-down prototype test system(no failure,slight pitting of the pinion,severe pitting of the pinion,2mm large gear tooth root crack,3.5mm large gear tooth root crack and3.5mm planetary gear Root cracks)are imported into the stacked noise reduction autoencoder for training,and compared with the performance of the fault diagnosis of shallow learning to verify the feasibility of the early fault diagnosis method of the wind power gearbox gear based on the stacked noise reduction autoencoder. |
PDF Full Text Request