Wind turbine planetary gearbox feature extraction and fault diagnosis using a deep-learning-based approach

Abstract

Planetary gearboxes are widely used in the drivetrain of wind turbines. Planetary gearbox fault diagnosis is very important for reducing the downtime and maintenance cost and improving the safety, reliability, and life span of the wind turbines. The wind energy industry is currently using condition monitoring systems to collect massive real-time data and conventional vibratory analysis as a standard method for planetary gearbox condition monitoring. As an attractive option to process big data for fault diagnosis, deep learning can automatically learn features that otherwise require much skill, time, and experience. This article presents a new deep-learning-based method for wind turbine planetary gearbox fault diagnosis developed by a large memory storage and retrieval neural network with dictionary learning. The developed approach can automatically extract self-learned fault features from raw vibration monitoring data and perform planetary gearbox fault diagnosis without supervised fine-tuning process. From the raw vibration monitoring data, a dictionary is first learned by a large memory storage and retrieval with dictionary learning network. Based on the learned dictionary, a sparse representation of the raw vibration signals is generated by shift-invariant sparse coding and input to a large memory storage and retrieval network classifier to obtain fault diagnosis results. The structure of the large memory storage and retrieval with dictionary learning is determined by optimal selection of the sliding box size to generate sub-patterns from the vibration data. The effectiveness of the presented method is tested and validated with a set of seeded fault vibration data collected at a planetary gearbox test rig in laboratory. The validation results have shown a promising planetary gearbox fault diagnosis performance with the presented method.

Dictionary Learning Approach to Monitoring of Wind Turbine Drivetrain Bearings

“…Tang et al [25] used shift-invariant sparse coding to generate a set of latent components that act as fault filters in a bearing or a gearbox. Moreover, studies by Ahmed et al [26] and He et al [27] proposed classification strategies that use the learned sparse representations on stacked autoencoders and large memory storage and retrieval neural networks, respectively. Further extensions of the work by Liu et al [22] had been developed by Wang et al [28] and Zhou et al [29], who used the same dictionary learning method with different classification strategies.…”

A combined approach of convolutional neural networks and machine learning for visual fault classification in photovoltaic modules

“…Various studies have recommended and demonstrated successfully the use of pre-trained network models. 31–37 In the present study, the feature extraction properties of several renowned CNN architectures like VGG 16, 18 GoogLeNet, 38 AlexNet, 17 and ResNet50 20 were evaluated with the help of machine learning classifiers. A brief description of the CNN models considered are described in this section.…”

Fault classification using convolutional neural networks and color channels for time-frequency analysis of acoustic emissions

“…They used auto-encoder residuals for component fault diagnosis and isolation. Another application in wind turbine was presented in He et al (2018) where DL was used for the condition monitoring of the wind turbine planetary gearbox. In this application, the fault features were obtained directly from raw vibration data using DL.…”