Design and implementation of automatic fault diagnosis system for wind turbine
Graphical abstract
Introduction
Energy is the material basis for the sustenance and development of human society. At present, coal, oil, natural gas, and nuclear energy constitute the world's main sources of energy. However, coal combustion causes considerable pollution, oil and natural gas supplies are being depleted, and the disposal costs of nuclear waste are much higher than construction costs. Thus, humankind is facing an energy crisis. Wind energy, as a type of clean energy, is becoming increasingly popular around the world [1], [2]. In recent years, China's wind power industry has witnessed rapid development. According to the statistics of the World Wind Energy Association (WWEA) published on February 25, 2019, the total installed capacity of wind power in the world reached 600 GW by the end of 2018, of which the new installed capacity was 53.9 GW. Furthermore, China, with a new installed of 25.9 GW, is the first country in the world whose installed capacity of wind power has exceeded 200 GW. China is the country with the largest installed capacity of wind power in the world. However, due to the short time for China to introduce wind power, the fault analysis technology for wind turbines is relatively backward, which has a lot of room for improvement. According to incomplete statistics, the average utilization rate of wind turbines in China is generally less than 95%, in which the high failure rate is a major factor, which leads to the maintenance cost becoming the main cost of wind farms. Therefore, reducing the maintenance costs is an important way to improve the operating efficiency of wind farms. In general, wind farms are located in harsh environments and face unsteady operating conditions. The signals detected by the sensors are affected by faults and various external factors. Moreover, the vibration signals of wind turbines are non-stationary and nonlinear. Wind turbine faults will not only affect electricity generation but also lead to serious safety issues, resulting in considerable economic losses. For the existing mature fault diagnosis method for wind turbines, it takes a lot of economic and manpower to better realize the fault diagnosis of the wind turbine. In recent years, a large number of researchers have attracted a lot of research on unstable and nonlinear signals.
Through steady development, fault diagnosis technology has emerged as an independent and comprehensive interdisciplinary information processing technology. It can be categorized into traditional diagnosis methods, mathematical diagnosis methods, and intelligent diagnosis methods. Traditional diagnosis methods include vibration detection technology, oil analysis technology, noise detection technology, and infrared temperature measurement technology [3]. Mathematical diagnosis methods include pattern recognition based on Bayesian decision criteria [4] and linear/non-linear discriminant functions, time series model diagnosis based on probability statistics, fault diagnosis based on distance criteria, the fuzzy diagnosis principle, gray system diagnosis, fault tree analysis, SVM [5], principal component analysis (PCA) [6], and wavelet analysis [7]. Intelligent diagnosis methods include fuzzy logic [8], expert systems, neural networks [9], and genetic algorithms. As the research of artificial intelligence becomes more and more mature, intelligent diagnostic methods enter the factory and gradually become the main direction of fault diagnosis. The SVM classifier can effectively solve the small sample and over-fitting problems, but the SVM needs to solve the support vector by means of quadratic programming, which is difficult to implement for large-scale training samples, and it is difficult for SVM to solve the multi-class problem. The main idea of PCA is to map n-dimensional features to k-dimension. This k-dimensional is a new orthogonal feature, also called principal component, which is a k-dimensional feature reconstructed on the basis of the original n-dimensional features. The significance of wavelet transform (WT) is that it can decompose the signal at different scales, and the selection of different scales can be determined according to different targets, which can realize the simultaneous representation of signals in the time domain and the frequency domain.
Fault signal processing and analysis methods have also witnessed considerable advancement and widespread use, from traditional analysis methods such as time domain waveform analysis, speed synchronization analysis, power spectrum analysis, detailed spectrum analysis, correlation analysis, coherence analysis, cepstrum analysis, and demodulation analysis, to new analysis methods such as wavelet analysis [10], Wigner–Ville technology [11], Hilbert demodulation, and other time-frequency analyses [12]. With the onset of the 21st century, researchers have increasingly focused on combinations of traditional or mathematical diagnostic methods with intelligent methods, such as the combination of a DWT and an ELM [14], [15]. Li et al. [15] proposed a fault detection and classification method for medium voltage DC shipborne power system combining wavelet transform multiresolution analysis technology with artificial neural network. The wavelet transform multiresolution analysis technique and Parseval theorem are used for different faults. The features are extracted and then the artificial neural network is used to automatically classify the fault types. Further, methods for fault diagnosis have been proposed on the basis of a combination of pattern recognition and artificial intelligence, fault feature extraction and separation based on empirical mode decomposition (EMD), and resonance demodulation.
Based on the theoretical analysis and the processing of the field data, this paper presents a new method of reasonable and higher accuracy for the signal of the non-stable state of the wind power generation, and can finally be used in the actual wind farm. On the basis of the fault diagnosis of wind power generation, a new hybrid diagnosis method is proposed in this paper. According to the unsteady and nonlinear characteristics of the acceleration vibration signal of wind power generation, combined with DWT, PSR and SVD, the fault feature extraction of wind turbine is realized. Finally, the classification is realized by using an ELM.
The remainder of this paper is organized as follows. A background of fault diagnosis and the whole fault diagnosis system construction of wind turbines are provided in Section 2. A sequential method consisting of a DWT, PSR, SVD, and an ELM is proposed in Section 3. Experimental results conducted on real wind turbine data are presented and comparison results with other methods are given in Section 4. Finally, Section 5 concludes the paper.
Section snippets
Problem description
At present, the operation and maintenance technology of wind turbines, especially the fault diagnosis technology, cannot effectively support the rapid development of the wind farm industry owing to the following three factors.
First, wind turbine faults are categorized into electrical faults and mechanical faults. Electrical faults are easy to locate and diagnose, but mechanical fault diagnosis needs to monitor the operation status of wind turbine units, analyze and process the operation data,
Algorithm
The block diagram of the proposed method for wind turbine fault diagnosis is shown in Fig. 4. As the environment of wind farms is often complex, the signals detected by vibration acceleration sensors on the wind turbines are usually non-linear and unstable. For such signals, we first establish a discrete wavelet transform. The discrete wavelet coefficients are obtained by dividing the original signals into several generations according to different scales. These coefficients reflect the
Experimental results
This study uses the data from an acceleration vibration sensor developed by Beijing Nenggao Automation Technology Co., Ltd. for a 2WM-class wind turbine. There are three main types of data, i.e., data in the normal state, gearbox fault state, and generator fault state. These data will be used to test the proposed wind turbine fault diagnosis method. The sampling frequency is 16,384 Hz and the rotational speed is 18 rpm. Before feature extraction, 1000 groups of data are randomly generated from
Conclusions
A framework on the fault diagnosis of wind power generation was proposed. Through the constructed fault diagnosis system, large amount of data of vibrations status of wind turbines was collected and processed for fault diagnosis of wind turbines. Because of the limitations of traditional non-steady state signals, traditional methods have difficulties in achieving fault diagnosis well. Therefore, this paper focuses on new method to resolve such problems. A new method is proposed through
Acknowledgement
All authors do the work of this paper.
Declaration of Competing Interests
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
Yu Pang was born in CHIFENG, Inner Mongolia, China, in 1981. He received the bachelor's degree from Beijing university of chemical technology. Now, he studies in Beijing Jiao Tong University. His-research interest include wind turbine fault diagnosis, machine learning, and big data analysis. Email: [email protected]
References (19)
- et al.
Wind turbine fault diagnosis based on Morlet wavelet transformation and Wigner-Ville distribution
Renewable Energy
(2010) - et al.
A sequential method using multiplicative extreme learning machine for epileptic seizure detection
Neurocomputing
(2016) - et al.
Intelligent fault diagnosis method of planetary gearboxes based on convolution neural network and discrete wavelet transform
Computers in Industry
(2019) - et al.
A survey on wind turbine condition monitoring and fault diagnosis—Part I: Components and subsystems
IEEE Transactions on Industrial Electronics
(2015) - et al.
A survey on wind turbine condition monitoring and fault diagnosis—Part II: Signals and signal processing methods
IEEE Transactions on Industrial Electronics
(2015) - et al.
Electrical Signature Analysis-Based Detection of External Bearing Faults in Electromechanical Drivetrains
IEEE Transactions on Industrial Electronics
(2017) - et al.
Concurrent Fault Diagnosis Based on Bayesian Discriminating Analysis and Time Series Analysis with Dimensionless Parameters
IEEE Sensors Journal
(2019) - et al.
A New Bearing Fault Diagnosis Method Based on Fine-to-Coarse Multiscale Permutation Entropy, Laplacian Score and SVM
IEEE Access
(2019) - et al.
Deep PCA Based Real-Time Incipient Fault Detection and Diagnosis Methodology for Electrical Drive in High-Speed Trains
IEEE Transactions on Vehicular Technology
(2018)
Cited by (0)
Yu Pang was born in CHIFENG, Inner Mongolia, China, in 1981. He received the bachelor's degree from Beijing university of chemical technology. Now, he studies in Beijing Jiao Tong University. His-research interest include wind turbine fault diagnosis, machine learning, and big data analysis. Email: [email protected]
Limin Jia is a Ph.D. supervisor, who works in state key lab of rail traffic control & safety of Beijing Jiao Tong University. His-research field is in intelligent control and intelligent theory, rail traffic intelligent control and safety system key technology development, and new energy system theory and technology research. Email: 12,114,[email protected]
Xuejia Zhang received bachelor degree from Beijing University of Chemical Technology in 2018. He is now a master degree candidate in Beijing University of Chemical Technology. His-research interest is fault diagnosis and machine learning. Email: [email protected]
Zhan Liu was born in Hubei province, China, in 1982. He received the bachelor's degree from Beijing university of chemical technology. Now, he works in Beijing NEGO Automation Technology CO., LTD. His-research interest include wind turbine fault diagnosis, machine learning, and big data analysis. Email: [email protected]
Dazi Li is currently a Full Professor and vice dean of College of Information Science and Technology, Beijing University of Chemical Technology, Beijing, China. Her research interests include machine learning and artificial intelligence, fault diagnosis, advanced process control, etc. She is currently an Associate Editor of ISA Transactions. Email: [email protected]
This paper is for CAEE special section SI-cps. Reviews processed and recommended for publication to the Editor-in-Chief by Guest Editor Dr. Zhihan Lv.