Hidden Markov model based principal component analysis for intelligent fault diagnosis of wind energy converter systems
Introduction
At the heart of the Wind Energy Conversion (WEC) systems is the power converter, which is used as a link between the machine and the grid [[1], [2], [3]]. Due to its influence on the power quality, the WEC Converter (WECC) should be inherently reliable and continuously available. It has been reported in Ref. [4] that 21 from 25 of the total failures in WECC are caused by the semiconductor. Thereby, the downtime becomes increasingly reliable on the WEC system size [5]. In this context, wind speed and power output of a WECC are used for overall rotor condition monitoring regardless of an increased blade surface roughness [6,7]. A spectral analysis of the nacelle oscillation has been successfully applied for the rotor blade supervision. On the same way, the issue of the double-fed induction generator (DFIG) blade imbalance has been addressed in Ref. [8] where the stator current of DFIG has been duly analyzed for extracting imbalanced faults features of the WEC under different wind speeds and imbalance coefficients [[9], [10], [11]]. Authors in Ref. [12] presented an unknown input observer to estimate faults in wind turbine converter assuming that the wind speed is unknown which in turn affects the rotational speed where its control is based on the converter torque. On the other hand and for the same purposes, a diagnosis formalism based on fuzzy prototypes is provided in Ref. [13]. Some other techniques are also duly considered and summarized in Ref. [14]. On the other hand, it has been proposed a deep learning machine for anomaly detection in the WEC gearbox and generator [15]. More recently, an artificial intelligence-based probabilistic anomaly detection approach has been used for a reliable WEC condition monitoring. This has been achieved by quantifying realistic uncertainties.
The WECC power is based on Insulated Gate Bipolar Transistor (IGBT). Nonetheless, one of the main factors of WECC faults is its periodic switching which affects the thermal cycle of different materials with different expansion coefficients; therefore, their life cycle decreases [16]. Besides, switching losses, due to the increase of the internal resistance, become increasingly important. In addition, the meteorological conditions, vibrations, dust and chemical products, under which the WEC operates, represent another source of faults in the converter [17]. Under these circumstances, a quite realistic WEC environment has been simulated and different experiments have been carried out.
The majority of fault detection and diagnosis (FDD) works have been focused on overall WEC faults diagnosis. Less much of the related literature on WECC has been reported. Therefore, in this work, a novel FDD framework based on Hidden Markov model (HMM) and principal component analysis (PCA) that is capable of detecting and identifying faults is developed. Features are appropriately extracted through PCA approach by which an optimal number of features is selected. Due to the need to develop a more sophisticated model that adequately takes into account the randomness of the operating environment, a well established probabilistic model based on HMM is used in classifying different faults that can be occurred in WEC power converters. The FDD performances using PCA-based HMM are illustrated through a simulated data collected from the WEC under different operating conditions. The application of the PCA-based HMM approach provides the high reliability and safety of the overall WEC system via the FDD of the converter.
The next sections of the paper are organized as follows: Section 2 briefly describes theoretically the background of PCA used in feature extraction and selection. Section 3 is devoted to the description of the machine learning technique using HMM. The simulation results that assess the performance of the proposed PCA-based HMM are presented in Section 4. In Section 5, some findings are drawn.
Section snippets
PCA-based feature extraction
PCA technique is mainly used for dimensional reduction. It is accomplished by determining a set of orthogonal vectors named as loading vectors which describe the most dominant features and the main trends in data. Let us consider a process that operates under healthy conditions with m sensors from which is collected n observations and regrouped in the data matrix X. The data is first shifted and scaled to zero mean and often in addition to unit variance, respectively, collected from a process
Description of Hidden Markov model technique
The Hidden Markov model (HMM) consists of a finite set of states , each of which is associated with a generally joint probability distribution of multivariate observables. Conceptually, HMM is based on a Markov chain. Only an external observation is visible at a hidden state [[18], [19], [20]]. Transitions between the states are commanded by a matrix of probabilities called transition probabilities, i.e.,where represents the state at the time t.
Fault detection and classification using PCA-based HMM technique
For the FDD purpose, the main steps of the proposed framework are as follows: i) Different measurements are recorded from the process under different operating conditions. The collected data represents healthy and different possible faulty scenarios that can be occurred in the process. It can be divided into two sets; one set is used for training and the other set is used for testing. ii) A PCA model is built using only the training data set where the system is working under normal operating
Description of wind turbines converter systems
In this study, a wind turbines converter system topology at variable speed is considered. This topology is based on a squired cage induction machine. The asynchronous machine is coupled to the turbine through a speed multiplier, (see Fig. 3). The variable speed operation of these turbines has become possible by static converters development and their control systems. Indeed, two static converters interfaced by a continuous bus are used. The connection of these converters to the grid is provided
Conclusion
In this paper, a machine learning-based Hidden Markov model (HMM) merged with principal component analysis (PCA) was proposed to deal with the problem of faults detection and diagnosis (FDD) in wind energy conversion converts (WECC) systems. The PCA model was applied in order to extract and select more efficient features to be used as observables in the HMM technique. Different operating conditions of the WECC were considered to show the robustness and the efficiency of the developed PCA-based
Acknowledgment
This work was made possible by NPRP grant NPRP9-330-2-140 from the Qatar National Research Fund (a member of Qatar Foundation). The statements made herein are solely the responsibility of the authors.
References (26)
- et al.
Condition monitoring and fault detection in wind turbines based on cointegration analysis of scada data
Renew. Energy
(2018) - et al.
Electrical & mechanical diagnostic indicators of wind turbine induction generator rotor faults
Renew. Energy
(2019) - et al.
Fault-tolerant wind turbine pitch control using adaptive sliding mode estimation
Renew. Energy
(2018) - et al.
A novel hybrid model based on vmd-wt and pca-bp-rbf neural network for short-term wind speed forecasting
Energy Convers. Manag.
(2019) - et al.
Real-time power switch fault diagnosis and fault-tolerant operation in a dfig-based wind energy system
Renew. Energy
(2018) - et al.
Current signature analysis to monitor dfig wind turbine generators: a case study
Renew. Energy
(2018) - et al.
Unknown input observer based scheme for detecting faults in a wind turbine converter
IFAC Proceedings Volumes
(2009) - et al.
Data–driven approach for wind turbine actuator and sensor fault detection and isolation
IFAC Proceedings Volumes
(2011) - et al.
A survey of fault diagnosis for onshore grid-connected converter in wind energy conversion systems
Renew. Sustain. Energy Rev.
(2016) - et al.
Anomaly detection and fault analysis of wind turbine components based on deep learning network
Renew. Energy
(2018)
A review on prognostic techniques for non-stationary and non-linear rotating systems
Mech. Syst. Signal Process.
Finding the best classification threshold in imbalanced classification
Big Data Research
Fault diagnosis and condition monitoring of wind turbines
Int. J. Adapt. Control Signal Process.
Cited by (117)
Ensemble learning framework for fleet-based anomaly detection using wind turbine drivetrain components vibration data.
2024, Engineering Applications of Artificial IntelligenceCorrelation warping radius tracking for condition monitoring of rolling bearings under varying operating conditions
2024, Mechanical Systems and Signal ProcessingPrincipal components-based hidden Markov model for automatic detection of whale vocalisations
2024, Journal of Marine SystemsMachine learning applications in health monitoring of renewable energy systems
2024, Renewable and Sustainable Energy ReviewsMultiscale monitoring of industrial chemical process using wavelet-entropy aided machine learning approach
2023, Process Safety and Environmental Protection