Fault Prediction and Diagnosis of Wind Turbine Generators Using SCADA Data

Yingying Zhao, Dongsheng Li, Ao Dong, Dahai Kang, Qin Lv, Li ShangEnergies 2017 · DOI

TL;DR

This paper introduces a cost-effective method for predicting wind turbine generator failures and diagnosing fault types using only readily available SCADA data. The approach achieves an 80% accuracy in predicting failures 18 days in advance and a 94% accuracy in diagnosing fault types, significantly reducing the need for expensive specialized sensors.

What Problem Does This Paper Address?

The rapidly expanding wind energy sector faces a significant challenge in managing operation and maintenance (O&M) costs. These costs can account for a substantial portion of the overall energy generation cost, and unexpected downtime due to component failures, particularly in generators, is a major contributor. While condition monitoring systems (CMSs) offer a solution through predictive maintenance, their high deployment costs, often requiring specialized, high-frequency sensors, limit their widespread adoption. This paper aims to bridge this gap by developing a predictive maintenance solution that leverages existing SCADA systems, thereby offering a low-cost alternative for improving wind turbine reliability and reducing O&M expenses.

Methods

The proposed solution integrates two key components: a prediction model and a diagnosis model.

  • Data Pre-processing: The process begins with extensive data analysis and pre-processing. This includes data cleaning, feature selection (identifying parameters relevant to generator health), feature reduction using Principal Component Analysis (PCA) to capture essential information while reducing redundancy, and data set balancing using the Synthetic Minority Over-sampling Technique (SMOTE) to address the inherent imbalance between normal and fault instances.
  • Prediction Model: This model focuses on predicting the Remaining Useful Life (RUL) of generators. It employs an unsupervised clustering method (DBScan) to identify healthy operational states. An "anomaly operation index" (AOI) is introduced to quantify the deviation from healthy states over time. The RUL is then estimated based on this AOI, aiming to provide operators with advance warning of impending failures.
  • Diagnosis Model: This model acts as a complementary tool to identify the specific type of generator fault when it occurs. It utilizes supervised classification techniques. The paper evaluates several classifiers, including Support Vector Machines (SVM), K-Nearest Neighbors (KNN), Artificial Neural Networks (ANN), and Naive Bayes, and combines them with SMOTE for improved performance on imbalanced data.

Datasets

The study utilizes data collected from two real-world wind power plants in China over an 18-month period. This dataset comprises:

  • SCADA Data: 58 parameters collected at a one-minute interval, including condition parameters (wind speed, direction, ambient temperature), health parameters (bearing and shaft temperatures, gearbox oil temperature), performance parameters (rotor speed, active power), and controlling parameters (PLC states).
  • Status Data: Information on fault status, including turbine number, fault cause, maintenance activities, and maintenance dates.

The paper does not introduce a new benchmark dataset but rather uses real-world data to evaluate its proposed method.

Key Insights and Ideas

  • Leveraging Existing Infrastructure: The core insight is that valuable predictive maintenance information can be extracted from widely available SCADA systems, eliminating the need for costly, specialized sensors. This significantly lowers the barrier to entry for predictive maintenance in wind farms.
  • Anomaly Operation Index (AOI): The introduction of the AOI provides a novel and intuitive way to visualize and quantify the degradation of a generator's operational state over time, serving as a crucial indicator for RUL prediction.
  • SMOTE for Imbalanced Data: The effective application of SMOTE is critical for handling the inherent class imbalance in fault diagnosis, where normal operational states vastly outnumber fault instances. This technique significantly improves the detection of rare fault events.
  • Integrated Prediction and Diagnosis: The paper demonstrates the value of an integrated approach, where prediction provides advance warning, and diagnosis offers specific insights for timely and accurate repairs.

Key Results

  • RUL Prediction: The proposed method can predict generator RUL 18 days in advance with approximately 80% accuracy.
  • Fault Diagnosis: When faults occur, the specific type of generator fault can be diagnosed with an accuracy of 94%.
  • Classifier Performance: After balancing the dataset using SMOTE, the SVM classifier demonstrated superior performance, particularly in predicting the True Positive Rate (TPR), achieving up to 96.34%.

Limitations and Future Work

  • Data Imbalance: While SMOTE is used, the inherent class imbalance remains a challenge, particularly for accurately predicting the True Positive Rate (TPR) in the initial stages. The paper notes that TNRs are generally higher than TPRs before balancing.
  • Generalizability: The solution was evaluated on two wind power plants in China. Further validation on a wider range of wind turbine models and geographical locations would be beneficial to assess its generalizability.
  • Future Work: The authors suggest further research into optimizing the over-sampling degree (N) for SMOTE to achieve a better balance between TPR, TNR, and ACC. They also imply that the integrated approach can be extended to other wind turbine components.

Summary updated

Abstract

The fast-growing wind power industry faces the challenge of reducing operation and maintenance (O&M) costs for wind power plants. Predictive maintenance is essential to improve wind turbine reliability and prolong operation time, thereby reducing the O&M cost for wind power plants. This study presents a solution for predictive maintenance of wind turbine generators. The proposed solution can: (1) predict the remaining useful life (RUL) of wind turbine generators before a fault occurs and (2) diagnose the state of the wind turbine generator when the fault occurs. Moreover, the proposed solution implies low-deployment costs because it relies solely on the information collected from the widely available supervisory control and data acquisition (SCADA) system. Extra sensing hardware is needless. The proposed solution has been deployed and evaluated in two real-world wind power plants located in China. The experimental study demonstrates that the RUL of the generators can be predicted 18 days ahead with about an 80% prediction accuracy. When faults occur, the specific type of generator fault can be diagnosed with an accuracy of 94%.
BibTeX
@inproceedings{zhao2017fault,
  author = {Yingying Zhao and Dongsheng Li and Ao Dong and Dahai Kang and Qin Lv and Li Shang},
  title = {{Fault Prediction and Diagnosis of Wind Turbine Generators Using SCADA Data}},
  booktitle = {Energies},
  year = {2017},
  doi = {10.3390/en10081210},
}
Figure