An Artificial Intelligence Neural Network Predictive Model for Anomaly Detection and Monitoring of Wind Turbines Using SCADA Data

• Published in: Applied artificial intelligence Vol. 36; no. 1
• Main Authors: Amini, Amin, Kanfoud, Jamil, Gan, Tat-Hean
• Format: Journal Article
• Language: English
• Published: Philadelphia Taylor & Francis 31.12.2022; Taylor & Francis Ltd; Taylor & Francis Group
• Subjects: Anomalies; Artificial intelligence; Artificial neural networks; Early warning systems; Effectiveness; Fault detection; Machine learning; Monitoring; Multilayers; Neural networks; New technology; Prediction models; Signal generation; Statistical analysis; Supervisory control and data acquisition; Troubleshooting; Wind turbines
• ISSN: 0883-9514; 1087-6545
• DOI: 10.1080/08839514.2022.2034718

The industry 4.0 has created a paradigm shift in how industrial equipment could be monitored and diagnosed with the help of emerging technologies such as artificial intelligence (AI). AI-driven troubleshooting tools play an important role in high-efficacy diagnosis and monitoring processes, especially for systems consisting of several components including wind turbines (WTs). The utilization of such approaches not only reduces the troubleshooting and diagnosis time but also enables fault prevention by predicting the behavior of different components and calculating the probability of near future failure. This not only decreases the costs of repair by providing constant component's monitoring and identifying faults' causes but also increases the efficacy of the apparatus by lowering the downtimes due to the AI-driven early warning system. This article evaluated, compared, and contrasted eight different artificial neural network (ANN) models for diagnosis and monitoring of WTs that predict the machinery's system failure based on internal components' sensor signals and generation temperature. This article employed a machine learning model approach with two hidden layers using multilayer linear regression to achieve its objective. The developed system predicted the output of the WT's generator temperature with an accuracy of 99.8% with 2 months in advance measurement prediction.