Predictive maintenance with industrial sensor data
Abstract
The Norwegian Ministry of Petroleum and Energy Commissions report shows that the government is making a large step closer to its ambition of allocating regions for 30,000 MW offshore wind via way of means of 2040. According to a report by IRENA, offshore wind operation and maintenance (O & M) costs make up a significant portion of the overall cost of electricity for offshore wind farms in G20 countries, ranging from 16-25%. To address this issue, it is essential to explore methods for improving operational reliability and reducing the maintenance costs of wind turbines. One promising approach is predictive maintenance, which involves leveraging data collected from sensors already equipped with the turbines to detect and address potential issues before they become more serious. Predictive maintenance is important in wind farms to reduce downtime and optimize the performance of wind turbines. Various rotating components in wind turbines make them complicated machinery, and if any of those parts fails, it can cause the entire turbine to shut down. This can result in lost revenue for the wind farm operator and lead to higher maintenance costs if the problem is not addressed quickly. This can be possible through a Supervisory Control and Data Acquisition (SCADA) system, which collects and analyzes data from various turbine components. We have developed a method for detecting and monitoring failures in critical components such as the gearbox and generator, based on historical SCADA data. Our approach utilizes machine learning models, specifically extreme gradient boosting (XGBoost), and has been tested on two real-world case studies involving eight different turbines. The outcomes show both the effectiveness and usefulness of our technique for boosting wind turbine reliability and minimizing maintenance costs.