Authors
Mahi Ayman, Mariam Elaskary, Nour Elgarhy, Zeina Tamer
Supervised by: Eng. Yomna Hassan, Eng. Maha Sayed
Publishing Date
January 3, 2022
Abstract
Institutions have been rapidly redirecting their investments away from fossil fuels, creating a favorable scenario for the generation of clean energy. The wind industry, especially, has seen an exponential increase in recent years. Early fault detection creates an alternative for operation and maintenance (O&M), helping forgo costs before they reach a catastrophic stage since preventive methods are used to stop malfunctions before their occurrence, improving turbine reliability and allowing for maximum utilization and better control of the wind farms. In this document, deep learning methods such as Autoencoders-LSTM (Long short-term memory) are used on time series data produced from SCADA to reach the optimal results.
1.1 Purpose of this document
This document highlights the features of the system, detailing its applications, functionalities and design so it would be easier for the target users to navigate and maintain it further down.
1.2 Scope of this document
This document provides the functional and nonfunctional requirements. It will further explain the main functionality of the system which is fault detection in wind turbines. It will also give a further description of the system architecture, the main objectives behind the system, the type of data used and the possible scenarios where the system will be needed and used.
1.3 System Overview
The proposed system has a few main steps, those are data acquisition and selection, processing, training model, and finally using a model on new data to detect faults and predict power.
The forecasted data and the SCADA data are compared. When major difference between the forecasted and acquired data occurs, the failure log is then checked during the period when the difference occurred. The model is then validated as whether a fault was close to its occurrence.
System implementation goes as follows: First, data is split into train and testing based on timestamp intervals. Second, is preprocessing the split data using MinMax Scaler takes place which transform features by scaling each feature to a [0,1] range, and PCA which is used for reducing the dimensionality of the dataset. Finally since the data the proposed system works with is time-series data, LSTM Auto encoder model will be used to detect anomalies.
A software accessible to on-field engineers would display the results reached by the model.
For this system, acquired signals from SCADA system will be regularly forecasted and the model shall be used for fault detection.
1.4 System Scope
The proposed system is a predictive model that aims to use time-series data, mostly provided by the SCADA system, to detect possible faults in wind turbines. The system shall forecast the power to be generated from the wind turbines by predicting necessary weather variables adherent to a specified dataset.