- |
Wind Turbine Fault Forensic Analysis of SCADA Data Using Machine Learning
![]() |
Juan José Cárdenas Araujo ITESTIT S.L., Spain WIND TURBINE FAULT FORENSIC ANALYSIS OF SCADA DATA USING MACHINE LEARNING Abstract ID: 451 Poster code: PO.075 | Download poster: ![]() ![]() |
Presenter's biography
Biographies are supplied directly by presenters at WindEurope 2016 and are published here uneditedJuan José Cárdenas A. is PhD from “Universitat Politècnica de Catalunya” and Electronic Engineer from “Universidad del Valle”. He is a data scientist with wide experience in using data mining techniques together with statistic while applied to energy sector. He has carried out research in energy efficiency enhancement for buildings and smart cities from a data-driven perspective. Nowadays, he is Researcher and Data Scientist in SmartIve, where he is looking for improving predictive algorithms for early fault detection and diagnosis of wind turbines.
Abstract
Wind Turbine Fault Forensic Analysis of SCADA Data Using Machine Learning
Introduction
One of the main tasks in O&M process is to find out the possible causes of a fault manifested by a specific alarm or a set of alarms that stops the wind turbine production. This process is crucial to reduce time of repair or detect more critic faults in earlier stages. Methodologies and tools that can support this type of process can benefit wind farm owners to increase availability and production and reduce costs. On the other hand, data availability from SCADA of the wind park has a great potential of information that can support this specific task, more when it is already available and using this data for fault diagnosis does not require any type of extra implementation or hardware installation in the wind turbine. However, due to the high number of available variables and data, analyzing them can be a high time consuming task and when just well-known related variables are analyzed hidden causes or not common causes cannot be or are hard to be found. For all these reasons, in this work, we present a methodology and tool, part of Smartive platform, that been fed by all available SCADA data and other source of information, can support fault forensic analysis in order to find the main causes of faults.
Approach
The proposed methodology is based on statistical and machine learning techniques in order to find out the most related variables with the fault under analysis. The most related variables are defined as those that most change their behavior previous to the specific fault and those that more can affect the precision of forecasting of a fault prognostic model. The first criterion is analyzed by means of hypothesis testing technique and information gain algorithms in order to preselect the inputs for the fault prognostic model. Then, after training and testing the obtained model, input relevance analysis into the model is carried out in order to find the final most relevant variables to the analyzed fault. Finally, a human expert analyzes the results in order to validate them and diagnosis the possible causes of the fault using the information given for the model. One of the most innovative features of our proposal is that there is no need for normality models of the variables under suspicion and the data is mainly processed with a daily time resolution.
Main body of abstract
We have been using this technique to analyze different wind parks around Spain with excellent results. In some cases, the tool has helped to find unexpired culprits and in other cases has helped to accelerate the process of analysis of faults based on data, reinforcing those suspicions that were proposed at the beginning by the human expert.
The proposed technique is built in SmartCast tool of the Smartive platform and makes part of its health estimation and failure prediction modules. The main aim of this tool is to know the root causes for faults and first trying to eliminate from the beginning in early stages of the fault. In case that it is not possible to anticipate the failure, the aim is to make easier for O&M personal to analyze critic faults/alarms and find out quicker the possible causes.
Some real cases are presented in order to show the technique and its effectiveness.
Conclusion
In this work, it has been presented a technique and tool, available in Smartive platform, that can exploit SCADA data in order to carry out a forensic analysis of faults, improving O&M process by reducing times of fault diagnosis, and detect critic faults in earlier stages.
Learning objectives
- Innovative techniques and tools for fault diagnosis using SCADA data.
- How to performance a fault forensic analysis using SCADA data.
- How this technique has been applied in successful practical cases.
Follow EWEA on: