Managing wind farms with new and diverse technologies while reducing operational costs is quite challenging. While SCADAs and Operation Centers focus on operational matters, plant and asset managers need reliable analytical tools to make faster and smarter decisions when dealing with unavailability events or failures. Analytical tools based on machine learning and big data are a key element to help AM and O&M teams improve renewable plants’ operational efficiency, by means of an early identification of problems and an optimization of actions in the farm.
Wind farms’ asset managers, on their mission of making sure that their performances accomplish the forecastings and that their operation is optimal, are facing an increasingly demanding environment. This environment can be summarized into three big challenges: a) portfolios to be managed with more plants and bigger wind turbines, which require a more complex maintenance; b) longer useful life of wind turbines than initially expected, resulting in an increase in problems at its beginning and end, which is known as the “bathtub curve” of the failure rate; and c) O&M cost reduction objectives in an environment of price uncertainty in wholesale markets.
So far, these managers have relied on the information provided by SCADAs and Remote Operation Centers to optimize the efficiency of wind farms. These tools are oriented towards immediate operation, managing alarms and anomalous signals with the main objective of avoiding unavailabilities. However, there are often inefficiencies and problems that require greater analytical capacity. This is where big data and machine learning tools can show their true potential.
Analytical tools based on machine learning and big data
These tools have the capacity to capture and integrate all sources of information: element signals, alarms, grid operator’s setpoints, weather stations, etc. They are also capable of incorporating and learning from expert knowledge, detecting uncommon operating situations, and identifying the root cause of problems and inefficiencies to quantify their impact. In addition, it is important that they have intuitive interfaces so that people can easily visualize the problems that have been detected, as well as APIs that allow their integration with the existing systems.
In order to properly operate, these tools need: a) scalable capture IT architecture with storage, computing capacity and interfaces; b) quality data flow; and c) algorithms capable of extracting useful information from them.
It is important that algorithms are designed with specific objectives in mind, and that they are trained for the specific operation of each wind turbine.
For the architecture, cloud-based tools (Google, Amazon, Azure, etc.) provide the basis of a scalable and cost-efficient infrastructure, optimized for the fluctuating loads of analysis processes.
Having high quality data is crucial to obtain valuable information, yet it is one of the main challenges. In fact, it is estimated that 60-70% of the time of a Data Science project is spent on data preparation. Gaps, inconsistencies, duplications, frozen data, etc. are part of the problems to be solved before starting any serious analytical process.
Algorithms are what allow the tool to get the most out of the data. To this end, it is important that they are designed with specific objectives in mind, and that they are trained for the specific operation of each wind turbine. For example, by designing algorithms to detect specific patterns of power limitation due to overheating. In this case, techniques such as neural networks –which can predict the temperature, power or intensity of a signal based on the model– are of great help in identifying emerging problems and quantifying their impact. To make it simple, the process of creating these algorithms is as follows:
In Isotrol, we have created the Bluence® Performance Management tool, which relies on advanced algorithms, to help asset managers identify problems that affect the efficiency of wind turbines. It analyses both the complete equipment and its main subsystems: generator, pitch, yaw, gearbox, etc. To make this possible, it combines machine learning techniques with digital twin ones to correct poor quality data, identify problems in early stages, quantify their impact and recommend actions based on data-driven findings.
In the image below, there is an example of how these algorithms work. The comparison between a signal and its prediction reveals that the root cause of a limitation is an overheating of the generator; in this case, there is no alarm signal from the manufacturer.
With Bluence® Performance Management, and through daily automated analysis, the efficiency of wind farms is improved by: 1) ensuring compliance with manufacturer specifications, for example, correcting limitations caused by faulty sensors; 2) reducing downtime and inefficiencies by an early detection; and 3) optimizing operations through predictive analysis and root cause identification, before major failures occur.
Conclusions
The mission of the AM and O&M teams is to ensure the performance and profitability of wind farms. To achieve this, it is crucial that these teams can detect problems and their root cause early, before they become worse and affect asset availability and energy production. Isotrol, based on its experience with information management of renewable plants, has developed Bluence® Performance Management to help asset managers in this task. An advanced analytical tool that automates data collection and correction, modeling and training of algorithms and their continuous exploitation.
Comments are closed.
