Reprinted with permission from Uptime® Magazine.
Machine learning is an approach of exploring and building algorithms that enable computers to continuously learn and adapt.
With utilities, unknown failures and the maintenance that goes into fixing them can add up at lightning speeds. A recent study conducted by GlobalData Power estimates that expenditures for wind turbine maintenance have been projected to rise from $9.25 billion to $17 billion by 2020. 1
The issues stretch past the deafening cost of maintenance, too. As of 2011, roughly $40 billion worth of wind equipment in the United States was out of warranty, placing the financial responsibility on the owner of the wind farms to provide a more cost-effective operation.
It seems inevitable that costs will continue to rise and wind farms will struggle to remain as functional assets to utilities. The problems within the industry will persist and the burden will continually be placed on the owners to replace and repair turbines on their farms. Additionally, unscheduled downtime caused by unwarranted failures adds to the losses and expenditures of these companies and the industry as a whole.
The inevitable failure of wind turbines and other assets on fleets raises the question: What if there was a way to predict these failures prior to their occurrence? Actually there is. Utility operators have begun outsourcing their problematic areas to machine learning.
Machine learning is an approach of exploring and building algorithms that enable computers to continuously learn and adapt. Machine learning and deep learning techniques, a subset of machine learning, are used to understand the reasoning behind algorithms and aid in learning complex patterns. Recently, Google’s DeepMind Lab team utilized deep learning mechanisms to beat a grand master in the ancient Chinese board game, Go. Go is a game where there are more potential moves than there are atoms within the universe. Because of this, new techniques in artificial intelligence are necessary to solve these types of problems. An approach that is more focused on reasoning and inference becomes essential in solving complex problems before they even exist. It is encouraging to know that a problem set with so many complexities can be solved. It means providing information like predictions of downtime eventually will become straightforward and second nature.
For utilities, utilizing machine learning means various things and can lead to a positive response across the board. It’s one thing to have data across a variety of platforms and another to codify the data, thus allowing for greater knowledge retention. There is a complex relationship between sets of data, so complex that the average analysis techniques seldom can notice a difference. With the codification of these data sets and utilization of machine learning techniques, visibility can grow, thus decreasing the need for humans to manually analyze each set for comparisons.
Figure 1: Organization of data for machine learning systems
With large quantities of data, systematic organization becomes key. There are four main sectors to understanding and organizing data within the utilities field: assets, data collection systems, analytic platforms and the output.
Data lakes are starting to become popular and are a concept where all the necessary information from data silos are gathered into one place. From a business perspective, this consolidation makes it more efficient to access all information, thus making it easier to provide and manage evidence and insights. This enables better results in a company employing predictive and reliability-centered maintenance programs.
Unlocking this information allows a utility to employ a more holistic solution across all of its business segments. The energy chain (i.e., generation, transmission and distribution, energy trading and risk management, and cybersecurity) now can be treated as one system where data is shared and analyzed, producing targeted, efficient results to utilities and consumers. Let’s further break down how artificial intelligence and machine learning technologies are providing actionable insights at each stage in the energy chain.
Figure 2: How machine learning can benefit different sectors within utilities
Smart meters are electronic devices that record the consumption of electric energy in one-hour intervals or less. Utilities collect terabytes of data per day, creating an ideal solution for fully utilizing machine learning within the field.
Energy disaggregation is where patterns of usage per appliance are gathered by deconstructing information from a single home sensor. This application requires the utilization of machine learning because thousands of energy “signatures” must be analyzed to find patterns of usage. The savings are immense, as homeowners can discover how much every appliance contributes to their energy bill. On the operations side, analysis of energy signatures predicts suspicious consumption values due to physically or digitally manipulated devices, sophisticated thefts, meter malfunctions and more.
Most forecasting solutions were not designed to manage the variability, complexity and volume of data that’s emerging in utilities in today’s world. Machine learning enables utilities to pick up on subtle patterns within data sets, enabling them to make unified and more accurate predictions across the board. This is a vital asset to the industry and consumers alike because if you can more accurately track the forecasting of energy, then pricing of power can universally improve.
Machine learning encourages learning from the past and adapts to the future to better strengthen and protect valued infrastructures.
Another facet to pay mind to is cybersecurity. A recent poll conducted by SAS cited cybersecurity as the number one benefit from machine learning— and with good reason. A main threat to the industry is the number of viruses that are created on a daily basis, which averages out at roughly 28,000. This creates the need to dynamically adapt to and learn from the ever-changing frontier of data and viruses alike. Machine learning encourages learning from the past and adapts to the future to better strengthen and protect valued infrastructures.
A specific example where machine learning has added value in the energy cycle is in the case of Invenergy, an energy company that owns and operates numerous wind turbine units in both the United States and Europe. The goal is simple: to provide advanced notice of gearbox failures on their wind turbines. Gearboxes are one of the most brittle components in a wind turbine and their failure accounts for roughly 85 percent of all wind turbine insurance claims.
These breakdowns easily could be reduced if there was a system in place, geared by machine learning, that notifies operators of failure prior to breakdown. Invenergy employs machine learning technologies at its facilities and is now able to get an advanced degradation warning 67 days in advance, as well as an indication (called a risk index) of impending catastrophic failures in the gearboxes 35 days in advance. This allows Invenergy to act with better knowledge and understanding of its fleet.
Figure 3: Results of Invenergy pilot
Whether you’re applying machine learning to better understand the operation of your assets or gain insight into your own consumption, there’s no denying its effectiveness within the industry. The ability to predict a failure before it happens is no longer a far-off wish. It’s happening now, thus enabling sustainability while keeping costs low.
- “Wind turbine maintenance costs to almost double by 2020,” Edie newsroom, Edie.net, 2015; https://www.edie.net/news/6/Win-turbine-maintenance-costs-to-nearly-doubl/
Reprinted with permission from NetexpressUSA Inc. d/b/a Uptime® Magazine. Copyright © 2017. All rights reserved. No part of this article may be reproduced or transmitted in any form or by any means without the prior express written consent of NetexpressUSA Inc. d/b/a Uptime® Magazine. Uptime® is a registered trademark of NetexpressUSA Inc. in the U.S. and several other countries. For more articles like this one, visit Uptime® Magazine online, http://www.uptimemagazine.com/