Almost Everything You Need To Know About Machine Learning

Published 19 December 2019

Offshore wind is waking up to the realisation that machine learning has much to offer the industry says Carla Shearer.

Carla Shearer, Summer Placement Engineer

The offshore wind industry, with a little help from us here at ORE Catapult, has been waking up to the realisation that machine learning and artificial intelligence (AI) has much to offer in terms of insights, performance prediction and maintenance enhancements for the UK’s fleet of offshore wind farms.

The Catapult has already completed a series of data pilot investigations to explore the value hidden within the large amounts of data being generated by each turbine and wind farm, and the Offshore Wind Innovation Hub (OWIH) has highlighted the benefits of cross-sector collaboration to increase understanding of how to mine large amounts of data as well as keeping it secure. In addition, there have also been several projects focussing on applying machine learning to offshore wind by companies in the industry over the past decade, and techniques for identifying leading edge blade erosion, filtering turbines alarms, and using data to predict power performance, are all in development.

Despite these forays into the digital world of wind turbines, machine learning is still not being used to its full potential. As we’ve previously discussed, the use of AI is revolutionising the way owner/operators and service providers approach maintenance and prediction issues. Until now, the development of such techniques has been restricted by the lack of understanding of what implementing machinery learning looks like. While many companies understand the benefits of data mining for insights, or the applications of automated prediction techniques, they are reluctant to share their data of pool resources out of a justified concern for security. A more thorough understanding of the methods used in machine learning and what the implementation of machine learning looks like is needed.

So here at the Catapult we’ve created a publicly-accessible machine learning wiki. Initially developed for our Data and Digitalisation team, the wiki serves as a starting point for understanding machine learning techniques – what the common approaches are and the technical understanding of how they work mathematically.

While very broad, the wiki has some more specific areas of focus such as the use of python for implementing machine learning and how machine learning can be and is being applied in the wind industry.


Machine Learning – The Lowdown

There is growing awareness that artificial intelligence (AI) is and will be useful in many ways, but machine learning can seem a mysterious and complex tool to use. Thanks to extensive developments and documentation in computer science, the biggest barrier to getting started with machine learning is public perception of its complexity. Trust in this “black box” technology also needs to be established.


What is Machine Learning?

Machine learning is a well-define subset of AI that learns from experience. Where any non-machine learning algorithm can classify items in a data set, if any of the incoming data is incorrect, mislabelled or incomplete then the result can be flawed and unreliable over time. This means machine learning can be used to make predictions, classify items or find trends that would take a long time to do by hand.


Type of Machine Learning

To start to understand the options within machine learning, this overview of different techniques may help. We also recommend this article that has a good explanation of one type of machine learning method.


Data Quality

When preparing data for machine learning, keep in mind that the quality of data input determines the quality (accuracy) of the output. Labelled, compete (not missing), extensive and accurate input is ideal, but rarely achievable.


Application Methods

Choosing the right method for your application is not always an obvious choice. Many papers that report on a machine learning project compare the accuracy of different methods for this reason. Luckily, this is where the ease of implementing machine learning comes in – preparing your data is the most important and lengthy part of it. At its simplest, training a model involves calling on a pre-prepared function in the language of your choice (python is very popular) and interpreting the results. No technical understanding required!


Of course, machine learning is not a magic tool that will answer all the questions you have or instantly provide insights that will solve your problems. While the implementation can be extremely simple, especially on well curated example datasets, the problems faced in industry are a lot more complex and may require extensive pre-and-post processing of data. Basic understanding and some curiosity go a long way in demystifying machine learning, and these complex problems in industry are being solved as we learn more.


Carla Shearer, Summer Placement Engineer

Carla joined the Data and Digitalisation team at ORE Catapult as an intern over summer 2019. She’s contributed to writing reports and deliverables with the team, as well as giving a hand where needed in creating data visualisation dashboards and cleaning up MATLAB and python code.

As a 4th year student in mechanical engineering at the University of Strathclyde, the work at the Catapult and her studies complement each other well. Her dissertation is a practical application of machine learning, and during the last few month at the Catapult she has worked on compiling resources for the team to learn more about machine learning.