David Socha, Teradata’s practice partner for Industrial IoT, kicks off a series of three blogs on machine learning and related techniques applied to asset management.Last month, my blog was on the changing role of the Information Technology (IT) Department in utility businesses. One of the things I discussed was the need for IT to become an enabler for innovation, helping the business to apply new techniques and capabilities to their problems …eh… opportunities to discover new value, writes Socha.
Over the next few blogs, I’m going to take a look at a few of those new-fangled IT innovations. This time, we’ll do a short introduction to some fashionable analytical techniques and discuss at a high level how one, in particular, can be applied in utilities today. Later, we’ll get into a little more detail on one of those techniques and also debunk a few of the myths surrounding their applicability to the industrial/engineering world.
Hype cycles and buzzwords
Today, no technology discussion is complete without throwing in terms such as Artificial Intelligence (AI); Machine Learning (ML) and Deep Learning (DL). All these fashionable techniques are at that point on the hype cycle where they’ll cure cancer, bring about world peace and solve world hunger too. Perhaps all at once. And of course, at the same time be a part of every affordable consumer product you could possibly imagine. AI is most certainly the worst victim of this. My favourite example is still the AI toothbrush1 from earlier this year.
On the flipside to all this madness, there is a plethora of information that describes in great detail the differences between AI, ML and DL. Some of it even makes sense. Though a lot has a somewhat sales-led tone. And therefore doesn’t.
Back in the real world
Rather than contribute further to the academic / sales material here, I want to instead consider a practical example of how one of those over-hyped magical technologies, Machine Learning really can be applied to an area of significant interest and expense in most utility businesses: Asset Management. So let’s get right to it. And start at the top. Put most simply, there are three aspects to applying any kind of analytics to Asset Management:
- there’s predicting things (often, but certainly not exclusively around when an asset is going to stop operating the way it should)
- there’s working out the consequences of that failure, balanced against the cost of false positives (eg is it - by whatever relevant measure - acceptable to have a certain number of failures or degradations versus the high cost of intervening too soon?)
- and there’s doing something about it (ie changing plans, practices, processes in light of new information)
Applying Machine Learning techniques can be an extremely valuable method in all three aspects - though it’s perhaps most applicable, and for now most often applied, to 1 and 2. We’ll get into the nitty-gritty of that next time. In anticipation of that though, let me do a very quick primer on Machine Learning for the uninitiated.
Defining machine learning, without Skynet
ML has been around as a concept since the 1960s. Traditionally, it has been described as the ability of computers to “learn without being programmed”. Which is all a bit scary-sounding and Skynet-esque, right? It is a fair enough description, but I think a better one might be along the lines of the ability of computers to discover insights without being explicitly told where to find them.
ML allows computers to study data sets and automatically find patterns, similarities, relationships, anomalies…whatever…that can then be used as reference prediction models. And as these models are used against new data, the computer can iteratively refine what it has learned, making the models more and more accurate. There are multiple categories of ML for different purposes (essentially Supervised, Unsupervised and Reinforcement Learning) and again, we’ll get into those in the next blog.
Back to asset management
For now, though, consider just a few of the possible applications of ML in asset businesses. Applying ML techniques to find patterns and iteratively refine predictions could mean an even more accurate prediction of when a waste water pump will fail; an even more explicit cost model for additional capital investment versus maintenance opex; or a more accurate identification of arcing across an HV insulator as seen by the camera on an inspection drone.
Worth finding out a little more? I think so.
See you next month
Over the course of the next two blogs, I will delve a little deeper into the types of Machine Learning and how they really can be applied to predicting the wayward behaviour of assets and determining optimum mitigation strategies. I’ll also take some time to consider and I hope refute some of the widely-held beliefs about ML and its applicability (or lack thereof) in the more-mathematical-than-statistical domains of science and engineering. Over the course of the three blogs, I hope to prove to you that ML is an exceptionally valuable tool, not only for detecting insider trading or identifying cat pictures on the internet but also in your own asset-centric, engineering world. It may not be The Messiah. But it’s not a very naughty boy2 either. See you then.
1 This isn’t actually AI. But you probably worked that out already.
2 My last couple of blogs have quoted some literary giants. So let’s stick with that theme once again, with this Monty Python classic.
About the author:
David Socha is Teradata’s Practice Partner for the Industrial Internet of Things (IoT). He began his career as a hands-on electrical distribution engineer, keeping the lights on in Central Scotland, before becoming a part of ScottishPower’s electricity retail deregulation programme in the late 1990s. After a period in IT Management and Consulting roles, David joined Teradata to found their International Utilities practice, later also taking on responsibilities in Smart Cities and the wider Industrial IoT sector.