How Artificial Intelligence can help Technicians maintain their Assets faster, better and in a safer way.

[Although I work for IBM and mention IBM solutions in this article, all sentences and opinions expressed here are my own]

I’ve recently had the chance to work with one of the most exciting new Watson IoT tools out there. It is called Equipment Maintenance Assistant, and uses artificial intelligence to help field technicians quickly diagnose their asset issues and find any relevant information needed to fix them.

You might remember how, back in 2011, Watson won the Jeopardy challenge against the show's best ever contestants. I acutely remember the Whoa! effect that struck me when I first saw a live repeat of the game in Orlando, 2 months after the TV show win. Most of us in the audience sensed that we were witnessing an important innovation that day. Imagine: in the early 2010’s, a machine answering in plain English questions asked in plain English (on random topics ranging from dialects spoken in ancient Greece to Isaac Newton) had never been seen before, really!

Well, it is exactly the same feeling that I recently experienced using Equipment Maintenance Assistant (EMA). Literally, EMA is: Watson Jeopardy brought to the Field Technician … and more.

Let me try to explain my excitement by describing how EMA works, what it is, and why all asset-intensive industries should consider using this type of tool.

How does Equipment Maintenance Assistant work?

When a field technician logs into EMA, 2 options are provided: Diagnosis or Query.

Let's first use EMA Diagnosis, with a 'Problematic Pump' example

Just imagine an industrial pump asset. Based on alerts sent by its 'Asset Performance Management' system, a reliability engineer just tasked a field technician to check and fix the pump. To keep things simple, assume the pump can display problematic symptoms like excessive temperature, vibration, pressure etc. Based on those symptoms and the reported alerts, the technician wants to swiftly find the most likely cause(s) of the pump problem, as well as the most relevant 'How to Fix' information related to the problem. The picture below summarises the flow that the technician is guided through by EMA Diagnosis:

• First EMA asks the technician: "What do you see?", providing a list of predefined symptoms associated to the pump.
• The technician selects an initial set of symptoms. EMA returns a list of probable causes associated to those symptoms, and asks if the technician has observed more.
• As the technician adds more symptoms, EMA updates the list of possible causes, each with a level of confidence (in %), the most likely one on top - e.g. Impeller Corrosion.
• The technician can then click on any of the causes and be provided with: (a) the repair instructions related to the given cause and symptoms, (b) the evidence supporting the diagnostic, (c) a list of any other supporting documents - with highlighted passages within the doc - that could help fix the issue considering the given 'problem situation' and diagnostic (more details on this in the Query part below).
• Once the technician is happy with the diagnostic and information provided, it is important that (s)he rates the EMA results - this enables the system to continuously improve as EMA retrains using the latest technicians inputs. As part of the feedback, the technician can 'Thumbs Up/Down' all the diagnostic causes and symptoms and/or add extra observations that will further feed EMA intelligence.

Now let's look at the EMA Query part

OK, so EMA Diagnosis has just indicated that the most likely cause of the pump issue is the corrosion of its impeller. Now, the technician can ask EMA Query questions in plain English to gather extra information that could help him fix the pump faster, better and more safely. By 'question', we mean anything related to e.g. safety instructions, tools or permits required, work steps, asset parts etc. By 'information', we mean any source of info that has been fed into the EMA corpus e.g. OEM manuals, work or inspection logs, maintenance procedures, web sites content, asset specifications, installation and operation guides, regulatory documents etc.

Let's keep using our 'Problematic Pump' use case to show an example of how the EMA Query part works (see picture below):

• The technician simply asks EMA: "Are gloves or glasses required to fix the Pump Impeller Corrosion issue?".
• EMA returns the most pertinent answers, each appearing as a card, with a confidence score. The top answer reveals a unique passage within the 'Safety Equipment Requirements' paragraph of the 'Pump HSE Risks, Safety and Equipment Sheet' manual. That's exactly what the technician was asking for!
• The technician can then open the document, online, with the passage highlighted within. As for the Diagnosis part, the technician should provide 'Thumbs Up/Down' feedback on EMA results to make the system better and better over time.

Note that, on top of the above 'Standalone EMA Query User Interface', EMA also provides an integration with Enterprise Asset Management systems like Maximo. That integration enables the technician to:

• Click a 'Seek Advice' button on any given work order, which results in:
• EMA providing a list of historical work orders that addressed a similar issue (on this, or all, or similar assets) and which information could help accelerate the work at hand.

Simple enough?

Well, maybe, but what happens in the background is actually very smart and involves some of the most advanced AI technologies out there - let's see what EMA has under its hood.

What is Equipment Maintenance Assistant?

EMA is available as a Cloud SaaS solution, and not only includes ...

• Watson Discovery Service (WDS): this is the 'Questions-Answers' engine. It uses Natural Language Processing/Understanding (NLP/NLU) techniques similar - but more mature - to those that made the 2011 Watson Jeopardy win possible.
• Watson Knowledge Studio (WKS): does your Industry use domain-specific jargon that only your experts talk and understand? Well, this is why you would use a so-called custom NLU model. WKS is the place where you'd build it, and WDS can use any custom NLU model you created in WKS. As an example, using a custom NLU model would allow EMA to understand that if one asks: "What SD needs I to get thy hotty P3 workiiin asap, man?" (typos intended!), the actual question really is: "What screwdriver should I use to fix the third pump heat issue quickly, sir?".
• Watson Assistant (WA): EMA comes with open APIs (as well as WDS and WKS). That means that using EMA's WA component, you can create chatbots using whatever statements the APIs provide.

... but also includes:

• Diagnosis Model Manager: this is really unique and is the engine that drives your 'Symptom - Cause - Repair' Diagnosis. This is where you define symptoms and causes, their interrelations (including a probability for each), and the associated repair instructions or other supporting documents. Using a fit-for-purpose probabilistic model that was developed in IBM Research Labs, EMA's cognitive Diagnosis (a) is very dynamic in the sense that the model is constantly re-trained using the feedback received, (b) integrates directly with EMA's Query engine, which enables technicians to further interrogate in plain English the corpus of information related to the symptoms and causes at hand.
• Knowledge Graph Manager: though this was not covered in this article, EMA Knowledge Graphs provide the ability to visually see relationships between assets and entities, e.g. see what location an asset sits in.

Why should you consider using this type of tool?

With its Query and Diagnosis capabilities, EMA enables field technicians to faster diagnose asset issues and to accelerate their 'relevant fix documentation' searches and gathering. This results in reducing the time spent on maintaining or fixing assets, thus improving key maintenance KPIs like Mean Time to Repair (MTTR), First Time Fix (FTF) or Mean Time Between Failure (MTBF). EMA also helps address knowledge gaps by providing all your technicians with recommendations that ensure the best repair - this is especially important for younger technicians as older, more experienced ones retire. By providing the most likely (and very unlikely) causes of a problem, the usage of EMA can also result in less replacement of asset spare parts that are actually in working order.

As a side effect of all this - faster, better, safer and more consistent repair -, your assets stay under maintenance for shorter periods of time, thus staying in 'up and running' state longer. This can result in considerable cost reductions for both maintenance and operations!

Thanks for your time, hope you enjoyed this article. If you would like to know more about EMA and how you could use it, do not hesitate to contact me via LinkedIn.

About the Author: Christophe Lucas is a fan of Miles Davis who likes good literature and finely crafted evolutionary sound waves. He loves discovering new things and regularly dreams of a world where - thanks to humans and AI - assets would be nicer to their surrounding environment and run in a safer and smarter way.