AI & industrial maintenance, a powerful match - Part 1

Published 2023-09-28 by Jean Silva (a 6.1 minute read) | Back to the main page

Introduction

The explosion of Artificial Intelligence and Machine Learning applications in the last 5 years has made its way through to many fields and equipment maintenance is not an exception. It is easy to understand that the prediction of a failure before it happens is an extremely appealing concept. Before production or safety or both are in any way impacted, some clever system will blow the horns with enough time to allow for proper planning for action to prevent a breakdown. A dream.

In a way, we're already living the dream but there's a few points that need to be addressed. Some of the technology that allows for all that AI makes possible is already here and it's in many ways rather mature. So why don’t we see it applied left and right? Well, let's dig in.

AI and Predictive Maintenance

Predictive maintenance (PdM) techniques have been around for a long, long time and in a way AI builds on those techniques and adds an additional and powerful level. With conventional PdM, each relevant parameter is collected from the equipment and looked at mostly independently from one another.

As an example, consider an hydraulic oil pump with both vibration sensors and an on-line hydraulic oil analysis monitor. Both methods will provide key insights about the state of the pump. Maybe oil iron levels are above a certain threshold, or a certain vibration frequency is above warning levels. Typical PdM will trigger actions based on actual measurable and usually analysis-independent conditions (means data is collected by a dedicated logger). Those conditions will be present when the damage is already there - with no chance for adjustment.

What AI adds to this scenario is that all parameters will now be considered at the same time including the historic component of the combination of them all. It's a multi-dimensional analysis that extends long into the past and that can easily include other not-so-common points like number of starts and stops and running hours. This way, potential problems can be detected much earlier when compared with PdM. So early, in fact, that possibly only minor adjustments are necessary to resume proper operation.

In short, PdM requires almost immediate recovery actions while AI provides room for proper planning of the repair or adjustment action.

All this power, not that much cost anymore

These techniques are not a replacement for existing maintenance types. What would be the point of an AI/ ML program on light bulbs? These new techniques are simply yet another tool to add to an organization's maintenance management arsenal. At this moment in time, he right candidates are key equipment for which a failure means impact to production and for which the cost of deploying an AI system is lower than the cost of a failure.

As with most tech though, the cost of deploying becomes lower and lower with time and thus the target application becomes more widespread. The technology becomes within reach of smaller companies.

Furthermore, when you associate the predictive capabilities of these systems as described above with an automatic call for action directly triggered on a CMMS/ ERP and then a feedback from the CMMS to inform the ML algorithm of repair-related information (like time to recover, fault description, etc), then the full loop from analysis to action to building an ever evolving body of knowledge is complete and the power of the technique becomes obvious. A fully automatic, evolving system - the only thing it doesn’t do is the maintenance work itself, but Elon is probably on it as I write this.

Current state of AI/ ML applications

At this moment there are a few suppliers that have commercially available products and services that implement solutions ranging from maintenance improvement to the overall manufacturing efficiency improvement, the latter takes into account a broader range of inputs like the continuous measurement of production output (think number of items made per unit of time, for example).

To imagine a fully automated AI overseer that guarantees maximum production is an extremely appealing idea. Yet, quite frankly, it is not easy to find an actual full-on application of these technologies. It all seems to be a little humble with a lot of praise for the potential of the concepts based on a number of pilot applications on more innovative companies. Other than perhaps a few maintenance improvement use-cases sometimes kickstarted by equipment manufacturers themselves, there are not that many reference projects in "normal" or "real" companies. But isn't it always the case with new technologies?

Conclusion

Clearly something powerful is happening that is bound to transform the way manufacturing operates, something that can actually save time and money and thus become a very important component of operations - but should you jump on to it at this moment? As with everything, it quite depends. What everyone on the maintenance field should be doing is to learn about these topics because some variation of what is currently available will most definitely dominate in the not so distant future.

If you are somehow involved as a supplier of data points as defined above - for example, you’re a pump manufacturer or a CMMS provider - you most definitely should be implementing connectors that allow your devices and systems to be linked to other devices or systems. Even if the protocol you work with doesn't survive, the underlying principles will, so it will all be time and money well spent.

If you're a maintenance manager and provided the financial conditions are there (I know, not always easy), I'd recommend running smaller pilot project(s) focused on improving the availability of critical equipment and then assess if the technologies, the budget and other aspects are actually worth the improvement, if any.

AI/ ML will undoubtedly be implemented on the production floor - it's just too good a concept to not translate into better, more efficient operations - as a maintenance manager, start now, start slow, keep updated, make informed decisions today and you will be prepared for that inevitable future. And that is the real challenge on all this - we're so busy taking care of the problems of the present that seldom do we have time to strategize for the future. If you want to be in the game 5 years from now, find the time, it will be worth it - good luck!

In part 2 on this topic (coming soon) we'll be addressing actual examples and results of industry implementation of AI projects including reference to some suppliers of these technologies. Stay tuned!