IFS acquires Falkonry AI to make it easier to find anomalies on the shop floor and beyond

Last week, enterprise applications provider IFS announced its acquisition of Falkonry AI, a startup that provides automated, self-learning anomaly detection across data from assets, machines, systems and industrial processes. This helps companies detect and investigate unusual behavior, which is often a precursor to equipment breakdowns or process failure. Yesterday I spoke to Christian Pedersen, Chief Product Officer at IFS, to find out what this means for customers of the company's IFS Cloud product set, which automate core operations, asset management and maintenance in industries such as manufacturing, energy, telecoms, defense, construction, logistics and facilities management.

The new acquisition, which is expected to formally close later this year, adds to IFS's existing set of AI-based tools, which provide simulation, scheduling and optimization to aid in planning what resources are needed, where and when. Falkonry's role is to discover the anomalous signals that indicate these best-laid plans are about to go awry. Pedersen explains:

You can put all these great plans in place, but then stuff happens. Stuff doesn't go exactly to plan. Historically, the way that the world has dealt with that has been really comprehensive machine learning projects, where you train machine learning to look for certain things and find these different things.

That's where Falkonry really comes in for IFS. Machine learning has its place, with supervised learning of the machine learning engines and training and all that stuff. But we've always thought that for the vast majority of the customers we work with, it is still too cumbersome to go through all that training.

That's where Falkonry have created this phenomenal self-learning capabilities in the engine, where instead of looking for what is wrong, it is basically identifying what is right. And then it's alerting us when there's something that is not as it normally is.

In other words, by analyzing the data produced during normal operations, Falkonry teaches itself what to expect when everything is working as it should. Once it has trained itself, if it then detects variations or anomalies that fall outside the norm, it sends an alert, which either generates an automated response, or goes to a human operator for further investigation. Pedersen gives a simple example based on data coming from a manufacturing shop floor, whether from a Manufacturing Execution System (MES) that controls the shop floor as a whole, or an individual Programmable Logic Controller (PLC) used to control equipment:

You have an MES or a PLC that runs on the shop floor that constantly streams data. The data is, of course, based on all the sensors that are on it. That machine works fine ... and because it has been working perfectly Falkonry has a clear view of what normal is for this machine right now.

Now suddenly, one of your vibration sensors starts creating more vibration. And vibration is the root of, I don't know, 80-90% of things that go wrong in a manufacturing machine. So they can now start identifying that 'Oh, we have more vibration here than we normally have.' And then it triggers an alert.

Of course, the first time you get an anomaly that you hadn't seen before, then, either through AI or human interaction, there'll be [a response] that says, 'Okay, because of this type of alert, this is the type of action that should then happen.'

Extending the reach of anomaly detection

Falkonry has come a long way in getting its product to a point where it can be harnessed for real-world applications, he adds, noting that he has been in touch with the company for the past two years. He goes on:

You're literally talking about, today, the Falkonry team going in, working with customers for a couple of weeks, before the system then starts self-learning based on the data. That's one of the simplest and easiest adoptions I've seen in this space at all, so we're very excited about it. We think it's a perfect fit for the spaces we're playing in, in asset management and manufacturing.

But those customers have still had to connect the AI engine to the underlying data and then define what happens next when it detects anomalies. Becoming part of IFS will change that, as he explains:

By bringing it into the IFS Cloud world, we will have these data connections out-of-the-box, because we're already building them with our MES systems, our asset systems and our IoT systems and all that stuff. And we have the outcome-based management of actions — should it trigger a work order, should it trigger a repair order, or should it trigger a change in your manufacturing schedule to send this manufacturing order to a different PLC than this one? All these different things. So again, it's one of these examples where we take a lot of the pain away from our customers in making things work together.

As well as the obvious use cases for anomaly detection in activities such as asset performance management, manufacturing execution systems, servitization, and configurable workflows, there are other potential use cases that IFS plans to explore. These build on its strategy of providing a unified data layer within IFS Cloud that brings together traditional business data and asset data with streaming data from the shop floor, IoT and elsewhere. He goes on:

The explosion of data is not stopping here. Because the next set of data that becomes really important from an AI perspective is the data on how people actually work, data about how processes are progressing. In other words, everything that goes on in a system ...

We are really looking to apply anomaly detection in areas where people don't think about it. Because now we can apply — again, back to the data layer — now we can apply anomaly detection in process progression. That's to say that, you have a manufacturing process that is not going as it normally would, we can use anomaly detection to identify that and come up with alerts early on. Or we can use anomaly detection in finance, and in many other areas ...

This is an area that really excites me, because nobody has applied anomaly detection in any way, shape, or form to these other scenarios — especially when you can start capturing this process data and so forth, or telemetry data, if you will, or even down to user data, like how will users actually interact?

A role for generative AI

Although this acquisition doesn't add any generative AI capabilities, that technology comes into the picture as one of the options when it comes to deciding what actions to take once an anomaly has been discovered. He explains:

We see generative AI as a natural progression, next step after anomaly detection, because that's where you should generate some action, or generate some material, or generate a work order.

Here again, IFS deploys the technology within its orchestration layer to provide the controls and safeguards customers need. This will become increasingly important as more specialized generative AI models emerge for specific industries or applications. He explains:

We are leveraging Microsoft Azure OpenAI, but we ... help them manage that in that layer. While we're leveraging some of the OpenAI scenarios out of the box, we also believe that the generative AI space with Large Language Models [LLMs] will have hundreds of different language models out there. They won't only come from OpenAI. So that last part of our AI strategy is to be able to manage these third party LLMs as well, that we don't even know what are yet, but we know we we have to enable our customers to manage it very responsibly.

Having this in the same orchestration layer as other AI technologies makes it possible to apply each where they're most appropriate. He adds:

Now you're in that world where you're combining rules-based automation with AI-based automation. That's another thing we bring to the table. We don't believe it's one or the other. Sometimes there are hard rules, because it can be regulatory and there could be other things. Other times it can be AI-based, what should actually happen. Most importantly, they need to interact with each other ...

We provide these capabilities out-of-the-box to our customers. So they don't have to go through expensive parallel projects, and then integrate a lot of different things.

My take

This is a useful addition to the growing portfolio of AI-based tools that IFS offers its customers. Anomaly detection is a long-established branch of AI, but Falkonry seem to have a fresh approach that makes it much simpler to harness the technology for a whole host of scenarios. Manufacturers I've spoken too often lament their lack of tools to usefully analyze all the data they have available from across their asset base and operations. They will appreciate the value of a tool that can analyze that data to make a practical impact on operational reliability, manufacturing quality or customer experience.