By now, you must have understood my slight obsession towards machine learning (ML). Well, it is pretty apparent, by the way; I tend to always probe into the ML usage and related aspects in my interviews.
Lately, during a conversation, I spoke to Oliver Sturrock, CTO at Fluke Digital Systems. Oliver leads the engineering unit at Fluke accountable for constructing their interconnected reliability framework, called Accelix. Their goal is to clamp machine learning health data reservoirs and facilitate predictive analysis.
We conversed about Fluke’s advancement and how they have been handling the tools because it hasn’t been that long since the whole setup shifted into the cloud.
Oliver affirmed that they have been working and stated a few facts on how ML helps in measurement of data combined with the right algorithms are helping in mimicking a person’s decades-long “tribal knowledge.” And how it has been enabling them to predict equipment failures and helping schedule resources and maintenance for care ahead of time.
Though there are considerations…
To begin with, the technology explanation can’t skim like magic. When you notice a black box, all you can do is react to what it decides.
That’s just the kind of scenario that interests numerous clients, forcing them to stick to their direct contact with data. In order to embrace a new solution, one needs to be able to trust them, and the quality of incoming data is the key part of the trust.
OEMs are printing big data clusters that describe what their assets look like when functioning normally. Systems like Fluke Accelix should be set to incorporate all that data into their model and correspond it to field situations, baselines at structure, etc.
But Oliver called out that there are a lot of environmental aspects to account for, from the difference in the performance to how well it has been maintained. So, Fluke aims to cover 80% of the course there with out-of-the-box data, and then the remaining 20% has to come from infield understanding.
This implies even if you encounter a black box, you need to work on it to make it as see-through as possible so that it’s not the box but you who gets the right to take the decision.
What I feel the only frustration companies would encounter or are bound to encounter is the proprietary data streams that don’t API. I believe it shouldn’t be that way. Instead, it should be built around APIs and interconnectedness for better results.
With that said, Oliver empathizes that Fluke has to encounter such an environment and how bad he feels that Fluke can’t possibly build every type of analytics and data monitoring that’s ever going to be needed. But he also asserts that he believes Fluke surely has a role in bringing all together at one place down the lane.
I believe there’s a very valiant future ahead of us. I value that the systems around me are getting safer, irrespective of where I go. Easier maintenance and reliable SAFER operating conditions are surely leading to a better world.