Is Machine Learning In Manufacturing A Joke?
One of the hottest buzzwords in any industry right now is artificial intelligence. In fact, trillions of dollars will be made by businesses over the course of the next decade who leverage this world-changing technology to disrupt their respective spaces. It offers the core solution to intimidatingly large data and complex ecosystems with revolving variables. AI and IoT are being viewed by many as a silver bullet to existing industry problems. Unfortunately, these solutions mostly aren’t here yet – for any industry – and smart businesses are forced to proceed forward with caution.
The manufacturing space is no exception to this phenomenon. Artificial intelligence and machine learning in manufacturing are talked about at every industrial conference; yet implemented on very few actual shop floors.
This groundswell would have you believe that this industry is on the verge of having every major manufacturer equipped with a fully-functioning smart shop floor; complete with Watson writing songs with Bob Dylan in the corner. Dare I say that this isn’t the case?
The Difference between an Algorithm and Machine Learning
Depending on which data scientist you argue with, artificial intelligence is a subset of machine learning (some would argue the opposite). It doesn’t matter though. Machine learning is the process by which machines learn from data and work out solutions. Artificial intelligence is a specific type of learning that mimics the processes of neurons and the methodology of the human brain.
This is important to understand because every software marketer in the manufacturing space is loading up trade show banners, display ads, and templated sales emails with this terminology. Why? It’s the industry’s hottest buzzword, and truthfully, these marketers and salespeople likely don’t even know that the technology they’re pushing actually isn’t anything close to actual machine learning or artificial intelligence.
So what’s the deal?
The truth is that most technology that is actively available today, or even implemented on shop floors, is primarily driven by algorithms; not true machine learning. What’s the difference?
The difference is simple. An algorithmic solution to a problem works like this.
- There is a bunch of data that gets dumped into an algorithm
- A data scientist or expert builds the algorithm according to rules that will deliver the desired outcome
- The data is organized, optimized, and parsed according to the way the rules were written
- The outcomes are directly a result of how the algorithm was written
A machine learning solution to a problem works like this.
- A bunch of data is dumped into a machine designed to solve a problem
- The machine learns from the data over time and writes its own rules to ultimately deliver the outcome that was desired.
In an algorithmic solution, a human must continually write rules and try to understand what’s important in a system, and what not, to generate an optimal outcome. As data gets bigger, the algorithm must grow, and the insights required to keep it up to date get harder and harder to scale effectively.
With machine learning, the machine learns as more and more data is accumulated and the machine works to understand trends and outcomes. It will effectively scale itself and write its own rules to optimize outcomes.
As you can probably tell, between the two, a machine learning solution will likely deliver far better outcomes over time, cost less to maintain, and offers the benefit of being self-sustaining. This makes it the desired choice between the two. Unfortunately, it’s currently not available in a way that any manufacturer can really implement to its fullest potential today.
What are Technology Providers Selling People Then?
As most of the smart guys in this space already know, there’s a lot of marketing behind this tech but very little real implementation. So, you might wonder, what are these major platforms and technology providers selling when they’re talking about machine learning and artificial intelligence?
In most cases, as you might guess, they are offering up sophisticated algorithmic solutions (or maybe unsophisticated, who knows). The truth is, the solutions are really only as good as the data that goes in. Algorithms written by humans, filled with imperfect (fake) data are unlikely to help many businesses.
Most manufacturers know this, that’s why few are pulling the trigger on these major implementations. In fact, most are happy to listen to Deloitte pitch the future of these projects but know deep down that the project would costs hundreds of thousands (if not millions) and would likely never work the way it worked in the slide presentation.
This Stuff Has to Happen in Stages
It is nearly impossible for most manufacturers today to go from where they’re at right now to a complete digital transformation with operational machine learning embedded in their manufacturing processes. They know this. Systems are disconnected, some data is still collected manually in spreadsheets, and even if the data is all centralized in one place, it isn’t currently being interpreted or distributed to the people that need it in a way in which they can use it.
This all means that the jump from where most manufacturers are today to a full-fledged smart factory 2.0 is a massive one. There have to be steps in between these stages.
Major integrators and consultants would be more than happy to quote you a price for taking you from where you are today to that smart factory promised above, but we both know what that will look like. Even if you can foot the bill, does anyone want to sign up for that project…. yeah, absolutely no one ever.
We Can Laugh Now, but we have to Start Taking Steps Forward
As comical as it can be to sit back and poke fun at some of the industry buzz around machine learning and artificial intelligence today, the actual future of this technology on factory floors is no laughing matter. It ultimately will be a solution that can solve complex problems for manufacturers.
The biggest challenge manufacturers face today is that they are largely unprepared for this future. This means they will be forced to play catch-up in the future or will ultimately fall behind when competitors outpace them in the race to smarten up. The act of catching up from where most manufacturers are today will be expensive, uncomfortable, and complex.
The solution to this problem is to begin taking the steps to improve manufacturing and shop floor processes now. This will make future implementations easier, faster, and less uncomfortable. The good news is that most of these early steps are affordable, short projects that deliver ROI right now.
What Projects Can Prepare Manufacturers for the Future?
Most manufacturers I speak with can do a small handful of simple projects now to set themselves up for future digital transformation and a smart shop floor in the future. These projects offer the ability to influence things like uptime and performance right now; while reducing waste and overtime.
All of these projects have one thing in common… centrally collecting data!
There are few things a manufacturing business can do right now that will prepare them for a smart future quite like the centralization and organization of machine data and information. Even if this is still manually collected, getting all of the data in one place – where it can be contextualized and combed for insights – is massively important.
Here are a few other small, simple projects that will help manufacturers enjoy short-term ROIwhile also moving them in the right direction towards real machine learning (complete with some links to where you can learn more about them).
- Reduce reliance on paper
- Centralize important data and make it accessible to all
- Analyzing data trends in real-time
- Removing focus on “acceptable” OEE numbers
I’ve seen firsthand just how beneficial these types of projects can be. They are low-cost, high ROI projects that manufacturers are usually happy to sign-up for because they are a welcome change to shop floor employees and quickly adopted by management.
Interested in learning more about how these projects can help you prepare for a future filled with machine learning? Take a look at our recent manufacturing analytics white paper.
Until then, enjoy the conferences. I’ll see you there 😉