There are many misconceptions about data, big data, AI, Industry4.0 and other terms often used as buzzwords. If you're not immersed in it, this creates confusion and makes it hard to see the forest for the trees. Hence this blog, to get a bit more practical.
The thread running under these pitfalls: AI only works on connected data. That is exactly what a MOM platform produces: data with context, not more loose numbers.
1. We need to collect all data!
While this is certainly true in a sense, there are some nuances to it. Yes, you don't want any data points or blind spots, because these could be relevant for a use-case that surfaces within 2 years. However, that's not where the pitfall lies.
What companies often run into is that they collect a lot of data because yes, data is the new gold, but once an analyst wants to solve a problem with it, things grind to a halt.
- Is this time series data anomalous because there was maintenance?
- Was there actually a breakdown?
And yes, there is an Excel file somewhere that contains a downtime with its cause. Another Excel has maintenance scheduled and yet another has production issues from the operators.
The engineers, analysts, whoever wants to do something meaningful, has to start by adding context, normalizing data, linking, labeling. Not to mention adding a bunch of assumptions. This alone takes weeks to months before any value is even created, namely investigating or analyzing a problem.
The result is one of two outcomes:
"We're giving up, AI and analytics aren't affordable"
OR
"Data analysts spend months on a small project"
So yes, collect all data, but provide all necessary context from the edge, the place where data is collected.
- Manage maintenance interventions digitally (and no, Excel doesn't count)
- Register downtime with root causes for later analysis
- Provide context, tags and log product codes, machines linked to your machine data.
By logging data with context, you save a lot of time for later analyses. Your next AI initiative will take days instead of months.
Strategy
At its core, the problem is often a lack of a digital strategy. Moving data from paper to Excel is digitizing something, but doesn't make you a digital company. Just because you no longer use paper doesn't mean you're a digital company. The difference lies in the overarching strategy. The realization that data is the greatest asset in your company.
2. Start with a use case
The goal of IIOT is not data logging. It's solving problems faster, gaining insights, and making informed strategic decisions. We don't digitalize for the sake of digitalization, but from a strategy to become more efficient.
Choose a first problem to solve:
- Predictive maintenance?
- Anomaly detection?
- Energy optimization?
- Quality prediction?
- Operator assistance?
Don't start setting up platforms blindly and expect all insights to flow from them. Ask for advice, get support!
No idea where to start?3. Wrong technology or architecture
Sticking with technology you've paid a lot for. This is one of the common pitfalls. If the technology doesn't scale, doesn't allow you to log contextual data, then you're on the wrong track.
Architecture
Ask for help to set up your data architecture. Get guidance but stay involved. Unified namespace is often used as a buzzword nowadays. There's also a lot of confusion around it, with one platform claiming to be UNS ready, while another claims to be a UNS itself. The UNS architecture is a solid principle and enables context collection (fewer data silos), but it's quite complex to start with without experience. It's not just 'installing an MQTT broker'.
Platform
Platform vendors also believe that everything can be done in their tool. Your solution is often a combination of tools and systems that are each good at something. We help choose and combine the right systems. Sometimes a piece of custom software can be the most cost-effective solution.
By testing systems daily, running into issues, and trying again, we can also help you make the right choices. Completely vendor independent.
After several Industry 4.0 projects, I learned it the hard way: data by itself is worthless.
Without structure and context, three years of logging temperatures, vibrations, speeds, ... is little more than digital noise, and you get to start over.
In my first ML projects, most of the time didn't go into the model, but into annotating, adding context, and normalizing.
Now I always critically examine the 'as-is' situation, so we can deliver real value faster.
Sometimes taking a step back is the fastest way forward.
Thomas Willemot
3 Things You Need to Do Now
If you want to successfully deploy AI in the future, you need to start logging data with meaning, structure, and relationships today.
1. Capture context at the source
- Add metadata directly during logging: machine, product code, shift, cause. Let edge devices and operators help.
- These MES modules can help you provide the context, starting with production monitoring.
2. Break down data silos
- Let systems talk to each other. One cohesive view of your processes is more valuable than ten separate snapshots. AI can only find patterns if the puzzle pieces fit together.
- AI can't do anything with isolated spreadsheets, and neither can your analysts.
3. Make data quality a strategic priority
- Assign data owners, monitor quality, and determine which data is truly important. More data isn't automatically better. Better data is better.
- Agree on how you name things and document it. If you do it right now, you can reuse the same data years later without starting over.
- Structure and label for reuse.
Result: Instead of months of data cleaning, your team (and your AI agents) can deliver insights in days or hours.
Learned something from this blog or have a different perspective? Feel free to reach out! 👋 Need help or spotted an error? Let me know, and I'll gladly fix it.