Industry Insights
& News
Expert perspectives on industrial automation, machine vision, and the future of smart manufacturing.
Smart Manufacturing in Practice: How AI, Managers and Maintenance Engineers Found the Real Root Cause
Smart manufacturing earns its keep on the day a line keeps stopping and nobody can agree why. It is not about adding more screens to the shop floor. It is about connecting machines, data and people so the right person sees the right evidence quickly.
The example below shows how that works in practice. A production manager, a maintenance engineer and an AI analytics layer work together to find the real root cause of a recurring fault, rather than the most obvious one.
The scenario is illustrative, based on the type of problem regularly seen on UK production lines. Names and figures are representative.
The Problem: A Line That Kept Stopping
A food manufacturer in the Midlands runs a high-speed flow-wrap and case-packing line. Over six weeks, OEE on the line fell from 78% to 63%. The main cause was a recurring “product jam at infeed” alarm, stopping the line 12 to 15 times per shift.
Everyone had a theory:
- The Production Manager, Sarah, believed it was an operator issue, as most stops seemed to happen on the night shift.
- The Maintenance Engineer, Dave, suspected the infeed photo-eye. He had already replaced it twice and re-tensioned the conveyor belt.
- The Quality team thought a new batch of packaging film was the problem.
Each fix worked for a day or two, then the fault came back. The team was treating symptoms, and the cost of lost output was rising every week.
How the Team Found the Answer
Instead of another round of guesswork, Sarah asked for one question to be answered with data: what is actually different when the line jams?
- Connect the data. The line’s PLC already logged alarms, servo data and production counts. These were linked via OPC UA with the OEE system, the vision inspection camera, the maintenance records (CMMS) and the shift rota. For the first time, all of it sat on one timeline.
- Let AI look for patterns. The AI model analysed six weeks of data, around 1,900 jam events. It compared conditions just before each jam with normal running.
- Rule things out. The AI showed the jam rate did not change after either photo-eye replacement, so the sensor was not the cause. Jams also occurred across all film batches, which cleared the packaging.
- Find the real signal. Two patterns stood out. Jams rarely happened in the first 90 minutes of running. And in the 30 seconds before a jam, the infeed servo’s following error rose sharply, while the vision system measured product spacing becoming uneven.
- Explain the night shift. Night shifts ran longest without a break, so the line spent more time in the “hot” condition. The operators were not the problem; run time was.
- Verify on the floor. The AI pointed Dave to the infeed drive train, not a single part. He took vibration readings and a thermal image after two hours of running. The infeed gearbox was running 25°C hotter than its twin, with a rising bearing vibration signature.
The Real Root Cause
The jams were caused by a failing bearing in the infeed gearbox. But the team kept asking “why” until they reached something they could prevent:
- Why did the line jam? Product spacing at the infeed became uneven.
- Why was spacing uneven? The infeed servo could not hold its position once the drive warmed up.
- Why could it not hold position? A bearing in the infeed gearbox was worn and adding friction as it heated.
- Why was the bearing worn? The gearbox had missed two scheduled lubrication services.
- Why were they missed? The site had moved from two shifts to three, but the planned maintenance interval was still set by calendar days, not machine run hours.
The root cause was a maintenance planning gap, not a sensor, an operator or the packaging film.
The fix:
- Dave replaced the bearing and serviced the gearbox in a planned 3-hour window.
- Lubrication and inspection intervals in the CMMS were changed from calendar days to run hours, fed directly from the PLC.
- An automatic alert was added: if the infeed servo following error or gearbox temperature trends upward, maintenance is notified before a jam occurs.
What the Manager Sees
All of this comes together on a single root-cause dashboard: the AI finding, the theories ruled out, the evidence charts, the 5 Whys and the action tracker.
| Theory | Raised by | Evidence | Verdict |
|---|---|---|---|
| Night-shift operators | Production | Tracks run time, not crew | Ruled out |
| Infeed photo-eye | Maintenance | No change after 2 swaps | Ruled out |
| Packaging film batch | Quality | Jams across all batches | Ruled out |
| Infeed drive train | AI analysis | Servo error + run-time pattern | Likely · verify |
| Action | Owner | Due | Status |
|---|---|---|---|
| Vibration + thermal check, infeed gearbox | Maintenance Engineer | Today | Done |
| Replace bearing, service gearbox (3 h window) | Maintenance Engineer | Sat AM | Planned |
| Switch PM intervals to PLC run hours in CMMS | Production Manager | Next week | In progress |
| Alert on servo error / gearbox temp trend | Controls / IT-OT | Next week | Not started |
The Technologies That Made It Possible
None of this needed a new factory. It needed the existing systems to talk to each other.
- Industrial automation: The PLC and servo drives were already recording the critical signals, such as following error, motor current and alarm timestamps.
- Industrial IoT and integration: OPC UA and Modbus connections brought PLC, OEE, vision and maintenance data onto one timeline.
- Machine vision: The inspection camera measured product spacing, giving an early warning that a human could not see at line speed.
- Data analytics and AI: The AI sifted thousands of events, ruled out false leads and highlighted the conditions that came before each failure.
- Traceability: Every stop, intervention and part change was linked to time, shift and batch, so the team could prove what changed and what did not.
AI Did Not Replace Anyone
The AI found the pattern in hours, but it did not fix the line. Each part of the team played a role:
| Who | What they contributed |
|---|---|
| Production Manager | Framed the right question, gave access to the data and owned the change to the maintenance process |
| AI analytics | Analysed six weeks of data, ruled out false leads and pointed to where to look |
| Maintenance Engineer | Used his experience to test the AI’s lead on the floor, confirmed the fault and carried out the repair |
| Operators | Were cleared of blame, and now get clear alerts instead of unexplained stops |
The AI narrowed the search. The engineer’s hands-on knowledge confirmed it. The manager made sure it would not happen again. That combination is what smart manufacturing really means.
The Results
Within four weeks of the fix:
- Infeed jams fell from 12–15 per shift to fewer than 2.
- OEE recovered from 63% to 80%, above where it started.
- No further photo-eyes or belts were replaced unnecessarily.
- Maintenance moved from reacting to breakdowns to acting on early warnings.
Lessons for Any Manufacturer
- The obvious cause is often wrong. Replacing the part that alarms is not the same as finding the root cause.
- Integration comes first. AI can only find patterns in data that is connected. Most factories already have the data; it is just sitting in separate systems.
- AI supports experts, it does not replace them. The best results come when AI insight meets engineering experience.
- Fix the system, not just the machine. The lasting win was changing how maintenance is planned.
Start Your Smart Manufacturing Journey
At Gomes Technologies, we connect PLCs, vision systems, OEE and maintenance data so your team can find root causes faster and prevent repeat failures. With over 18 years in industrial automation across aerospace, food, automotive and general manufacturing, we understand both the data and the machines.
Have a recurring fault you cannot get to the bottom of? Contact us to talk it through.