Contact
By /

Case study · Manufacturing

Plant productivity, calculated automatically, line by line

How much each line produces compared with what it could. Why it loses performance. And when it is about to make a defective part, before it happens. With AI built on the data your plant already generates.

Real-time OEE

AI predictive quality

Recycled material under control

The starting point

Everyone knew which line underperformed.
Nobody knew why.

A plastic packaging manufacturer runs several production areas (injection, blow molding and printing) across multiple shifts. The data was there: output, scrap, quality. But it sat split between the ERP and the plant’s in-house MES, with different structures that were hard to reconcile.
Calculating productivity was manual work. Spreadsheets, cross-checks, hours. And the result came too late, once the shift had ended and the deviation had already turned into scrap.

The challenge

Output, scrap and quality data spread across ERP, MES and machines, with no shared definitions.

Productivity calculated by hand and too late to act on.

Performance drops were visible, but their causes were not: downtime, cycle time, inactive cavities or dropped parts.

Reactive quality: defective parts were only detected once they had been made.

A new injection blow molding line that had to be under control from day one.

What we proposed

We built a single, automated data model that combines ERP, MES and machine data. On top of it, we set up a total productivity KPI (actual output against theoretical maximum output), broken down by its drivers, along with scrap and quality tracking for each production area. We then connected the new line in real time and deployed a machine learning model that sends an alert before the machine starts producing defective parts.

What changed

From reconstructing the last shift to correcting the current one

Before

Productivity calculated by hand, days later

ERP and MES with different structures that could not be reconciled

Performance drops were visible, their causes were not

Defective parts discovered at the end of the line

Now

Productivity and scrap refreshed every hour, with no manual calculations

One model combining ERP, MES and machine data, with validation checks and error alerts

Every loss broken down: downtime, cycle time, cavities and dropped parts

Alerts before the machine produces defective parts

connected production areas (injection, printing and blow molding)
0
quality statuses under control (non-conforming, quarantined, scrapped and recovered)
0
AI model that learns every day
0

The value delivered

What your plant gains, shift by shift

Plant productivity, every hour

Actual output against theoretical maximum output, line by line, without a single manual calculation.

The root cause of every loss

Downtime, cycle time, active cavities and dropped parts, broken down to show exactly where performance is lost.

Scrap measured at the source

Kilograms and units for each production area, including non-conformities, quarantined, scrapped and recovered product.

One version of plant data

ERP, MES and machine data in a single model, with the same definitions for every area and shift.

Data that validates itself

Every load goes through data quality checks and flags any issue before it reaches the dashboard.

AI predictive quality

The model predicts when the machine is about to produce defective parts and alerts the production team in time.

What the client says

We knew one line underperformed, but not why. Now we see the productivity of every line every hour, with the cause of each loss, and the system warns us before the machine starts making defective parts. We have gone from explaining scrap to preventing it.

Operations Management

Packaging manufacturer

Common questions

Frequently asked questions

What is OEE and how can it be calculated automatically?
OEE (Overall Equipment Effectiveness) measures how much of the planned production time a line spends making good parts at its theoretical speed. It is calculated as Availability × Performance × Quality. To automate it, machine downtime, cycle times, active cavities, non-conformities and scrap need to live in a single model that refreshes on its own. At BertIA, an AI consultancy based in Barcelona and Madrid, we build it on the data the plant already generates in its ERP, MES and machines.
The MES records what happens on the machine, but it does not combine that with ERP data or break each loss down by its cause. A productivity dashboard brings ERP, MES and machine data together with the same definitions for every area and shift. That shifts the conversation on the shop floor from “this line is underperforming” to “this line is losing performance on cycle time during the night shift”.
A machine learning model learns from machine and quality history which combinations of variables come before defective production, and sends an alert before it happens. It is validated on data it has never seen before going live, retrained every day with new production data, and it shows the main causes behind each alert. The model raises the alert; the decision stays with the production team.
No. The data layer is built on Azure and connects to existing systems and to the data the machines already generate, without replacing anything or stopping production. The MES and ERP keep doing their job; the productivity model reads them, combines them and validates every load automatically.
A plant that already records production in an MES or ERP, access to those sources and one pilot line. One or two diagnostic sessions on site are enough to map lines, data sources and where performance is being lost. The pilot puts automated OEE in place on one line before rolling it out to the rest.
Yes, as long as plant data is unified and governed. On top of a single model combining ERP, MES and machine data, a conversational agent in Microsoft Teams answers questions about productivity, scrap or downtime using the same figures as the dashboard.

Your next project starts here

Email us
Call us