Atmora Tech

Industry

Line data that reaches the ERP within the shift

Getting the plant floor and the ERP to agree on what actually happened during the shift.

Context

Technology in Manufacturing

Our manufacturing work sits at the ISA-95 boundary: collecting data from PLCs, historians and quality stations, turning it into events with meaning, and delivering it to MES, ERP and planning without a spreadsheet in the middle. We have connected machines that speak OPC UA and machines whose only interface is a serial port and a printed manual from 1998, and the second category is usually the majority.

The measure of success is that a supervisor and a finance controller can look at the same production number without arguing. That means reconciling counts, scrap and downtime reasons at the source rather than patching them in a report. A unified namespace will not fix a line whose instruments disagree: if the ultrasonic level gauge and the weighbridge differ by four per cent, spend the money on calibration first, because no amount of streaming architecture repairs a wrong measurement.

37 min
unplanned downtime removed per line per week at Kestrel Industrial
6
controller vendors bridged into a single tag model at Vantara Metals
2.1B
tag readings ingested per month with store-and-forward at the edge

Challenges

What actually keeps this sector awake

Not generic disruption — the specific constraints that shape every technology decision here.

  • A PLC estate spanning twenty-five years

    Five controller vendors, three fieldbus generations, and cells whose vendor support ended a decade ago. Any change risks stopping production, so the safe default becomes another isolated data island and another manual transcription step.

  • MES and ERP telling different stories

    Production booked against a work order in ERP does not match what the line counted, because scrap, rework and partial batches are recorded in different places with different rules. Month-end becomes an investigation rather than a close.

  • Unplanned downtime with no usable history

    Fault codes are captured but reasons are entered as free text, so the top failure category is 'other'. Predictive maintenance projects stall because there is no labelled failure history to learn from, only alarms.

  • Product-level emissions and compliance reporting

    Customers and regimes such as CBAM ask for embedded emissions per product, not per site. That requires bill of materials data joined to metered energy at the process step, which almost no plant records at that granularity today.

Approach

How we address them

  • A unified namespace that operations can read

    MQTT Sparkplug and OPC UA into one hierarchical tag model, edge gateways with store-and-forward so a network outage does not lose an hour of production, and a naming convention agreed with the people who maintain the line.

  • MES and ERP reconciliation at source

    One definition of a good part, scrap and rework, applied at the point of capture and posted to ERP as it happens. Variance reports show the specific station and time window, so investigations take minutes rather than a day of exports.

  • Condition monitoring with an honest alarm budget

    We agree an acceptable false alarm rate with maintenance before modelling, because a predictor that cries wolf twice a week gets muted in month two. Vibration and current signature models are validated against real failures, not synthetic labels.

  • Product carbon accounting from real meters

    Sub-metering at the process step joined to routing and BOM data, with allocation rules documented and reproducible. Output is a per-product figure your customer's auditor can trace back to a meter reading and a timestamp.

Services

Relevant practices

Products

Platforms used in this sector

Proof

Work in this sector

Start a project

Tell us what is
breaking.

We reply within one working day, and the first call is with an engineer who would actually work on it — not an account manager. If we are not the right studio for the problem, we will say so on that call.

Start a project