Shop Floor Control and MES Integration

Where the plan meets the machines. This is the boundary between ERP time (hours and days) and factory time (seconds and milliseconds).

The ISA-95 model

The standard vocabulary for the whole ERP↔factory boundary. Learn it — every integration conversation uses these level numbers.

Level 4  ERP / Business planning and logistics
         Orders, materials, costs, months–days
              ▲  production orders down │ confirmations up
              ▼
Level 3  MES / MOM — Manufacturing Operations Management
         Dispatching, execution, tracking, quality, days–hours
              ▲                        │
              ▼
Level 2  SCADA / HMI — supervisory control, seconds
              ▲                        │
              ▼
Level 1  PLC / DCS — sensing and manipulating, milliseconds
              ▲                        │
              ▼
Level 0  The physical process — machines, sensors, actuators

The rule: never let an ERP talk directly to Level 1. ERPs are not real-time systems, and a PLC cannot wait for a database transaction. Level 3 is the buffer that translates between the two worlds.

What ERP sends down

Data	Purpose
Production/process order	What to make, how much, when
BOM components and quantities	What to consume
Routing/recipe operations	The steps
Work centre / resource assignment	Where
Quality inspection plan	What to check
Batch and serial requirements	Traceability
Documents	Drawings, work instructions, SOPs
Priority and sequence	What to run first

What MES sends back

Data	Purpose
Operation start/finish	Progress and actual times
Yield and scrap quantities	Confirmation
Scrap reason codes	Analysis
Actual material consumption	Real, not backflushed
Batch/serial genealogy	Traceability
Labour time by operator	Cost and payroll
Machine downtime and reason	OEE
Quality results	Inspection lot results
Process values	Temperature, pressure, cycle time — for batch records

The integration contract

ERP ──▶ MES:   ProductionOrderDownload   (create/change/cancel/release)
               MaterialMasterSync
               BOMSync, RoutingSync
MES ──▶ ERP:   OperationConfirmation     (yield, scrap, times)
               MaterialConsumption
               GoodsReceipt
               DowntimeEvent
               QualityResult

Rules for this integration

Idempotency on every message. Factory networks drop; MES retries. A duplicated confirmation posts duplicate cost and duplicate consumption.
MES must run offline. The factory does not stop because the ERP is being patched. Queue confirmations locally and replay in order.
Order of operations matters on replay. Confirmation for op 20 before op 10 must either be accepted or explicitly rejected — decide which.
Reconcile daily. MES-reported quantities vs ERP-posted quantities, per order. Publish the differences every morning.
The ERP owns the financial posting. MES confirms physical facts; ERP books cost and inventory.
Version the message contract. MES upgrades and ERP upgrades happen on different schedules.
Backpressure. A high-speed line can generate confirmations faster than the ERP can post them. Aggregate at the MES (confirm per hour, not per unit) rather than melting the ERP.

When you need an MES

Signal	Verdict
Operators confirm on paper, entered once a day	ERP alone may do
You need real-time WIP visibility	MES
Machine data collection / OEE	MES
Electronic batch records (pharma)	MES, mandatory
Complex routing with dynamic dispatching	MES
Detailed labour tracking per operator	MES
Quality data collected at each operation	MES
Paperless factory / 21 CFR Part 11	MES

Many mid-size manufacturers run ERP-only shop floor control with tablets for confirmation. That’s a legitimate choice, and often the right one.

Data collection methods

Method	Notes
Barcode scanning	Cheapest, most reliable, still the default
RFID	No line of sight, bulk reading
Touchscreen terminals	Confirmations, downtime reasons, quality
Direct machine interface	OPC UA, MQTT, Modbus — counts, cycle times, alarms
IoT sensors	Temperature, vibration, energy
Vision systems	Automated inspection
Manual entry	Last resort, and the least accurate

OPC UA is the dominant industrial interoperability standard for machine data; MQTT with Sparkplug B is common for lightweight IIoT telemetry. If your integration touches machines, these two names will come up.

Downtime tracking and OEE

downtime_event(
  id, work_center_id, start_time, end_time,
  category,      -- PLANNED | UNPLANNED
  reason_code,   -- BREAKDOWN | CHANGEOVER | NO_MATERIAL | NO_OPERATOR
                 -- QUALITY_ISSUE | MAINTENANCE | BREAK | NO_ORDER
  sub_reason, machine_code, comment,
  operator_id, order_id
)

Reason codes are the whole value. “The line was down 4 hours” is useless. “Down 4 hours: 2.5h waiting for material, 1h changeover, 0.5h breakdown” is an action plan — and it says the fix is in the warehouse, not in maintenance.

OEE calculation is covered in Manufacturing Types. The critical design point: you cannot compute OEE without downtime reasons. Capture them from the start.

Electronic batch records (pharma/food)

Regulated process industries need a complete, signed record of every batch:

Every material lot consumed, with its own certificate of analysis
Every process parameter with timestamps (temperature profile, mixing time, pH)
Every operator action, with electronic signature
Every deviation, with investigation and disposition
Equipment used and its calibration/cleaning status
Final QA release signature

Review by exception is the modern approach: rather than a human reading a 400-page batch record, the system flags only the parameters that went out of specification. It’s what makes electronic batch records economically viable, and it’s a real design requirement — you must define acceptable ranges for every recorded parameter.

Regulatory frame: 21 CFR Part 11, EU GMP Annex 11, GAMP 5 for validation. Systems in this space are validated, meaning a documented IQ/OQ/PQ exercise, change control on every deployment, and no untested hotfixes. Plan your release process accordingly.

Andon and visual management

Andon — an operator signals a problem; the line stops and help is summoned. Digital andon routes the alert by reason and escalates on a timer.
Digital shop-floor boards — real-time output vs plan, quality, downtime
Takt indicators — are we on pace?

Small builds, disproportionate operational value. A digital andon board that shows current output vs target is frequently the most-loved thing a developer delivers on a manufacturing project.

Labour tracking

labour_confirmation(
  operator_id, order_id, operation_number,
  activity_type, start_time, end_time, duration,
  quantity_produced, indirect_reason
)

Complications:

One operator, several machines — how do you split the time?
Several operators, one operation — team confirmations
Indirect time — cleaning, meetings, training, waiting. Charged to a cost centre, not an order
Payroll linkage — attendance vs order-confirmed time must reconcile, and the gap is “indirect”

Be careful here. Labour tracking is individual performance monitoring. In many jurisdictions it requires works-council agreement or employee consultation, and in the EU it engages GDPR. Design it with the ability to aggregate and anonymise, and involve HR before you build it.

Maintenance integration (EAM/PM)

Production and maintenance compete for the same machines:

Preventive maintenance consumes capacity and must be in the schedule, not a surprise
Breakdown creates an unplanned maintenance order and blocks the work centre
Condition-based / predictive uses IoT sensor data to trigger maintenance before failure
Maintenance orders consume spare parts from inventory and labour from technicians

The ERP must model machine availability as the intersection of production shifts and maintenance windows. Planning production into a scheduled maintenance slot is a common and avoidable error.