MESkit Simulator — AI-Driven Production Simulation for Training, Testing, and Learning

• MES training without a factory — Learn how a Manufacturing Execution System works by running a complete production cycle: define a product, configure a route, generate units, move them through workstations, and analyze quality results.
• ISA-95 education — The Simulator exercises every level of the ISA-95 hierarchy: physical assets (Level 0-2), product and process definitions (Level 3), production execution (Level 3), and quality operations (Level 3). Students see the standard in action, not just in a diagram.
• OEE fundamentals — Understand how Availability, Performance, and Quality combine into Overall Equipment Effectiveness. The Simulator generates the machine downtime, cycle time variance, and quality data needed to compute OEE.
• Quality management training — Watch defect patterns emerge, see how the Quality Analyst agent detects yield drops and defect clusters, and practice root cause analysis on simulated production data.
• Support team onboarding — New support engineers run simulated production to learn what normal operations look like, then encounter simulated problems (machine faults, quality drops) to practice troubleshooting.
• Dashboard and analytics testing — Generate realistic production data quickly to verify that charts, metrics, alerts, and reports display correctly under various conditions.
• AI agent evaluation — Observe how the Quality Analyst, Operator Assistant, and Production Planner agents respond to production events. Evaluate agent accuracy, response time, and decision quality.
• Client demos and proofs of concept — Show a live MES with realistic production data flowing through it. More compelling than screenshots or slide decks.

When you press Start, the Simulator Agent wakes up and enters a decision loop:

1. Read state — The agent queries the current shop floor configuration, product definitions, routes, and WIP status.
2. Generate units — Creates batches of units at the head of production lines, simulating incoming material with auto-assigned serial numbers.
3. Move WIP — Advances units through their route steps, respecting workstation capacity. If a station is backed up, the agent slows intake — just like a real production line.
4. Quality decisions — At pass/fail gates, the agent decides outcomes contextually: base yield around 95%, degrading over time, with realistic defect clustering. Failed units are scrapped or flagged for rework.
5. Machine events — Periodically toggles machine statuses: introduces faults after extended runtime, simulates repairs, and models maintenance windows.
6. Repeat — Waits for the configured tick interval, then reads state again and makes the next set of decisions.

Every action flows through the tool layer to Supabase, triggering Realtime updates that push to all connected clients. The live ticker scrolls with each event. Other agents — like the Quality Analyst — react to the data in real time.

The Simulator is a standalone add-on that calls the MES from the outside — like a test client or an external integration. It is not embedded in the MES core. This isolation is enforced by design:

• No simulation columns in MES tables — There is no is_simulated flag on units, no source column. The data model does not know or care whether data came from the Simulator, a human, or an MQTT sensor.
• No simulation conditionals in MES code — The tool layer, agent runtime, Realtime subscriptions, and UI never check if (simulation). MES behavior is identical whether the Simulator is running or not.
• Separate file structure — All Simulator code lives in its own directories. Deleting these files leaves the MES fully functional — it compiles and runs without changes.
• Public API only — The Simulator calls the same tool registry functions as the UI and the Operator Assistant. No internal shortcuts, no special access.
• OEE and analytics belong to the MES — Features like machine downtime tracking, cycle time measurement, and OEE calculation are MES features that work with manual data, MQTT data, or Simulator data equally.

Three input sources, one MES

The MES accepts data from any source through the tool layer. The Simulator is just one of them:

Dashboards, analytics, Realtime subscriptions, and agent alerts all work identically regardless of input source. No code changes needed to switch between them.

The top bar provides simulation controls:

Speed settings

Speed controls how fast the agent makes decisions. All speeds produce the same quality of data — faster speeds generate it more quickly.

The Simulator exercises the complete ISA-95 hierarchy. Here is what each level looks like during a simulation run:

Running a simulation from start to finish exercises every ISA-95 concept in the MESkit data model. For students, this provides hands-on experience that maps directly to textbook material and industry practice.

OEE is the gold standard metric for manufacturing efficiency. It combines three factors — and the Simulator generates data for all of them:

OEE = Availability × Performance × Quality. World-class manufacturing targets 85% OEE. The Simulator typically generates data in the 70-80% range, providing realistic scenarios for analysis and improvement exercises.

The Simulator does not work in isolation. It creates conditions that other agents respond to:

• Quality Analyst — Monitors production data generated by the Simulator. When yield drops below threshold or defects cluster, the Quality Analyst fires alerts in the ticker and chat panel. The Simulator deliberately creates these patterns so users can observe how proactive quality monitoring works.
• Operator Assistant — Stays available in the chat panel while the simulation runs. Users can ask “What is stuck at Assembly?” or “How is yield today?” and get answers based on live simulation data. This demonstrates the conversational MES interface.
• Production Planner — Uses simulation data for capacity analysis. “How long to build 500 units at current throughput?” works because the Simulator has been generating real throughput data. This shows how AI assists production planning.
• Anomaly Monitor — Reacts to sensor telemetry that the Simulator generates. Temperature drift, cycle time increases, and vibration anomalies trigger predictive maintenance alerts. This demonstrates the MESkit North Star: agent-driven predictive response.

Step 1 — Set up the shop floor (Build Mode)

Create at least one manufacturing line with workstations and machines. You can use the UI or the Operator Assistant:

"Create a line called Assembly"
"Add 5 workstations to Assembly"
"Add a machine called Solder Station to workstation 3"

Step 2 — Configure a product (Configure Mode)

Create a part number, configure a serial algorithm, and design a route through your workstations:

"Create part number Smartphone X"
"Configure serial algorithm for Smartphone X with prefix SMX- and 5 digits"
"Create a route for Smartphone X through all Assembly workstations"

Step 3 — Start the simulation (Run Mode)

Switch to Run Mode and press the Start button in the top bar. The Simulator Agent will begin generating units and moving them through your route. Watch the live ticker for events.

Step 4 — Observe and interact

While the simulation runs, you can:

• Watch units flow through workstations in the Run Mode UI
• Ask the Operator Assistant about WIP status, yield, or specific units
• See Quality Analyst alerts when defect patterns emerge
• Switch to Monitor Mode to view dashboards with live data
• Adjust speed to generate data faster or watch more carefully

Step 5 — Analyze results

After pausing or completing a simulation run, use Monitor Mode to analyze:

• Throughput charts — units completed over time
• Yield summary — pass/fail ratios per workstation
• WIP distribution — where bottlenecks formed
• Unit history — drill into any serial number to see its full route
• Quality events — defect codes, scrap rates, failure patterns

Step 6 — Reset and iterate

Press Reset to clear all production data and start fresh. Your shop floor and product configuration are preserved. Change the route, add workstations, or adjust the product definition, then run again to compare results.

The Simulator reads your existing configuration. Before pressing Start, make sure you have:

1. MESkit installed and running — see the Getting Started guide.
2. A shop floor (Build Mode) — at least one line with workstations and machines.
3. A product (Configure Mode) — at least one part number with a serial algorithm configured.
4. A route (Configure Mode) — at least one route with steps assigned to your workstations.

No factory, no hardware, no MQTT broker. Everything runs locally with a browser and a Supabase project.

Future releases will include scenario profiles that modify the Simulator’s behavior without code changes. Each scenario targets a different learning or testing objective:

Key facts

• The Simulator is a standalone add-on — it does not modify, extend, or depend on the MES core. Removing it leaves MESkit fully functional.
• The Simulator Agent uses the same tool layer as human operators and the Operator Assistant — no separate simulation engine.
• Every simulated event (unit creation, WIP movement, quality check, machine fault) flows through Supabase and triggers Realtime updates.
• Quality decisions are contextual, not random — yield degrades over time and defects cluster realistically.
• Speed control lets you run the simulation at 1x, 2x, 5x, or 10x — from watchable pace to rapid data generation.
• The data the Simulator generates is structurally identical to real production data — dashboards, analytics, and OEE calculations work the same way.
• No factory, no hardware, no MQTT broker needed — everything runs locally with a browser and a Supabase project.

Mini FAQ