Each machine on this virtual line reports a noisy vibration/temperature signal to a lightweight anomaly-scoring model, the same idea behind real Industry 4.0 predictive-maintenance and digital-twin systems:
risk_i(t) = σ( k1·wear_i(t) + k2·|vibration_i(t)| − k3 )
wear_i(t+dt) = wear_i(t) + speed · rate_i · dt
OEE = mean_i[ (1 − wear_i) · running_i ] × 100%
- Production speed — scales how fast every machine accumulates mechanical wear, mirroring how higher throughput targets stress real equipment faster.
- Sensor noise — jitters the simulated vibration/temperature readings; too much noise makes the AI model's risk score jumpy and can trigger false-positive maintenance calls, just as it does on a real shop floor.
- Risk threshold — the probability above which the model calls predictive maintenance before failure; the translucent "digital twin" copy floating above each machine shows the model's current risk estimate, separate from the physical wear it is predicting.
- Inject fault — spikes one random machine's wear instantly, so you can watch the AI catch (or miss) a sudden anomaly.