Smart Manufacturing ML-MAS — Control Center

Loading…

Key performance indicators
Fill rate
—
Avg delay
—
Resilience score
—
disrupted ÷ normal
Throughput
—
units satisfied
Full system architecture — decentralised multi-agent supply chain
Hover to tilt · Animated lines = live data flow · All 12 agents shown · Click nodes for detail
Forecasting Layer Demand Forecasting Agent · RandomForest DisruptionEngine 4 types · Stochastic Monitoring event stream Q-Agent RL Reinforcement Learning Q-table 20×20×7 · ε-greedy Input Layer Customer Demand Trigger Demand Adjustments Feedback Triggers AUTONOMOUS DECISION LAYER A2A MessageBus — Priority Queue · ALERT > WARNING > INFO · Pub/Sub · Event-Driven StockRisk · RouteChange SupplierSwitch · DemandAdj OrderManagement Agent · Orchestrator RECEIVED→COMPLETE FSM Inventory Agent Global View · Branch A/B/C Wh A · Wh B · Wh C Production Agent RL Policy Driven Actions: [20..200] units Procurement Agent · Validates qty Disruption buffers Logistics Agent · 300 cap Route change alert Inventory Available? YES Fulfillment Agent · Confirms Last Mile Delivery Agent Customer Delivery ✓ NO Supplier Discovery Agent 5 suppliers · Scored Supplier Switch Decentralised Supplier Network Node 1·2·3·4·5 bid Smart Contract Agreement Escrow · Auto-execute MULTI-WAREHOUSE SYSTEM — BRANCH A/B/C Warehouse A Primary · 300 cap Branch A target Transfer time: 1 step Warehouse C Bulk · 500 cap Lowest transfer cost Transfer time: 3 steps Warehouse B Secondary · 300 cap Regional node Transfer time: 2 steps DISTRIBUTION LAYER Distribution Hub Agent · Sorting · Routing Consolidates bulk → last-mile Feedback: Delivery Time · Cost · Satisfaction → Q-Agent policy update LEGEND RL / Intelligence layer Disruption engine Order management Warehouse / fulfillment Production / supplier Supplier network A2A message / policy Disruption alert (pink) Physical flow (teal) Animated = live data movement 12 autonomous agents · A2A MessageBus · Branch A/B/C · Smart Contracts · RL Policy Broadcast
Scenario comparison — RL cost efficiency vs baseline
Fill rate & cost — scenario comparison with SLA threshold
RL system achieves comparable fill rate at substantially lower total cost — the measurable value of reinforcement learning.
Training progress
RL learning curve
Fill rate & delay per episode
Last episode — supply chain
Demand vs satisfied supply
Inventory & costs
Inventory levels with safety stock threshold
Cost breakdown per step
Disruption analysis
Disruption type distribution
Resilience radar — normal vs disrupted
Agent activity log — final episode · 12 agents
Disruption event log
StepTypeDescriptionDurationSeverity