What is AIMAS?
AIMAS (AI multi-agent system for autonomous decision-making in the manufacturing industry) will build an industrial decision ecosystem in which real company data feed specialised agents, each with its own analytical capabilities, coordinated through orchestration mechanisms.
The industrial challenge
Manufacturing companies, and SMEs in particular, still rely on manual or semi-manual planning tools that cannot react to disturbances, demand changes or quality deviations. AIMAS continuously monitors the plant and generates alternative, optimised decisions for production planning, resource allocation and quality control.
Project objectives
Conceptual and technological framework
Define a reference architecture for an AI multi-agent system applied to manufacturing operations.
Interoperable intelligent agents
Develop agents that analyse real-time data, collaborate and coordinate to propose optimised decisions autonomously.
Validation in real pilots
Deploy and evaluate the AIMAS ecosystem in industrial environments with quantitative and qualitative KPIs.
New scientific and technical knowledge
Generate publications, methods, software and guidelines for industrial transfer and adoption.
Starting hypothesis
Deploying an interoperable AI multi-agent ecosystem in manufacturing environments, able to analyse operational and quality data in real time and coordinate autonomous decisions, will significantly improve key performance indicators.
Service level
More reliable deliveries.
Resource utilisation
Better use of productive capacity.
Quality
Autonomous, real-time quality control.
Resilience
Agile response to disruptions.
Conceptual architecture in five layers
Physical & Data Acquisition
Machines, sensors, IoT, operators, MES, ERP, maintenance, quality, inventory, orders, historical and real-time data. Agents cannot make useful decisions unless they are connected to the reality of the company.
Data Management & Communication
APIs, middleware, interoperability, data models, synchronisation, data quality and semantic models that turn heterogeneous sources into consistent information for the agents.
Intelligent Multi-Agent Layer
The core of AIMAS: specialised agents for planning, scheduling, capacity, quality, maintenance, procurement, inventory, resources, performance and disruptions, coordinated by an orchestrator.
Analytics & Optimisation
Operations Research, mathematical programming, ML, DL, RL, heuristics, metaheuristics, simulation, forecasting and multi-objective optimisation that give each agent its intelligence.
Explainability & Decision Support
Model-level, agent-level and system-level XAI: why a decision is recommended, which agents took part, which constraints mattered and which KPIs change.
The AIMAS agents
Each agent combines a decision capability with cooperative behaviour: it perceives information, proposes actions, interacts with other agents and escalates conflicts when needed.
Planning AgentProduction planning and re-planning.
- Inputs
- Orders, capacity, calendars, inventory, due dates
- Methods
- Mathematical programming, heuristics, forecasting, simulation
Scheduling AgentDetailed scheduling and sequencing of operations.
- Inputs
- Orders, routings, processing times, machine availability
- Methods
- MILP, heuristics, metaheuristics, reinforcement learning
Capacity AgentLoad and capacity analysis.
- Inputs
- Expected load, resources, shifts, availability
- Methods
- Simulation, optimisation, what-if analysis
Quality AgentQuality control and prediction.
- Inputs
- Process variables, inspections, non-conformities, historical data
- Methods
- Machine learning, deep learning, anomaly detection
Predictive Maintenance AgentFailure prediction and maintenance planning support.
- Inputs
- Equipment condition, failure history, maintenance plans
- Methods
- Survival models, machine learning
Procurement AgentPurchasing and supply decisions.
- Inputs
- Requirements, inventory, suppliers, lead times, costs
- Methods
- Optimisation, forecasting, supplier selection
Inventory AgentInventory management and replenishment needs.
- Inputs
- Stock, demand, consumption, lead times
- Methods
- Forecasting, stochastic optimisation, reinforcement learning
Resource Allocation AgentAllocation of machines, operators and resources.
- Inputs
- Resources, skills, availability, priorities
- Methods
- Optimisation, assignment models, heuristics
Disruption Management AgentResponse to disruptions and unforeseen events.
- Inputs
- Events, breakdowns, delays, demand changes
- Methods
- Rules, optimisation, simulation, scenarios
Performance Analysis AgentKPI, efficiency and cost analysis.
- Inputs
- KPIs, plans, costs, performance
- Methods
- KPI analytics, anomaly detection, multi-criteria analysis
The orchestrator
Planning “We must produce 1,000 units this week.”
Scheduling “I need machine M2 for 30 hours.”
Maintenance “M2 has a high failure risk and should stop.”
Procurement “There is not enough raw material.”
Orchestrator
Identifies dependencies between agents, resolves conflicts and generates coherent scenarios for the whole production system: from locally optimal decisions to a globally consistent one.