Leader
Mauris et nulla tellus. Nulla consectetur placerat nunc placerat bibendum.
Hosting Facilities
Mauris et nulla tellus. Nulla consectetur placerat nunc placerat bibendum.
AlleyOPP
Automation
Autonomous AI for energy optimization in parts production
Challenge: Automated energy management for parts production
This proposal delivers a vendor-agnostic shopfloor module for real-time energy optimization in high-intensity aluminum casting and automotive production. Advancing from TRL 6 to a live TRL+7 prototype at LTH Castings, the solution replaces legacy PID controls with a hierarchical, cyber-physical architecture. By feeding a hybrid Digital Twin (physics + LSTM) into a patented two-stage Reinforcement Learning engine, the system dynamically synchronizes equipment states—like washing machines, furnace pre-heating—with master production schedules. The module autonomously learns complex consumption profiles to achieve a verifiable 15% energy reduction per part, slashing costs and carbon footprint.
Background
LTH Castings, a top European automotive supplier specializing in high-pressure aluminum die-casting, faces immense pressure to cut its carbon footprint and navigate rising energy costs. To achieve its goal of carbon neutrality, the group integrated into the European “metaFacturing” initiative. The project transitions shop floors from legacy, static equipment controls to advanced data-driven automation. By providing critical production data for near-real-time analytics and intuitive operator dashboards, the project bridges the gap between complex industrial data and actionable shop floor insights—driving transparency, traceability, and drastic energy reductions.
Solution
The BeChained AI solution is a vendor-neutral shop floor module utilizing a hierarchical, cyber-physical architecture. A hardware-agnostic data gateway normalizes real-time PLC/SCADA data via OPC-UA to feed a hybrid digital twin. This Twin fuses physics-based thermal models with an LLM-RAG created neural network and LSTM to simulate real-world behavior. Using this environment, a patented Reinforcement Learning (RL) engine optimizes machinery states. First, a heuristic algorithm filters out a safe set of control policies. Second, a Soft Actor-Critic (SAC) RL agent selects the optimal strategy, calculating the precise, energy-minimal curves needed to maintain melt quality during pauses. In aluminum casting, it masters the non-linear thermal dynamics of holding furnaces during high-power pre-heating and stand-by modes, eliminating hours of wasted energy. Fitting into AID4SME goals, this provides a centralized, automated energy optimization that the facility currently lacks. It replaces slow, manual modeling with autonomous learning that adapts to production plans & external conditions, delivering a solution that slashes costs without impact throughput.
Objectives
- Drastically cut energy use: Achieve a verifiable reduction in energy consumption per produced part of 15%.
- Implement intelligent automation: Replace legacy, static control systems (like PID) with a closed-loop, datadriven reinforcement learning engine.
- Optimize asset states: Synchronize high-power equipment states (start-up, shut-down, stand-by) dynamically with master production schedules.
- Maintain operational integrity: Lower carbon footprints and costs without compromising production throughput or quality.
- Advance technology readiness: Elevate the existing energy optimization platform from TRL 6 to a validated TRL 7 prototype within a live industrial environment.
KPIs achieved
Development
Mauris et nulla tellus. Nulla consectetur placerat nunc placerat bibendum. Sed sit amet consequat ex. Pellentesque id odio felis. Etiam tellus dui, maximus non vehicula ut, suscipit vel ex. Maecenas quis lorem ut sem aliquet fermentum. Duis vel nulla hendrerit augue aliquet posuere. In erat arcu, accumsan ut rhoncus in, fringilla at elit. Morbi non gravida ligula. Mauris et nulla tellus. Nulla consectetur placerat nunc placerat bibendum. Sed sit amet consequat ex. Pellentesque id odio felis. Etiam tellus dui, maximus non vehicula ut, suscipit vel ex. Maecenas quis lorem ut sem aliquet fermentum. Duis vel nulla hendrerit augue aliquet posuere. In erat arcu, accumsan ut rhoncus in, fringilla at elit. Morbi non gravida ligula.
Feedback
Mauris et nulla tellus. Nulla consectetur placerat nunc placerat bibendum. Sed sit amet consequat ex. Pellentesque id odio felis. Etiam tellus dui, maximus non vehicula ut, suscipit vel ex. Maecenas quis lorem ut sem aliquet fermentum.
Mauris et nulla tellus. Nulla consectetur placerat nunc placerat bibendum. Sed sit amet consequat ex. Pellentesque id odio felis. Etiam tellus dui, maximus non vehicula ut, suscipit vel ex. Maecenas quis lorem ut sem aliquet fermentum.
Mauris et nulla tellus. Nulla consectetur placerat nunc placerat bibendum. Sed sit amet consequat ex. Pellentesque id odio felis. Etiam tellus dui, maximus non vehicula ut, suscipit vel ex. Maecenas quis lorem ut sem aliquet fermentum.