ECOLE · Project

Self-Improving Optimization Algorithms That Get Smarter With Every Engineering Problem They Solve

digitalPrototypeTRL 3Thin data (2/5)

Imagine you hire an engineer who gets better at solving problems the more projects they work on — they remember what worked before and apply those lessons to new challenges. ECOLE built computer algorithms that do exactly that: optimization software that accumulates "experience" from past engineering projects and uses machine learning to get faster and more accurate over time. The more problems you throw at it, the smarter it gets. Honda and NEC were directly involved, applying this to car design and IT systems.

By the numbers

consortium partners

industry partners in the consortium

62%

industry partner ratio

countries involved (DE, NL, UK)

50%

ESR time spent at industry partners

total project deliverables

The business problem

What needed solving

Engineering companies face optimization problems that keep getting more complex — more parameters, tighter constraints, shorter deadlines. Every time they tackle a new design or process challenge, their optimization software starts from scratch, ignoring everything learned from similar past problems. Expert engineers carry this experience in their heads, but that knowledge retires when they do.

The solution

What was built

ECOLE produced research algorithms that combine evolutionary optimization with machine learning, enabling the software to accumulate and reuse experience across different engineering optimization tasks. The project delivered 21 outputs including trained doctoral researchers with hands-on industry experience at Honda and NEC, along with the underlying algorithmic methods published in academic literature.

Audience

Who needs this

Automotive OEMs and Tier-1 suppliers with complex multi-objective vehicle design optimizationIT infrastructure companies optimizing large-scale network and resource allocationAerospace manufacturers running computational design optimizationProcess manufacturers tuning multi-parameter production linesEngineering simulation software vendors looking to add learning capabilities

Business applications

Who can put this to work

Automotive

enterprise

Target: Car manufacturers and automotive component suppliers with complex design optimization tasks

If you are an automotive manufacturer dealing with increasingly complex vehicle design — balancing weight, safety, fuel efficiency and cost across thousands of parameters — this project developed optimization algorithms that learn from every past design iteration. With Honda Research Institute Europe directly involved in the research, these methods were tested against real automotive engineering challenges. The algorithms improve with each use, meaning your 100th optimization run is significantly faster than your first.

IT Infrastructure & Cloud Computing

enterprise

Target: IT companies managing large-scale network and system optimization

If you are an IT infrastructure provider struggling with resource allocation, load balancing, or network configuration problems that grow more complex as you scale — this project developed learning-based optimization methods with direct input from NEC Europe. These algorithms remember solutions from previous optimization tasks and transfer that knowledge to new but related problems, reducing computation time for recurring operational decisions.

Advanced Manufacturing

mid-size

Target: Manufacturing firms with complex production process optimization needs

If you are a manufacturer running multi-parameter optimization on production lines — tuning temperatures, pressures, speeds, and material compositions — this project built algorithms that accumulate experience across different optimization runs. Instead of starting from scratch each time you face a new production challenge, the system draws on learned patterns from previous problems, cutting the time needed to find optimal settings.

Frequently asked

Quick answers

What would it cost to implement these optimization algorithms?

ECOLE was a doctoral training network (MSCA-ITN), so there is no off-the-shelf product with a price tag. The outputs are research algorithms and trained researchers. Adoption would likely require hiring graduates from the programme or commissioning custom development based on published methods. Licensing terms would need to be negotiated with the University of Birmingham or the specific partner that developed the relevant algorithm.

Can these algorithms handle industrial-scale problems?

The consortium included 5 industry partners (62% industry ratio), with Honda Research Institute Europe and NEC Europe directly involved in the research. ESRs spent 50% of their training time at non-academic partners, which means the methods were developed with real industrial problems in mind. However, as a training network, the primary outputs are research publications and trained PhDs rather than production-ready software.

Who owns the intellectual property from this project?

IP from MSCA-ITN projects typically follows institutional agreements between the consortium members. With 8 partners across 3 countries (UK, Germany, Netherlands), IP ownership likely varies by specific algorithm or method. Contact the University of Birmingham as coordinator for clarification on licensing specific research outputs.

How does this differ from existing commercial optimization software?

The key differentiator is the 'experience' concept — unlike standard optimization tools that start fresh each time, ECOLE algorithms learn across problems and over time. They combine evolutionary algorithms with machine learning to transfer knowledge between different optimization tasks. This is a research-stage capability not yet available in commercial solvers.

What is the current status and timeline for availability?

The project closed in March 2022. The research results exist as academic publications and PhD theses. There is no commercial product timeline. Companies interested in these methods would need to engage directly with the research groups at the University of Birmingham or Leiden University, or recruit the trained researchers who now carry this expertise.

Which industries were these methods actually tested in?

Based on the consortium composition, the methods were applied in automotive engineering (Honda Research Institute Europe) and ICT systems (NEC Europe). The EuroSciVoc classification confirms automotive engineering and machine learning as the primary domains. Other industrial applications are plausible but not documented in available project data.

Consortium

Who built it

The ECOLE consortium is notably industry-heavy for an academic training network, with 5 out of 8 partners (62%) coming from industry. The two headline industrial partners — Honda Research Institute Europe (automotive) and NEC Europe (ICT) — are major R&D players, which gives the research real-world grounding. The academic side is anchored by the University of Birmingham and Leiden University, both ranked in the top 150 globally. With partners in 3 countries (Germany, Netherlands, UK), the network covers key European R&D hubs. However, there are zero SMEs in the consortium, which means the methods were developed for enterprise-scale problems and may need adaptation for smaller companies. The fact that doctoral researchers spent 50% of their time at industry sites is a strong signal that the algorithms were tested against genuine engineering challenges, not just academic benchmarks.

THE UNIVERSITY OF BIRMINGHAMCoordinator · UK
UNIVERSITEIT LEIDENparticipant · NL
Honda Research Institute Europe GmbHparticipant · DE
NEC LABORATORIES EUROPE GMBHparticipant · DE
NEC EUROPE LTDparticipant · UK
BAYERISCHE MOTOREN WERKE AKTIENGESELLSCHAFTpartner · DE

How to reach the team

University of Birmingham, Computer Science department — look for the ECOLE principal investigator via the project website or university staff directory

Order a report on this consortium See THE UNIVERSITY OF BIRMINGHAM profile →

Next steps

Talk to the team behind this work.

Want to know if ECOLE's self-learning optimization methods could cut your engineering design time? SciTransfer can connect you with the right researcher from this consortium.

Order a report on this topic More in Digital & ICT