Living Labs
AIM Lab

AIM Lab (Artificial Intelligence in Manufacturing Lab) was established in September 2021 as a strategic partnership between the Faculty of Electrical Engineering and Computer Science at VŠB-TUO and ŠKODA AUTO. The laboratory serves as a unique research and educational hub in Industry 4.0, bridging the gap between innovative academia and real-world industrial practice.

Our core focus lies in applied artificial intelligence, machine learning, advanced industrial data analytics, and mathematical optimization for manufacturing and logistics. The team concentrates on predictive maintenance models, storage and production simulations, industrial data mining, and smart mobility technologies.

Beyond applied R&D, AIM Lab plays a vital role in education by training students across bachelor’s, master’s, and doctoral programs through hands-on industrial projects. By turning theoretical methods into deployable solutions, the lab actively accelerates digital transformation across the modern industrial ecosystem.

Industry
Services & Equipment

Our cooperation includes:

AI readiness and feasibility audits

Consulting audits that evaluate data, infrastructure, and use-case maturity, then propose a practical roadmap for deploying AI in the partner organisation.

Applied computer vision for manufacturing quality control and operational monitoring, including classification, tracking, and automated visual inspection workflows.

Support for building consortia and preparing proposals under schemes such as TA ČR and Horizon Europe, from concept to project execution.

Education for industrial partners, from practitioner workshops to structured training in machine learning, data mining, and neural-network methods.

Contract R&D covering custom predictive models for industrial and commercial partners, from architecture design through training and evaluation on partner data.

Data-engineering support that turns heterogeneous industrial data into structured datasets suitable for modelling, monitoring, and production ML pipelines.

Methods for continual learning from data streams, including concept-drift detection and memory architectures that prevent catastrophic forgetting in production models.

The lab deploys and adapts generative language models on local infrastructure so that company data stay on-premise, including retrieval-augmented generation and semantic analysis of internal documents.

Algorithms and dashboards for industrial telemetry that detect anomalies, estimate remaining useful life, and report operational status in real time.

Simulation of storage, production, and logistics flows so partners can evaluate layout and sequencing decisions before physical deployment.

Industrial optimization of robotic trajectories using computational geometry, machine learning, and multi-objective methods to raise transfer efficiency and shorten cycle times.

Equipment & facilities

High-Performance AI Computing Cluster

High-VRAM nodes: 2× NVIDIA RTX A6000 (48 GB GDDR6 ECC) per node, 96 GB total VRAM per server — for large transformer training, memory-intensive batch jobs, and complex fine-tuning. High-throughput nodes: 2× NVIDIA GeForce RTX 4090 (24 GB GDDR6X) per node, 48 GB total VRAM per server — for rapid prototyping, computer vision, genetic algorithms, and high-frequency inference.

Artificial Intelligence

Standard ML tooling (Python virtual environments, Scikit-learn, JupyterLab) plus an integrated framework for running on-premise generative models using Ollama and Open WebUI.

Artificial Intelligence

Ultra-fast scratch storage (/spad0): 12 TB local NVMe for active dataset streaming during training. Local persistent storage (/local_storage): 12 TB SSD for working directories, code, and checkpoints. Network storage (/storage1 & /storage2): over 1.3 PB combined capacity (2× 679 TB) for shared datasets, backups, and archiving.

Artificial Intelligence

TOP Publications

Chakraborty P., Bandyopadhyay A., Bhattacharyya S., Platos J.
IndiVNet: A region adaptive semantic image segmentation for autonomous driving in unstructured environments
IndiVNet: A region adaptive semantic image segmentation for autonomous driving in unstructured environments
Scientific Reports, 16 (1), art. no. 2468, 2026.
Phien N. N., Anh D. T., Platos J.
T2S-EDNN: A Segmentation-Driven Specialized Ensemble Framework with Cluster-Adaptive Routing for Robust Chaotic Time Series Forecasting
T2S-EDNN: A Segmentation-Driven Specialized Ensemble Framework with Cluster-Adaptive Routing for Robust Chaotic Time Series Forecasting
IEEE Access, 2026.
Rashidi S. F., Olfati M., Mirjalili S., Grosan C., Platoš J., Snášel V.
A hybrid DEA–fuzzy clustering approach for accurate reference set identification
A hybrid DEA–fuzzy clustering approach for accurate reference set identification
Machine Learning with Applications, 100818, 2025.
Ali A., Snášel V., Platoš J.
Health-FedNet: A privacy-preserving federated learning framework for scalable and secure healthcare analytics
Health-FedNet: A privacy-preserving federated learning framework for scalable and secure healthcare analytics
Results in Engineering, 106484, 2025.
Svoboda R., Basterrech S., Kozal J., Platoš J., Woźniak M.
A natural gas consumption forecasting system for continual learning scenarios based on Hoeffding trees with change point detection mechanism
A natural gas consumption forecasting system for continual learning scenarios based on Hoeffding trees with change point detection mechanism
Knowledge-Based Systems, 304, 112482, 2024.

Contact Person