VIBES — Smart-Livestock Cloud–Edge AI Platform

Hierarchical cloud–edge platform and embedded intelligent services for smart livestock farming (Korea–Czech joint R&D)

VIBES developed a hierarchical cloud–edge platform with embedded intelligent services for smart livestock farming (Korean title: 스마트 축산용 계층형 클라우드-엣지 플랫폼 및 임베디드 지능형 서비스 솔루션 개발), under the 2019 Korea–Czech International Joint Technology Development program (bilateral co-funding, 2019–2022). The system combines AI, IoT, and video to diagnose and predict farm-environment and animal-growth conditions, and manages them through tiered cloud–edge services.

Consortium

  • Korea: Gluesys (lead), KETI, and Korea National Open University (KNOU)
  • Czech Republic: Iterait (lead), Brno University of Technology, and the Czech University of Life Sciences

Concept

The platform is split across an AI-Cloud Platform — a Control Server for hierarchical farm/edge management plus an AI Server for training and model deployment — and multiple AI-Edge Systems that collect IoT and video data and run on-site inference, complemented by a portable meter for body-temperature and growth data. This tiering distributes compute efficiently and keeps services running under network delays or outages, while AI analysis drives environment, growth, and fire-management services with forecast-based alerts.

System architecture: a tiered AI-Cloud Platform (Control Server + AI Server) with multiple AI-Edge Systems and a portable meter, delivering environment, growth, and fire-management services.

KETI’s major contributions

  • A hierarchical AI-edge platform and AI model-management system — an AI service interface, an AI model repository, and a container-deployment service built on Kubernetes — that packages AI as microservices and offloads them from cloud to edge, including embedded accelerators (ARM, NVIDIA Xavier)
  • An environment-prediction AI server and module — time-series preprocessing (refinement, outlier cleaning, imputation) and AutoML for regression, forecasting, and classification — reaching 84.96% average prediction accuracy against a 75% target
  • An edge-resource monitoring dashboard built on Prometheus, with a Xavier node exporter for real-time CPU/GPU/memory monitoring of embedded edge nodes
KETI's environment-prediction AI server: time-series analysis, visualization, and AutoML over the collected farm sensor data.
KETI's hierarchical AI model-management platform: an AI interface, model repository, and container deployment over Kubernetes, offloading AI from cloud to edge.

Outcomes: field-validated on commercial pig farms (Sangju and Gunwi, Korea), the project exceeded its quantitative targets and contributed an international standard proposal to ISO/IEC JTC1 SC29 WG7 (MPEG-IoMT, m61001) on time-series data analysis for smart farming.

Funding: Korea–Czech International Joint Technology Development Program
Period: 2019 – 2022
Role: KETI PI (Korean participating institution; Gluesys was the Korean lead)