Focus on Second Open Call winners: AI-Powered Antenna Quality Control System – Antenna-I

WHAT ABOUT THE EXPERIMENT?

The project aims to improve the quality, reliability, and cost-efficiency of IoT sensor manufacturing by introducing an AI-powered antenna assembly verification system at Correlation Systems’ production facility in Czechia. In line with AI REDGIO 5.0 objectives, the project leverages AI-at-the-edge, open hardware, and edge-to-cloud continuum to address a critical challenge: ensuring correct antenna-to-sensor assembly in small-scale, high-mix production environments.

Correlation Systems is an Israeli SME specializing in the development of mobile phone-detecting sensors for crowd analytics. While its sensors incorporate various antenna types and are used across Europe and Asia, the current quality assurance process for antenna assembly is manual and error-prone. This is particularly problematic when antennas are improperly mounted, leading to degraded wireless performance and a 20% return rate. Due to the cost and complexity of traditional solutions (e.g., anechoic chambers or radiation pattern testers), these defects often go undetected.

The project introduces Antenna-I, a smart, cost-effective system that utilizes ESP32-based receivers to capture the radiation pattern of each assembled sensor and analyze it using AI models. These models compare the measured RSSI profile with expected patterns to verify correct antenna assembly. This solution enables accurate, real-time analysis using edge AI, while relying on cloud-based models for more complex cases.

Antenna-I consists of two core AI models: one for automatic antenna type recognition and another for evaluating assembly quality. These models will be trained using data from at least three antenna types (two external, one internal), and will function with a maximum of 10 ESP32 receivers while achieving at least 90% prediction accuracy within a 3 dB margin of deviation. The complete system, including hardware, software, and setup, will cost under €150, offering a highly accessible solution compared to commercial alternatives costing thousands of euros.

A notable innovation lies in the departure from vision-based inspection systems. Instead of relying on camera input and image processing, Antenna-I evaluates RF signals directly, making it more robust in variable lighting and assembly conditions. This shift reduces computational demands and allows the AI models to run on resource-constrained ESP32 devices.

Expected results include:

• High prediction accuracy: ≥90% accuracy in detecting antenna assembly quality, within a 3 dB deviation.

• Fast training: Each antenna type will require ≤30 minutes of training to reach high prediction performance.

• Minimal hardware: System will operate accurately with ≤10 ESP32 receivers, reducing cost and complexity.

• Edge-first deployment: AI will run on ESP32 for real-time checks (≤5 min), with cloud fallback for complex cases (≤10 min).

• Improved defect detection: AI will outperform manual inspection by reducing false negatives from 20% to 5% within one year.

• Non-expert usability: The system will feature an intuitive interface suitable for non-technical operators.

• Low-cost implementation: Total system cost will stay below €150.

Antenna-I aligns well with AI REDGIO 5.0’s goals by enabling flexible, and sustainable manufacturing through edge AI. Project outcomes will be shared via open-source platforms, and commercialized as a turnkey solution for other sensor manufacturers.

WHAT IS THE EXPECTED IMPACT?

The Antenna-I project will significantly advance the digitalisation of Correlation Systems’ (CS) sensor manufacturing operations, particularly in its Czech Republic site where production remains manual and small-scale. Current manual quality checks contribute to a 20% sensor return rate due to antenna-related faults. By integrating an AI-based antenna assembly verification system at the edge, CS expects to reduce this return rate to 5% within one year, improving both product reliability and operational efficiency.

Technologically, it will upskill internal teams by enabling the development and deployment of AI models on ESP32 microcontrollers. This knowledge will support future transitions of other cloud-based analytics, such as nationality detection in crowd monitoring sensors, to edge devices. A key KPI is completing this migration within six months post-project.

Economically, the project will reduce costs associated with returns, repairs, and replacements, while enabling fast deployment across CS’s global manufacturing sites. The full system will be implemented for under €150 per unit and rolled out to additional locations within three months of project completion.

Additionally, CS plans to commercialize Antenna-I as a turnkey quality control solution for other SMEs, supporting broader industrial adoption. Participation in AI REDGIO 5.0 will also strengthen CS’s visibility and innovation reputation, with a KPI of 5,000 online engagements.