Abstract
Detecting leaks in hermetically sealed cooling equipment is crucial for the refrigeration industry, as manufacturing defects may lead to refrigerant loss or even render cooling systems completely inoperable. Currently, in the refrigeration sector, leak verification in hermetic compressors is typically performed through manual human inspections, which can potentially cause worker fatigue and exhaustion. In this working paper, we tackle the problem of automatically detecting manufacturing defects in hermetic compressors using computer vision-based solutions. More specifically, we design an AI-based approach that integrates optical flow algorithms, image processing filtering, and neural networks to detect bubbles when the compressors are submerged in a water tank. By experimentally analyzing the leakage detection problem on preliminary real image compressor videos, our computer system was able to automatically detect failures in damaged compressors in an accurate and efficient way.
Content
![](https://www.cambridge.org/engage/api-gateway/miir/assets/orp/resource/item/64b6746fb605c6803bec9773/largeThumb/leak-detection-in-hermetic-compressors-using-computer-vision-and-artificial-intelligence.jpg)