Enhancing leak detection and localization in end-of-line leakage stations through the integration of infrared and acoustic technology with computer vision

• A comparison between IR and acoustic technologies for leak detection was conducted. • Tailored vision-based algorithms enhanced detection and localization performance. • Acoustic approach achieved robust and real-time leak detection in EOL testing. • IR offered high spatial precision but was sensitive to environmental conditions. End-of-line (EOL) leakage testing is crucial in household appliance manufacturing to ensure product safety. However, conventional methods often lack accurate leak localization. This study systematically evaluated and compared infrared (IR) and acoustic technologies for air leak detection and localization under controlled EOL scenarios, considering varying operational parameters, such as leak rate, camera inclination, and camera-to-leak system distance. The results revealed that the IR camera achieved high spatial localization accuracy under optimal acquisition conditions, but its performance was sensitive to environmental variations, particularly ambient temperature, resulting in low and variable detection rates. Acoustic technology maintained consistently high detection rates (100 %) across all tested scenarios, although its localization accuracy was lower than that of the IR system. To enhance leak detection, customized vision-based multi-criteria frameworks were developed for both technologies. For IR data, the algorithm extracted complementary spatiotemporal indicators, including temporal variance, entropy variation, and the frequency of extreme temperature values, and combined them into a unified scoring map that estimated leak probability at the pixel level. This approach enabled automated leak detection without manually defined regions of interest, providing interpretable visual outputs and achieving high precision (up to 0.79) despite moderate prediction rates (∼35 %). For acoustic data, the algorithm analysed spatial intensity consistency and temporal variance, and performed frequency-domain validation, enabling real-time detection and localization of all leaks, outperforming manual inspection. Overall, the acoustic technology combined with the proposed algorithm provided the most consistent and timely performance, while IR imaging offered higher positional accuracy under controlled conditions. These results demonstrate the suitability of both technologies for automated EOL leak testing in household appliance manufacturing and suggest that a hybrid approach could further improve air leak detection and localization.