Machine learning-assisted optical fiber specklegram sensor for early and spatially distributed water leak detection and localization

We harness the change in boundary conditions of a no-core fiber (NCF) for high-sensitivity water leak detection and leak-spot localization, using a cost-effective charge-coupled device (CCD) camera and machine learning analysis. The CCD camera is utilized as an interrogation unit for capturing the specklegram images at the end-face of the NCF. Water leaks, which alter the ambient refractive index and thus the boundary conditions at the more sensitive uncoated sections of the NCF, induce noticeable shifts in specklegram images generated by multimodal interference within the NCF, and these shifts are detected using a convolutional neural network. Remarkably, the sensor exhibits high sensitivity, detecting water volumes as small as 0.1 mL and identifying leak spots 1 cm apart. Moreover, the presented simulation results support the experimental findings, enhancing the study’s robustness and providing a comprehensive, low-cost, and efficient approach to leak detection.