Thermal Anomaly Monitoring and Detection System for Industrial Equipment Based on Computer Vision

This work presents the design and prototype of a computer vision system for the automatic analysis ofthermographic images of industrial equipment such as motors, bearings and furnaces. The proposed pipelineperforms preprocessing (contrast adjustment and denoising), region-of-interest (ROI) segmentation and hot-spotlocalization, optical character recognition (OCR) of the color-scale values to convert color levels into temperatures in°C, and comparison against equipment-specific, configurable operating limits. The thermal condition is classified asnormal, warning or critical, and each analysis is stored in a database and may trigger decision-support alerts. As anevolution, the measurement history is proposed as a means of identifying heating trends and supporting predictivemaintenance. The prototype was evaluated on 534 thermal images: in 40% of them the extracted temperature showedan error of 0.1 °C, and the classical segmentation (no neural networks) correctly located the region of interest inpractically all images, although delimiting a smaller area than desired.