APPLICATION OF ARTIFICIAL INTELLIGENCE FOR IR IMAGE PROCESSING IN REMOTE TEMPERATURE MONITORING
Saša Milić, Miša Kožicić, EPS JSC BELGRADE, HPPS ĐERDAP, KLADOVO, SERBIA · 2024
Today's technological advancements offer a range of industrial applications for infrared (IR) cameras to quickly capture moving industrial objects, providing information about the temperature distributions of these objects. The IR cameras can detect heated and/or overheated surfaces in real time, enabling rapid fault detection and the potential monitoring of its gradient for future predictions. The paper describes the methodology for the application of artificial intelligence in the analysis of infrared images for remote temperature monitoring. The proposed methodology should improve the existing measurement system. The improvement would be based on replacing the current optical measurement assembly with an infrared industrial camera and a real-time image analysis measurement algorithm for detecting overheated rotor poles of the hydro generator. A basic presentation of several approaches in the application of machine learning models for image and video processing is given. Two fundamental concepts and one detection algorithm are presented in detail: a concept based on convolutional neural networks, a concept based on two types of autoencoder networks, and the YOLO algorithm for the classification and detection of complex objects. The advantages and disadvantages of the proposed concepts and detection algorithm are analyzed in detail to practically improve the existing monitoring system and take into account the key requirements: improving the functionality while ensuring the required reliability and ease of maintenance.