AP25793497 – Automated Pipeline Leak Detection Using Thermal Imagery from Fixed-Wing Drones and Convolutional Neural Networks

Objective of the project – The goal of the project is to develop and implement an intelligent automated system for detecting leaks in pipelines in Kazakhstan using thermal images obtained from fixed-wing drones and CNN. The system is aimed at enhancing the security of infrastructure, reducing environmental risks, and optimizing pipeline management.

Relevance: The relevance of this project is driven by the growing need for effective and automated leak detection systems in extensive pipeline infrastructures. Traditional monitoring methods are often inefficient, while leaks can lead to significant financial losses and environmental damage. The use of UAVs with thermal imaging and CNN-based analysis offers a modern solution for accurate and rapid detection of anomalies. These technologies enable high-quality data collection over large areas and improve detection accuracy while reducing false alarms. In Kazakhstan, the large-scale pipeline network and reliance on foreign technologies highlight the urgent need for domestic, secure, and cost-effective solutions. Therefore, the project is highly relevant as it enhances infrastructure safety, supports environmental protection, and contributes to technological independence and economic security.

Scientific supervisor: Master of Technical Sciences, Lecturer, Olzhas Nauryzabekovich Akylbekov

Expected and achieved results: The project successfully defined its main objectives and established clear success metrics for evaluating system performance. Key system characteristics and operational constraints were identified, including pipeline types, environmental conditions, and the use of UAVs and sensors. A comprehensive technical specification (ToR) was developed, covering functional and technical requirements, system architecture, interfaces, data requirements, and information security aspects. User groups, application scenarios, and access roles were clearly defined to support system usability and management. A structured plan for required competencies, team composition, and potential partnerships was outlined. Detailed requirements for hardware, software, and algorithms were formulated, including video and thermal data processing and machine learning models. A project roadmap and acceptance criteria were established, including a requirements matrix, risk register, and data source identification. Key domain concepts were defined, such as thermal signatures of leaks, environmental and geographic conditions, and pipeline characteristics. Expert knowledge was analyzed to establish criteria for interpreting anomalies in thermal images and identifying leaks. Formal knowledge representation methods were selected, including ontology, data models, feature sets, annotation rules, and decision-making procedures. In addition, data models and machine learning algorithms, including convolutional neural networks, were defined along with the data flow architecture and detection logic. As a result, robust data structures and algorithmic approaches were developed, forming a reliable foundation for a high-accuracy leak detection system under various operating conditions.