1 January 2019 in TECHNICAL SCIENCES AND TECHNOLOGIES
CTChernihiv National University of Technology
Chernihiv National University of Technology
FIFounder Of The Company Incomvision
Founder Of The Company Incomvision
Urgency of the research. The existence of the need to improve the accuracy of research areas, including in hard-toreach areas, updates the direction of finding new ways of automatic localization for the task of the robot about collecting the information surrounding it for its own localization and building the environment map. Target setting. Existing methods for determining the location of a moving platform in space have a significant error, which is unacceptable for use in devices designed to operate without human intervention. Actual scientific researches and issues analysis. Research trends show satisfactory results of the introduction of new algorithms based on neural networks, but most of the solutions are designed for large moving platforms, while for small platforms, such as UAVs, solutions are not enough. Uninvestigated parts of general matters defining. Works on the automatic determination of the location of a moving platform often demonstrate the results of experiments that were performed in laboratory conditions. The question arises: how the system will behave when tested in real conditions? The research objective. It is supposed to implement a system of automatic movement of the platform based on modern hardware and localization algorithms. The statement of basic materials. Considered general information for methods of localization and mapping of SLAM, the general algorithm for the operation of methods of SLAM and their mathematical representation are presented. Sensors, that are used in localization tasks, are described and two main SLAM methods with a detailed description are presented. The results of the simulation of the ORB-SLAM2 method were presented. Conclusions. The methods for localizing a moving platform in space and algorithm for the operation of the SLAM methods were reviewed in this paper. The features of SLAM methods and their development prospects are presented.
14 January 2026 in TECHNICAL SCIENCES AND TECHNOLOGIES
The paper presents a review and comparative analysis of modern artificial intelligence models suitable for real-time quality control of toroidal transformer winding. It is shown that existing AI solutions for transformer diagnostics and electronic product quality control are mainly focused on post-production (offline) analysis and do not address the specifics of the winding process. More than 25 neural network architectures were evaluated according to model size (≤ 200 MB), accuracy, computational load, and relative efficiency metrics. The most promising models were identified: EfficientNetV2-S and YOLOv11l as primary options for classification and object detection respectively, and EfficientNetB0 with YOLOv8n as lightweight alternatives. A universal four-stage scheme for integrating AI into a winding quality control system is defined and analyzed. The obtained results provide a scientific basis for developing low-cost intelligent monitoring systems for toroidal transformer winding machines, suitable for laboratory, educational, and small-batch production.