Zaretskyi M. Models and methods of intelligent information technology for assessing the functional state of sewage pipes

The dissertation is devoted to solving the actual scientific and applied problem of automating the process of assessing the functional state of sewage pipelines based on video inspection data through the development of information technology for intelligent data analysis. One of the complex tasks that the dissertation aims to research is the creation of intelligent information technology for machine learning for automatic data analysis of sewer pipe video inspection under conditions of limited training data volume and resources available for system training and maintenance. At the same time, solving this task involves the necessity of analyzing spatio-temporal contextual information. The object of the study is the process of assessing the functional state of sewer pipes. The subject of the study is models and methods of information technology for assessing the functional state of sewer pipes. Based on the conducted analysis of methods and approaches to sewer pipe inspection, it has been established that video cameras are the primary source of information, as video inspection requires the least overhead costs, and the informativeness of the obtained data ensures the detection of most important types of defects. The main trend in the development of modern data analysis technologies for video inspection is associated with the implementation of machine vision and artificial intelligence technologies, although these technologies do not yet provide a sufficient level of automation due to the inadequate reliability of the formed reports. Therefore, their improvement is an important direction of research. Based on the analysis of the sewer pipe video inspection process, the need for creating a new intelligent information technology for assessing the functional state of sewer pipes by using the ideas and methods of model composition, contrastive learning, noise-resistant coding, and information-extreme intelligent technology (IEI-technology) has been substantiated Thus, the main direction for enhancing the effectiveness of models for evaluating the functional state of sewer pipes lies in adapting the input mathematical description and constructing classification decision rules during the learning process. These rules are intended for context recognition, defect recognition in individual contexts, and water level change assessment, all achieved through maximizing efficiency criteria. For the first time, a method has been developed for training a hierarchical feature extractor, which, unlike known ones, ensures an increase in the functional efficiency of decision rules under conditions of noisy and imbalanced training data by combining the principles of knowledge transfer, siamese networks, contrastive self-learning, and self correcting binary codes. The obtained scientific results in the form of informational and software provision have been implemented during the training of personnel and conducting of sewer network inspections at KP "Miskvodokanal" SMR (Sumy city), during the development of an intelligent video monitoring system for infrastructure objects at the Limited Liability Company “SMART TECHNOLOGIES “TITUL”” (Sumy city), during the development of software for a machine vision module for a remotely controlled mobile platform at the Limited Liability Company “NORD TREK” (Shostka city, Sumy region), during the development of a classification analysis system for sewer pipeline video inspection data at the company Molfar.AI sp. z o.o., (Gdansk, Poland), and also in the educational process of the Department of Computer Science of Sumy State University while teaching the discipline “Introduction to Data Science”.