NSTU Scientists are developing a unified stochastic platform of a new generation. It is designed to assess and predict the reliability, as well as the remaining life of technical systems for various purposes, including transport, energy and production facilities. The work is carried out within the framework of the grant The Russian Science Foundation.
Modern transport and industrial systems have become so complex and energy-saturated that the classic approach of "we service according to regulations and respond to breakage" has ceased to be economically and technologically justified, says Boris Malozemov, project manager, Associate Professor of the Department of Electrotechnical Complexes at NSTU-NETI, Candidate of Technical Sciences. Equipment failure often does not occur suddenly, usually conditions accumulate gradually through the degradation of batteries, electric drives, power electronics, and mechanical components, and the degradation is stochastic (random) and depends on the operating mode. That is why the task of reliably estimating the remaining resource and early prediction of failures becomes critically important. The platform is designed as a response to the gap between the requirement of high reliability and limited observations in real—world operation - it will combine diagnostics, forecasting and decision support in a single digital circuit.
"In the context of our development, the concept of "unified" is important — we are not talking about a highly specialized program for a single facility, but about a modular digital environment that can be customized for different classes of equipment and integrated into industrial circuits. The platform performs three key functions: firstly, it turns operational data and work scenarios into formal models of degradation and risk, and secondly, it links these models with digital counterparts so that the resource assessment is not "on average for the hospital", but for a specific facility and its modes. And thirdly, it adds intelligent forecasting modules, including self—learning neural network components, so that the system "adjusts" as data accumulates," Boris Malozemov said.
According to the scientist, the stochastic approach is based on the recognition that a huge number of random factors affect wear and failures. For the same equipment with the same rated load, the degradation trajectory may vary due to temperature conditions, energy quality, vibrations, microdefects, human factors, and incompleteness of measurements. Therefore, the forecast point (one digit of the service life) is often misleading. Stochastics provides a more practical form of response: the probability of failure in time, confidence intervals of the resource, risk profiles for operational scenarios, as well as resistance to incomplete/noisy data are all key features of the platform.
"It's like a weather forecast for technology. It is important to us not only whether it will rain or not, but with what probability, in what range and under what conditions the risk becomes unacceptable, in order to manage maintenance, managing risks rather than responding to accidents," explains the scientist.
The uniqueness of the development lies in the hybrid architecture. The platform combines stochastic modeling, physico-chemical/multiphysical models of aging (thermal, electrical, mechanical, chemical processes) and ML modules that are trained on operational data and achieve accuracy where the physical model inevitably simplifies reality.
The practical effect is the transition from scheduled maintenance to predictive (repairs and replacements are not "on the calendar", but according to risk and resource), reduction of unplanned downtime, justification of modernization/replacement schedules and optimization of lifecycle costs. The key effect is to increase safety and reduce accidents due to early detection of signs of failures and assessment of equipment limit conditions.
The target consumers of the platform being developed at NSTU-NETI are urban and intercity electric transport operators, machine-building and energy enterprises, as well as companies engaged in monitoring and automation.
The project is designed for two years. It is currently at the stage of deploying fundamental modules: scientists are laying the mathematical core and architecture of the digital twin in order to further "build muscle" — to connect new classes of equipment, ML modules, integration interfaces. This year, it is planned to create a "skeleton" of the platform — a library of basic stochastic models for estimating residual resource, prototypes of digital twins, a launch IT version, and a structured database of operational scenarios. In 2027, it is planned to launch a fully functional version of the platform, create high-precision hybrid models and test them in real operation.
The project "Unified Stochastic Platform for assessing and predicting the reliability and remaining resource of technical systems in transport and industry" was included in the list of supported projects following the results of the 2025 competition for grants from the Russian Science Foundation.