PNIPU: Program for automatic assessment of the technical condition of the external walls of brick buildings

Main article: Data mining Data mining

2025: Presentation of the program for automatic assessment of the technical condition of the exterior walls of brick buildings

Perm Polytechnic scientists have automated the assessment of building safety using artificial intelligence. The university announced this on December 26, 2025.

𝓲

Фото: PNIPU

In the context of a massive aging of housing stock and an acute shortage of qualified personnel construction industries , there is a direct threat to the safety of millions of people. Many building structures gradually lose strength and hidden defects accumulate, creating the risk of sudden collapses. Traditional assessment methods, such as visual inspection, are time-consuming financial and costly, making mass surveys difficult in conditions of staffing shortages and limited resources. Perm Polytech scientists have developed an artificial intelligence program to automatically assess the technical condition of the exterior walls of brick buildings. It classifies the degree of wear with high accuracy (84%), minimizing the risk of missing emergencies.

The global problem of aging infrastructure is what most of the world's developed countries face. The structures of buildings, especially those erected decades ago, gradually wear out under the influence of time, environment and operational loads. At the beginning of 2024, about 70 thousand houses, in which more than a million people live, were officially declared emergency in Russia.

Regular monitoring and timely detection of defects are the key to preventing disasters. However, a number of problems interfere with a systematic and high-quality examination: the high cost of examinations, the labor intensity of processes and the "human factor." As a result, hidden defects accumulate, which go unnoticed until they reach the critical stage, requiring no longer planned repairs, but emergency recovery work.

The traditional method of evaluation, in which an engineer visually inspects a building and gives a verdict based on his experience, is considered not only slow, but often subjective. For the expert systems and probabilistic models that exist today, diagnostics also require complex manual configuration of rules. This often requires the participation of highly qualified specialists of two profiles: experienced civil engineers who can formalize their experience, and programmers who can translate this knowledge into digital systems. All these conditions make it long and expensive to check the condition of buildings.

Scientists at the Perm Polytechnic University have developed a software solution based on artificial intelligence that can automate the routine process of determining the technical condition of the outer walls of brick buildings.

At the first stage, they analyzed and digitized the experience gained during field surveys of houses. Data collection was carried out from the archives of expert organizations, technical reports and own field research. From this, scientists collected a training sample describing the facades, each of which was divided into key elements: a basement, the main field of the wall, bridges.

As parameters, 18 critical signs were used, such as the width of the cracks, the amount of deviation of the wall from the vertical, the actual strength of the masonry and others. The final result of the analysis was the assignment of one of the four categories of state according to GOST: normative, operable, limited operable or emergency.

To create an intelligent system, scientists tested five different algorithmic machine learning for neural networks. The most promising results were shown by the AutoGluon library. She independently performed a search of various algorithms, combined the best approaches, adapting to the characteristics of specific data, which made it possible to achieve maximum accuracy without prolonged manual tuning.

The next step was the creation of a program structure that will analyze information at several levels. First, the intelligent system receives and processes the original building status data. The second level of the program receives the results of the analysis and calculates complex relationships between different parameters of wall states. At this stage, combined effects from a variety of factors are identified. The third forms the final assessment: compliance with four possible categories of technical condition, - said Galina Kashevarova, Doctor of Technical Sciences, Professor of the Department of Construction Structures and Computational Mechanics, PNIPU.

This program structure allows you to consistently process information: from the primary analysis of individual parameters to the identification of complex dependencies to the classification of the building.

The training of the program took place in several stages. At first, she studied 65% of the data we uploaded - this was the main educational sample. Then, intermediate checks were carried out by 20% to make sure that the program does not adapt to specific examples, but reveals patterns itself. The final stage took place on the remaining 35% of the data that the program saw for the first time. To combat retraining (when the model "remembers" examples, but cannot reveal general patterns), a method was used to exclude neurons. This approach significantly increased the reliability of the program when working with new buildings that had not previously been encountered, - said Sergey Krylov, graduate student of the Department of Construction Structures and Computational Mechanics at PNIPU.

The training results showed high efficiency on training data: accuracy reached 92.3%. In the final sample, which was not used in the training, the accuracy was 84.62%.

To determine the results, scientists also analyzed program errors and identified an important feature. In cases of imprecise category definition, it usually chooses an adjacent but stricter category. For example, if a building is actually operational, the program can define it as restricted. This "understatement" of the category is considered acceptable, since the main task of development is not to miss the emergency state.

The practical application of this program will allow preliminary assessment of thousands of buildings as soon as possible, which is especially valuable for organizing planned repairs and prompt response after emergencies, such as earthquakes or hurricanes.