PERM POLYTECH UNIVERSITY SCIENTISTS’ NEURAL NETWORK PREDICTS FOUNDATION SETTLEMENTS OF BUILDINGS

14 September 2021 127

The development was carried out with the financial support of the Russian Foundation for Basic Research.

The development was made by a young scientist Jan Ofrichter and his academic advisor, professor of the Department of Construction Operations and Geotechnics, Doctor of Technical Sciences Andrei Ponomarev.

— To predict the settlement of piles in the soil, it is necessary to take into account many factors: soil characteristics, geometric dimensions of the pile and the method of its installation. Research is carried out using direct and indirect methods. The former require financial costs, while the latter require additional calculations. Therefore, we used a different approach: we built an artificial neural network and trained it to predict direct tests based on the data of cost-effective indirect ones, — says the author of the project, assistant, Research Assistant at the Department of Construction Operations and Geotechnics of the Perm Polytech University, Jan Ofrichter.

The development will allow not to carry out additional research in the future. Instead, you will need to populate the existing database. Due to the ability of neural networks to process complex dependencies, it is possible to use the results of various tests in a complex. This approach will allow studying a larger number of soil characterictics and significantly improve prediction accuracy.

To get the result, the neural network uses data from direct tests, static probing, vertical load and pile characterictics. Scientists have tested many neural network models to find the most efficient architecture. As a result, they got a network of 4 hidden layers, each of which consists of 200 neurons. The researchers carried out machine learning with a backpropagation algorithm using a database of already existing objects.

The results presented by the neural network turned out to be 15-20% more accurate than the precipitation data using traditional methods. According to scientists, the development has improved the accuracy of the prediction using less information.

Print:

Read all news

Same News

March

2023

Perm Polytech enrolls for English-taught programs "Information Technology and Big Data analysis" and "Computer Systems and Communication Engineering"

Partner of Perm National Research Polytechnic University (PNRPU) - Shandong consulting company in the field of international education "Bridge of the Russian language" is implementing a joint cooperation project on the English-taught programs "Information Technology and Big Data analysis" and "Computer Systems and Communication Engineering". These programs are provided by the Department of Information Technology and Automated Systems (ITAS), Faculty of Electrical Engineering of PNRPU.

November

2022

TECHNOLOGY OF PNRPU SCIENTISTS WILL RESTORE VALUABLE ELEMENTS FROM PRODUCTION WASTE

Now more and more attention is paid to the process of restoration of nitrogen and phosphorus from industrial and agricultural waste. This is connected with the need to implement the concept of sustainable development of society, based on the circular economy. These elements are biogenic, vital for plant growth. Phosphorus reserves are limited, and the process of obtaining nitrogen is energy-consuming. Potential sources of secondary raw materials are water waste from many industries: the fertilizer industry, pharmaceutical products, semiconductor production and agriculture. The recycling of nitrogen and phosphorus allows not only to reduce the consumption of non-renewable mineral resources, but also to reduce the saturation of natural water bodies with biogenic elements, the excess of which leads to swamping of water bodies and to a decrease in the amount of drinking water. To solve this problem, PNRPU scientists have developed a new approach that has increased the efficiency of extracting ammonium (a form of nitrogen) and phosphate (a form of phosphorus) from industrial waste water to 98-99%. The traditional method does not achieve such results. Moreover, the technology of Perm scientists will make it possible to produce domestically produced long-acting fertilizer based on struvite (a phosphate mineral) and will contribute to the import substitution of foreign analogues.

November

2022

THE MODEL CREATED BY PNRPU SCIENTISTS WILL PREDICT DEFECTS IN LASER POWDER SURFACING

The method of laser powder surfacing is used to restore worn-out or damaged metal parts. A powerful fiber laser creates a molten weld pool, and then metal powder is automatically fed into it. A common problem that occurs in this process is the deformation of the part, being repaired, under the action of internal stresses. This leads to a defect called warping. One of the aims of developing a technological process for the repair of parts by laser powder surfacing is the prediction of warping defects. The determination of the optimal defect-free surfacing mode for specific parts experimentally is unprofitable due to high financial and time costs and often even impossible. The researchers of the PNRPU together with JSC “UEC-Aviadvigatel” developed a mathematical model for calculating warping during laser powder surfacing to solve this problem. The scientists were able to predict with high accuracy the behavior of the part during the surfacing process thanks to numerical simulations. In the future, the development will help in choosing the optimal set of parameters for repairing parts with the most complicated parts.