In the May issue of the Scientific Reports journal, a research by an international team of scientists from Saudi Arabia, Iraq, India, and Russia was published. Its topic was the advanced features of flowmeters. The team also included our fellow-countryman, Deputy Head of the SUSU Research Laboratory for Self-Validating Sensors, Systems, and Advanced Instrumentation Vladimir Sinitsin.
This research covers the use of machine learning for modelling of heat transfer in nanofluids.
“The traditional methods of CFD modelling yield accurate results, but require huge computational capacity,” explains Vladimir Sinitsin. “CFD modelling is computer-aided simulation of the behaviour of fluids with the solving of Navier-Stokes equations and heat transfer for complex systems (for example: pipes with nanofluids in them). This requires very powerful servers and much time. Modelling of heat exchange in just one heat-exchanging unit could take up no less than two to four weeks! Our method’s innovation is that we have combined CFD methods and the algorithms of machine learning: GPR, KNN and MLP. It’s as if you switch from hand calculations to smart predicting: faster, cheaper, but with the save level of accuracy.”
The results are impressive: the GPR and KNN models predict the temperature distribution to the accuracy higher than 99.8%. This is a true breakthrough for manufacturers of flowmeters and transducers!
For instance, say, a company is designing a new transducer for chemical production. Earlier, weeks of expensive CFD simulations would have been needed to calibrate it. The models proposed by the team of scientists, which Vladimir Sinitsin is also part of, allow to obtain similar data within mere hours, and at minimum expenses.
Of special interest is the use of nanofluids with CuO particles, which improve the heat transfer significantly. CuO is copper oxide (Cu is copper, and O is oxygen). These nanoparticles are added to fluids (water, oil) in order to increase the thermal conductivity by 20–30% and improve the flow stability.
“We saw a clear parabolic profile of temperature in the pipe,” the scientist notes. “This totally complies with the flow physics and proves that the models work properly. For engineers, such data are a gold mine: they help optimize the design of transducers and improve their accuracy.”
This development can be used in power generation, petrochemistry, and climate control systems. The team and Vladimir Sinitsin are planning on adapting the models for other types of fluids and complex industrial systems. This step opens up a path towards a real revolution in instrumentation engineering.