PNIPU: Method of Developing a Virtual Diesel Boiling Point Analyzer Using Neural Networks

Main article: Neural networks (neural networks)

2023: Introduction of a method for developing a virtual diesel boiling point analyzer using neural networks

The development of scientists from the Perm Polytechnic will simplify quality control in oil refining technology. The university announced this on December 15, 2023.

In the oil refining industry, the volume and quality of products produced are the most important indicators of operational efficiency. They depend on compliance with the required process parameters. Oil samples are taken and analyzed in the laboratory to determine quality parameters, however, this is a labor-intensive and expensive process. An alternative method was the use of virtual sensors, which allow you to quickly receive and reliably transmit product information. Scientists of the Perm Polytechnic University have created a method for developing a virtual analyzer of the boiling temperature of diesel fuel using neural networks.

All refineries are based on process plants in which oil is divided into fractions (gasoline, kerosene, diesel) and then processed or used as components of marketable petroleum products. This is how almost all components of motor fuels, lubricating oils, raw materials for secondary processes of oil refining and petrochemical industries are produced.

Proper process control is required to obtain a quality product. Virtual analyzers, using the results of direct measurements of parameters (temperature, flow, pressure), function as conventional measuring devices and can be used in computer control systems in conjunction with them. Thus, they allow the operator to control the quality of products during the process using a computer at a certain stage of production, "said Anna Streltsova, a graduate of the master's degree in the Department of Equipment and Automation of Chemical Industries at PNIPU.

When processing oil, the raw material is heated, at which it is important to control the increased temperature. Perm Polytechnic scientists have built a virtual analyzer to control the boiling point of diesel fuel. This temperature determines the completeness of its evaporation. At too high values, some fractions do not have time to evaporate and remain in the liquid phase in the form of droplets and a film, which leads to increased carbon formation and wear of the equipment. The virtual sensor will allow you to quickly and continuously monitor the entire heating process in the production of diesel fuel.

The operation of the virtual analyzer is based on a mathematical model that reflects the relationship of process variables and boiling temperature as an indicator of quality, for example, in the production of diesel fuel. A feature of the proposed method is a two-stage approach. At the first stage, we built a neural network model of the relationship, at the second - based on the results of a computational experiment with the resulting neural network, we built a regression model that predicts the values ​ ​ of the quality indicator of the oil product produced. The developed regression model is installed in the computer production control system, - said Alexander Shumikhin, scientific director, doctor of technical sciences, professor of the Department of Equipment and Automation of Chemical Industries at PNIPU.

When building a virtual sensor, polytechnics took into account such input data as oil density, fuel oil consumption, temperature of petroleum product vapors and oil itself. These values are supplied to the input of the trained model, and at the output the boiling point of diesel fuel is obtained. This control method is much more profitable than tool analyzers, easier and cheaper to implement and maintain. It is also much faster than normal laboratory analysis, since the virtual analyzer shows the state of the oil product continuously with the update of values ​ ​ in the control system at any given time interval.

According to experimental data obtained on real physical equipment, the scientists proved that the resulting regression model is statistically adequate to this data. This means that the developed method of building virtual analyzers is promising for use in the oil refining industry. The researchers note that the models can be adapted to new experimental data, which makes them applicable to other indicators of the quality of petroleum products. The virtual analyzer of PNIPU scientists to determine the boiling point of diesel fuel will simplify the technology of monitoring oil quality indicators. Monitoring this stage at the enterprise will reduce the risks of equipment wear and deterioration in the quality of marketable petroleum products. The development will increase the speed and accuracy of information acquisition, which will significantly affect the efficiency and economic benefits of crude oil processing processes.