ORLEN Unipetrol RPA
ORLEN Unipetrol RPA, s.r.o. is a leading Czech producer of refinery, petrochemical, and agrochemical raw materials, which is also reflected in the abbreviation RPA: refinery, petrochemistry, agrochemistry. The company mainly supplies the market with motor fuels, heating oils, asphalts, liquefied petroleum products, oil hydrogenates, other refinery products, olefins and aromatics, agrochemicals, carbon black and sorbents, and polyolefins (high-density polyethylene, polypropylene).
Starting situation and description of the problem
For high-quality production of the products in the company's portfolio, it is essential to maintain their prescribed composition. To achieve this, Unipetrol regularly performs analytical tests during the production process for the purposes of in-process and final inspection. This process of taking samples and analyzing them in laboratories is demanding both in terms of time and human resources, and it also tends to be costly financially.
Unipetrol's requirements
Unipetrol is therefore looking for a solution in the form of software that would enable immediate prediction of the qualitative parameters of streams (semi-products or final products) on the basis of commonly available process parameters. The required tool, a so-called virtual sensor, is to take the form of a system called an automated neural network, with a strong emphasis on being able to export the calculation algorithm (the trained network) as code into other software, such as MS Excel. Given the fact that all potential suppliers offered only alternative solutions, Unipetrol chose, as its so-called virtual analyzer solution, the software Statistica Automated Neural Networks CZ + Code Generator.
Implementation process
The software installation was a completely simple process, which Unipetrol's own representatives managed themselves. The installation was preceded by a two-day training on the Neural Networks module, which took place while the tender process was still ongoing.
The most complex process is selecting the data with which the neural network will work.
Ing. Jiří Schöngut, CSc., Head of the Process Development Department, Unipetrol RPA, s.r.o.
Cooperation with the supplier
The cooperation with the supplier was, I would say, exemplary. During testing, only a few minor errors appeared in the export, such as incorrect syntax during code generation. The supplier's team reacted and stepped in to fix them immediately and, above all, helpfully. All support was also provided to our complete satisfaction.
Ing. Jiří Schöngut, CSc.
Final situation
The software, as a tool for creating a neural network model that serves to instantly estimate the composition of the manufactured product, currently helps the company significantly to reduce the frequency of sampling, stabilize quality, and maximize the production of the desired quality. The procedure itself is as follows: a neural network model created in the Statistica software on the basis of historical data, then implemented in the Excel program, uses the available process data and calculates a prediction of the qualitative parameters (e.g. composition) of the monitored stream. Thanks to data processed in this way, the operator of the given process can immediately take any steps needed to correct the technological process and does not have to wait for the lengthy process of laboratory analysis. A timely response thus significantly reduces losses in the given process. Statistica Automated Neural Networks CZ + Code Generator saves Unipetrol on the order of millions of CZK per year.