SIBUR presents its approach to applying AI in applied R&D

23 April 2026. Scientists within the SIBUR PolyLab R&D ecosystem are using artificial intelligence to manage the properties of materials and finished products. This was announced by Artur Aslanyan, Head of Process and Service Development at SIBUR PolyLab.

At the idea generation stage, modern digital solutions help analyse large volumes of global patents, scientific publications, news and technical regulatory documentation, identify promising areas and application niches for polymer solutions, and shift from a catch-up development model to a predictive one. Instead of manual analysis, SIBUR PolyLab experts use LLM-based tools that structure knowledge and accelerate decision-making.

In addition, SIBUR scientists work with large datasets generated at every stage of polymer production — from pyrolysis and synthesis to extrusion, laboratory testing, and conversion into finished products. These datasets include production line parameters, formulations, test results, and data from SIBUR PolyLab centres. Based on this information, predictive models are developed to establish relationships between composition, processing conditions, and the final material properties.

One of the most illustrative examples is a formulation generator for polypropylene films. Based on target performance requirements (strength, transparency, modulus of elasticity, and other parameters) and taking into account a specific production line, the model creates the optimal multilayer film structure, calculates component ratios, layer thicknesses and expected material properties.

A similar approach is now being applied in another area: the development of a model for predicting product colour parameters when recycled feedstock is incorporated. The model analyses historical data on recycled material characteristics and their share in virgin polymers, enabling preliminary assessment of the impact of recycled content on the appearance and quality of the final product. The model is currently being trained to improve forecasting accuracy.

Another focus area is the management of finished product properties. Solutions aimed at protecting products against counterfeiting by using special markers and digital control tools accessible to all participants across the value chain — from processors to end consumers — are currently under development.

Artur Aslanyan, Head of Process and Service Development at SIBUR PolyLab, commented:

“We see artificial intelligence as a practical tool that is already transforming development approaches at SIBUR, enabling not only process acceleration but also highly accurate prediction of end-product properties. The key challenges in AI implementation remain trust in model outputs, the handling of confidential data, and the quality of source information. To address these challenges, we focus on employee training, embedding AI tools into daily workflows, developing internal RAG solutions, and automating data collection.”

The development of AI in applied research is supported by SIBUR’s systematic digitalisation of its scientific divisions. At the SIBUR Innovations R&D centre, hypotheses related to AI-based accelerated catalyst and new material design are being tested: models analyse large experimental datasets, identify relationships between composition, structure and substance properties, help predict catalytic system efficiency, and reduce the volume of labour-intensive laboratory testing. This approach makes it possible to move from lengthy trial-and-error solution selection towards a more precise and manageable scientific search, creating a foundation for faster commercialisation of technological solutions and strengthening the Company’s in-house research capabilities.