(96c) Advancing Petrochemical Processes Inferential Analysis with Eigen's ML Transfomers

Eigen Control Inc. is leveraging the power of Transformer models in Inferential Analysis of Product Properties in distillation and reactor processes. Transformer models are a cutting-edge development in artificial intelligence (AI). We believe it represents a significant shift from traditional empirical based inferential methods which enhances efficiency, scalability, and precision of product quality inferential predictions.

Historically, the concept of Transformers gained momentum with the seminal 2017 research paper "Attention Is All You Need" by the Google Brain team. This paper was instrumental in revolutionizing sequence-to-sequence learning, setting a new precedent in the AI realm. Subsequent adaptations and advancements, such as the 2018 GPT framework, emphasized the self-attention mechanism's role, scalability, and parallelization—each contributing to the complex architectural evolutions and industry agnostic practical applications seen today, such as that of ChatGPT from OpenAI.

The application of Transformer models in petrochemical processing control presents a paradigm shift in handling sequence-to-sequence data and understanding intricate process dependencies. These models, recognized for their self-attention mechanisms, make them particularly suited for the time-sensitive and data-rich environment of refinery operations. The efficiency of Transformers stems from their parallel architecture, allowing real-time processing even on standard low-power computers commonly used in plant control networks. Moreover, their scalability is demonstrated in their adeptness at managing growing data volumes and model complexities, essential for adapting to highly nonlinear processes inherent in refinery control.

This presentation will explore the application and customization of Transformer technology to separation processes, particularly as it applies to a Crude Unit and an NGL Fractionation column. A significant challenge lies in merging Transformer architecture with chemical engineering expertise, encompassing a) equipment and instrument design, b) process design, and c) operations. Additionally, we face the hurdle of incorporating classical mathematics and control theory, such as Kalman filters, for data validation and integrity in real-time applications. We will also discuss the challenges and opportunities presented by the software infrastructure supporting Transformer technology in a control environment. This includes the development and integration pipeline crucial for enhancing the resilience, security, and availability of Transformer-based inferential applications for closed-loop control.

In conclusion, we will present the outcomes of an actual deployment in an NGL Fractionation plant, highlighting the superior accuracy of Transformer-based model results as a significant advantage of this emerging technology.