An overview of adoption and maintenance strategies of modern AI-based APC sponsored by the AFPM OPCAT committee.
By Allison Buenemann, Product Marketing Manager, Imubit
American Fuel and Petrochemical Manufacturers (AFPM) recently published a white paper on the Maintenance and Adoption of AI-based APC. This paper was a collaborative effort between AFPM’s Operational Planning, Controls, and Automation Technologies (OPCAT) committee and Imubit Senior Product Manager Aaron Durke. You can read the full white paper, which goes into depth on the following areas.
Requirements for a New Approach to Long-standing Industry Challenges
New AI solutions are overcoming the shortcomings of traditional linear or hybrid-modeling based process optimization and control solutions. Implementing these new technologies has required vendors to prove out rigorous change management procedures to ensure safe implementation, build in Explainability and AutoML capabilities for maximum adoption amongst a range of skill sets and user personas, and develop processes for model evaluation and retraining to ensure ongoing value delivery. The combination of these functionalities is integral to demystifying technically sophisticated solutions addressing operational complexities like feedstock shifts, price volatility, and everchanging environmental and regulatory conditions.
The New Approach
Deep Learning Process Control® (DLPC) uses multiple categories of AI to optimize complex industrial processes. It starts by utilizing historical plant data to create accurate neural network models that reflect the actual plant process dynamics and constraints. This holistic data process model is then put through hundreds of millions of rigorous simulations where reinforcement learning is used to train the model to make optimal decisions to all possible scenarios. This approach overcomes pitfalls of traditional technologies by effectively handling model mismatch and adapting to real-world complexities, like how humans understand and adapt to imperfect information.
Explainability and AutoML are Necessities for Adoption
When AI is used for closed loop optimization and control it’s not being used by data scientists or machine learning experts. Rather, it’s in the hands of domain experts with a broad range of skillsets and programming familiarity. When you’re talking about operators, process engineers, process control groups, and anyone who’s ever worked in plant operations, you know that a black box model will never be sufficient. These personas require a level of trust in the models they’re using – and rightly so, given the high-risk environments they’re operating.
With this understanding, AI technology providers working in the refining sector are unpacking the black box, using techniques like explainability and AutoML to help consumers of model outputs to understand what decisions the model will make in different scenarios, and why. Some of the most effective embodiments of these techniques come in the form of interactive data visualizations, for example allowing users to perform “what-if” scenarios to test model response to a variety of disturbance and manipulated variable changes. These interactive applications confirm nonlinear dynamics, such as gain flipping which occurs in the transition from under to over cracking (Figure 1), and sometimes uncover new dynamics that prompt teams to investigate, collaborate, and learn together.
Figure 1: Interactive gain histograms are an explainability feature of Imubit’s model evaluation technology.
Explainability in AI models builds trust by making AI decisions transparent and interpretable. AutoML empowers domain experts to develop and optimize AI models efficiently. Collaborative functionalities built into software break down organizational silos and drive teams towards a single, unified view of the process, controls, reliability, and market economic situations. Deep Learning Process Control leverages these techniques to provide an end-to-end solution that integrates AI into complex industrial processes.
Want to dive into the details of how closed loop AI optimization is transforming the refining sector? Download the full AFPM white paper.
