Back to AI Research

AI Research

Explaining Process Control Optimisation Recommendat... | AI Research

Key Takeaways

  • Explaining Process Control Optimisation Recommendations via GradientSHAP and Implicit Differentiation Modern industrial processes rely on automated optimisat...
  • Automated optimisation is increasingly adopted in industrial processes, yet a trust gap persists between engineers who design these algorithms and operators who must act on their recommendations.
  • Explainable AI methods like SHAP (SHapley Additive exPlanations) have transformed interpretability for machine learning predictions; optimisation outputs could benefit from similar techniques.
  • Our approach leverages IFT to compute exact parameter sensitivities $\partial p^*/\partial x$ from the optimality conditions, enabling efficient GradientSHAP computation.
  • We validate on industrial scenarios and present feedback from domain experts on generated explanations.
Paper AbstractExpand

Automated optimisation is increasingly adopted in industrial processes, yet a trust gap persists between engineers who design these algorithms and operators who must act on their recommendations. Explainable AI methods like SHAP (SHapley Additive exPlanations) have transformed interpretability for machine learning predictions; optimisation outputs could benefit from similar techniques. We present an approach that integrates Implicit Function Theorem (IFT) based sensitivity analysis with SHAP attribution and narrative generation via Large Language Models (LLM), producing explanations tailored for operators. Our approach leverages IFT to compute exact parameter sensitivities $\partial p^*/\partial x$ from the optimality conditions, enabling efficient GradientSHAP computation. For an industrial High Pressure Grinding Roll (HPGR) control optimisation problem with 22 features, we achieve equivalent SHAP attributions (correlation $>$0.99 with KernelSHAP) with over 40$\times$ speedup, enabling real-time natural language explanations. We validate on industrial scenarios and present feedback from domain experts on generated explanations.

Explaining Process Control Optimisation Recommendations via GradientSHAP and Implicit Differentiation
Modern industrial processes rely on automated optimisation to improve efficiency, but a "trust gap" often exists between the engineers who design these systems and the operators who must implement their suggestions. When an algorithm recommends a change—such as adjusting pressure or speed in mineral processing—operators need to understand the "why" behind the decision to ensure safety and performance. This paper introduces a new framework that combines mathematical sensitivity analysis with Large Language Models (LLMs) to provide clear, human-readable explanations for these automated recommendations.

Bridging the Gap with Explainable AI

The researchers aim to make optimisation algorithms transparent by applying techniques typically used for machine learning models. They focus on the High Pressure Grinding Roll (HPGR), a piece of equipment used in mineral processing that requires constant adjustments to hydraulic pressure and roll speed. By using SHAP (SHapley Additive exPlanations), a method that assigns a value to each input feature based on its contribution to a decision, the team can break down exactly which factors—such as feed material properties or production targets—drove a specific change in the machine's settings.

Efficient Sensitivity Analysis

A major challenge in explaining optimisation is that traditional methods like KernelSHAP are computationally expensive, often requiring hundreds of repeated calculations to produce a stable result. To solve this, the authors use the Implicit Function Theorem (IFT). By mathematically differentiating through the optimality conditions of the control problem, they can calculate the exact sensitivity of the output to changes in the input in a single step. This approach allows them to use GradientSHAP, which is significantly faster than traditional sampling methods. In their tests, this method achieved over a 40x speedup while maintaining a correlation of over 0.99 with standard SHAP values.

Translating Data into Narratives

Once the system calculates the numerical importance of each input, it passes this information to an LLM. The LLM acts as a translator, converting raw data into a narrative that an operator can easily understand. Feedback from domain experts highlighted that operators prefer actionable summaries over complex technical breakdowns. Consequently, the system is designed to lead with a one-sentence summary, use familiar units like percentages, and avoid showing raw, confusing SHAP values. This ensures the output is practical for daily plant operations.

Current Limitations and Future Directions

While the approach is effective for the HPGR process, it currently has some limitations. The methodology is designed for unconstrained optimisation, meaning it works best when the optimal settings fall within the normal operating range. It does not yet account for complex scenarios where the system hits hard safety or equipment constraints. Additionally, the current model requires the underlying process to be mathematically differentiable. Future work will focus on extending this framework to handle constrained problems and conducting formal user studies to further refine how these explanations support operator decision-making.

Comments (0)

No comments yet

Be the first to share your thoughts!