Abstract:
In industrial process plants, optimization of the Condensate Recovery Factor (CRF) is critical with regard to energy efficiency and sustainability since it quantifies the reuse of condensate- a byproduct of a steam system, to reduce waste of resources and cost of operations. Despite the developments in IoT-enabled data gathering, there are challenges in transforming minute-scale steam production and the succeeding condensate recovery information into actionable conclusions. Traditional frameworks struggle with changing operational complexities such as varying rates of steam flow and system delay, whereas black-box machine learning methods lacks transparency, which hinders trust and practical implementation. This research addresses these gaps by suggesting a hybrid framework that combines physics-informed machine learning with domain knowledge. A Fourier-inspired neural network structure is employed to represent temporal condensate recovery cycle patterns, assisted with genetic algorithms for hyperparameter adaptation. To bridge the gap between automation and human judgement, Explainable Artificial Intelligence (XAI) techniques like SHAP and LIME are leveraged to explain the model behavior, thus enabling engineers to compare predictions with operational limitations. The approach focuses on interpretability without compromising on predictive power, ensuring that the insights gained are consistent with industrial best practices. By harmonizing IoT data streams with collaborative human-AI analysis, this work advances data-driven decision-making in steam system operations. It illustrates how Industry 4.0 technologies can transform underutilized datasets and convert it into strategic assets, driving sustainability through smarter resource utilization. The outcomes highlights the relevance of embedding domain knowledge into an AI system, and providing a scalable solution for industries to achieve operational excellence amidst the intricacies of digital innovations transformation.
