Machine learning-guided design of ORC working fluids for waste heat recovery

Abstract

Identifying sustainable working fluid alternatives without compromising system performance is critical for advancing the deployment of Organic Rankine Cycle (ORC) technology in renewable energy systems. This study presents a novel computer-aided molecular and process design (CAMPD) methodology developed to design next-generation working fluids for geothermal waste heat recovery in Rankine cycles with different process configurations. The methodology simplifies the complex problem of co-optimizing fluids and cycles into two distinct steps, allowing for the use of detailed cycles and thermodynamic models while effectively managing computational demands. First, flowsheets and operating conditions are optimized for a selection of established fluids using rigorous thermodynamics. Predictive models are then constructed from the resulting optimization data to correlate molecular structure with cycle performance. In the second step, these models facilitate the screening and design of novel fluids without requiring comprehensive characterization. Additional models are employed to predict safety and environmental impact properties, allowing a final multi-objective optimization that seeks to design fluids that effectively balance cycle’s performance, safety and environmental impact. A case study is performed for dual-pressure organic Rankine cycle (ORC) and organic flash cycle (OFC) configurations. The proposed methodology identifies several promising working fluids, including hydrochlorofluoroolefins (HCFOs), with predicted net power outputs of around 16 MW while satisfying the imposed environmental and safety criteria. These computationally designed molecules provide a basis for future thermophysical characterization and experimental validation.