Design and Algorithm-based optimisation of Axial ORC turbine and transient cycles incorporating novel machine Learning tools

The flue gas stacks of industrial steam boilers can be considered an untapped waste heat source, which is characterised as highly intermittent. Although Organic Rankine Cycles pose strong potential to reuse such low-grade heat, the component and system levels analysis of ORCs to efficiently utilise these highly intermittent heat sources in a techno-economic fashion is still an unanswered research question. Such a holistic approach ultimately expedites the commercial adoption of ORCs to utilise a broader range of waste heat sources achieving the highest possible techno-economic benefits. To answer this research question, emphasising scale ORCs that employ axial flow turbines owing to their scalability and superior isentropic efficiency, this thesis undertakes turbine and cycle configuration optimisation by integrating the Craig and Cox loss model to simulate a small-scale axial flow ORC turbine. The transient waste heat of an actual industrial steam boiler stack was employed as a heat source to investigate ten novel cycle configurations. The optimisation was undertaken using parametric, metaheuristic (nature-inspired) and mathematics-based optimisers. Artificial Neural Networks (ANNs) and genetic algorithms (GAs)-based on the loss model led to an optimised turbine configuration that improved turbine total-to-static efficiency and cycle efficiency by 5.2% and 0.24%, respectively. The recuperative cycle proved the optimal thermodynamic configuration, with a 26.5% increase in mean power generation. Furthermore, a multi-objective analysis revealed the recuperative cycle integrated with an air preheater as the optimum thermo-economic configuration, with a 48.9% improvement in the combined overall value of the multiple objectives, including the Specific Investment Cost and mean power, achieving the final payback within 1.72 years. The ideal configuration was observed as a strong function of the Levelized cost of fuel and electricity prices. Application of a mathematical technique based on the non-linear programming by quadratic Lagrangian algorithm was validated for single- and multi-objective cycle configuration optimisations, providing results comparable to the well-established metaheuristic-based genetic algorithm, with a computational efficiency of greater than one order of magnitude. The overall approach of the direct loss model, artificial neural network- and genetic algorithm-based turbine optimisation, parametric cycle pre-optimisation, mathematical technique-based component optimisation and payback evaluation can be considered a blueprint for the future evaluation and design of organic Rankine cycles utilising transient waste heat sources