Underground Thermal Energy Storage: new solutions and sustainability

Latent heat thermal energy storage (LHTES) can buffer mismatches between heat supply and demand, but its performance is controlled by phase-change transport in complex geometries and materials. This thesis develops and applies an integrated CFD + machine-learning (ML) framework to quantify how fin geometry, fin material, and nano-enhanced PCM (NEPCM) jointly influence charging and discharging of a shell-and-tube unit. Melting/solidification is modeled with an enthalpy–porosity formulation and Boussinesq buoyancy; NEPCM is represented by homogeneous, temperature-dependent effective properties. Four heat-exchange layouts are examined—plain tube, annular fins, vertical fins, and a helical coil—across practical PCM fractions (20–25%) and candidate fin materials/coatings, including copper and graphene-coated aluminum. The CFD data train accurate surrogates for melt fraction and stored energy, enabling rapid design sweeps and multi-objective trade studies. Time-resolved simulations recover the canonical sequence of wall-conduction, convection-aided melting, and a conduction-limited solidification tail, and yield a robust geometry ranking: helical > vertical fins > annular fins > plain tube. Increasing fin length and thickness reduces solid-side resistance and advances the onset of natural convection; within the studied range, vertical fins of 22 mm length and 5 mm thickness cut melting time by up to ~38% versus shorter/thinner variants, with diminishing returns beyond ~20–30 fins due to thermal-field overlap and reduced flow space. Material studies show copper maximizes heat spreading, while graphene-coated aluminum offers a favorable performance–mass–cost compromise. For NEPCM, low-loading carbon fillers (e.g., MWCNT at ~0.6 wt%) reduce simulated melting time by ~44% relative to pure PCM and outperform common oxides at similar loading. A techno-economic map (normalized heat stored vs. normalized cost) shows a clear “knee” at 20–25% PCM: this band captures most of the achievable capacity at modest cost, whereas higher fractions bring diminishing returns and greater risk of conduction bottlenecks late in the cycle. The ML surrogates reproduce CFD trends with high fidelity (R² ≈ 0.99 overall; helical cases ≈ 0.98), enabling design clustering (fast/high-rate archetypes centered on vertical fins at 20–25% PCM) and near-instant prediction for optimization and control. Limitations include the homogeneous NEPCM assumption and fixed boundary conditions; reported gains should be viewed as upper bounds absent dispersion loss, viscosity-induced pumping penalties, or sedimentation over cycling. Overall, the framework provides a physics-grounded, data-efficient pathway to prioritize prototypes—helical coils or moderate-count vertical fins coupled with low-loading CNT-based NEPCM near 20–25% PCM—and a reusable surrogate for system-level studies. These insights inform the design of next-generation LHTES modules aligned with decarbonization targets and operational reliability.