An initial screening of adsorbent materials revealed that composite adsorbents, made of a porous host matrix impregnated with a hygroscopic salt, are the most promising materials for low-temperature adsorption (T<50 °C).
A wet-impregnation method for preparing composite adsorbent materials with controlled hygroscopic salt content was developed.
An optimized composite adsorbent consisting of mesoporous silica gel impregnated with CaCl2, with water adsorption capacity >25% by weight and a regeneration temperature of 50 °C was prepared and characterised.
A novel vapour-permeable resin was formulated and its associated 3D printing technique was optimised.
The best TPMS-based HEX structures were defined by adjusting key design parameters (porosity, thickness, and unit cell length) .
COMSOL Multiphysics modelling was performed for transient simulations of H&M transfer, to support the design of the adsorber HEX .
A prototype of a TPMS-based adsorbent HEX, involving the design and 3D printing of the HEX prototype (overall volume ~2 L) and preliminary functional testing to evaluate its hydraulics, seals, and overall robustness was performed.
A flow channel (pipe) with micro-fin internal surface that ensures seamless switching between evaporation and condensation modes within the same heat exchanger was developed.
The initial design for new refrigerant pumps optimized to handle the liquid phase of the refrigerant with minimal cavitation and reduced energy consumption was undertaken.
An external evaporator and condenser (tubular with external fins) with optimized heat transfer surfaces (micro-fin for evaporation and condensation processes) for the integration into the sorption pump was developed.
A preliminary analysis performed showed that the current cooling architecture does not allow to cool the racks with a water temperature higher than 45 °C, thus an adsorption chiller cannot be run for central data centre cooling (BSC use case). A change in the cooling circuit is needed.
The initial design of an adsorption chiller able to work with a temperature of 48-55 °C has been completed. The MODBUS TCP/IP protocol has been agreed for the integration of the chiller in the overall architecture. The racks will be initially simulated using an electric heater matching the rack rejected power.
A prototype for rack-integrated data centre cooling has been adapted and sent to IQ. The preliminary testing with two-phase flow is in progress (WP3).
The initial design of an adsorption chiller able to work with a temperature of 55-60 °C has been completed for rack-integrated data centre cooling . The MODBUS TCP/IP protocol has been agreed for the integration of the chiller in the overall architecture.
A first prototype of the ontology is based on the simulation components found in the pilot data centres.
An initial ICT architecture to achieve a proper system integration where simulations, monitoring and most probable control set points can optimise the thermal performance of the data centres has been defined. Virtual tests have also been defined and the platform is waiting for the synthetic data.
A literature review focusing on privacy-enhancing technologies suitable for federated learning such as Homomorphic Encryption (HE) and Differential Privacy (DP) is underway. Advanced AI methods—particularly Multi-Agent Reinforcement Learning (MARL)—as a foundation for decentralized control and optimization strategies are ongoing.
A generic Modbus plugin to communicate with the cooling devices has been developed and tested.
A specific module for the adsorption chiller implementing different sensors and knobs and using Modbus underneath has been developed.
A first implementation of the cooling monitor, which periodically get data from the adsorption chiller has been developed. Test with a simulated chiller has been performed.
A report plugin which uses OpenMetrics format, Prometheus, Grafana and a timeseries DB to store and visualize data from the cooling sensors has been developed.
Synthetic datasets are being curated to emulate real-world behaviour where live data is sparse or incomplete. This data will serve as the foundation for model training, evaluation, and early prototyping for the high-level control and building system integration.
A prototype implementation in Python using Keras/TensorFlow is being drafted for testing on available synthetic datasets.
