Mihail-Bogdan Carutasiu

Hybrid heating systems combining air-source heat pumps (ASHP) and gas boilers (GB) require advanced control to achieve both high energy efficiency and low operating costs. Digital Twins (DT) have emerged as a promising solution for supporting such control strategies, although their practical deployment remains challenging due to the integration of heterogeneous physical models, real-time data streams, and predictive decision-making algorithms.,This paper presents the HESTIA digital twin framework, a control-oriented architecture developed for a hybrid HVAC experimental platform integrating an air-source heat pump, a condensing gas boiler, a stratified puffer tank, while also allowing photovoltaic generation integration. The proposed framework follows a four-layer architecture comprising the physical system, data acquisition, model and decision, and visualization parts. At its core, the framework combines three predictive models: a Physics-Informed Neural Network (PINN) for heat pump performance forecast, a semi-empirical gas boiler model, and a three-node stratified thermal model of the puffer tank. The framework's subsystems are validated using external operational datasets to demonstrate feasibility prior to live-deployment. The system’s control is implemented through a seven-mode finite state machine (FSM) that dynamically adapts the operating strategy according to energy prices, photovoltaic production, and system operating conditions.,The ASHP PINN was trained using 7,389 operational measurements and 3,060 EN 14511 certified reference points collected from publicly available datasets. The resulting model achieved an R² of 0.796 for coefficient of performance prediction, while thermal output and electrical power estimation reached R² values of 0.957 and 0.917, respectively. Moreover, the embedded thermodynamic constraints successfully prevented physically inconsistent predictions, with no violations of the Carnot efficiency limit observed during testing. The semi-empirical gas boiler model reproduced manufacturer-declared efficiencies with deviations below 2% under both condensing and non-condensing operating conditions.,The proposed framework establishes the computational foundation of the HESTIA digital twin and provides a modular architecture that can support future real-time deployment, predictive control, and experimental validation as the physical platform becomes fully operational.

This paper presents the HESTIA Digital Twin (DT), a control-oriented digital-physical architecture for real-time state estimation of a multi-source hybrid HVAC system. This experimental infrastructure integrates an air-source heat pump (ASHP), a methane gas condensing boiler, a puffer storage tank, and a rooftop photovoltaic array. The DT is organized as a four-layer hierarchical architecture encompassing physical instrumentation, data acquisition, model computation, and data visualization. The Model and Decision Layer is composed of three independent predictive sub-models: a Physics-Informed Neural Network (PINN) for ASHP performance prediction, a semi-empirical model for the gas boiler analysis, and three-node stratified ordinary differential equation model for the puffer tank. For the PV energy production data from the inverted was considered. A total seven-mode finite state machine encodes the economic dispatch logic, dynamically adapting the heat pump competitiveness threshold to prevailing electricity and gas tariffs. The ASHP PINN enforces four structural thermodynamic constraints, and the model was trained on 7,389 datapoints and 3,060 EN 14511 certified references obtained from two open-source databases. On the held-out test set, the model achieves R² = 0.796, MAE = 0.351, and RMSE = 0.515 for COP prediction, R² = 0.957 (MAE = 0.219 kW) for thermal output, and R² = 0.917 (MAE = 0.126 kW) for electrical power consumption, with zero violations of the Carnot upper bound recorded across all test samples. The gas boiler model reproduces manufacturer- declared efficiencies of 97% and 105% at the rated condensing and non-condensing operating conditions, with deviations below 2%.

This paper presents the HESTIA Digital Twin (DT), a control-oriented digital-physical architecture for real-time state estimation of a multi-source hybrid HVAC system. This experimental infrastructure integrates an air-source heat pump (ASHP), a methane gas condensing boiler, a puffer storage tank, and a rooftop photovoltaic array. The DT is organized as a four-layer hierarchical architecture encompassing physical instrumentation, data acquisition, model computation, and data visualization. The Model and Decision Layer is composed of three independent predictive sub-models: a Physics-Informed Neural Network (PINN) for ASHP performance prediction, a semi-empirical model for the gas boiler analysis, and three-node stratified ordinary differential equation model for the puffer tank. For the PV energy production data from the inverted was considered. A total seven-mode finite state machine encodes the economic dispatch logic, dynamically adapting the heat pump competitiveness threshold to prevailing electricity and gas tariffs. The ASHP PINN enforces four structural thermodynamic constraints, and the model was trained on 7,389 datapoints and 3,060 EN 14511 certified references obtained from two open-source databases. On the held-out test set, the model achieves R² = 0.796, MAE = 0.351, and RMSE = 0.515 for COP prediction, R² = 0.957 (MAE = 0.219 kW) for thermal output, and R² = 0.917 (MAE = 0.126 kW) for electrical power consumption, with zero violations of the Carnot upper bound recorded across all test samples. The gas boiler model reproduces manufacturer- declared efficiencies of 97% and 105% at the rated condensing and non-condensing operating conditions, with deviations below 2%.

Reducing consumption of energy and emissions of greenhouse gas emissions is a critical worldwide problem, with the buildings’ sector accounting for up to 44% of fossil fuel consumption. Residential buildings, in particular, have great technical potential for energy savings. Our study addresses this by creating a model for real-time energy efficiency analysis of building insulation using infrared imagery and computer vision. The model monitors and offers important insights of energy transfer by continually evaluating a building’s thermal envelope. Additionally, an experimental stand comprising of a miniature model house made from different materials and insulation types, an IR camera and a Raspberry Pi were considered. It will provide data for an AI based model, which employs a robust and already proven architecture, YOLOv8 namely, used to determine the thermal characteristics of the analysed materials and the energy performances respectively.

The importance of mitigating energy intensity in the building sector is widely recognized by most stakeholders, and different strategies are proposed for minimizing fossil fuel dependency. This paper presents a complex study assessing the possibility of integrating an air-source heat pump (ASHP) into the retrofitting process of a typical Romanian residential building. Mathematical simulations were conducted to estimate the energy reduction potential of a complex renewable-energy-based HVAC system. The results showed that a more than 50% increase in energy efficiency could be obtained by using the right insulation materials, and the building’s fossil fuel dependency could be curbed by 75% by integrating the proposed renewable energy systems.

An optimization model that may be applied to analyze building retrofit strategies is presented in this research. The aim of this research paper is to identify the optimal thermal envelope configuration that will assure the minimum energy requirement for heating in the case of a residential building, while also considering price restrictions obtained through a specific market survey. To achieve this, several values for the following parameters are considered: thermal insulation materials’ conductivities and thicknesses, windows’ overall heat transfer coefficients and total solar energy transmittance and doors’ thermal proprieties. Additionally, this paper presents a method used to find the best option from among the available heat pumps that could cover most of the energy requirements for heating and domestic hot water systems, also considering the products’ prices. The proposed method is based on a Non-dominated Sorting Genetic Algorithm II (NSGA-II) model developed in the Pymoo (Multi-Objective Optimization in Python) library. The result shows that the energy requirement for heating can be reduced by up to approximately 75% compared to that obtained in the case of a non-insulated building by using suitable insulation materials and doors and windows with superior thermal proprieties chosen by the NSGA-II.

Numerical investigation for turbulent flow of nanofluid (Ethylene glycol/water-Al2O3) is carried out with a volume fraction of 0.5%, 1%, and 1.5%. The diameter of solid nanoparticles is equal to 20 nm. The range of Reynolds number in this simulation is from 6300 to 27500 through a three-dimensional steel corrugated spirally tube. For enhancement of the heat transfer performance in internal flow through the horizontal tube, the tube is combined with aluminum twist tape of a thickness of 0.8 mm and a width of 10 mm along the spirally corrugated tube. In the present work, the same twisted ratios TR=(y/w) are implemented for corrugated tube and tape. The predicted results showed that there is a clear increase in the values of heat transfer and the pressure drop in the direction of flow. Furthermore, the friction factor is increased because of the grooved and twisted tape which rise the resistance of the fluid flow. The values of Nusselt number are increased with the VOF of nanofluid. These numerical results are compared with the experiment found in the literature and showed a good agreement.