Modeling and Simulation of Fluid Flow and Energy Systems

Abstract

Engineering and science-related problems become more complicated as human knowledge evolves. This complication includes apparatus geometry and operational environment such as extreme variations in pressure and temperature. Analytical solution for such problems needs many assumptions that underestimate the problem under study and could lead to unrealistic results. Moreover, an experimental setup for a certain problem is constrained by the prototype size and each experiment is set up for certain operating conditions. This leads to building up many setups to deal with changes in size and operating conditions and, therefore, the prototype validation becomes very expensive and time-consuming. This calls for modeling and simulation approaches to deal with such engineering problems with the powerful computational capabilities available nowadays. Real-world patterns and processes are roughly modeled by scientific models. They may be refuted because they are representations, which are by definition imperfect. Models, however, are quite helpful for a variety of reasons. They first give us a method to comprehend procedures that would otherwise be incomprehensible to us. They also give scientists a base on which to build new research and theories. Finally, modeling and simulation reduce the time and cost of prototyping.

1. Introduction

Modeling is a small-scale representation of a process or a device, using mathematical descriptions implemented by a computer program to predict the behavior of a system at different operating conditions and at any size scale of a process or a system. Simulation is the process of representing the results of the model in a clear and understandable format like figures, tables, and animation. Simulation is used to examine the function of a device that is not subjected to experimentation. A good example of this is modeling and simulation of energetic materials (explosives) where experimentation is almost impossible due to its hazardous effects and very expensive cost [1, 2].

Simulation provides evidence for decision-makers to proceed to industrial and commercial production of the products. Verification and validation of the model are very important and for that sufficient data and information are used to prove the result of any model. Modeling and Simulation have many advantages since they reduce the time and cost of real experimentation, are suitable for severe operation conditions (very low and high temperature, very low and high pressure) in which real experimentation is very expensive, and give information that increases the confidence of the stakeholders in the product.

This chapter discusses some of the engineering applications related to our own experiences. Three subjects are covered; fluid flow and energy systems. Each subject is discussed in three categories. Namely, CFD packages, object-oriented programming, procedural programming, and the most recent usage of Artificial Intelligence (AI) for fluid flow and energy systems.

ANSYS-FLUENT uses the finite volume method to solve the governing equations of fluid flow and heat transfer. This involves dividing the domain of the simulation into a series of small, interconnected control volumes or cells. The values of the variables (such as velocity and temperature) are then calculated at the centroids of these cells. ANSYS-FLUENT uses iterative algorithms to solve these equations and arrive at a converged solution.

Artificial Intelligence (AI) has recently been used in the modeling and simulation of fluid and energy systems [3, 4, 5, 6, 7, 8]. The term ML is also used synonymously. Machine learning (ML) arguably has four types: supervised learning, unsupervised learning, reinforcement learning (RL), and deep learning (DL). In supervised learning, the model learns from labeled data. Hence, the input and output are known. Whilst in unsupervised learning, the output is not known and the data are unlabeled. The model learns the similarities and correlations in the unlabeled data. On the other hand, RL, the third type of machine learning [3, 4], attains long-term objectives by direct contact with the environment, without prior knowledge. The RL agent assesses the state of the environment and selects the best course of action. The RL agent is rewarded positively if it does the right action. A negative is created if the agent commits a mistake. In RL, the model learns through trial and error and does not require a big dataset [5].

Finally, DL requires big data and is where the biological brain is mimicked through the concept of neurons. A neuron is the building block of any biological brain. By connecting multiple neurons in a network, an artificial neural network is generated and could learn from data. Figure 1 shows the basic concept of such a network.

Figure 1.

An artificial neural network configuration showing input, hidden, and output layers [4].

Summary of all four AI (or ML) configurations and methods is shown in Figure 2.

Figure 2.

The four types of ML: Supervised, unsupervised, reinforcement, and deep learning methods [4].

2. Modeling of fluid flow

Fluid flow modeling is very important since it is widely used for many industries that are not limited to aero-planes, vehicles, air conditioning, propulsion, bio-fluids, nano-fluids military projectiles, and missiles. Fluid flow is classified as continuum flow or molecular flow at a low Knudsen number in which molecules are moving freely in its domain. Other five categories of classification include; laminar or turbulent, steady or unsteady, compressible or incompressible, viscous or nonviscous, and rotational or irrotational flow.

2.1 Computational fluid dynamics (CFD)

Computational fluid dynamics (CFD) became the most popular tool used to model fluid flow due to increase in computing capabilities offered by both main frame super computers and even personal computers that could deal with complex geometry and operating conditions. Computational fluid dynamics modeling or CFD is based on the principles of fluid mechanics, utilizing numerical methods and algorithms to solve problems that involve fluid flows.

CFD is classified into difference method (FDM), finite volume method (FVM), finite element method (FEM), lattice Boltzmann method (LBM), and smoothed particle hydrodynamics (SPH). CFD tools include many packages that are commercially distributed like ANSYS-ANSYS-FLUENT, Crew, COMSOL, and PATRAN or Open source CFD tools like Open Foam, Palabos, and MFIX. Another possibility for CFD is the direct numerical solution of fluid flow differential equations by procedural programming or by using object-oriented programs like MATLAB or MODELICA.

CFD tools include many software that are commercially distributed like ANSYS-FLUENT, Crew, and PATRAN. Another possibility for CFD is the direct numerical solution of fluid flow differential equations by procedural programming like C++, MATLAB, and Fortran or by using object-oriented programs like MATLAB or MODELICA. ANSYS-FLUENT is a general-purpose computational fluid dynamics (CFD) software offered by ANSYS, which is used to model fluid flow, heat, and mass transfer, chemical reactions, and more. ANSYS-FLUENT offers a modern, user-friendly interface that streamlines the CFD process from pre- to post-processing within a single window workflow.

Many examples of using ANSYS-FLUENT will be discussed throughout following paragraphs. ANSYS-FLUENT was used to study the flow of hydrogen in the scramjet engine combustor with different type’s fuel injectors at varied velocity, pressure, and temperature as shown in Figure 3. The researcher showed that Pylon injector gave better results for stabilized mixing of air-fuel at any Mach number over other types used in the simulation [9]. Mesh and typical results are shown in Figure 4. It is very important to investigate the mesh sensitivity (mesh independence or mesh convergence) for each modeling problem since this will increase the confidence in the results and so became one of the most important steps in the simulation. Mesh sensitivity provides information on the errors associated with mesh size and type as errors tend to zero as mesh size tends to zero but this will lead to unwanted extra computations that are necessary for model validation. The idea is to start with coarse mesh then this could be refined by the ANSYS-FLUENT program at any time during modeling and simulation.

Figure 3.

Scramjet propulsion system [9].

Figure 4.

Scramjet propulsion system mesh and mass fraction of O₂ [9].

Another example is where ANSYS-FLUENTCFD package is used to model underwater complex geometry vehicle to figure out the parameters that could be applied to the open-shelf depth setting TUV. The TUV predesign was validated by ANSYS-FLUENT to reduce the cost of repeated prototypes and shorter calculation time when using direct numerical simulation. Figure 5 showed the model and physical photo of TUV system.

Figure 5.

Model and physical photo of TUV system [10].

The simulation with ANSYS-FLUENT started with 2D modeling, meshing independence test, specifying boundary conditions on geometry, setting the turbulent model and material, and getting results as shown in Figure 6. There are no differences between the fluid dynamic simulation of deep-setting TUV properties and the empirical formula. The reliability of simulation results and the simulation results are intuitive within an acceptable range. The deep-setting model and programming can then be created. For pre-post design quantitative analysis, the earlier design stage can be leveraged to shorten design cycles and conduct qualitative research. Consequently, it can lower development costs.

Figure 6.

Pressure distribution of TUV system [10].

Also, ANSYS-FLUENT could be used to investigate the accuracy of the k-ε models applied to certain applications, for example, the accuracy of k-ε model applied to turbine 99 draft tube as shown in Figure 7. Governing equations are given in Eqs. (1) and (2). Two different grid concentrations near the wall y + 1 and y + 50 were used to study the flow behavior. Discussion is based on graphical results and by comparing numerical simulations and experiments in operational mode. It has been investigated how these models relate to the first-order upwind, second-order upwind, third-order upwind (QUICK), and power law schemes.

Figure 7.

Geometry and mesh of draft tube [11].

2.2 Modeling of fluid flow using object-oriented and procedural programming

For simple geometry one and two dimensions modeling, researchers preferred to implement simple programming to save the time during scope stage of the project that imposed many changes during project evolution. Simple programming could be elaborated by using procedural languages like C++ and Fortran or by more representable, easy interactive graphical interface, and object-oriented software like MATLAB and TRNSYS.

2.3 Modeling of fluid flow using object-oriented and procedural programming

For procedural programming, the system is described by a set of differential equations or matrix then solved simultaneously at given initial or boundary conditions. To model the vacuum pressure for the 340 m vacuum chambers at SESAME light source storage ring using molecular pump as shown in Figure 8 [12], the researcher had two options either to use a 3D modeling that required 3–6 months of modeling the complicated chamber or to use one-dimensional matrix computation with 5 days handwritten program at early stage of the project with daily changes.

Figure 8.

One cell of SESAME storage ring [12].

To evaluate the pressure value through unit cell, MATLAB program was developed to solve the linear system of balanced equations of three consecutive connected elements. One cell is divided into 152 elements, each one is approximately 11 cm in length except those elements where pump is located 20 cm long, and the total length is approximately 16.3 m.

The governing continuity equations:

$${\mathbf{C}}_{\mathbf{i}}\left({\mathbf{P}}_{\mathbf{i}\mathbf{-}\mathbf{1}}\mathbf{-}{\mathbf{P}}_{\mathbf{i}}\right)\mathbf{+}{\mathbf{C}}_{\mathbf{i}\mathbf{+}\mathbf{1}}\left({\mathbf{P}}_{\mathbf{i}\mathbf{+}\mathbf{1}}\mathbf{-}{\mathbf{P}}_{\mathbf{i}}\right)\mathbf{+}{\mathbf{Q}}_{\mathbf{i}}\mathbf{=}{\mathbf{S}}_{\mathbf{i}}{\mathbf{P}}_{\mathbf{i}}\left(\mathrm{i}=1,2,\dots ,\mathrm{n}=152\right)$$E1

with the following periodic boundary condition:

$${\mathrm{P}}_1={\mathrm{P}}_{152}$$

where:

P_i − 1, P_i, P_i + 1 unknown pressure of the elements i − 1, i, and i + 1.

C_i and C_i + 1 conductance value i and i − 1 and between element i and i + 1.

evaluated for CO molecules using MOLFLOW.

S_i is the pumping speed of element i

Q_i is the total gas load of element i

The results of MATLAB program shown in Figure 9 are compared with 3D MOLFLOW software, and experimental data with very good agreement as shown in Table 1.

Figure 9.

MATLAB vacuum pressure inside one cell of SESAME storage ring [12].

Dose (Ah)	Current (mA)	Model’s results Pressure (mbar)			Measurements Results (mbar)
		MATLAB	VACCALC	MOLFLOW	With Beam	WO Beam
0.01	10	—	—	1.26x10^−7	2x10^−8	3x10^−10
100	100	1.26x10^−9 (400 mA)	1.29x10^−9 (400 mA)	1.4x10^−9	3x10^−9	1x10^−10
500	300	—	—	1.05x10^−9	5x10^−10	5x10^−11
1000	400	—	—	8.75x10^−10	2x10^−10	1.5x10^−11

Table 1.

Model’s results VS measurement results [13].

One crucial step toward low-carbon environmental protection, energy conservation, and emission reduction development goals, as well as toward meeting the requirements of sustainable development plans, is clean heating. Another industrial use of orient project MATLAB/SIMULINK for modeling of stratified water tank in order to increase the precision of the stratified hot water storage tank model and simplify the stratified water tank modeling process. The three nodes stratified water tank and SIMULINK model are shown in Figure 10.

Figure 10.

Three nodes stratified water tank and SIMULINK model [14].

It is confirmed that this procedure is accurate and correct by comparing it to a platform simulation run on the TRNSYS system. A six-node stratified hot water storage tank is also created using this technique. The effect of various inlet and exit water flow rates and temperatures on the stratification effect of the hot water storage tank is examined through the modeling of 8 distinct working situations. Sample of results is shown in Figure 11.

Figure 11.

Results of SIMULINK model at working condition one [14].

The SIMULINK showed that stratification of the hot water storage tank is more likely to occur when the intake water flow on the collector side is higher than the return water flow on the load side. The higher the temperature difference between the inlet water temperature on the collector side and the return water temperature on the load side, the more conducive it is for the stratification of the hot water storage tank when the inlet water flow rate on the collector side and the return water flow rate on the load side.

2.4 Modeling of fluid flow systems by artificial intelligence

Fluid flow has two regimes. Depending on the speed of the flow, which is characterized by the dimensionless Reynolds number, these regimes are called laminar and turbulent. Simulation of fluid flow, in either regime, is called computational fluid dynamics (CFD). Especially in the turbulent regime, CFD has always been an expensive simulation if a high-accuracy solution is required. This is due to the complicated partial differential (PDE) solved in such simulations, Navier-Stokes eqs. A trade-off decision between accuracy and speed is made in such problems. If a full resolution and the highest accuracy are required, the simulation is called direct numerical simulation (DNS). That is, Navier-Stokes equations are completely solved to the finer mesh scale. However, DNS is sometimes impossible and, therefore, other methods are traditionally used. These methods use coarser meshes but reduce the accuracy of the solution. The methods, namely, are Reynolds averaged Navier-Stokes (RANS) [15] and large-eddy simulation (LES) [16].

To overcome CFD simulations, pure ML models have been used [17, 18, 19, 20, 21, 22]. However, there could be some caveats that should be looked at when going the route of pure ML models. First off, ML techniques like deep learning frequently cost a lot to train and need a lot of data. Therefore, it is critical to pinpoint instances when ML excels above established, decades-old techniques and may be more precise and effective. The method used to create the training data and whether the cost is included for benchmarking raise further concerns. Transfer learning is a potential field in this context [23]. It is also important to remember that there are deep learning alternatives that could be better suited for specific jobs. Having this said, it is worth noting that a 700× faster than simulation speed could be obtained when such pure ML models are used [17]. Figure 12 illustrates the general approaches ML used in fluid flow.

Figure 12.

In the context of direct numerical simulations, turbulence modeling, and reduced-order models, a summary of some of the most pertinent areas where machine learning might improve CFD [24].

The reduced order model (ROM) depends on the fact that even complicated flows frequently display a few prominent coherent structures that might provide basic but useful information about the flow. In order to characterize the fluid in a lower-dimensional, lower-fidelity manner, ROMs explain the development of these coherent structures. By providing optimization and control tasks that need several model iterations or quick model predictions, ROMs act as a quick surrogate model for the more expensive CFD approaches discussed above. This efficiency comes at the expense of generality since ROMs are designed for a particular flow arrangement, offering tremendous acceleration but with a constrained scope of use. In general, the lack of generalization was a major drawback in pure ML models. A hybrid approach that combines machine learning with CFD in a single framework shows more promising results where physics is predicted accurately and the simulation speed is increased [25].

CFD is essential for many science and engineering applications. It is, for example, at the core of climate simulation and modeling and for understanding climate change. However, the available and used paradigm widely used is purely physics-based. That is, the known physical equations (partial differential equations; PDEs) are solved through numerical differentiation and integration. This is very computationally expensive. In the recent literature, data-driven approaches have been used. This includes deep learning. One main drawback of deep learning is that it is physics-illiterate. That is, it is mainly a statistical method that has little or no knowledge of fluid flow physics.

Therefore, another paradigm of fluid flow simulation is the physics-informed neural network approach (PINN). A crucial step in improving AI for physical sciences is creating deep learning techniques that can systematically incorporate physical rules. PINNs are the result of intersecting the two approaches; data-driven and physics-based. On top of utilizing the system’s physical equations, they use the data-driven supervised neural network to learn the physics. Therefore, they get both advantages. Being consistent with the physics and remain a data-driven approach. Moreover, they could extrapolate beyond the range of available data, as they benefit from being physics-informed. As a result, PINNs are suitable for applications where data are limited. This is particularly appealing for many science and engineering problems. Where data are already scarce.

Such examples of PINN are the modeling of turbulent flow [20] or the prediction of sea surface temperature [26, 27] or the temperature of lakes [28]. Or even in fluid animation applications (such as gaming) where the physical consistency constraint is less stringent [29].

3. Modeling of energy system

Energy is the capacity to carry out work. People have figured out how to transform energy from one form to another and then use it to accomplish tasks, making modern civilization possible. Walking and bicycling, driving cars on roads and boats through water, cooking meals on stoves, making ice in freezers, lighting our homes and workplaces, producing goods, and sending astronauts into space all require the usage of energy. There are many forms of energy: 99, light, motion, electrical, chemical, and gravitational. There are several energy sources, however, they may all be categorized into one of two groups.

Renewable energy and nonrenewable sources of energy. Large energy systems like power plants, wind turbines1, or even nuclear plants are very expensive and so modeling the output of each one is very important to give complete information to decision-makers to avoid misconceptions and low performance of plant before installation. CFD packages, object-oriented software like MATLAB/SIMULINK, TRNSYS, and procedural computing were used to carry out energy performance modeling for many projects to prove the feasibility and technical studies.

3.1 Modeling of energy system by CFD

Solar energy system is widely investigated by researchers with many tools, ANSYS-FLUENT was used to investigate the thermodynamic properties and efficiency of the solar system. For example, the thermal and aerodynamic processes occurring in a solar air heater with L-shaped fins that absorb light are modeled numerically [30]. For the Russian city of Samara, the analysis is conducted using an ANSYS-FLUENT Solver with an inbuilt solar calculator. CFD model and Grid are shown in Figure 13.

Figure 13.

CFD model and grid [30].

During the CFD analysis, it was determined how design (the step between the fins) and technological factors (the Reynolds number) affected heat-exchange processes and flow aerodynamics. The thermal and aerodynamic properties of a smooth (un-finned) surface were compared with the findings of a physical experiment in order to confirm the accuracy of the results produced by the CFD model. The dependences determined analytically were also compared with the data derived from the developed model for model verification. A good convergence of the data produced by computer simulation with the outcomes of the physical experiment and the analytical solution is clear from the provided dependencies, which suggest that the ANSYS-FLUENT Solver’s parameters were chosen correctly.

The velocity distribution, pressure, and other aspects of the airflow in the solar air-heater box are determined using pictorial outlines in Figure 14. In another work, the ANSYS-FLUENT software for operational efficiency of solar air heater model was created whose suitability was confirmed using experimental data. The operational efficiency of the solar air heater has been found to be at its highest for the end of the calendar autumn at an angle of inclination of α = 60° to the earth [31]. Typical results was shown in Figure 15.

Figure 14.

Velocity contour and velocity vector at Re = 1200 [30].

Figure 15.

Temperature contours of the light-absorbing surface (contour 1) and the air at a distance of 14 mm from the surface of the glass (contour 2) at u_in = 1.5 m/s [31].

The researchers have determined how the ribs affect the heat transfer between the air and the light-absorbing surface. The immediate area of the ribs is where vortices and reverse-current zones occur, giving the temperature contour its curvilinear appearance. Research on the SAH’s aerodynamic properties has revealed that the ribs have less of an impact than the smooth, light-absorbing surface on the amount of energy used to push air through the air duct [31].

3.2 Modeling of energy system by object-oriented and procedural software

The MATLAB/SIMULINK is widely used to model and validation of Photovoltaic solar cell to produce electricity, pumping system, and many electric devices like electric vehicles. Every important role in using MATLAB/SIMULINK software is to model PV cells (solar cells), PV modules (solar modules), and PV arrays (solar arrays) as well as the analysis of how PV performance changes as a result of changes in various parameters, including temperature, solar radiation, reverse saturation current, series resistance, and shunt resistance [32]. PV electric circuit and model using MATLAB/SIMULINK are shown in Figure 16. The PV and IV characteristics curve for the PV array is usually given by this simulation.

Figure 16.

PV electric circuit and SIMULINK model [32].

Another important object-oriented tool to simulate energy systems is TRNSYS (Transient System Simulation Tool), in which, most simulations are geared toward evaluating the performance of thermal and electrical energy systems.

There are two components to TRNSYS. The first is a kernel-based engine that analyzes the input file, solves the system iteratively, checks for convergence, and graphs system variables. In addition, the kernel offers tools for, among other things, calculating thermophysical characteristics, inverting matrices, running linear regressions, and interpolating external data files. The second component of TRNSYS is a sizable library of parts, each of which simulates the operation of a different system component. About 150 models are included in the standard library, ranging from wind turbines to electrolyzers, from pumps to multizone structures, from weather data processors to economics routines, and from basic HVAC equipment to cutting-edge developing technologies. Models are built so that users can change already-existing components or create their own features. Excellent example is to use TRNSYS program to analyze both experimental data and theoretical predictions about the performance of a Tri-Generation cooling system using two adsorption chillers. This study aims to analyze the performance of two adsorption chillers and the effects of operational and design parameters on the tri-generation cooling system performance. The system, which has a 240 m² parabolic trough solar matrix, was created and confirmed at the Mutah University in Jordan. The trough matrix heated the thermal oil to 260°C and generated 13.7 bar of superheated steam at 210°C. By using the engine’s 120°C steam to evaporate brackish water, the power cycle is finished. Each hour, distillation unit produces 150 L of distilled water. The distillation unit rejects heat and is kept in a thermally isolated hydraulic storage tank used to run a special two-stage air-cooled adsorption chiller with a 10 kW cooling capacity. TRNSYS graphical model is shown in Figure 17.

Figure 17.

TRNSYS simulation for solar tri-generation system [33].

The results of TRNSYS include the normalized capacity and COP of adsorption chillers.

One tool preferred by some researchers is to use procedural programming capabilities of MATLAB to model and simulate the performance of energy such astwo-stage adsorption chiller with heat recovery. The chiller schematic used in the simulation is shown in Figure 18. The performance of the proposed two-stage adsorption chiller with and without heat recovery utilizing an activated carbon/methanol combination was modeled and simulated using the MATLAB program. The experimental results carried out by Millennium Industries were used to validate the simulation model results. The 10th-order differential equations that made up the model were utilized to determine the adsorption isotherm and adsorption kinetics, while six of them were used to forecast bed, evaporator, and condenser temperatures.

Figure 18.

Schematic of two-stage adsorption chiller [34].

The energy balance equation used in the simulation:

$$\begin{array}{l}\frac{\mathrm{d}}{\mathrm{dt}}\left\{\left({\mathrm{M}}_{\mathrm{ac}}\ast {\mathrm{Cp}}_{\mathrm{ac}}+{\mathrm{M}}_{\mathrm{ac}}\ast {\mathrm{Cp}}_{\mathrm{m}}\ast {\mathrm{x}}_{\mathrm{ads}}+{\mathrm{M}}_{\mathrm{Al}}\ast {\mathrm{Cp}}_{\mathrm{Al}}+{\mathrm{M}}_{\mathrm{cu}}\ast {\mathrm{Cp}}_{\mathrm{cu}}\right){\mathrm{T}}_{\mathrm{ads}}\right\}\\ =M_{\mathit{ac}}\ast Q_{\mathit{st}}\ast \frac{d{x}_{\mathit{ads}}}{\mathit{dt}}-M_{\mathit{ac}}\ast {\mathit{Cp}}_m\ast \left(T_{\mathit{ads}}-T_{\mathit{eva}}\right)\ast \frac{d{x}_{\mathit{ads}}}{\mathit{dt}}+M{c}_w\ast C_{\mathit{pw}}\ast \left(T_{\mathit{ci}}-T_{\mathit{co}}\right)\end{array}$$E2

The left side represents the required sensible heat transfer of activated carbon, the metallic part of bed, and the methanol amount absorbed by activated carbon. The first part of right side represents the adsorption heat. The second term represents heat flow from evaporator to adsorbent bed. The third term represents the heat transfer by cooling water. The outlet temperature of the cooling water is calculated by Log Mean Temperature Difference (LMTD) (Figure 19):

Figure 19.

Chiller experimental versus model cooling capacity [34].

For evaporator:

$$\left(M_{\mathit{em}}\left(t\right)\ast {\mathit{Cp}}_m+M_{\mathit{eva}}\ast {\mathit{Cp}}_{\mathit{eva}}\right)\frac{\mathit{\text{dTeva}}}{\mathit{dt}}=-M_{\mathit{ac}}\ast h_{\mathit{fg}}\ast \frac{d{x}_{\mathit{ads}}}{\mathit{dt}}+M_{\mathit{chw}}\ast C_{\mathit{pw}}\ast \left(T_{\mathit{chi}}-T_{\mathit{cho}}\right)$$E3

For condenser:

$$\left(M_{\mathit{cm}}\ast {\mathit{Cp}}_m+M_{\mathit{con}}\ast {\mathit{Cp}}_{\mathit{con}}\right)\frac{\mathit{\text{dTcon}}}{\mathit{dt}}=-M_{\mathit{ac}}\ast h_{\mathit{fg}}\ast \frac{d{x}_{\mathit{des}}}{\mathit{dt}}+M_{\mathit{con}}\ast C_{\mathit{pw}}\ast \left(T_{\mathit{\text{coni}}}-T_{\mathit{\text{cono}}}\right)$$E4

it is made clear that the findings of the simulation model for the two-stage air-cooled chiller are well compared with the experimental data in terms of cooling capacity (6.7 kW for the model compared with 6.14 kW from experimental result at same conditions) as shown in Figure 20. The coefficient of performance (COP) prediction is quite similar to the value provided by the Carnot cycle operating under the same conditions. To improve cooling capacity and COP, the model optimized the switching, adsorption/desorption, and heat recovery times. The model gives both heat recovery and chiller operation modes, the temporal history of the bed, evaporator, and condenser temperatures as shown in Figure 21. The study is significant since it served as the foundation for future built chiller series, which is one of the important needs for simulation in energy industries.

Figure 20.

Temporal history of bed, evaporator, and condenser temperatures [34].

Figure 21.

Effect of activated carbon mass on chiller average cooling capacity and COP [35].

Model could be used also to carryout parametric studies like chiller size, evaporator, and condenser size. Activated carbon mass, bed size, and mass flow rates for cooling, heating, chiller, and condenser are also studied using the model to determine how they affect cooling capacity and COP [35]. Some result of simulation parametric results are shown in Figure 21.

One of the interesting and user-friendly environments is an object-oriented programming offered by MODELICA, in which a built in graphical library is used to model the energy system operation and efficiency for large-scale plant as shown in Figure 22. The model in MODELICA program is shown in Figure 23. Results include power generated, solar irradiation, and control strategy for power plant [36].

Figure 22.

Schematic of a parabolic trough solar thermal power plant with TES [36].

Figure 23.

Model a parabolic trough solar thermal power plant with TES [36].

3.3 Modeling of energy by artificial intelligence

Since 2000, AI has attracted more attention, particularly since 2014, when the number of articles on ML skyrocketed. And in 6 years, ten times greater than they were in 2014. Despite the fact that several publications in various sectors of energy have offered novel applications of AI and ML in that area, the breadth of energy consumption and its respective domains make it clear that no single book, article, or source could possibly cover all of these topics. However, by looking at the frequency of the AI and ML energy-related keywords in the literature, four areas could be identified: (1) AI-based applications for effectiveness and utilization, (2) machine learning for prediction, (3) algorithm and pattern recognition for learning systems, (4) and administration and transportation of energy sources. Moreover, from 2012 to 2014, the majority of research focused on various fuel technologies, including hydrogen, ethanol, and biofuels. The majority of the literature was dominated by ideas like sustainability and sustainable development during the course of the following 2 years. A lot of attention was also paid to issues, including environmental effects, climate change, greenhouse gas emissions, carbon dioxide, and life cycle analysis. Additionally, attention has been drawn to clean energy sources like solar and wind. AI ideas started to appear around the beginning of 2016. In 2017, forecasting, decision-making, and smart grids were the main topics of study. Later, in 2018, ML approaches for energy use and renewable energy developed. Table 2 shows the main applications of ML in the energy sector [5].

Goals	Systems and sections	Techniques and models
Decision making	Solar power	Numerical model
Forecasting	Wind power	Time series analysis
Sustainable development	Stochastic systems	Uncertainty analysis
Energy management	Air conditioning	Fuzzy analysis
Environmental impact	Smart power grid	Genetic algorithm
Performance assessment	Electric vehicle	Support vector
Impact assessment (Social, technical, economic, etc.)	Power transmission network	Multi-objective optimization
	Renewable energy	Random forest
	Energy storage	Decision tree
	Buildings	Artificial neural network
	Internet of things	Pattern recognition
	Energy consumption	Ant colony optimization
		Regression analysis
		Particle swarm optimization
		Data mining
		Deep learning

Table 2.

Main applications of ML in the energy sector.

Another work related to ML in energy systems has compared how well extreme gradient boosting (XGBoost) performs when used in conjunction with artificial neural network (ANN) and degree-day ordinary least square (OLS) regression to create predictive energy models. The effectiveness is tested of each method in terms of the precision of the produced energy models using datasets from cooling electricity and heating gas. Their findings show that XGBoost generates extremely precise predictive energy models. Although XGBoost outperformed ANN and degree-day OLS regression in terms of model accuracy, more research with different energy datasets is necessary for the future to make more definitive claims about their general effectiveness [6].

Special attention should be given to ML applications for renewable energies. The importance of renewable energy sources, such as wind and solar energy, in supplying the world’s energy demands is rising. For energy system operators, their fluctuation and unpredictable nature provide serious difficulties. For the system to be stable and reliable, accurate forecasting of renewable energy output is essential. ML and DL algorithms have shown promise as methods for forecasting renewable energy. For a baseline model, traditional ML techniques like linear regression have been successfully applied to the forecasting of renewable energy. It is straightforward, simple to understand, and uses fewer computer resources. However, it could be unable to capture the complex patterns and non-linear correlations present in the data. For forecasting renewable energy, linear regression has been proven to be inferior to the random forest, SVMs, and XGBoost models. Even in the face of noise and outliers, these models are capable of handling non-linear connections, complicated data, and the ability to make reliable predictions. To handle irregularly spaced time series data and identify patterns and trends to produce precise forecasts, specialist time series forecasting techniques like autoregressive models, moving averages, and RNNs are better choices [7].

ANN is also used to predict buildings’ energy consumption in terms of heating and cooling electricity consumption. Simulation results were compared of building performance to actual measurements while employing ANN to forecast building energy performance. Energy usage data from the case building for a week was used for training and testing the ANN. With a mean absolute error of 0.9%, the results demonstrate a satisfactory match with the mathematical model [8].

4. Conclusions

Modeling and simulation of systems in mechanical engineering both sub-disciplinary of applied mechanics [37, 38, 39, 40] or thermal systems “as described in this chapter” became more popular in industry and scientific research due to accelerating in computing power and graphical interface of most of the simulation tools offered commercially or free open access software. CFDs software with all its five techniques are the most interesting for many researchers due to its flexibility and due to the fact that the high cost of experimentation of fluid and energy systems. Modular energy plant system modeling and object-oriented software like MATLAB/SIMULINK, MODELICA or TRNSYS had been widely used last two decades. Libraries are expanded and updated to include more objects representing energy system to fulfill the needs of industry and research for validation of their systems prior to going for prototyping and understanding variant system operating conditions and system efficiency and alternatives for system components. Procedural programming is still in use by some researchers who had the chance to write their own codes for specific system and could be modified to another system with minimum effort by the researcher. Special systems and environments required writing code from scratch since these systems could not be modeled by CFD or object-oriented programming.

Moreover, AI techniques have been used to address the problems of modeling and simulation of fluid and energy systems. In our data-centric era, the application of such techniques will just continue and evolve. Many algorithms have been used across the data-driven modeling and simulation of engineering problems. The availability of data and whether the amount of data is considered a “big data” or not are vital in deciding on the approach taken. Deep learning (DL) neural network (NN) is suitable where data is abundant and considered big data. Whilst ML algorithms are used when the data is comparatively limited. Another interesting topic in the field of modeling and simulation is the physics-informed neural networks (PINNs). PINNs use the data to learn while respecting the physics of the problem at the same time. As physical equations are involved, PINNs usually require less data than what a typical NN would for the same problem. It also serves as a good extrapolator in regimes outside the convex hull of training data. Finally, ML and DL are showing promising results in the modeling and simulation of engineering problems. The investigation of such techniques is still at the early stages and more work needs to be done to fully exploit their capabilities. It is preferable to include a conclusion(s) section, which will summarize the content of the book chapter.

Acknowledgments

I would like to thank my colleagues, at Mechanical Engineering Department of Al-Zaytoonah University, who provided me with their remarkable research in the field of fluid and energy, which highly contributes to the completion of this chapter.