Waste heat recovery (WHR) can be classified in several ways; one way is to sort based on the utilisation method, which can be either passive or active. Passive utilisation means that exchange heat at the same or lower temperature level, such as using heat exchangers for internal waste heat recovery (IWHR). It utilises the heat for certain applications or processes such as district heating or thermal energy storage (TES). Active utilisation means that the waste heat can be converted to another form of energy, such as Mechanical Vapour Compression (MVC), sorption/absorption heat pump/chiller, or waste heat to power (WHP) using thermodynamic cycles or thermoelectric technology [9]. Another way of classifying waste heat can be either by its source composition or its temperature, in many cases, waste heat is categorised into high (T > 650 °C), medium (232 < T> 650 °C), and low (T < 232 °C) [9].

In many thermal processes, especially in aluminium and steel furnaces, thermal power and heat fluctuations are present mainly in mass flow rates and temperatures. High fluctuations in temperature and mass flow rates constrain the system design, especially in WHP applications such as using an ORC, because some organic fluids decompose at high temperatures. On the contrary, low temperatures have a risk of producing wet vapours that would affect the expander's efficiency and service life due to possible erosion. These challenges complicate or, in some cases, fail to find a suitable design, especially with high ranges of mass flow and temperature fluctuations, to comply with safe operation and optimal output. Furthermore, transient input of waste heat also leads the system to operate in off-design conditions, which decreases the efficiency of the system significantly and creates some vibrational loads on the rotating shafts and uncertainties in operation. A review study has been conducted thoroughly on the aluminium industry, outlining the aluminium manufacturing process and the state of the art technologies used in the sector, including the waste heat recovery technology [10]. However, the study has not pointed out the challenges and the characteristics of the variable waste heat associated with the holding and remelting furnaces.

Several studies were conducted to minimise the effect of the flue gases' thermal fluctuations for WHP systems. The resulting suggestions can be classified into two main categories shown in Figure 1:

1. modifying stream control, which can be implemented to both the waste heat stream and the recovery cycle working fluid stream,

2. using an intermediate TES system to absorb and damp the thermal fluctuations [11].

Jiménez-Arreola and his co-workers [12] assessed the potential of generating power through an ORC in an off-design operation from fluctuating exhaust gases exiting a steam generator plant. This study compared utilising internal combustion engine (ICE) exhaust gases waste heat in operating an ORC with two candidate evaporator heat exchangers (HEX), double pipe HEX and plate HEX. The results indicated that the double pipe HEX design has twice the heat transfer area as the plate HEX has due to the change in the heat transfer characteristics. Also, they found that when the ICE load is reduced to 60% of the full load, the ORC engine does not respond by the same amount of reduction; it measures 72% of the ORC full load [12]. Nevertheless, utilising temporal source waste heat in a part-load manner without thermal storage will result in inefficient energy conversion and infeasible waste heat recovery.

Another techno-economic analysis was performed for retrofitting an ORC to three industrial waste heat applications that are corresponding to three operational scenarios: (1) hot air exhaust from clinker cooling with temperature fluctuations only, (2) rolling mill reheating furnace exhaust gases with mass-flow fluctuation, and (3) electric arc furnace exhaust gases with both temperature and mass flow fluctuations [13]. The study compared different solutions for each case scenario based on CO₂ savings and the Levelized Cost of Electricity (LCOE). With regard to the first case (WHR from clinker cooling), a comparison was considered of integrating an intermediate oil loop, air injection to the source heat, heat source by-pass, latent heat storage using two tanks, and a single sensible heat thermal storage. This comparison suggested that implementing latent heat storage had the highest CO₂ savings and the lowest LCOE. As for the rolling mill reheating furnace, only by-passing access mass-flow and two-tank thermal storage were considered. By-passing 75% of the highest mass flow showed better results in terms of CO₂ savings. In contrast, the electric arc furnace with both temperature and mass fluctuations, implementing an oil loop without storage, had the best LCOE followed by the two-tank sensible TES. The latent heat TES showed better recovery resulting in higher CO₂ savings [13].

An important study of designing and comparing two ORC direct evaporators for a WHR considering the dynamic behaviour of flue gases from a diesel engine was conducted by [14]. The study focused on optimising an ORC evaporator at a design point and its behaviour under the variability of the source considering thermal inertia. Phase change materials (PCM) as a latent heat TES were previously proposed for furnaces having high-temperature fluctuations with quasi-steady mass flow rates, such as electric arc furnaces in [11],[13],[14]. Although, another study proposed using an aluminium alloy based TES to recover a quasi-constant temperature with significant variability in mass flow rate from a reheating furnace [17].

Figure 1

Solutions for thermal power fluctuation suggested by researchers [11]

The TES technology has emerged to be one of the attractive topics for researchers in the last decades due to its potential of contributing to many applications, including renewable technology, industrial WHR, heating & cooling, heat pumps, buildings, and even transportation. TES systems vary in their methods, technologies, functionality, capacities and applications. There are three leading technologies in TES systems: sensible, latent and thermochemical heat storage, each of which has its unique characteristics. For all TES technologies, properties of the materials should be taken into consideration for selection to ensure good performance, durability and safety.

Sensible heat storage is the most mature and widely implemented method compared to other methods due to its simplicity, where the material, either solid or liquid, stores the energy in its specific heat capacity. The materials experience no phase change in this process of heating/cooling, and the materials used in this method are usually water, oils, molten salts, sand, rocks or concrete. The low cost of these materials and their abundant presence allowed it to be the cheapest technology.

Although many investigations were conducted assessing sensible thermal storages, yet there was a lack of studies conducted on industrial waste heat recovery. This is due to several issues associated with sensible heat storage. One of the main issues is that the heat transfer is limited to the specific heat where no phase change occurs to increase the capacity of the heat absorbed by the material; thus, large heat storages are required. Another issue is that although fluctuation of the power is decreased, the temperature outlet of the sensible TES is usually dynamic, especially in high fluctuation input. Furthermore, processes such as heating furnaces have high-temperature ranges, which can limit candidate materials.

Several research papers studied temporal waste heat recovery from industrial processes by integrating TES of several technologies. A study proposed a novel mathematical model for simulating transient conditions in a concrete passive sensible TES that shows alignment with experimental data within 2% of error [18]. Another study provided experimental and computational analysis for a packed-bed TES to recover waste heat up to 525 °C at an industrial scale [19]. Moreover, a practical case study of a multi-product batch process in a textile plant investigated both a direct heat recovery using pinch point analysis as well as an indirect heat recovery [20]. The indirect heat recovery proposed consisted of using a closed intermediate loop and heat storage. This method enabled the recovery of 5% of the total energy consumed daily in the plant.

Reverberatory gas-fired furnaces, as shown in Figure 2, are common in the aluminium industry for remelting and metal holding, and alloying.

Figure 2

Reverberatory gas-fired metal holding furnace

Fuel and air are supplied to the furnace burners within 5–10% of excess air for complete combustion. There are mainly two types of burners used in reverberatory gas-fired furnaces, standard cold-air burners or regenerative burners. Although the regenerative burners are more efficient than the former, using them for insufficient loads will result in inefficient performance [21].

Standard reverberatory gas-fired remelting furnaces expel about 30–50% of their high-quality energy through the chimney [22]. Figure 3 illustrates the heat flow schematic in a typical molten-metal holding furnace.

Figure 3

Energy losses in a reverberatory remelting furnace

The main source of energy comes from the combustion of fuel which is natural gas in this case, and the molten metal consumes a small portion of the total energy to keep the metal in its molten condition at a specific set point. The rest of the input fuel energy will leave the furnace as waste heat through the furnace walls, frequent door opening and flue gases. Flue gases leave the furnace at a temperature around 700 ?C before they get cooled by entrained cold air. The flue gas waste heat has high energy potential because of its high temperature. The number of burners and their loads depends on the furnace metal-batch capacity and the main task: melting, holding, and alloying.

Figure 4

Temporal flow and temperature measurements for three furnaces over a single batch

The furnace operation is controlled by a feedback control system that receives a set-point temperature from the metal thermocouple, located 60 cm below the metal surface, to switch the flame of the burner on and off. Therefore, the measured flue gas flow rate and the temperature vary with time, as shown in Figure 4 for three furnaces, F1, F2, and F3. The exhaust gas temperature and species concentration measurements are taken from a tab located between the furnace and its chimney exit before it mixes with the atmospheric air.

This study considers flue gases collected from three aluminium-holding furnaces operated in batch mode. Several fluctuation management techniques are implemented to determine the best alternative in achieving a smooth output of waste heat that includes: (1) an optimisation model of scheduling a time shift between batches, (2) temporal cold-air injection to the flue gas, and (3) integrating sensible thermal storage. The heat output will be used to model an ORC and RO plant to utilise the excess heat for power and water production. Also, the analysis will be extended to include the utilisation of remelting furnaces. This approach provides the following contributions:

Flue gas exergy code is written on MATLAB to estimate the maximum available work that can be delivered at the dead state condition where the aluminium plant is located. The total exergy flow is calculated based on the measured flue temperature and flow rate collected from three metal holding furnaces. If temperature and flow rates are varied, as shown in Figure 3, then exergy is varied accordingly.

The specific exergy for steady-state steady flow (SSSF), where kinetic and potential energies are neglected, can be expressed as:

$e_x=(h-h_o)-T_o(s-s_o)$ (1)

This is the maximum available work per unit mass that can be produced by a flow at temperature T and dumped to a dead state at temperature T_o[K]. The specific enthalpy h and specific entropy s are dynamically calculated using the measured heat source temperature. The total exergy flow ${\dot{E}}_x$ is:

${\dot{E}}_x={\dot{m}}_{flue}{e}_x$ (2)

As it can be seen from this equation, the resultant exergy fluctuation will be even higher than each term in the equation due to the multiplication.

The collected stream of flue gas has high variability in temperature as well as mass flow. In this assessment, the stream is injected with cold air to help cooling the stream to suit the ORC as well as observing the stream output in terms of mass flow. The stable temperature and mass flow will enable to utilise the waste heat directly to generate power by the ORC. This can be done through a control system that measures the amount of air required to reach a specific temperature. Air injection will reduce the maximum temperature; however, it will not reduce the fluctuation significantly. The fluctuated flue gas flow rate will mix with a fluctuated cold air stream that maintains the total amount of energy with lower quality because of the low average temperature of the resultant mixture of hot and cold flow streams.

The main goal of this method is to optimise the operation time shift between a cluster of three furnaces to generate a total temporal flue gas flow rate and temperature that is much smoother than a single furnace. It is possible to minimise either the standard deviation or the difference between the peaks amplitudes of the total fluctuated signal. This method is novel, and up to the knowledge of the authors, it has not been studied so far. The particle swarm optimisation (PSO) algorithm is one of the candidate methods for such optimisation because of its feature and applicability in similar power-and energy-related techniques, as explained in the following paragraphs. The expected output of this algorithm is to answer this question: what is the best operation time-shift among the three furnaces to get the most smoothed results?

PSO is considered one of the evolutionary algorithms based on artificial intelligence; therefore, it is widely applied in engineering and scientific research [23]. The PSO was already implemented in the field of power generation, and power systems accounting for 5.8% of the total PSO applications with more than 39 papers were published [17], [18]. The concept of this model is inspired by the phenomena of a bird flock flying in search of food where each particle position corresponds to a possible solution, and thereby, a flock of "n" particles will iterate and move in search of an optimal solution. The conditions of the particles are altered in the influence of three factors: the own particle inertia, the particle's most optimal position, the swarm's most optimal position [26]. The following equations change the position and speed of a particle:

$v_{id}^{k+1}=w{v}_{id}^k+c_1{r}_1^k(p{{}_{bes{t}_{id}}}^k-x_{id}^k)+c_2{r}_2^k(g{{}_{bes{t}_{id}}}^k-x_{id}^k)$ (3)

$x_{id}^{k+1}=x_{id}^k+v_{id}^{k+1}$ (4)

where: $v_{id}^k$ and $x_{id}^k$ are the velocity and position of the i^th particle, with d and k indicating the dimension and iteration, respectively; p_best and g_best represent the personal best position and the global best position. Among the constants that regulate the equation in terms of particle movements, w represents the inertia of particles attained in the previous position, while c₁ and c₂ are the acceleration constants and are set to be 2 in most applications. Also, $r_1^k$ and $r_2^k$ represent random numbers in the range of 0 to 1 [23]. Therefore, applying this method will explore the optimum operating time delay between the furnaces by trying two objective functions at a time: minimising the standard deviation and minimising the difference between the lower and the higher peaks value. Therefore, applying the PSO model will be the most suitable as it enables exploring the optimum operating-time shift between furnaces. It will be performed by shifting the input profiles in a systematic and directed manner instead of the genetic algorithm that is based on altering several variables each iteration in a semi-randomised manner. The set-up of the model is summarised in Table 1. The algorithm code is developed on MATLAB.

The main objective of the computation fluid dynamic (CFD) modelling is to investigate and design the appropriate sensible TES solid structure that will be able to diffuse the waste heat fluctuation in the solid structure and produce uniform temperature. This solid structure comprises a solid block that has the flue gas and HTF running in tubes installed in it. The solid block must be semi-infinite enough with the appropriate thermal diffusivity transport phenomena to damp the dynamic waste heat components provided by the flue gas tubes to a uniform temperature before it reaches the HTF tubes. Therefore, the sizes of the solid block, flue gas tubes, HTF tubes, and the distance between tubes as well as the solid thermal diffusivity, must be investigated and optimised to maximise the gain of heat recovery. Obviously, it is a complex engineering problem that can be investigated either experimentally or using a sophisticated CFD model. The experimental investigation is significantly expensive and exhaustive, not only cost-wise but also time-wise. The availability of a high-speed computing cluster allows us to develop several CFD models to reach the appropriate TES block design.

3D transient CFD model was developed on STARCCM+ platform. As shown in Table 2, the appropriate physics is implemented to simulate the fluid flow and heat transfer governed by the laws of mass, momentum, and energy conservation equations (5)–(8). These physical laws include fluid flow turbulence, heat conduction, convection and radiation. In addition, an implicit transient time step was used for the transient simulations. Cylindrical coordinates (r,θ, z) are used to describe the continuity, momentum and energy equations for the flow running in the tubes. Cartesian coordinates (x,y,z) are used to describe the transient conduction heat transfer in the solid block. The following are the mathematical representations of the governing equations; where the mass and momentum equations are presented respectively:

$\frac{\mathrm{\partial }\overline{\rho }}{\mathrm{\partial }t}+\mathrm{\nabla }\dot{}\left(\overline{\rho }\tilde{\overrightarrow{v}}\right)=0$ (5)

$\frac{\partial \left(\overline{\rho }\widetilde{\stackrel{\rightharpoonup }{v}}\right)}{\partial t}+\nabla \left(\widetilde{\stackrel{\rightharpoonup }{v}}·\nabla \widetilde{\stackrel{\rightharpoonup }{v}}\right)=-\nabla p+\nabla ·(\overline{\tau )}+\overline{\rho }\stackrel{\rightharpoonup }{g}$ (6)

where $\stackrel{\rightharpoonup }{g}$denotes gravity acceleration. The stress tensor $\overline{\overline{\tau }}$ is given by:

$\overline{\overline{\tau }}=\mu \left[\left(\nabla \tilde{\overrightarrow{v}}+\nabla {\tilde{\overrightarrow{v}}}^T\right)-\frac{2}{3}\nabla ·\tilde{\overrightarrow{v}}I\right]$ (7)

where I is the unit tensor, and:

$\frac{\partial T}{\partial t}+\widetilde{\stackrel{\rightharpoonup }{v}}·\nabla T=\alpha \left({\nabla }^{2}T\right)+\varnothing$ (8)

where ∅ is the viscous dissipation function.

The energy equation of the heat transfer in the solid block is given by:

$\frac{\mathrm{\partial }T}{\mathrm{\partial }t}=\alpha (\frac{{\mathrm{\partial }}^{2}T}{\mathrm{\partial }{x}^2}+\frac{{\mathrm{\partial }}^{2}T}{\mathrm{\partial }{y}^2}+\frac{{\mathrm{\partial }}^{2}T}{\mathrm{\partial }{z}^2})$ (9)

Certain boundary and initial conditions are assigned in the three computational domains, solid block, flue gas tubes and oil tubes to solve the mathematical problem. As for the momentum equation, the no-slip condition is set on the tubes' boundaries. For the TES solid block, adiabatic walls were applied on the wall's boundaries. Furthermore, the thermal properties of the solid blocks were evaluated based on both the literature [27] and the thermophysical database JMatPro. As for HTF, a commercial fluid Dowtherm A was used, and the thermal properties were evaluated based on the datasheet provided by the oil manufacturer [20], [21]. Table 2 and Figure 6 summarise the specifications and properties of the modelling set-up.

Figure 6

CFD sub-model setup

Regarding the accuracy and stability of the developed models, a sensitivity analysis was conducted with three mesh sizes to ensure mesh size independence. In addition, in the transient models, a sensitivity analysis was conducted on the number of iterations for each time step as well as varying the time step for the simulations to prove that the results are independent of the solving time resolution.

Organic Rankine cycle (ORC) is used to convert the flue gas waste energy into useful electricity. The TES medium is required to damp the flue gas energy fluctuation and smoothly operate the ORC. The produced electricity can be used to operate the reverse osmosis (RO) plant to supply the cast-house with the desalinated water required for direct chill (DC) casting mould. After implementing the three models of ORC using IPSEpro software, the efficiencies and output power were the highest when isopentane was used as the working fluid, and a recuperate heat exchanger was added. Figure 7a shows a layout of a simple ORC with its corresponding T–s state. The liquid fluid is first compressed by a pump from state 1 to 2 and after that, evaporated by the heat source from state 2 to 3, reaching to saturated vapour. The saturated vapour goes through an expander in which a shaft is driven to generate power from stage 3–4. Finally, the exhaust vapour goes through a heat sink (mainly a water-cooled condenser) in stage 4–1. Another configuration is shown in Figure 7b, which uses a heat exchanger to preheat the stream coming out of the pump in stage 2–5 by utilising the exhaust heat exiting of the expander in stage 4–6.

Figure 7

Layout of: simple ORC (a) and recuperated ORC (b)

The mass balance of the ORC components is given by:

$\sum {\dot{m}}_{in}=\sum {\dot{m}}_{out}$ (10)

and the energy balance is given by:

$\sum \dot{Q}+\sum \dot{W}=\sum {\dot{m}}_f{h}_f-\sum {\dot{m}}_{in}{h}_{in}$ (11a)

${\dot{W}}_{in}=\dot{m}(h_2-h_1),{\dot{W}}_{out}=\dot{m}(h_3-h_4),{\dot{Q}}_{in}=\dot{m}(h_3-h_2),{\dot{Q}}_{out}=\dot{m}(h_4-h_1)$

For the ORC cycle:

$\sum \dot{Q}+\sum \dot{W}=0$ (11b)

The thermal efficiency is the ratio of the net power output to the input heat rate which can be expressed by:

$\eta =\frac{{\dot{W}}_{out}+{\dot{W}}_{in}}{{\dot{Q}}_{in}}$ (12)

The RO is a process of water desalination based on membrane technology where the membrane allows water molecules to penetrate and blocks the salts from entering. This is done using the high-pressure flow of water supplied by powerful pumps, and this process can be expressed by the equations below.

Mass/volume balance:

$V_f=V_p+V_b$ (13)

Salt balance:

$V_f{S}_f=V_p{S}_p+V_b{S}_b$ (14)

Salt rejection:

$SR=(1-\frac{S_p}{S_f})\times 100$ (15)

Specific power consumption:

$w=\frac{V_f\times max.pressureapplied}{V_p\times Mech.\zeta }$ (16)

where: V is the volume, S is the salt concentration, and subscripts f, p, b denote feed, permeate, and brine, respectively.

Figure 8 shows the schematic of the RO plant equipped with an energy recovery component that recovers the pressure of the brine to be used in the feed water stream based on positive displacement.

Figure 8

RO water desalination plant for waste heat recovery