Modelling steel strip heating within an annealing furnace
 Stephen W. Taylor^{1}Email authorView ORCID ID profile and
 Shixiao Wang^{1}
DOI: 10.1186/s4073601700307
© The Author(s) 2017
Received: 15 August 2016
Accepted: 12 April 2017
Published: 27 April 2017
Abstract
Annealing furnaces are used to heat steel in order to change its chemical structure. In this paper we model an electric radiant furnace. One of the major defects in steel strips processed in such furnaces is a wavelike pattern near the edges of the strip, apparently due to extra heating near the edges. The aim of the paper is to model this effect and provide a way to calculate the elevated temperatures near the edges. We analyse two processes that are suspected to contribute to uneven heating. The modelling involves an asymptotic analysis of the effect of heat flux at the edges and a detailed analysis of the integral equations associated with radiant heat transfer in the furnace.
Keywords
Radiant heat transfer Annealing furnace Asymptotic analysisIntroduction
The high temperatures within a steel annealing furnace preclude any reliable way to take measurements of the temperature; hence the need for mathematical models so that the temperature can be computed. We model an electric radiant annealing furnace with length of order 100 metres through which strips of steel sheet pass at speeds of up to 130 metres per minute in order to achieve the strip temperatures required for annealing. A schematic diagram of the furnace is shown in Fig. 4. The temperature along the furnace is controlled by varying the power supplied to the heating elements and the line speed through the furnace is reduced for strips of large thickness and width in order to achieve the required temperatures within the steel strips. At the beginning of the annealing–coating line there is an automatic welding process which welds the beginning of a new coil of steel sheet to the end of its predecessor, allowing the line to run continuously.
Occasionally the edges of the strip may take a wavelike shape after passing through the furnace and this seems to be a result of extra heating at the edges of the strip. This hypothesis is supported by a COMSOL^{®;} model of the system [1, 2] which shows a trend in increasing steel strip temperatures closer to the edges. The goal of this paper is to gain a better understanding of the nonuniform heating of the strip across its width.
The furnace has already been modelled in a recent MathematicsinIndustry Study Group (MISG) meeting [5]. However the model developed in that meeting was based on an assumption of uniform heating across the width of the strip and is thus unsuitable for explaining such defects. There is a very limited amount of modelling of such furnaces in the literature. Apart from the papers already cited, perhaps the closest work is [10] which also takes into account the radiative heat transfer within a mulltizone annealing furnace. However, although the model in [10] is more detailed than that given in [5], it also makes the approximation that the strip temperature does not vary across its width. Other related models concern an electric furnace model for crystal formation in the papers by PérezGrande et al. [7], Sauermann et al. [8], Teodorczyk and Januszkiewicz [9].
Because of the high temperatures within the furnace, radiant heat transfer is the primary mode of heat transfer. This is discussed briefly in the MISG paper [5], but for a complete discussion we cite some standard texts by Incropera and DeWitt [4], Modest [6], Siegel and Howell [3].

the region of space occupied by the strip is$$ \mathcal{S}=\{(x,y,z): 0\le x\le L, w/2\le y\le w, 0\le z\le h\}; $$(2)

L is the length of the furnace;

x measures distance from the point of entry of the strip into the furnace, z is a distance coordinate in the vertical direction and y is a distance coordinate across the strip;

v is the velocity of the strip through the furnace;

w and h are respectively the width and thickness of the strip.

ρ _{ S }, C _{ S } and k _{ S } are the strip’s density, specific heat capacity and heat conductivity respectively.
The functions w and h are typically piecewise constant functions of x and t and v can vary with time, but in this paper we limit our analysis to the desirable steady state operation of the furnace for which these variables are constant.
Mathematically, it is also appropriate to specify a boundary condition where the strip exits the furnace at x=L. One could propose a model leading to an appropriate boundary condition there. However we will see soon that heat conduction in the steel strip in the direction of the xaxis is very small, which means that the term involving \(\frac {\partial ^{2} u}{\partial x^{2}}\) can be neglected everywhere except in a small boundary layer near x=L. Physically, heat is not conducted quickly enough for the temperature of the part of the strip that has already left the furnace to affect the temperature within the furnace. Thus, any boundary effect at x=L is not expected to be significant and thus we do not attempt to model it.
for (x,z)∈(0,L)×(0,h). The incoming surface heat fluxes ϕ _{ a }, ϕ _{ b }, ϕ _{ c } and ϕ _{ d } are determined by considering an energy balance of the radiation within the furnace.
which was the justification in [5] for neglecting the heat conduction terms for the x and y directions in (1). Note however that boundary conditions must be satisfied, so one expects boundary layers near y=±w/2 where the boundary conditions are satisfied. We wish to investigate these particular boundary layers to see how much they contribute to edge heating. Thus we depart from the analysis in [5] by retaining the terms involving \(\frac {\partial ^{2}u}{\partial y^{2}}\). However, as in [5], we neglect the term involving \(\frac {\partial ^{2}u}{\partial x^{2}}\) and any associated boundary layer for reasons discussed earlier.
The physical problem has reflectional symmetry through the xz plane, so we assume that \(\Phi _{E}^{+}=\Phi _{E}^{}=\Phi _{E}.\)
We wish to investigate two effects that could lead to edge heating of the strip. The first is the creation of a boundary layer near the edges y=±w/2 due to the boundary condition (11) there. We do this in Section 2. The second effect is a variation of Φ _{ S } in the direction of the yaxis that might explain extra heating near the edges. This requires a detailed analysis of the radiation heat transfer problem to calculate Φ _{ S }. We do this in Section 3. For the boundary layer analysis of Section 2 we use a simple approximation for Φ _{ S } that is independent of y.
Analytical treatment of edge heating
where ε _{ S }≈0.2 and ε _{ W }≈0.9 are the emissivities of the strip and furnace materials respectively, σ=5.670×10^{−8}Wm^{−2}K^{−4} is the Stefan–Boltzmann constant and p is the sum of height and width of a crosssection of the space inside the furnace. The temperature of the furnace walls and heating elements is assumed to be the same and is given by T _{ W }. We note that this flux does not vary across the width of the strip. Smaller emissivities are associated with more reflective surfaces, which lead to a greater amount of reflection of radiant heat energy arriving at a surface.
Φ _{ E }, the heat flux absorbed at the edges, is expected to be greater than Φ _{ S } because the steel strips are formed by cold rolling of steel which results in a rougher, less reflective surface at the edges.
We limit our analysis to the steady state operation of the furnace. This simplifies the analysis because it allows us to approximate the heat flux Φ _{ S } using the power supplied to the heating elements. For the nonsteady state operation, one needs to take into account the heat dynamics that occur near the inner surface of the furnace walls, which are coupled to the dynamics of radiant heat transfer and heat transfer within the steel strip. For steady state operation, one can simply use the fact that the furnace walls are very good insulators and neglect the heat lost through them.
We thus seek steady state solutions of Eqs. (8), (10) and (11). In order to get a closedform expression for the solution, we assume in Section 2.1 that ρ _{ S }, C _{ S }, k, Φ _{ S } and Φ _{ E } are all constant. We analyse the more general case for which these quantities are not constant in Section 2.2.
2.1 The case of constant ρ _{ S } C _{ S }, k, Φ _{ S } and Φ _{ E }
Here, \(\tilde {T}_{0}=T_{0} \frac {hv \rho _{S} C_{S}}{\Phi _{S} L}\) and δ is given by (5). In these equations, \(0<\tilde {x}<1\) and \(1/2<\tilde {y}<1/2\).
We seek the steady state solution of the whole system (13)–(15), so we set \(\Tilde {T}=\tilde {T}_{0}+2\tilde {x}+\tilde {T_{2}}\) and see that \(\tilde {T}_{2}\) must satisfy
2.2 The case of variable ρ _{ S }, C _{ S }, k, Φ _{ S } and Φ _{ E }
To allow for such variations, we assume that ρ _{ S }, C _{ S } and k _{ S } are known functions of the strip’s temperature. Further, because our system is at equilibrium, Φ _{ S } and Φ _{ E } are assumed to be known functions of x which can be calculated by measuring the power supplied to the heating elements in the vicinity of a distance x along the furnace.
and this corresponds to a solution outside boundary layers.
We use similar approximations for C _{ S }(T) and k _{ S }(T).
χ must also satisfy an “initial” condition, χ(0,ζ)=0, and a matching condition, \(\chi (\tilde {x},\zeta)\to 0\) as ζ→∞.
where \(g(\zeta,\tilde {x})=\frac {\partial \psi }{\partial \tilde {x}}=e^{\zeta ^{2}/4\tilde {x}}/\sqrt {\pi \tilde {x}}\) and ψ is given by Eq. (25).
Furnace radiative heat transfer analysis
3.1 Assumptions
 1.
The heating elements are distributed uniformly over the top and bottom inner surfaces and the the density of the input electric power is specified as a constant.
 2.
All surfaces are considered opaque gray. All surfaces emit and reflect radiation diffusely; the typical emissivity of the furnace wall surface and the heating element is ε _{ W }=ε _{ E }=0.9 and of the strip is ε _{ S }=0.2. For an opaque gray surface, the reflectivity ρ and emissivity ε are related by ρ=1−ε.
 3.
Temperature changes within the furnace are gradual and radiative and thus convective heat transfer along the length of the furnace can be ignored. The strip temperature at the entry of the furnace is at room temperature and it can reach up to 700 °C at the last heating stage, which is still significantly lower than the temperature of wall surface and the heating elements. Considering that the radiative power is proportional to the fourth power of the temperature, the dominant radiation is from the wall surfaces and the heating elements.
These assumptions simplify the analysis and are reasonable for a furnace with brick covered wall and a steel strip with rough surface finishing. For a steel strip with smooth surface finishing, a partly specular reflection model shall be considered.
We use these assumptions to develop a two dimensional model of the temperature distribution within the furnace. The model is two dimensional only in the sense that it relies on the approximation that there is only a gradual variation of temperature in the direction of the moving strip.
We are interested in temperature variations across the strip and for this we must solve a system of integral equations for the radiative and reflective heat exchange between surfaces within the furnace.
3.2 Mathematical model
For a diffuse surface, it is well known that the net radiation method can be used to analyse the heat transfer. This method is discussed in many texts on thermal radiation such as the works of Modest [6] and of Siegel and Howell [3]. The method, which involves an energy conservation argument for the absorption, emission and reflection of radiation inside an enclosure, results in an integral equation.
Note that d x denotes the differential strip element which, due to the longitudinal symmetry, is infinite in the x _{3} direction.
for parallel elements, where β are the angles shown in Fig. 5, d is the perpendicular distance between the two parallel elements, see [3].
for a surface where the input power flux is given. The integral equation uniquely determines the outgoing heat flux q(x).
3.3 Numerical procedure
In general such integral equations do not have a closed form solution so a numerical method is needed to find an approximate solution. The integral equation is linear and the discretised equation is a linear system and can be solved by a standard LU decomposition. To numerically solve (48) and (49), all of the surfaces, which because of the assumption of longitudinal symmetry of the problem, are one dimensional domains, are divided into sufficiently small intervals of equal length and the q(x) is assigned on the nodes q(x _{ i }). A standard trapezoid method is applied to integrate (48) and (49) numerically, resulting in a linear system with q(x _{ i }) as unknowns.
 1.
The k(x,x ^{′}) function has a discontinuity at the corner points of the wall arising from the two different formulas for the parallel and perpendicular elements. Thus, the numerical integration is performed on the individual planar surfaces. This gives a total of six planar surfaces including four wall surfaces and two strip surfaces.
 2.
For a node x located on a furnace wall, the kernel k(x,x ^{′}) is only a piecewise smooth function of x ^{′} due to the presence of the steel strip and its shadow effect.
This can be observed from Fig. 6, showing a node x exposed only to partial heat flux emitted from another wall surface. The kernel k(x,x ^{′}) has jumps at \(\mathbf {x}^{\prime }_{\mathbf {1}}\) and \(\mathbf {x}^{\prime }_{\mathbf {2}} \), each of which lies between two neighboring nodes. The exact positions of \(\mathbf {x}^{\prime }_{\mathbf {1}}\) and \(\mathbf {x}^{\prime }_{\mathbf {2}} \) can be found from the geometric relation. The numerical integration is performed only on the viewable portion of the relevant subintervals bounded by \(\mathbf {x}^{\prime }_{\mathbf {1}}\) (or \(\mathbf {x}^{\prime }_{\mathbf {2}} \)) and one of the two neighboring nodes.  3.
Due to the singularity of the kernel k(x,x ^{′}), when two nodes on the neighboring wall are sufficiently close, the variation of k(x,x ^{′}) on a single element is significant for whatever small step size has been chosen. To overcome this singularity, the following approximation is applied to these nodes.
Assuming a linear distribution of the q(x) on the element \((\mathbf {x}^{\prime }_{\mathbf {i}}, \mathbf {x}^{\prime }_{\mathbf {i}+\mathbf {1}})\), we may estimate the integral in (48) and (49) as$$ \begin{aligned} \int_{\mathbf{x}_{i}^{\prime}}^{\mathbf{x}_{i+1}^{\prime}} q(\mathbf{x}^{\prime}) k(\mathbf{x},\mathbf{x}^{\prime}) d\mathbf{x}^{\prime} \approx \frac{1}{h } \int_{\mathbf{x}_{i}^{\prime}}^{\mathbf{x}_{i+1}^{\prime}} ((q(\mathbf{x}_{i+1}^{\prime})\\q(\mathbf{x}_{i}^{\prime})) \mathbf{x}^{\prime}  \mathbf{x}_{i}^{\prime}+ q(\mathbf{x}_{i}^{\prime})h) k(\mathbf{x},\mathbf{x}^{\prime}) d\mathbf{x}^{\prime} \end{aligned} $$(50)where \(h=\mathbf {x}^{\prime }_{\mathbf {i}} \mathbf {x}^{\prime }_{\mathbf {i}+\mathbf {1}} \) is the step size. This can be written in terms of q(x _{ i+1 }) and q(x _{ i })$$\begin{array}{@{}rcl@{}} \int_{\mathbf{x}_{i}^{\prime}}^{\mathbf{x}_{i+1}^{\prime}} q(\mathbf{x}^{\prime}) k(\mathbf{x},\mathbf{x}^{\prime}) d\mathbf{x}^{\prime} \approx C_{i}^{(1)} q(\mathbf{x}_{i}^{\prime}) + C_{i}^{(2)} q(\mathbf{x}_{i+1}^{\prime}) \end{array} $$where the coefficients \(C_{i}^{(1)}\) and \(C_{i}^{(2)}\) are determined by$$\begin{array}{@{}rcl@{}} C_{i}^{(1)}= \frac{1}{h}\int_{\mathbf{x}_{i}^{\prime}}^{\mathbf{x}_{i+1}^{\prime}}(h  \mathbf{x}^{\prime}x_{i}^{\prime}) k(\mathbf{x},\mathbf{x}^{\prime}) d\mathbf{x}^{\prime}. \end{array} $$(51)and$$\begin{array}{@{}rcl@{}} C_{i}^{(2)}= \frac{1}{ h}\int_{\mathbf{x_{i}^{\prime}}}^{\mathbf{x_{i+1}^{\prime}}}  \mathbf{x}^{\prime}\mathbf{x}_{\mathbf{i}}^{\prime} k(\mathbf{x},\mathbf{x}^{\prime}) d\mathbf{x}^{\prime}, \end{array} $$(52)respectively. Adding the l corner elements leads to$$ {\begin{aligned} {}{ \sum_{i=1}^{l}} \frac{1}{h } \int_{\mathbf{x}_{\mathbf{i}}^{\prime}}^{\mathbf{x}_{\mathbf{i}+\mathbf{1}}^{\prime}} ((q(\mathbf{x}_{\mathbf{i}+\mathbf{1}}^{\prime})q(\mathbf{x}_{\mathbf{i}}^{\prime})) \mathbf{x}^{\prime}  \mathbf{x}_{\mathbf{i}}^{\prime}+ q(\mathbf{x}_{\mathbf{i}}^{\prime})h) k(\mathbf{x},\mathbf{x}^{\prime}) d\mathbf{x}^{\prime} \\ = C_{1}^{(1)}q(\mathbf{x}_{\mathbf{1}}^{\prime}) +C_{l}^{(2)}q(\mathbf{x}_{\mathbf{l}+\mathbf{1}}^{\prime}) + \sum_{i=2}^{l1} (C_{i}^{(1)}+C_{i1}^{(2)})q(\mathbf{x}_{\mathbf{i}}^{\prime}) \end{aligned}} $$(53)l will be chosen to cover all elements affected by the kernel singularity.
 4.There are two nodes belonging to two neighbouring walls intersecting at a corner point denoted by \(\phantom {\dot {i}\!}\mathbf {x}_{\mathbf {S}_{\mathbf {1}}}\) and \(\phantom {\dot {i}\!}\mathbf {x}_{\mathbf {S}_{\mathbf {2}}}\) where S _{ 1 } and S _{ 2 } indicate the surfaces the nodes belong to. The integration with respect to x ^{′} over the surface S _{ 2 } for \( \mathbf {x} = \mathbf {x}_{\mathbf {S}_{\mathbf {1}}} \phantom {\dot {i}\!}\) should be estimated by calculating its value at a nearby point x _{ ε }∈S _{ 1 }, \(\mathbf {x}_{\epsilon } \sim \mathbf {x}_{\mathbf {S}_{\mathbf {1}}}\phantom {\dot {i}\!} \), and then passing to the limit as \(\mathbf {x}_{\epsilon } \rightarrow \mathbf {x}_{\mathbf {S}_{\mathbf {1}}} \). It can be shown that$$ \frac{1}{2}q(\mathbf{x}_{\mathbf{S}_{\mathbf{1}} }) ={\lim}_{\mathbf{x}_{\epsilon} \rightarrow \mathbf{x}_{\mathbf{S}_{\mathbf{1}} }}\int_{\mathbf{S}_{\mathbf{2}} } q(\mathbf{x})k(\mathbf{x}_{{\epsilon}},\mathbf{x}^{\prime}) d\mathbf{x}^{\prime}. $$(54)
Remark: Equation (54) has a clear physical meaning: A differential element at node \(\phantom {\dot {i}\!}\mathbf {x}_{\mathbf {S}_{\mathbf {1}}}\) on surface S _{ 1 } receives half of the total heat flux which is emitted from a neighborhood of \(\phantom {\dot {i}\!}\mathbf {x}_{\mathbf {S}_{\mathbf {2}}}\) on surface S _{ 2 }.
3.4 The isothermal surface case
Thus, the heat transfers between the surfaces can be calculated by using the view factor. By solving this problem with the numerical method developed in this article, we found that q(x) is actually always varying with the location. However, the variations of q(x) across the wall or strip are not significant. This is particularly true for small ε _{ S }, as in such cases, the strip reflectance ρ _{ S }≈1 and the radiated energy from the wall is well reflected by the strip, and thus q(x) becomes less dependent on the location.
This is not a surprising result because the view factors of the strip to various wall locations are very different and the portion of the heat energy emitted from the wall which is eventually absorbed by the strip is largely determined by the relative geometric position, or in other words, by the view factor.
Through these comparisons, we found that the numerical results correctly reflected the geometrical effects due to the view factor and matched very well with formula (55) at the limit case ε _{ S }≈0 as expected. This validates the numerical method.
3.5 The strip temperature distribution
 1.
The power input density of the top and bottom heating elements is specified as a constant p=1.294×10^{4} Wm ^{−2}, which is typical in this applications.
 2.
The side wall surface is considered as a perfect thermal insulated surface.
 3.
ε _{ W }=ε _{ E }=0.9 and ε _{ S }=0.2, where ε _{ E } denotes the emissivity of the heating elements.
 4.
Find the strip surface temperature distribution.
 1.
The strip temperature 5 0 0°C.
The numerical results are shown in Fig. 10. It was found that (a)The temperature of the electric power element varies from 949°C to 983°C. The temperature variation is about 3%.
 (b)
The temperature of the side wall varies from 949°C to 956°C. The temperature variation is about 0.7%.
 (c)
The strip net power influx varies from 2.186×10^{4} W m ^{−2} to 2.216×10^{4} W m ^{−2}. The net influx variation is about 1.3%.
 (a)
 2.
Temperature T _{ S }=2 0°C.
The numerical results are shown in Fig. 11. It was found that (a)The temperature distributions of the heating element and side wall are similar to the case T _{ S }=500 with a slightly lower temperature range.
 (b)The net heat influx of the strip is indeed very similar to the case T _{ S }=500. Figure 12 shows the comparison of the two results.
 (a)
Concluding remarks
We have analysed in detail two effects that contribute to extra heating of the steel strip at its edges.
The numerical results of Section 3 indicate that the geometrical effect due to view factors would account for an elevated temperature at the edges of about 7°C.
The analysis of Section 2 took into account the fact that the edges of the strip are really surfaces themselves. Although these surfaces are small, they contribute significantly to temperature increases at the edges because the rate of heat conduction away from the edges is slow. If one assumes that the edges are smooth and have an emissivity of about 0.2, the same as the larger surfaces of the steel, then this effect would result in temperature elevations of about 9°C at the edges. In reality the edges are much more rough than the rest of the strip’s surface. The actual temperature elevation is proportional to the emissivity, so an emissivity of 0.5, for example, would contribute to a temperature elevation of about 22°C near the edges. Moreover these elevated temperatures occur within about 1 cm of the edge of the strip, resulting in potentially damaging high temperature gradients.
Declarations
Authors’ contributions
SWT was responsible for Section 2, SW was responsible for Section 3. Both authors read and approved the final manuscript.
Competing interests
The authors declare that they have no competing interests.
Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Open Access This article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made.
Authors’ Affiliations
References
 Depree, N, Sneyd, J, Taylor, S, Taylor, MP, Chen, JJJ, Wang, S, O’Connor, M: Development and validation of models for annealing furnace control from heat transfer fundamentals. Comput. Chem. Eng. 34(11), 1849–1853 (2010).View ArticleGoogle Scholar
 Depree, N, Taylor, MP, Chen, JJJ, Sneyd, J, Taylor, S, Wang, S: Development of a threedimensional heat transfer model for continuous annealing of steel strip. Ind. Eng. Chem. Res. 51(4), 1790–1795 (2012).View ArticleGoogle Scholar
 Howell, JR, Siegel, R: Thermal Radiation Heat Transfer. 5th. CRC Press, Boca Raton, Fla. (2011).Google Scholar
 Incropera, FP, DeWitt, DP: Introduction to Heat Transfer. 4th. John Wiley and Sons, New York (2002).Google Scholar
 McGuinness, M, Taylor, SW: Strip Temperature in a Metal Coating Line Annealing Furnace. In: Proceedings of the 2004 MathematicsinIndustry Study Group (2004). http://www.mathsinindustry.org/miis/41/.
 Modest, MF: Radiative Heat Transfer. 2nd edn. Academic Press, SAN DIEGO, CA (2003).MATHGoogle Scholar
 PérezGrande, I, Rivas, D, de Pablo, V: A global thermal analysis of multizone resistance furnaces with specular and diffuse samples. J. Crystal Growth. 246, 37–54 (2002).View ArticleGoogle Scholar
 Sauermann, H, Stenzel, CH, Keesmann, S, Bonduelle, B: Highstability control of multizone furnaces using optical fibre thermometers. Cryst. Res. Technol. 36(12), 1329–1343 (2001).View ArticleGoogle Scholar
 Teodorczyk, T, Januszkiewicz, KT: Computer simulation of electric multizone tube furnaces. Adv. Eng. Softw. 30, 121–126 (1999).View ArticleGoogle Scholar
 Zareba, S, Wolff, A, Jelali, M: Mathematical modelling and parameter identification of a stainless steel annealing furnace. Simul. Model. Prac. Theory. 60, 15–39 (2016).View ArticleGoogle Scholar