5¢4u F.mswyVol. 24, pp.

441--449
Pc~amoa Press Ltd., I~0. Pdated in Great Britain

A CONVOLUTION MODEL OF ROCK BED

THERMAL STORAGE UNITS
E. F. SOWELL~and R. L. CURRY~
California State University, Fullerton, CA 92634, U.S.A.
(Received 13 July 1979; accepted 11 December 1979)

Abstraet--A method is presented whereby a packed-bed thermal storage unit is dynamically modeled for
bi-directional flow and arbitrary input flow stream temperature variations. The method is based on the principle of
calculating the output temperature as the sum of earlier input temperatures, each multiplied by a predetermined
"response factor", i.e. discrete convolution. A computer implementationof the scheme, in the form of a subroutine
for a widely used solar simulation program (TRNSYS) is described and numerical results compared with other
models. Also, a method for efficientcomputationof the required response factors is described; this solution is for a
triangular input pulse, previously unreported, although the solution method is also applicable for other input
functions. This solution requires a single integration of a known function which is easily carried out numericallyto
the required precision.

INTRODUCTION and
The transfer of heat to and from fluids flowing through ~ = p~ALCb/(rhC~). (4)
packed beds has been of industrial importance for some
time, and has therefore been the subject of numerous The widely known model described by Duffle[6], for
investigations[l--A]. Moreover, it is well known that example, reduces (1) and (2) to a set of simultaneous
identical partial differential equations govern a number first-order, ordinary differential equations by segmenting
of other physicalphenomena, such as transient heating the bed. This model is known to require a large number
of duct walls,' and the processes within chemical of segments and short-time step for reasonable accuracy,
reactors[5]. However, most of the solutions to these which leads to excessive solution times for long-term
equations presented in the literature are for the bed and simulation[7]. The Solar Energy Laboratory at the Uni-
exit fluid temperature in response to a step change in the versity of Wisconsin simplified the problem by assuming
inlet fluid temperature. infinite fluid-bed heat transfer (NTU-+~), and thus
More recently, there has been a need for general arrived at a more efficient model for incorporation in the
mathematical models, suitable for computer simulation TRNSYS program[8] (this model is referred to here as
of systems which include packed beds. Such models the Hughes-Klein-Close model). Hughes et al.[9] show
present a more challenging problem, since they must that this model may be accurate enough for long-term
predict outflow fluid temperature for arbitrary, temporal simulation studies of some systems with finite NTU. The
variations in inlet temperature. In many .system simula- principal problem with this model is that it can be
tions, such as of solar heating systems, the model must misleading for simulations of systems which are sensitive
also allow for flow reversal at arbitrary times. Addition- to the sharpness of bed stratification; the actual smearing
ally, there is great importance placed on computational of input temperature profiles is lost completely and sup-
efficiency and accuracy, since often simulations are car- planted with artificial distortion related to discretization
ried out for exceedingly long times (several thousand error rather than physical factors. Hughes discusses
hours). From this point of view, the mathematical solu- several other models in[10], including one in which the
tions presented in the literature are only the beginning of fluid temperature is allowed to vary while flowing
a suitable simulation algorithm. through a uniform bed segment. Later, Mumma and
All simulation algorithms for packed beds presented to Marvin [7] reported detailed numerical experiments with
date are based on finite-difference representation of the this model (referred to here as the Hughes/Mumma-
governing equations, which are Marvin model) and were impressed with the relatively
large time and distance steps which it allowed without
OT
O(--X[L)= NTU(Tb - T) (1) appreciable loss of accuracy for the step response.
Hughes reports even superior performance of a variation
aT~ in which the bed segment temperature is allowed to vary
~(--~) = N T U f T - Tb) (2)
linearly as it exchanges energy with the fluid, von
Fuchs[ll] has carried out numerical studies with an
where
improved version of this model and has shown it to be
N T U = h~AL/(mC/), (3) considerably better with respect to prediction of the bed
profile several hours after a step change in the input.
These models have several attractive features and
tEngineering Division. have many applications. For example, they allow arbi-
~Mathematics Department. trary inlet flow variations, arbitrary initial bed profiles,

441
442 E. F. SOWELLand R. L. CURRY

and variable, bidirectional flow• However, they still factors ri of the bed. They depend upon the physical
require the solution of a set of differential equations properties of the bed, and the mass flowrate. The actual
during the simulation, and it develops that computation shape of the response curve, and hence the r/s, can be
time is controlled by stability and accuracy con- found by solution of eqns (1) and (2), but will be deferred
siderations. In this connection it should be noted that the until the use of these factors is explained.
tests reported by Hughes and Mumma were for a step Since eqns (1) and (2) constitute a linear, invariant
change in inlet temperature. In this work it has been system (provided mass flowrate is constant), super-
found that a more realistic sequence, such as inlet tem- position applies, and it is apparent that the outlet tem-
perature varying as the daily insolation profile, followed perature due to any number of sequential inlet pulses is
by flow reversal for heat extraction, is a more severe test the sum of their individual effects. For example, the
and usually requires smaller time steps than suggested by output at time 28 due to an input unit pulse centered at
Mumma. time 0 and followed by another at time 8 is
It is the purpose of this paper to present an alternative
simulation model for packed beds which is potentially Ti(28) = Tbi + r, + r2.
more accurate and efficient than any of the models
reported to date. Its accuracy derives from the fact that This is so since r~ is the output effect of an input unit
it employs a predetermined solution to the partial pulse 8 earlier, and r2 is the effect due to one 28 earlier.
differential equations, which can be determined with The linearity property also ensures that a pulse of
great precision. Its efficiency derives from the fact that height "a" will result in an outlet effect of (ar3 at i8 time
differentials are effectively removed from the simulation units later. It is therefore evident that a series of over-
equations• However, it must be observed that this model lapping triangular pulses at the inlet of height Tk.i-~ and
is based upon the convolution principle from linear sys- widths 28 at times k = 0, 1,2... results in outlet tem-
tem theory[12], and therefore has certain limitations• For peratures at 8 times increments of
example, in its simplest form, it is restricted to constant,
unidirectional flow in an initially-relaxed bed. On the
T~.i = T~ + ~, r~(T~_~.j_, - T~j). (5)
other hand, it is shown that by coupling the method with i=o
bed segmentation, flow reversal can be accommodated
while retaining most of the computational efficiency. It is The summation continues until the rl becomes negligibly
notable that this technique, applied to conduction in small. Mathematically, this process is the discrete form
solids, has virtually replaced finite-difference for- of the convolution integral[12].
mulations in building heat transfer simulations in recent As can be seen in Fig. 2, a series of overlapping
years [13-15]. triangular pulses constitute a continuous input tem-
perature profile composed of straight line segments. It is
THE CONVOLUTIONPRINCIPLE therefore evident that eqn (5) represents the exact solu-
The convolution model is based on the principle of tion to eqns (1) and (2) at discrete times for an initial bed
linear super-position, which suggests that the outlet temperature of Tbj and an inlet temperature profile con-
temperature at any instant is the sum of effects of all sisting of a piecewiseqinear curve, provided only that the
earlier inlet temperature disturbances. The process can r~'s are exact. This is the basic convolution model.
be most easily understood after first considering the It is notable that the convolution model is free from
response of the bed to a single inlet temperature dis- discretization error during the simulation process. Also,
turbance, usually taken as a unit triangular pulse. Figure there is no question of stability; the simulation time step,
I shows the bed with a constant flow and an initially which must equal 8, can be as large as desired, provided
uniform temperature, Tbj. The inlet flow temperature is only that 8 must be short enough to allow satisfactory
then disturbed in the form of a triangular pulse of unit representation of input profiles using a piecewise linear
height above Tbi.t It is apparent that the output tem- curve with breakpoints at this interval. This is in contrast
perature will also rise and fall, although this response to the finite-difference models which require selection of
will be later in time and somewhat "smeared". The time step based on stability considerations, often neces-
amplitudes of the output response curve at intervals of 8, sitating short time steps even for smooth expected in-
the half-width of the input pulse, are called the response puts.
For unidirectional flow, eqn (5) is all that is required. It
fThe index j is introduced for consistency of nomenclature, is suitable for computer programming, or hand cal-
since it is required for segmented beds. culations once the r,'s are known. One improvement,

T/-I
Rock bed
/

-~ o 0 ~ 26B 3B

t---> t--->
Fig. I. Response of bed to unit triangular pulse.
 A convolution model of rock bed thermal storage units 443

Equation (8) employs the convention that the jth column

of the T array contains the outlet flow temperature of
the jth bed segment.
In the complete, bidirectional convolution model, the
initial bed profile Tb is specified and used to initialize all
historical values of fluid temperature, T. Then as long as
flow continues in the same direction, only eqn (6) is
applied. Upon flow reversal, eqn (8) is used to recalculate
Fig. 2. Representationof inlet temperatureby series of triangular
Tb and reinitialize T.
pulses.
CALCULATIONOFTHERESPONSEFACTORS
however, is usually desired in order to reduce the num- Response factors are the ordinates of the curve r(t)
ber of terms in the summation, and hence, the number of for t =0,8,28... where r(t) satisfies eqns (1) and (2)
historical values of the temperature to be saved. with the initial and boundary conditions of
A property of the response factors is that, due to the
exponential character of the response curve at large T(0) = Tb(0) = 0,
times, they tend to a geometric series, so that after m
T(t) = (t/8)(U(t + 8)- 2U(t)+ U(t - 8)), (10)
terms
where U(t) = unit step function.
r,+dr, ----CR, T(t) is seen to be a unit triangular pulse at time zero,
with a base half-width of 8. There are several ways of
called the common ratio. It can then be shown that eqn
obtaining the solution r(t), including
(5) becomes
(1) A finite-difference solution, for example, using the
Hughes/Mumma-Marvin model.
Tk.j = Tbi + ~ r~T~_,.j_, - Tbj) + CR( Tk_,.j - Tbj), (2) Integration of one of the available solutions for a
i=0
step input, e.g. Anzelius [2] or Schumann[l].
(6)
(3) Apply a Laplace transformation to eqns (1) and (2)
where and invert the resulting complex series solution term-by-
term.
r~=ro
(4) Develop the solution directly for the triangular
~=r~-CRrH; i>0. (7) pulse.
All of these approaches have been tried. The first is
This shows that by using a single historical value of the straightforward, but solution time and accuracy are
outlet temperature, one can significantly reduce the difficult to control. The second was programmed using
required number of terms in the summation. It can also double numerical integration of the Anzelius solution;
be shown[15] that by using more historical outlet values, for large NTU the solution time became unacceptable.
further reduction in the inlet temperature suinmation can The third method results in a series which does not
be achieved. converge for many values of NTU of interest. The first
cue BtDntF.C~OSALCONVOLUTIONMOD~L three methods are therefore not recommended.
The fourth method, developed in this work, is based
The previous development assumes an initially relaxed
upon solution of eqns (l) and (2) directly for the trian-
bed. It is not directly useful for bidirectional beds, since
gular pulse boundary condition at the inlet. The solution,
upon flow reversal arbitrary bed profiles may exist. In
details of which are given in the appendix, expresses the
order to allow calculations after flow reversal, it is
response function r(r) in terms of an integral of a known
necessary to segment the bed which permits calculation
function, i.e.
of this profile.
Equation (6) applies also to the segmented bed, with
the understanding that j is the index of the bed segment, r(¢) = exp (- NTU){ f(¢) + ~ N T u w f( ~ -4~TU2 )
ranging from 1 to N, the number of segments, and that
the r~ now are for a bed length L/N. The outlet tem-
x exp (- ~2/4NTU)I,(71) d *lI (l l)
perature of one segment becomes the inlet temperature
to the next. where f(¢) is the triangular pulse described by eqn (10), T
By applying eqn (6) at each time step, the inlet and is dimensionless time (t + 8)/~c, and 7/is a dummy vari-
outlet fluid temperatures are known for each bed seg- able of integration. This solution has been programmed
ment. This allows recalculation of the bed profile at any for evaluation using Romberg integration; this approach
desired time by means of the equation yields acceptable times for a wide range of NTU and ~'c.
Typical results are shown in Table I.
Tb~= bTk~-(b - 1) Tk.j-,, (8)
TRNSYS IMPLEMENTATION
where Although the basic algorithm is simple, efficient im-
b = 1/(1 - exp (NTU/N)). (9) plementation is not straightforward. This is especially
444 E. F. SOWELLand R. L. CuRRy

Table 1. Rock bed response factor

IA+'C ~*00 t~.+0 11,.00 F~. I0 .U0 10, ~0 ~,00 ~.00 I~,00

,J

P .5S7350E'0| ,7%S?~2E-01 ,I+RbSpF-OI *|~4~Rq[-II *+H~l~t-I)2 *ll~lE-~l~ ,4n~OISFo02 *mTOH~IF-O~ .|~4~F-n~

3 *|O?20AFPO0 *SO3~IgE-OI ,p?7||flV*OI *~o0]F*'il *t7~q3q~*01 .gT~=+F-';~ ,+m47R4F-O] *~7~I+~F-O~ ,|1071%Ko~
4 *|3P§41E~O0 o76q621E-01 .44S~ilF-0| *i~laNflFO"O ,4RPqp3r-o| .t73~4E-~l *|~7S~FOOU *~7~F-O] ,6110fl~F-~2
.IS~?06F'+00 *qgO4q2E'~1 .610R5~-0| ,|aqRg?EO~O *9201&OF-01 *~U2R4RE-ql *PP~RRHF000 .?k6~*FoO] *P~qP~+F-+|
h *1445')~F'100 .|lZIqS(no+ .7~41177E+'0| *+04147E0^0 *|03104F000 ,~624~?E-+| *~4R0~?¢000 .1~7;4mF000 .+qg++~F-~l
? *12|Sq'lE+~00 *||$0)8[00+ *~h778flF-Ol *|k~|fl7EQ~O .!~k~O6FO00 *q4999HE-h| ,lnfl~3|FO00 .l~++~qFO00 .RqguFqF-~I
*92~|qqE'O| *1088~4E000 ,~lk?87F'Ol *|~q146E000 .1~44~PF000 ,11700nE00~ .|~0246F000 *l~8~qqF000 *17~444F00~
q ,bS+le, O(oO| *g63k09(-OI ,ql7hB6F-Ol *~Tk145E°"| ,|3~37r000 ,l~70qqEOuO ,442003F-0| ,1060~+F000 *1~]47~F0~
I0 ,43?36?[*0| ,806|?8[-0| ,R7~+$4[-0| *~7~|0[-,'| .|0017N~000 ,|240~4[0il0 ,|~P334F-O! .]074~F000 *|5180~F000
II ,2700||F-0| o642439E-01 .7q4~30F*O| *|1~|3qF*~1 *6840q~F-+| .1|0474E+~0 .4~K333F-02 *~I+~IF-O| *]?22P~FO00
12 *16~27~E-01 *490659[*0) .6q7335r*O| *4q~PO2E"!~ .4~7~4~F-0~ *~07977['~| *|oS67~F'0~ *~?~F~F-O| *]D]qO?F~O0
|3 ,goo94qE-O? ,360q33('01 ,~qPIOSF*OI *1~974~Eo!~2 ,~4~q7&r-o| .~9S3~3(-01 .p?2124F-03 ,)4~11~F*01 ,7~4qp4r-nl
14 .49~66"E'02 *?$676§E'0] .4RR|81F*O| ,470qO3F-q3 *131q4|F-O| *4q99?~E'~| .410767F-04 .~q2q~F-02 *4427P~F-~|
|~ .26~870[-02 *|?724~['0| *3q|900E-O| *|~9130['~3 *6fl~qA4F*O~ *+]q7~%['~| .677|~4F-05 *~]O~14F-OZ ,?~50t~-~|
]~ ,I]pg~TFo02 .11g062(-01 ,307~7SF-0| ,2~1734E'!14 .31~=74P-02 *Plq31~fo~l *|+Off|IF-05 .7+511~F-03 *|3~8~gF-~l
17 ,6Selg6E-03 *700161E'0~ ,73~662F'0] *;m4~7~E-~5 ,14Pn46F-O~ ,|35|~nE-o| ,|+~027F-06 .2PB~F-O~ ,6718a2F-~2
Ip ,3172.~=E-03 ,4q9702[-07 .I?71~|E'O] *E~OOfi6E-~E ,609~4F-03 *Tqs)PIE-02 .I~E69~F-07 *+~97~F-04 .~131~Fo~2
19 *|6ql~4E-03 ,313427[-0~ ,130730F*0! *~g~??OF-~l~ *~50~3F-03 .4531~4E-02 *lq~233F-08 *|7800kF-04 .I]70Fn;o~2
?e ,bS~54~E-04 *]420i?E'OP .q4830~F-02 *[>2~30E-~6 ,qS?~?iF*04 ,p4A]4~[-n2 *p,5?0qF-09 *~5||UqFo05 o5~?S|~*~
21 .30P§OuE'04 .116S04E'0~ .67717~F-02 *~RhSE'~7 .37~n04~-04 *|313q~Eo~2 *pO402?F']O *lnTqp~F-05 *PP~21?F-~
?2 ,13~1~¢E-04 *69~P35E-03 ,476423F-02 ,~4053[-~0 .l~?4p4r-04 ,67424~[-113 .l~lOq2F-|| *p44~IE-~6 ,A36q~?F-~4
~3 *~SqT~bE'O~ *4049|4['0+ ,330~7qFo0~ *i~PlgqE'+;8 ,4~7~7~F-05 *0360~7E'03 *|~740F-|2 *~7|30F'07 ,3~|1|F-~4
P~ ,?5111~EoOS *p33~95[-0~ .226476F-02 o|~?134F*~9 .|6?~p4r*oS .1629~E*03 *l~Olq4F-l~ .1~4F-07 ,103~n4F*~4
P~ *lOg21|E-O~ ol32686(-0~ .1~0207E-02 *~q108E-!O .5S940~F-06 .770|~1E-04 *1109~8F-|4 ,?I311AF-OB *041703Fo~5
?~ ,434|3~E-06 o744570E-04 ,IOP4BOF*O~ ,~147]?E-I! .]01Ap4F-06 ,~5P~[-~4 ,~1~02|F-|6 .4~0~1F-~9 o|+@7?QF-~
?R *?~SG7E-OT *226|22('04 ,444076E'0~ o|7436|E'1~ o|7R~O|F-O? .70~R~0[o05 .4140~1F-|8 .l?7k;~F-|0 .qq~171F-~?
~ *?R+6+!E*O? *|72S31E'04 o28ROO4F-03 *~!A|gF'|3 ,537~44F-08 *+055?flE-05 ,P?44~2F-|O *~14Q7n[-|| *p~7474F-I?
31 o42461~E'Og o34RA]|E*OE *117~60E'03 *~qPT46E'I5 *4~q7~EF-oq *~3q2n7F'~6 .|~7040F-~1 *¢%4~7~F-13 *?!~?F-~8
32 *16~3~1F-08 .183445~-0~ .744436F*04 ,]'?~R3E'I$ *|~0440F'Oq *P~O4akE-06 .~+~64F-Z3 *~50|1~Fo14 *~8|340F-~
34 ,~3p|?~[*Oq *493~q3E*06 *~RgqN~F'04 *~16q~RE'|? ,99PR?q~-|| ,3500~1[-~7 *?0092~-~5 .1~F'|5 *37~TP~F'IO
~ *HS~892E-|O *2q304gE'O~ *17R~73F-04 ,3~077E-]0 ,~b6?g%r-!| .t367~+[-u7 .1~707B~-~ ,?~)O~F-|~ ,~?TFTF-II

37 *|1431~F-10 .648006E-07 .6~76S~!-05 *~74767~-;0 .)R3414r-|P ,l~7n~E-~l~ .p~5~5|F-2~ ,~791~1E-18 *%~q4~?~*t~

30 *)47877[-II .|6|536E-07 ,~4?~?tF*05 ,1~0S65E-~| .I|Rk|7F-|3 ,~73~?[-~q .kTqO~?F-3~ ,~7|?7F-20 o~?b6~)F-I~

40 .52~4~8E-|2 oTq?744E*O~ *144970~-0S *|~9?3|K-~2 .~9~=44;-]4 *~9~RPmE-]O .~1~603Fo33 ,lOI4n~F-20 .~|4~)Fol4
61 ,18~3|~(-|2 .37|094E-0~ .A6|O33F*06 ,I',47~?E*~) .724~c-]5 *+$616bE-10 *|4%913F*3~ *1747P1F-~I .1~01~F-14
4] *~474E*13 .9~4664E'0~ *PqS?43F-06 *~39F*~5 o42047qr-|6 ,447~nE-]| .?A~R!|F~7 *]7800qF-2~ .fl~77~F-t~

47 .307910E-|5 ,481~O|E-In *33fl?4pFo07 *EtOII2E-29 ,1~2;n6~*18 *#1~)~5E-13 .O+~O00FO01 ,P~!7~+F*~7 .I~I~TF-I~

40 ,|0361~E-15 ,p??q23(-I~ .lq~g?r-()7 ,E~0134~-~0 .?76n~4~-lq *?05~qSE-13 ,O00000FO0| *316~I~F-~R ,]~74F-1~

~rl o11~377E-16 .4970|6E')1 .k?A~?6F*OB *?~4p~qE-~2 *L~71PqF-~0 *P~3|~7E-14 .000000F~0] *~SPT~F-30 *7173~g-;1

~TU 1.00 |*~0 1.00 ~*~,0 ;~,00 ?.~+0 "~*00 ~.00 "~,~0

I *26+39qE+00 *l?4339EOAfl .14PS3~EO~O *]e~4~lfO~O ,~90n~4rO00 .~3~3+4[++0 ,4;49~+000 ,~017~F000 ,PT+834FO00

*?S49q+E*Ol *119747E000 ,108z33ro00 ,E40)~2E-'! *I+4~?IFO00 *l?qOalEOOO .7%~43~*01 *~?k~TFOOO .p~+++POOO
+ *19338)[-01 *804796E001 ,RIO~9?F-OI *407BO6E't~ .lf16~4?EO00 .12317~E000 *l?ShTPE-02 *)15i+~FflO0 .1487?~F080
4 .46?6p+E-O? *$3|S69E'01 ,600~41E*01 ,+~lm2+E'~3 ,5560790-01 ,7933qlE'+l ,?mTq13F*04 ,4mJ+m~F-Ol .84~+4~F-01

,230S?IE-03 ,722681E-0! o3?0??~E-01 .I+~?P3E*+5 ,126Oq?F-OI .~SRB14E-O| ,7~07~3F-08 .kll~e+F-++ .?i5710F-+l

8 *101§34E'04 *8q4Q4IE-OP °16~64F'01 *+t0663E'~6 ,24ql~3F'02 *q32?P~('+? *]mlAPqg-)l *m+3~F-+3 ,4~09000-n2
9 .+@0140E-05 *Sfi04lO('O~ ,IIR3OAP-01 ,|~m~QSE-Oq *|05q47F-O+ ,~179m0[-i+2 ,I.3950F°13 .1771P~F°03 ,|~74~PF-~2
]A .41062]E'06 *3484q|('0? *R+g420E'02 ,t;+I6PE'11 *4407~9F'03 ,+?7t?3e-+a ,000000F001 ,+Bt+TmF-04 ,7738q+[-n]

13 *297TN4['OR *~O~O70E*++ *2gPISOF'O2 *]o~|SIE'IS *2R~73¥'04 *40171~['03 *~O0000FOOX *P77Nt~F-Ob *4flQR49F-n4

|4 *S6~SffR[-O9 °492~20[-00 °?03q36F*02 *l+~q~flE-16 *111403F*04 *?0~7;?E'03 *DO0000FO01 *~PO+?mF*Oh *|773~tF-q4
18 *100200E-09 ,298;RSE'O] *141qM6E'02 *~+O~4gE'IR *4?qgqRf'OS *|044~[*~3 ,0000000001 *~40q44F-+7 *EtB~61P-n5
16 .1900~4(-|0 .180011E*03 ,gB4112E-03 .+O0000EOCI .163"08F005 ,~5~46E*~4 *OnOOOOgOO) ,I]O~7~F-O? ,~40++?F-n~
IT .]SP44SE*II .108~32E'03 *680617F003 *+POOOOEO~| ,61~7r*06 ,7622~qE-04 ,000000[001 ,+~gq?OF-~B ,~7707F-~6
18 .~S36H3(-12 ,648~|7E-04 .46q456F-03 ,~O000EO~l ,731~72[-06 ,1~9844E-04 *O+O000FO0| *7~4747F-~q .~OP~F-n6
lq ,ll+3+?(-l~ ,38?340[-04 ,3279Q4[-03 ,~+0000[001 ,SST~)3F-O? ,#3R3AqE-+5 ,000000[00] ,t~7+qNF-09 ,168667F*~
~( *21PBTTE-13 .230653[*06 *22|6gGF-03 *r+OOOOEO01 ,]|5~IOFo07 *N|180~(*O5 *O00000FO+| *3mO~E-l+ ,~?NI13F-~7
21 *380S1~(-14 ,136961(-n4 ,1GI833E*03 *CnOOOOEO~| .lJ~i~(-07 ,151363E-+~ *O~O000FOO| *"%50q~F-11 .~10~4~-n7
22 .670~10~-I~ .811109E-05 ,10~76~F-03 ,C~0000(0~1 *4|?m6SF-08 *7305~0E-06 .~O0000FO01 .100127F-11 *44*+l?F-~+
23 *llqSle[-IS ,6?9148[-0S ,707733E-04 *O~O000EOOI *|S0~69F-08 .3S0717('"6 *000000~00! .~|R~PnF-|2 .IS+~OPF-o8
?4 ,OlnlOqE+O| ,282376(-0~ ,48179~F-04 .¢~0000[00! ,530~7F-09 .)67510E-06 .o+O00nFOO] ,~11~71F-13 .~4]lgF-~g
25 *004000[fl01 ,!66030(-0~ ,32¥400F-04 *CO0000FO0| *192~78P-09 *?q62~GE-07 *O00000gOOl ,1~67~F-13 0]677~--L19
20 *00~000[001 *974PI3E'06 .22~100[-04 .0n0000(001 ,6ROI|~F-IO ,~767qlE-07 *000000F00| .+~Oq++F-l+ *~4200F-10
~? *OOnO00((01 *S?0448~-06 ,l~O+~|F'04 ,GO0000EO~I *23q~44F-lO .I??S+?('~7 *~0000~E00| *~9301+F-15 .1~I~0?F-10
2e ,OlO00fl[nOl ,333576(-00 01017|6[-04 ,000000[00] .840349f-11 ,R331~3E*~+ *OO000flEOOl ,l~8]nl+-|S *%Q246+E-+I
29 *OIIO00EOOI *lg646?E'O~ ,~86767F-OS .~0000~0~1 *793~t7~-11 ,38q44qE-qR ,000000F00| *3q~t4~F-16 *lq|~F-tl
3n *OOflO00[OO| .113?36[-fl6 ,4~30S6F-0§ *GOOAOOEOC| *|OIqN]P-ll .1813~?E*0~ *O00000FQOI ,815~+PF-l? ,~176k?F-+2
31 *000000(~0| *688~46[-07 *O00000FO01 *000000[0~| .3S2~44F-12 *~41~4E'~9 *O00000FOOI *]~87~qF-|7 ,Iq7704F-t~
3? ,OO40~OEq01 *000000[00! *O00000EO01 .000000E001 *12|q83F-12 ,~89~77E-~9 *+NOOOOFOOi .+*4~mmF-18 *~q+i+F-++
33 *0000~9[+0| *O00flO0[O0! *O00000EO01 *CnO000[O01 .417~?~*|3 .17903~(-09 *000000;001 *70107qF-lq *!qq34~F-l+
34 *004000(a0| *000000(00| *000000000| .CnOooOE001 *1400710-13 .++$21)E-|0 .0+00001001 .I+Iq+++-19 *80~P+6P-)~
3S *UOOOUfl(nO| *000000(00| *O0000N(O0| ,+nOOOO[U+| *4RG~q4p*]4 *+78196(-10 *O000AOEO01 .PP+?TPF-~+ .IqT+~1r-14

36 oOOPOAqE~O| *O00000EO0| *O00000POOI ,~NGoOOEO~ ] *|~S~PqF-14 ,17~8~(-I+ .DNOOOGE~OI *~TPIN~F-2t .~1%05~F-I~

37 *000000[001 *000000(00) *OOOO00+OOl .+AO000EO+l *fi+OPS~+-15 .Te?71RE-II .nnflOOOEOOI .113q*~F-+t .I+IP++E*15
38 *00000~¢~01 *000000(001 ,OOO~OOE@OI ,~nO0~O[Olll *|Rg~F-15 *~%RO71E-I] *~+O00OFO01 * ~ ? ~ E - P 2 o~9~OA?F-I~
 A convolutionmodel of rock bed thermal storage units 445

Table 1. (Contd)

3g oO00000g~OI ,000000(001 ,000000[001 ,+O0000ffOt'l *6~66qnF*16 *1623q~[011 .flAOOOnFO01 ,~oOnOnFO01 ,la~%g~F-16

40 ,O0~OnO(OOI ,O00000(ooI ,O00nOO[O01 ,~nOflO0[OO| .213~9S~016 .7343?7(-12 ,no0000£O01 ,AOOnnnFO01 .~O~F-~7
41 *0000flo[:~0| *O00000E00l ,000U00FOOI ,(n~0OOEO+q ,O00~0F00l ,~313g~[-I? .0~000nFO01 ,q)00n~n~O0l .17|5~7~-17
47 .UO0000[~Ol .O00000[OOl .000000[001 *LQOnO0[OCl .OOOQOOFO0| ,1492~?(*I~ .ono000[O01 ,O00OOOFOOI .~Fq47F-18
43 .OOOO~EnO| ,O0000O[OOl .O0000OFOOl .(nOOOO£O~'l *000000¢001 .670314E-13 ,~nOOOn~O0! .nOO~O~FOO) ,1~8~-18
44 *OOO00O[~O| .000000(001 .O00qOOFOO] .~cO000[OQI .O00fl~O¢O0| .QO00~O[OOI .OOOANO[OOI .+O000OFO01 .4~IOOIF*IQ
+S ,000OO+[+OI ,000000[OOl .ooo00m[OOl .PnoooO[O~! .000oon~OOl .nOOOoO[OOI ,noooooFool ,O00OOOFOOI .I~SI~&F-IQ
40 *OOOO,+e(l,Ol *OOO000EOO} .O00OOOEO0| *C+OOOOEOnl *O00nO0 001 ,nooonnEOql .~oo0oo[o01 ,+~OOOnFOO) .OO000OPOOI
47 .oooO+O(~O] ,000000(001 .OO0000FOOI .CnOOO0[O+| ,O0000OFO0| .+O00oo[OOl .oPO00o¢oOl .i)NOflfl~FO01 .OOflOO~F~l)l
4P .OO~O0~(~Ol .O00000(OOl .O00OO0~ol ,¢~ODOOFO,I .O00~OOrOOl ,~OOO~o[O~l .O00000[OOl *OflOO~nFO01 ,O00OOOFO~l
+9 .O0~OCO[~O] ,000000[00! ,OOOOOOFOOI ,+"0~00~0"II *O00flO0~O01 * 0 0 0 0 9 0 [ 0 ~I *OOO000FO01 .+oononFo01 .OQO000FOA!
~O .O0~OIiA[~O] .000000(061 .000000[001 .~¢'O00O[O ~ ] .OOO~OOO01 .flO00O~[O01 .ONOOOAFO0| *OAOOOOFO0| .~OOOOO~O~l
CR ~.I?R 0.~8~ 0.67S 0, 0+~ 0.337 ~. 4 S 0 O,OIl O.)gq 0*303

GSUM |*OOflO~OOOO OsqqqOQQQO94 OeQqO~Q~PR5 ]*C~OflOOOOlO r*qqqqq9~9~] ]*O000OO00NO |*O00000OO00 0.OOQqq~qq~g I*NODO000O~I

true in the context of the TRNSYS program which is times. If flow is from top to bottom, the elements in the
oriented toward differential equation solving. The com- selected row are updated left-to-right; reverse indexing is
plications encountered include allowing for flow reversal, done for bottom to top flow. The last calculated tem-
time-step size smaller than the response factor base and perature is transferred to the appropriate TRNSYS out-
multiple calls within a time step. An approach to these put element.
difficulties is discussed below. There are several conditions for which the above
The TRNSYS subroutine employs two basic arrays, T sequence must be modified. On the first call in the
for the fluid temperature and TB for the bed tem- simulation, all calculations are skipped to get the routine
peratures. T is 2-dimensional with each column in step with the TRNSYS time variable. Also, on calls
representing a bed segment and each row representing a for which time is not a multiple of the response factor
time in increments of 8. The first and last columns are time base 8, inlet temperatures are averaged and control
used to store inlet and outlet fluid temperatures. The returned to the calling routine, skipping the rock bed
column indexing convention is such that the ,/th column updating. Thus, the TRNSYS time step can be anything
of T always contain~ the temperature of the fluid leaving smaller than 8 and energy will not be lost; however, the
the segment whose bed temperature is stored in TB]. bed outlet temperature remains unchanged over several
Row indexing is done by a "wrap-around" pointer whose TRNSYS time steps.
value is decremented when time advances by 8; this The most perplexing problem derives from the
allows programming of eqn (6) without moving data in TRNSYS policy of multiple calls in a time step. Consider
the T array. the case when the first call in a time step indicates a flow
Before the T array is recalculated on a call, a flow reversal. This precipitates the recalculation of TB, and
check is made. On a call in which the flow is detected to reinitialization of the T array. Now if on a subsequent
be the reverse of the last calculation, the bed tem- call in the same time step the flow shows to be back in
perature vector TB is updated using eqn (8) and the T the original direction, the data from the previous time
values from the previous calculations. These TB values step, needed for properly proceeding, has been lost. This
are then loaded into all elements of the corresponding is a likely circumstance often occurring as the TRNSYS
columns of the T array, thus properly initializing the bed simulator attempts to converge the entire system. To
for the new flow direction. deal with this, the entire T array and current pointer
After all conditions are checked and adjustments made values are copied into auxiliary variables, so that T can
for reversal, the row pointer is decremented and the fluid be restored if required. This greatly complicates the
outlet temperatures are calculated by applying eqn (6) N logic, increases running time due to data transfer, and

Table 2. Rock bed parameters for test cases

2
Frontal Area, A 2.5 m

Length , L 4.0 m

Effective Density, 0 1680 kg/m 2

Volumetric Heat Transfer Coefficient, h v 5432 kJ/hr-m 3-c

Rock S p e c i f i c Heat, Cb 0.837 k J / k g -C

F l u i d S p e c i f i c Heat, C f 1.012 kJ/kg-C
Mass Flow Rate, mf 1512 kg/hr
NTU 35.5
Time C o n s t a n t , ¢ C 9.19 hr
446 E. F. SOWELLand R. L. Cu~Y

doubles the required storage. However, it appears to be N (hrs)

the only way to deal with this problem in the TRNSYS t Convolution
2 Hughes/Mumma-
15
15
0.50
005
context. Marvin
3 Hughes-Klein- 15 0.05
Close
MODELCOMPARISONS I00 4 Hughes/Mumma- 150 0,00S
Marvin
Twenty-two cases were run in order to test and com-
pare the Convolution model with various finite difference t.) 8 0
models. The study was aimed at assessing accuracy and g
efficiency under expected usage patterns. To this end ~ ~o
each run represented a 72-hr simulation of the bed E
studies by Mumma and described in Table 2, with the
charging and flushing sequence described in Fig. 3 which
is somewhat typical of a solar heating system. 20
Figure 4 summarizes performance of the models in
terms of outlet temperature profiles after flow reversal.
The cases selected for plotting here were for the com-
binations of bed divisions and time step size required for Time, fir

reasonable accuracy of each model. In all cases, 15 Fig. 4. Response of finite-differencerock bed models to reverse
divisions seemed to adequately represent the bed. The flow test sequence.
Hughes/Mumma-Marvin and Hughes-Klein-Close
models both require a time step of about 0.05 hr with this the Hughes/Mumma-Marvin model is compared with the
segment size. It is seen, however, that the Convolution Hughes-Klein-Close model, with various segmentation
model does quite well with 0.5 hr time step. For com- and time step. First it is observed that with a given
parison, the most accurate solution attempted, using the segmentation, e.g. 15, the Hughes/Mumma-Marvin
Hughes/Mumma-Marvin model with N = 150 and t = model will predict above or below the correct profile
0.005, is also shown. It will be observed that any of the depending on the time step. Second it is observed that
models are within 2-4°C of this solution throughout the when segmentation is too coarse, the predicted profile is
interval. The close agreement is highly dependent upon flatter than it should be; this is an artificial "smearing" of

! ; TO
choice of number of segments and time step for the finite the temperature fronts introduced by the model. That
difference models. This is demonstrated in Fig. 5, where Hughes was able to show good long term simulation
results with such a model is probably an indication that
his system was insensitive to the discharge temperature.
Current results suggest that the finite difference models
should be used with caution, especially in the study of
temperature-sensitive components such as heat pumps
and control elements.
20 , i i J , i i • i i i I I i I I I I I I I i I )
12 24 Table 3 gives an indication of model relative efficien-
Time, hr cies. The numbers presented are estimates of execution
time, derived by subtracting the machine time for a
1512 I zero-hour simulation run from that for the 72-hr simula-
Topf l o w tion. To the extent possible, printing was equalized for
kg/hr I
, l l t l l l I I 3 I I I ] I At
00~ , , , , ~ i 12 N (hrs)
24
1 Hughes/Murnma- 150 0005
Marvin
2 HughestMumma- 15 0005
Marvin
20 3 Hughes/Mumma- 15 0050
Bottom temperature ioo Marvin
4 Hughes/Mumrna- 15 0500
/-3 Marvin
C ~ //-5 5 Hughes- 50 0010
80 ~ . _ ~ . ~ / / ~ - I Klein-
~. ~ ~ r 2 Close
O ~ ~ , ~ , ~ , = , , , = ~ , ~ , , = , , , t = I . f ~-~/~../-4 6 Hughes- 5 0.050

, , i i i , i , i i i , i i i I I I I I I I i 1 , t i
0 12 24 15 20 25 30

Fig. 3. Flow-temperature sequence for reverse flow testing of Time, hr

rock bed models. Fig. 5. Performance of three rock bed models with flow reversal.
 A convolution model of rock bed thermal storage units 447

Table 3. Rock bed model run-time comparisons*

Time Step (Hrs)
S e g m e n t s Model**
0.005 0.01 0.05 0.1 0.5

51 44 35
HKC 66 48 36
C

290 51 43 35
15 HKC 603 89 61 33
C 70 60

HMM 243 78
SO HKC 71S
C

HMM 876 45S

150 HKC
C

* Times are in seconds for the execution step, CDC31S0.

** HMM = Hughes/Mumma-Marvin; HKC = H u g h e s - K l e i n - C l o s e ;
C = C o n v o l u t i o n model

each model. From these run times, one can draw ten- of energy, this method will be preferred over the finite
tative conclusions about the model efficiencies. For low difference methods.
accuracy, i.e. small numbers of segments and long step
sizes, the Hughes/Mumma-Marvin and Hughes-Klein-
Close models are about equivalent, and more efficient Acknowledgements--This work has been supported by the Solar
than the current Convolution implementation. As the Heating and Cooling Research and Development Branch, Office
of Conservation and Solar Applications, U.S. Dept. of Energy.
accuracy requirement goes up, however, the convolution Programming has been done by Donald K. Bell, Sharon Kouns
model is superior; this is so since its accuracy is depen- and Nina Hostetler. Computations were carried out with the
dent upon the pre-determined response factors, rather facilities of California State University, Fullerton, Computer
than on reduction of time step size during the simulation. Center.
It is notable that even if other parts of a system simula-
tion necessitate reduction of step size, the Convolution
NOMENCLATURE
model solution time goes up only slightly. It is also true
A bed frontal area, m2
that the Convolution model will not lose accuracy under b see eqn (9)
these circumstances. Cb bed solid particle specific beat, kJ/kg-°C
C! fluidspecific beat, IO/kr°C
CR common ratio of successive terms for large t
CONCLUSIONS h~ volumetric heat transfer coefficient, kJ/hr-m3-°C
/ dummy index for convolution procedure
A method has been presented whereby the outlet j subscript representing bed segment
temperature from a rock bed storage unit is calculated k subscript representing time increments of
exactly for piecewise linear variation of inlet tem- L bed length, m
perature, without integration of finite difference equa- rh fluidmass flow rate, kg/m
N number of bed segments
tions. This is done by a discrete convolution process NTU number of transfer units (see eqn 3)
between historical values of the inlet temperature and ri response factor: fraction of input pulse which appears at
the curve which represents bed response to a unit trian- outlet i8 later
gular pulse. An efficient method of pre-calculating this modified response factor (see eqn 7)
response at discrete points has also been presented. The T fluidtemperature, °C
Tb bed solid particle temperature, °C
Convolution method will find application in long-term t time, hr
simulation of systems involving packed beds as well as U(t) unit step function at time t
other devices represented by the same equations. To this X bed position coordinate, m
end, an implementation as a TRNSYS subroutine has 8 response factor time base, hr (see Fig. I)
dummy variable of integration
been discussed. Results presented suggest that in simu- Pe bed effective density, kg/m3
lations which require economical, accurate prediction of 7 dimensionless time, (t + 8)/~',.
bed exit temperatures, as opposed to simple conservation r~ bed time constant hr (see eqn 4)
448 E.F. SOWELLand R. L. CURRY

Appendix (where different from the text) Substituting into (AI)-(A4) yields
c number of transfer units (NTU)
[ input temperature profile, C Ou
--= cv, (A7)
Ii modified Bessers function of order 1 Ox
t dimensionless time (tire) Ov
U modified fluid temperature, C = cu, (A8)
Ux partial derivative of U with respect to X
v modified bed temperature, C u(x, 0) = 0, (A9)
x dimensionless distance (x/L)
y, z, r dummy variables of integration u(O, t) = [(t) exp (ct). {AIO)

Now (A7) is differentiated with respect to t and this result is

REFERENCES combined with (A8) to yield
1. T. E. W. Schumann, Heat transfer: a liquid flowing through a
porous prism. J. Franklin lnstit. 208, 405-416 (1929). 02U
- cZu. (A11)
2. A. Anzelius, Uber erwarming vermittels durchstromender Ot c~x
medien. Z. Angew Math. Mech. 6, 291-294 (1926).
3. M. Riaz, Analytical solutions for single and two-phase Assuming that u is sufficiently smooth, (A11) may be integrated
models of packed-bed thermal storage system. J. Heat from 0 to t to obtain
Transfer, Trans. ASME 99C, 489--492 (1977).
4. C. C. Furnas, Heat transfer from a gas stream to a bed of
broken solids. Trans. Am. Inst. Chem. Engr. 24, 142-169
(1929).
ux(x, t) - ux(x, 0) = c 2
fo' u(x, y) dy. (A 12)

5. W. M. Kays and A. L. London. Compact Heat Exchangers, For u sufficiently smooth, (A9) implies that
Chap. 3. McGraw-Hill, New York (1955).
6. J. A. Duffle and W. A. Beckman, Solar Energy Thermal ux(x,O) =0. (Al3)
Processes. Wiley, New York (1974).
7. S. A. Mumma and W. C. Marvin, A method of simulating the Equation (AI3) is used in (Al2) to obtain
performance of a pebble bed thermal energy storage and
recovery system. ASME paper 76-HT-73, presented at
ASME-AICHE Heat Transfer Conf., St. Louis, Aug. 1976.
8. TRNSYS--A transient simulation program. Sol Energy Lab,
ux(x, t) = c z
I/ J
u{x, y) dy. (A141

Univ. of Wisconsin, Engr. Expt. Station Rep. 38, March Equation (AI4) is integrated from 0 to x to obtain
1975, as revised 10175.
9. P. J. Hughes, S. Klein and D. J. Close, Packed bed thermal
storage models for solar air heating and cooling systems. J.
Heat Transfer, ASME Trans. C98, 336-338 (1976).
u(x,t)-u(O,t)=c
2
fo fo
x t
u(z,y)dydz. (A15)

10. P. J. Hughes, The design and predicted performance of Equation (AI0) is used in (AI5) to obtain
Arlington House. M.S. Thesis, Univ. of Wisconsin-Madison,
Dept. Mechanical Engineering (1975).
u(x,t)=/(t)exp(ct)+c 2
II. G. F. yon Fuchs, A rock bed computer model. Presented at
the Solar Energy Storage Options Conf., Trinity Univ., San fo fo ' u(z,y)dydz. (AI6)

Antonio, TX, 18--22 March 1979. The unique solution to (AI6) may be expressed in terms of the
12. W. T. Thompson, Laplace Trans[ormation, pp. 38--39. Pren-
Neumann series as
tice-Hall, Englewood Cliffs, N.J. (1960).
13. D.G. Stephenson and G. P. Mitalas, Cooling load calculation C2n+2 /'x t
by thermal response factor method. ASHRAE Trans. 73,
111.1.1-1.7, Pt. I (i%7).
a(x't)=t")exp(ct)+ o Jo fo (x-z).
14. T. Kusuda, Thermal response factors for multilayer struc- × (t - y)"[(y) exp (cy) dy dz. (AI7)
tures of various heat conduction systems. ASHRAE Trans.
75,246--271, Pt. I (1%9). The integration in (AI7) is performed with respect to z to obtain
15. B. A. Peary, A note on response factors and conduction
transfer functions. Paper presented at ASHRAE Semi- x 2n+2 n+l
+ C X
annual Meeting, Atlanta, 1968, AT-78-6 No. 3; also in u(x,t)=/(t)exp(ct) ~, t ~ v .
n=o[n.) (nt I!
ASHRAE Trans. 84, Pt. I (1978).

APPENDIX
x
fo' (t - y)"/(y) exp (cy) dy. (AI81

Equations (1) and (2) can be rewritten for convenience as

For x and t > 0 a substitution is made in (AI8):
OT = c(Tb - T), (AI)
Ox 2cxlt2(t _ y),2 = r, (AI9)
and
0Tb = c(T - Tb), (A2) 4c2x(t - y) = ?. (A20)
at
T(0, t) = [(t), (A3) The result is
T(x, 0) = 0, (A4)

and it is assumed that f(t) is continuous and /(0)=0. The a,x. ,) :(,) oxp (ct) + .2 ,. ,.. l) fo
solution is facilitated by a change of variables defined by
×[ t r2 2 exp ct r2 r2 " r d r (A21)
T = u(x, t) exp (- cx - ct), (A5)

Tb = v(x, t) exp (- cx - ct). (A6) Following common notation we let It(Z) denote the modified
 A convolution model of rock bed thermal storage units 449

Bessers function of order one, where At the end of the bed, where x = l, (A24) simplifies to

T(1,t)=exp(-c){f(t)+ ff"°~:/(t- ~'c2)

I,(Z) = \~-] ~"=o( n ~ I)' (A22)
×exp(~c )ll(r)dr}. (A25)
The use of I~ allows (A21) to be rewritten as
For the problem at hand, the function / is a triangular unit pulse
u(x,t) =exp(ct){f(t) +fo~"~'"~,(t - ~--~c2) centered at t = 8.
r2 The reader will observe that the problem has now been
xexp(~xc)l,(r)dr }. (A23) reduced to one of integration of a known function, and yields a
solution thai provides the necessary factors for the convolution
process. This is in contrast to the finite difference models which
(A23) is multiplied by exp ( - c x - ct) to obtain involve integration of unknown functions during the simulation.
The integration in (A25) is readily done using the Romberg
T(x,t) =exp(- cx){/(t) +fo2'~x'"'2[(t - ~--~c2) scheme. It has been programmed for evaluation of the response
factors necessary as input parameters for the TRNSYS subrou-
×exp(~c )l,(r) dr}. (A24) tine.

SE Vol. 24, No. 5--C