You are on page 1of 9

International Journal of Food Science and Technology (1990) 25, 148-156

Lethality calculation for thermal processes with


different heating and cooling rates

Q . T. PHAM, Meat Industry Research Institute of New Zealand, Hamilton,


New Zealand

Summary
In many thermal processes, the time constants fh and fc for the heating and cooling
periods differ. This can cause difficultiesin calculating the process lethality. This paper
presents a simple method for taking into account this change in the time constant. The
author’s formula method is tested against simulated data from finite-difference
computations, both forfh=f, and forfh#fc, and found to be at least as accurate as the
best previous formula method.

Keywords
Error minimization, heatinglcooling time-constant inequality, sterilization, thermal
processing.

Introduction
Canned foods are sterilized by heating at a constant temperature for a certain period,
then cooling in another medium. Sterilization (and nutrient destruction) takes place
both during heating and cooling, due to the thermal inertia of the can and its contents.
The lethality for a given combination of product, can and thermal process is
conventionally found by processing a can of product in the laboratory and monitoring
the can thermal centre temperature. From the time-temperature curve, the sterilizing
effect can be calculated using either direct integration or some short-cut method. The
latter generally consists of sets of tables relating some features of the temperature
curve, usually g (the residual difference between heating medium temperature and
can centre temperature) and j (the lag factor), to the sterilizing effect in an approximate
manner. These tables are referred to, somewhat misleadingly, as ‘formula methods’
although usually no formula is given. An exception is the method of Pham (1987), which
does consist of algebraic formulae.
In some of the formula methods, the sterilizing values of the heating and cooling
periods are calculated separately (Ball, 1923; Ball & Olson, 1957; Hayakawa, 1970;
Griffin et al., 1971; Steele & Board, 1979); in others, only the total sterilizing value is
given (Stumbo, 1953,1973; Stumbo & Longley, 1966;Jenetal., 1971; Pham, 1987). The
latter are more convenient to use but cannot be applied to situations where measured
values of the time constants fh and fc differ (Gillespy, 1951; Hayakawa, 1977). This

Author’s address: Meat Industry Research Institute of New Zealand, PO Box 617, Hamilton, New
Zealand.
Thermal process lethality calculation 149

paper presents a modification that overcomes this drawback. The formula method of
Pham (1987) will be tested with and without this modification, using the methodology of
Smith & Tung (1982).

Theory
When a product is suddenly exposed to a different ambient temperature, its
temperature will, after an intitial period, follow a logarithmic relationship with time.
For heating:

and for cooling:

log10 ( jc;-T;c) ) = -Ifc

It is usually assumed that fh and fc have the same value, f, for a given product and
container. Methods presented elsewhere (e.g. Pham, 1987) will then give the value of
W= Ulf, from which the sterilizing value U for the whole process can be calculated.
In practice however, measured values of fc andfh often differ. For steam heating and
water cooling it has been found that fc=1.3 fh. The reasons for this have been discussed
by Olson & Jackson (1942), Gillespy (1951), Hayakawa (1977) and others. They
include: temperature measurement errors; heating or cooling ended before logarithmic
period is reached; experimenter’s subjectivity; finite surface resistance to heat transfer;
changes in product properties; and convection within product.
When fh # fc, U can be calculated from:
u = w h f h + wcfc (2)
where w h and Wc are the dimensionless sterilizing values for the heating and cooling
periods.
Formulae will now be presented for calculating w h and W,.
High sterilizing value range: for w + Wc > 1, Pham (1987) gave:
h

w h = - loglo ( g / z ) - 0.613 + (0.074 - 0.102/jh)N1 (3)


Wc = - 0.04 + 0.088 jc + 0.177 jc N 2 + 0.177 jc N 2 (4)
Since wh and Wcare calculated separately, they can be inserted into equation (2) to
give U.
Low sterilizing value range: for 0.04 < W < 1.5, Pham (1987) gave:
glz = a + bjc (5)
where

loglo a = - W + Al + A2e -2.7 V% (6)


-2.1 m
loglo b = - W + B1 + B2e (7)
150 Q. T, Pham

Ni
A1 = - 0.71 - 0.41 - e-0.5s/N2
N2
+
A2 = 2.14 N$ 0.60 N$/N1 - 0.26 N: - 1.24 N1 + 1.02 (9)
B1 = m)
0.31 + 0.55 + 0.61 fii - 1.86
f l 2 (10)
B2 = (0.91 Ni - 3.18 - 0.755) m
N1 )- 1.38 Nt + 2.55 N1 + 1.52 (11)
Equations (5) to (11) enable g to be calculated directly from W (=Wh+Wc) and the
reverse operation to be done by trial and error. However, Wh and Wc are not given
separately. In the range W = 1 .O to 1.5, either the high sterilizing value equations or the
low sterilizing value equations can be used. Justification for this is given in Appendix 1.
Equation (4) suggests that a similar equation for Wcmay hold for the low sterilizing
value range, as long as Wc is replaced by WclOg/" to account for the fact that at the
beginning of cooling, the product is at temperature Tg(= Th-g) and not Th (as assumed
for equation [4]).
Temperature curves were generated by computer using basic heat conduction
equations, as described by Pham (1987). The same assumptions were made as in that
paper. These curves were then used for lethality calculation, as proposed by Jen et al.
(1971) and Stumbo (1973). The values of WClOg'*thus calculated were then related toj,
N1 and N2 using a forward stepwise multiple-linear regression. (A forward stepwise
multiple-linear regression is one where each explanatory variable is tried in the gression
equation and retained only if its presence reduces the error sum-of-square by a given
value, normally four mean-squared errors.)
The equation obtained was:
W, = (- 0.08 + 0.12~'+ 0.13jN2 + 0.19N2) (12)
with variance accounted for as R2 = 0.96. The similarity between equations (12) and (4)
is apparent (in equation [4], g = 0) but it must be kept in mind that they apply to
different W ranges. U can now be calculated by rewriting equation (2) as:
u = (W - Wc)fh + wcfc (13)
with W being given by equations (5) to (11) and Wc by equation (12).

Error analysis
In the high sterilizing value range, the error in equations (3) and (4) was given by
Pham (1987) as +3%. In the low sterilizing value range, the extra error introduced by
equation (12) can be calculated as follows.
Over the rangej > 0.6, W = 0.04 to 1.5, N1 = 0.1 to 1, N2/N1= 0.5 to 1 which covers
all practical situations (Pham, 1987), the error range in Wc/W(more precisely, the 95%
+
confidence interval) as calculated by equation (12) is 0.09 (Fig. 1). To calculate the
resulting error in U , equation (13) is differentiated:

Equation (14) predicts that, because 6(WJW) is about 0.1 at most, the extra
uncertainty in U introduced by using equations (12) and (13) is less than 1/10 the relative
difference betweenfh andf,. For example, iffh and fc differ by lo%, equations (12) and
(13) increase the uncertainty in the calculated value of U by less than k1%.This is
Thermal process lethality calculation 151

-12 -8 -4 0 4 8 12

% error

Figure 1. Distribution of errors in W J W with W, calculated by equation (12).

acceptable in view of normal experimental uncertainties of +lo% to k30%, even under


well-controlled conditions (Lenz & Lund, 1977). If there is a marked difference
betweenfh andf, (say 20% or more), uncontrolled factors must have a large influence
and the experimental uncertainty would very likely be larger than Lenz & Lund’s
figures.
In all equations of this paper, the sterilizing value has been measured in terms of U
or W. The reader is reminded that F-values are easily calculated from U by:
F = u 1O(Th - Tr)/z (15)

Testing of method
A finite-difference program was used to stimulate the variation in centre temperature of
a can of food. Pure conduction was assumed to take place, without the complication of
convection. The can’s surface temperature was assumed to come instantaneously to the
temperature of the heating or cooling medium (infinite heat transfer coefficient). These
are standard assumptions used in thermal processing calculations (Pham, 1987). The
finite-difference calculation used an 8x8 grid for a quarter can and the Crank-Nicolson
(central difference) stepping method.
To use formula methods to calculate sterilizing values, the parametersf and j must
be found from the simulated curves. In particular, an asymptote must be fitted to the
linear portion of the log( Th- T) or log( T- T,) vs. time curve, so that the slope llfcan be
found (equation [l]).Since there is a gradual merging of the linear and non-linear
sections of the log temperature-time curve, a decision must be made as to when the
linear portion actually occurs. This decision must be made in a way that realistically
simulates the reasoning method of a human operator.
A version of a method introduced by Smith & Tung ( 1 9 8 2 ) was used. A linear
regression was carried out on log(residua1temperature) vs. time and the regression time
values were compared with the measured time values. If the maximum error in time was
152 Q. T. Pham

-10 -5 5 10 15

Z error

Figure 2. Distribution of errors in W for Pham's (1987)method, fh = fc.

larger than 0.1% of the time range, earlier parts of the data were discarded and the
procedure repeated. When errors were all less than 0.1% ,j andfwere calculated from
the regression line.
To ensure consistency with previous tests (Smith & Tung, 1982), the same range of
parameters was used in the present tests: z = 10°C; T h = 120°C; g= 5, 0.5 and 0.05"C;
Ti=20, 45 and 70°C; Tc=20"C; UD=0.3, 0.5, 0.7, 1.0, 1.4, 2.0 and 3.3. In
dimensionless terms, N1= 0.1,O. 13and 0.2; N2 = 0.1; g / z = 0.5,0.05 and 0.005; and L/D
as above.

Case of constant f-value


Over the range of parameters investigated, the mean error in U and W was 2.1%,
standard deviation 1.8% and range -5.7% (overprediction) to +11.8% (underpredic-
tion). The error histogram (Fig. 2) shows a typical bell-shaped distribution.
Figure 3 shows how the error varied with UD ratio at different values of g / z . The
magnitude of the errors did not vary appreciably with the value of A$ = z/(Th - Ti);in
other words, the value of the initial product temperature did not affect the accuracy of
the formula, so the errors for different values of this parameter were averaged.
Errors were largest for large g/z, and a maximum occurred at an WD of about 1.0.
Lenz & Lund (1977) showed that processes with large g/z values are inherently
uncertain with regard to their sterilizing effects, with uncertainties of 25-30% being
expected forg/z = 0.5. This is because a large part of the sterilizing effect occurs during
the cooling period, and errors in the timing of this period or in the thermal properties of
the product can lead to large uncertainties in U-values.
Thus, the (relatively large) errors in the formula at g/z = 0.5 are still much smaller
than errors due to physical uncertainties. At g / z = 0.05 or less, which is more common,
errors lay between - 1% (overprediction) and +4% (underprediction), which is very
Thermal process lethality calculation 153

10 ’

I
0
L
I
a, 5.
K

0’

-5
0 1 2 3 4

L/D
Figure 3. Errors in Pham’s (1987) method for different UD ratios. Ti = 20-70°C; T, = 20°C; T,,
= 120°C; z = 10°C.

satisfactory. Uncertainties due to measurement in this region were about 10-15°/0


(Lenz & Lund, 1977).
Compared with other formula methods previously tested by Smith & Tung (1982),
Pham’s (1987) formula had about the same accuracy as Stumbo’s (1973) method and
was more accurate than the methods of Ball (1923), Hayakawa (1970) and Steele &
Board (1979) (Fig. 4).Stumbo’s method also had larger errors near LID = 1and at large

40 .

30 ’
L
P
20’
b\“


10

-10 I
0 1 2 3 4

L/D
Figure 4. Errors in lethality using five ‘formula’ methods for g/z = 0.5. All curves except that for
Pham’s (1987) method were taken from Smith & Tung (1982).
154 Q.T . Pham

g/zvalues. The similarity between Pham’s method and Stumbo’s method is due to the
similarity in their derivation, the major difference being the smoothing algebraic
expressions used for Pham’s method.

Case f h f f,
To simulate cases where fh differs from fc, the finite-difference simulation was run
with the effective thermal diffusivity of the product being changed when going from
heating to cooling.
Simulated temperature curves were generated using the following range of
parameters: N 1 = 0.10r0.2; N2 = 0.l;gIz =0.5,0.05 and0.005;UD = 0.3,1.0and3.3
(a total of 18combinations). First, 18curves were obtained withfh = f,. Then, 18curves
were obtained with fh < f c (by decreasing the thermal diffusivity by 20% on cooling).
Finally, 18 curves were obtained with fh > fc (by increasing the thermal diffusivity by
20% on cooling). Thus, we end up with a set of 18 curves withfh = fc and a set of 36
curves with fh # fc.
Both sets were processed using Pham’s (1987) method without modification; then
the setfh # fc was processed with this paper’s modification to correct for the effect offh
# fc. Error statistics are presented in Table 1.
If no correction was applied for the change inf-value, large errors arose (row 2 of
Table 1). If the effect of this change was corrected for using the modification of this
paper, errors become almost identical to the case of unchangingf-value (rows 1and 3 of
Table 1).Thus, the modification reliably accounted for the effect of a change inf-value.

Table 1. Error statistics for simulated temperature curves withfh = fc and for curves withf, # fc

Error statistics (”/.)


Case Method Mean s.d. Minimum Maximum

fh = fc Pham (1987) 2.3 4.6 -5.7 + 11.4


fh # fc Pham (1987) uncorrected forfh # fc 3.0 7.0 -10.0 +23.0
fh # fc Pham (1987) corrected for fh # fc 2.3 4.6 -6.7 +11.9

Conclusions
For the case fh = f,, Pham’s (1987) formula method has been shown to be at least as
accurate as the best previous method.
For cases wherefh andf, differ, the modification presented in this paper was shown
to lead to negligible extra error, about 1%for a 20% difference betweenfh andf,. The
modification was applied to Pham’s (1987) method, but obviously it can be used
together with any of the methods mentioned in the introduction.
In common with most previous works in the field, a conduction-heating model has
been used to derive the formulae. For convection-heating cases the lag factors tend to
be smaller and errors subsequently less (Smith & Tung, 1982).
Thermal process lethality calculation 155

References
Ball, C.O. (1923). ThermalProcess Timefor Canned 692-698.
Food. Washington, DC. National Research Lenz, M.K. & Lund, D.B. (1977). The lethality-
Council. Bulletin 7-1 (37). Fourier number method: confidence intervals for
Ball, C.O. & Olson, F.C.W. (1957). Sferilization in calculated lethality and mass average retention of
Food Technology. New York: McGraw-Hill. conduction-heatingcanned foods. Journal of Food
Gillespy, T.G. (1951). Estimation of sterilizing Science, 42, 1002-1007.
values of processes as applied to canned foods. Olson, F.C.W. & Jackson, J.M. (1942). Heating
Journal of Science of Food and Agriculture, 2, curves: theory and practical applications.
107-112. Industrial and Engineering Chemistry, 34, 337-
Griffin, R.C., Herndon, D.H. & Ball, C.O. (1971). 341.
Use of computer derived tables to calculate steril- Pham, Q.T. (1987). Calculation of thermal process
izing processes for packaged foods. 3. Application lethality for conduction-heated cans. Journal of
to cooling curves. Food Technology, 25(2), 134- Food Science, 52,967-974.
143. Smith, T. & Tung, M.A. (1982). Comparison of
Hayakawa, K.-I. (1970). Experimental formulas for formula methods for calculating thermal process
accurate estimation of transient temperature of lethaiity. Journal of Food Science, 47,626-630.
food and their application to thermal process Steele, R.J. &Board, P.W. (1979). Thermal process
evaluation. Food Technology, 24(12), 1407-1418. calculations using sterilizing ratios. Journal of
Hayakawa, K . 4 . (1977). Mathematical methods for Food Technology, 14,227-235.
estimating proper thermal processes and their Stumbo, C.R. (1953). Thermobacferiology in Food
computer implementation. Advances in Food Processing. New York: Academic Press.
Research, 23, 75-141. Stumbo, C.R. (1973). Thermobacteriology in Food
Jen, Y., Manson, J.E., Stumbo, C.R. & Zahradnik, Processing, 2nd edn. New York: Academic Press.
J.W. (1971). A procedure for estimating steriliza- Stumbo, C.R. & Longley, R.D. (1966). New param-
tion and quality factor degradation in thermally eters for process calculation. Food Technology,
processed foods. Journal of Food Science, 36, 20(3), 1OP-113.

Appendix 1: Consistency of low Wand high W equations


Since Pham’s (1987) method consists of two sets of equations, one for the high Wrange and one for the low W
range, the question arises as to whether the two sets agree at intermediate W-values.The author suggested
that the value W = 1.0 be used as demarcation between the two ranges (although the low Wequation remains
valid up to W = 1.5). This suggestion will be tested as follows: for W = 1.0, g/z is calculated according to the
low Wformula. Usingthisglz value, avalue W’ iscalculated according to the high Wformula. The difference:
e = (W’ - W)/W x 100%
gives the percentage discrepancy in calculated lethality between the two sets of equations at W = 1.0.
This calculation was carried out for the following range of parameters: N 1 = 0.1 to 1.0; N2 = 0.1 to 1.0;
N2/N, = 0.5 to 1.0; j = 0.5 to 2.0.
For most lethality calculationsN2 5 0.3 in which case the discrepanciesranged from -2% to 0%. Slightly
increased discrepancies, -5% to +1%, occurred at larger N2-values. However this happened only when
z-values are large and the physical process itself is subject to large inherent uncertainties, due to
chemical/biological destruction continuing even after the can has been cooled.
If W = 1.5 rather than 1 .O was chosen as the intermediate value, discrepanciesranged from 0 to +2% for
N 2 5 0.3 and from -3% to +2% for the entire N2 range.
In conclusion, discrepancies between the low Wand high W equations are negligible for W = 1.0 to 1.5,
and any value of W within that range can be used as the changeover value.
156 Q.T. Pham

Appendix 2: Nomenclature
D can diameter, m
f time for difference between product centre and ambient temperatures to be reduced by a
factor of 10, s
F sterilizing value measured as equivalent time at a reference temperature, s
g T h - TB,"C
i lag factor
L can height, m
N1 z/(Th - Ti)
NZ -
z/(Th Tc)
t time, s
T product temperature, "C
T,, T,, Th, Ti temperatures of cooling medium, can centre at end of heating, heating medium, and product
at start of process, "C
T, reference temperature (121.1"C)
LJ sterilizing value at thermal centre of can, i.e. equivalent time at Th, s
W UV
2 temperature change that would cause a 10-fold change in the sterilization rate, "C
Subscripts
C cooling period
h heating period

(Received 11 May, accepted 25 October 1989)

You might also like