Professional Documents
Culture Documents
www.elsevier.com/locate/compag
Abstract
Keywords: Fruit and vegetable storage; Horticultural crop; PI controller; Automation; Mini-chambers
0168-1699/03/$ - see front matter. Crown Copyright # 2003 Published by Elsevier Science B.V. All rights
reserved.
doi:10.1016/S0168-1699(03)00005-X
24 N.R. Markarian et al. / Computers and Electronics in Agriculture 39 (2003) 23 /37
1. Introduction
dCc
Vc gi gp vleak (Cc Ca ) (1)
dt
t
t
Qinj Kc emeasured s
tic g
k0
emeasuredk (2)
The monitoring system consists of three main components: PCs, data acquisition/
switch units (DAQs) and instrumentation (gas analyzers, valves, thermocouples).
Control of the MCs in each refrigeration room was designed to be independent, with
a separate PC and DAQ. Two gas analyzers (Siemens, Ultramat 23) are shared
among the three computerized systems. These analyzers enable the automatic
sampling of the gas in each MC.
Due to the large amount of dead time in the process, two PI controllers were used
for the control of CO2 and O2, respectively. In a general form, the proportional and
integral parts can be combined as presented in Eq. (3) (Hughes, 1988).
t
K
Cout Kc e c
ti g e dtm
0
(3)
where Cout is the controller output; e, the error, the difference between the set point
and the process variable; Kc is the proportional gain of the controller; m , the
constant setting for the control element when e/0; t is time; and ti is the integral
time.
The constant setting m has to meet the static or average load of the process. It
improves the performance at startup, without waiting for integral action. Switching
processes require on-line estimation of m (Sigrimis et al., 2000). The value or
function of Kc is dependent on the process variable, the facility and the
instrumentation used to regulate it. In a dynamic system, the function representing
Kc must be customized to encompass the dynamics of the system.
N.R. Markarian et al. / Computers and Electronics in Agriculture 39 (2003) 23 /37 27
The actual digital implementation of the algorithm was adapted for the process.
For the CO2 control, Cout (s) represents the CO2 scrubbing time from the chamber,
whereas Cout (s) for O2 control represents the injection time of air into the chamber.
The integral time (ti) was set to 48 h (172,800 s), which implies that the system is
expected to re-stabilize within 48 h following a disturbance. Due to increased
aggressiveness of the process, a ti of B/48 h resulted in poor system stability. The
integral component of the controller generates an output based on the error history
that has accumulated in the controllers integrator. These include errors due to
physiological changes within the stored produce.
Due to the multivariable nature of the process, the coupling effect of CO2 and O2
was investigated (Beaudry et al., 1999). Tests were performed using three empty MCs
to determine the behavior of chamber CO2 and O2 concentrations as a function of
scrubbing time. Tests demonstrated that the oxygen level increased negligibly during
the scrubbing process.
2.5. Determination of Kc
The proportional gains (Kc) for O2 and CO2 represent the gain that must be used
with the error to draw the control output close to the set point. For this specific
system, Kc_O2 (s l1) represents the time to effectively inject 1 volume unit of O2 into
the chamber. Kc_CO2 (s l1) represents the time to effectively scrub 1 volume unit of
CO2 from the chamber. The exponential behavior of the dilution process complicates
the determination of Kc.
Kc_O2 was determined using a model, which consists of a closed container with a
certain volume of stored produce, in which a gas with a certain concentration is
injected. A mass balance equation can then be derived as a first order linear
differential equation (Eq. (4)).
d(PV O2 ) PV O2
Cin Fin Fout (4)
dt V
where Cin is the concentration of the incoming gas (l l 1, i.e. 21% O2); Fin, the flow
rate of the incoming gas (l s 1); Fout, the flow rate of the outgoing gas (l s 1);
PV _O2 is the O2 process variable or amount of O2 in the MC (l); and V is the free air
volume (l).
This model contains several assumptions: (1) the gas in the chamber is uniformly
mixed, thus the outgoing gas has the same concentration as the gas in the container;
(2) the outgoing gas has the same flow rate as the incoming gas (i.e. Fout /Fin); (3)
the gas in the chamber is incompressible; (4) there is no infiltration due to leaks, or so
little that it would be negligible for short injection times; and (5) Cin and Fin are
constant.
28 N.R. Markarian et al. / Computers and Electronics in Agriculture 39 (2003) 23 /37
With these assumptions, Eq. (4) can be solved into Eq. (5).
where PV _O20 is the initial amount of O2 in the MC (l); and t is time (s).
The time to effectively inject 1 volume unit of O2 into the chamber is obtained by
simplifying Eq. (4) as shown in Eq. (6).
dt 1
Kc O2 (6)
d(PV O2 ) PV O2
Fout Cin
V
Experimental testing was used to validate the proposed model. Three empty MCs
were set up. Nitrogen gas was used to flush the MCs to 0% O2 and CO2. The
duration of air injection and gas analysis was noted. The experiment was performed
for the range of oxygen used in CA storage (0 /8% O2).
Kc_CO2 determination was performed experimentally only. Three empty MCs
were set up. Pure CO2 was injected until the CO2 level in the MCs reached :/20%.
The scrubbers were turned on. The duration of scrubbing and gas analysis was
noted. Average CO2 concentrations in the MCs versus time were plotted as shown in
Fig. 1. Average scrubbing rate was computed using the time required to remove a
given amount of CO2. The scrubbing rate versus the amount of CO2 in the chamber
was then plotted (Fig. 2) and a regression equation was fit to the data (Eq. (7)).
Fig. 1. Average CO2 levels in the mini-chambers versus time from experimental results obtained for the
determination of the scrubbing rate of the scrubbers.
N.R. Markarian et al. / Computers and Electronics in Agriculture 39 (2003) 23 /37 29
d(PV CO2 )
0:1717(PV CO2 1:1412 ) (R2 0:9941) (7)
dt
where (d(PV _CO2)/dt ) is the rate of CO2 removal with respect to time (l s 1); and
PV _CO2 is the CO2 process variable or amount of CO2 in the MC (l).
The time to effectively scrub 1 volume unit of CO2 from the chamber is obtained
by inverting Eq. (7) as shown in Eq. (8).
dt 1
Kc CO2 (8)
d(PV CO2 ) a PV CO2 b
where a and b are the constants found in Eq. (7). Clearly, Eq. (6) and Eq. (8) do not
include a respiration rate variable and therefore can be used for any type of produce.
Figs. 3 and 4 show the controller output computation and process control
routines, respectively. Once gas concentrations in each MC are acquired, the
30 N.R. Markarian et al. / Computers and Electronics in Agriculture 39 (2003) 23 /37
The monitoring system was tested for two consecutive storage seasons (1999 /2000
and 2000/2001) for different types of produce: apples (Bissonnette et al., 2001),
broccoli (DeEll et al., 2001) and cauliflower (Demian et al., 2000). The 1999 tests
included only produce in the 110l polyethylene rigid plastic chambers. The full
facility was used to store different varieties of apple and squash at different
concentrations, temperatures and MC volumes during the 2000/2001 season, with
gas concentrations ranging from 1.0 to 4.5% O2 and from 1.0 to 8.5% CO2.
Variance analysis was performed using a GLM procedure (SAS Institute, 1988) to
compare the average errors and stability over time for O2 and CO2. Data was
collected for a 60-day storage period with nine MCs consisting of three MCs of three
volumes (110, 285 and 500 l). Differences in mean average errors were tested using
Duncans multiple range test (SAS Institute, 1988). Conditions in both types of MCs,
such as the quantity of stored produce per unit volume, O2 pull-down period, air
circulation and gas control, were representative of commercial storage room
conditions.
N.R. Markarian et al. / Computers and Electronics in Agriculture 39 (2003) 23 /37 33
Fig. 5. Comparison of results obtained from experiments performed to determine Kc_O2 and to results
obtained using the derived model for Kc_O2.
Percentage of O2 in the MCs obtained from the experimental tests versus time of
injection was plotted along with results from the model (Fig. 5). The experimental
data fits the model with an average error of /0.24%. Concentrations from the
experimental data (Fig. 5) that are lower than the predicted most likely occur from
our models assumption of perfect mixing, which may not be the case in reality. The
air entrance and outlet in the MCs are physically close together, which may result in
higher than average concentrations near the outlet. If necessary, this difference could
be decreased either by increasing the distance between the air inlet and outlet, or by
increasing the air circulation rate in the MCs.
Fig. 6 shows an example of the behavior of the controller while storing Spartan
apples for 60 days in a 110 l MC and for set points of 1.5% for both CO2 and O2. The
small variations encountered were likely due to the electrical calibration of the inputs
and reduced scrubber efficiencies over time. Occasional software failure during the
storage period also caused some variations in controller output. Such temporary
stoppages caused a decrease in O2 and an increase in CO2 due to produce respiration
(Fig. 6). In each case, the software was restarted to resume the control process. In
addition, some of the larger variations in Fig. 6(b) representing sudden O2 depletion
are most likely due to physiological changes, such as increased respiration by the
stored commodity. Once lower O2 readings were measured, the controller attempted
34 N.R. Markarian et al. / Computers and Electronics in Agriculture 39 (2003) 23 /37
Fig. 6. Example of controller performance for a storage period of 60 days while storing Spartan apples: (a)
CO2 at setpoint of 1.5%; (b) O2 at setpoint of 1.5%.
to readjust the O2 concentration to the set point causing an over shoot and
eventually settling near the set point. This example demonstrates the ability of the
integral component of the PI controller to readjust the O2 concentration after a
sudden disturbance.
Table 1
Average of differences between the set points and process variables for a 60-day storage period (four
measurements per day) in three different volume MCs containing apples: (a) CO2 (% or litre of CO2 100
l 1 of air); (b) O2 (% or litre of O2 100 l 1 of air)
Rep Average errors (110 l) Average errors (285 l) Average errors (500 l)
The number in parenthesis represents the S.D. (%). Data within the same row with the same superscript
are not significantly different at a B/0.05 according to the Duncan multiple range test.
N.R. Markarian et al. / Computers and Electronics in Agriculture 39 (2003) 23 /37 35
Table 1 shows the average errors for CO2 and O2 concentrations for a 60-day
apple storage period (four measurements per day) in three different volume MCs
(110, 285 and 500 l). Maximum average errors (absolute values) obtained in any
mini-chamber were 0.0035 (110 l) and 0.0016 (110 l) for the CO2 and O2
concentrations, respectively. These relatively low differences between the set points
and the process variables demonstrate the ability of the controller to maintain the
desired set points.
There was no significant difference between the means of average errors for the
three treatments for O2 concentrations (Table 1b) (F4,2155 /0.50; P /0.608).
However, a significant difference was found between the means of average errors
obtained for CO2 (Table 1a) for the 110 and 500 l MCs (F4,2155 /2.56; P /0.0078).
The average concentration of CO2 in the 110 l MCs were maintained at a value
0.002% lower than the set point and at a value 0.002% higher than the set point for
the 500 l MCs (Table 1a). This difference, although very small can be explained by
the fact that all MCs had the same type and capacity of CO2 scrubber and the
scrubbing time required to decrease the CO2 concentration by a given amount is
longer for a MC with larger volume. Hence, when very precise control of CO2 is
desired, a larger capacity scrubber should be used for larger MCs. Nonetheless, the
users (Demian et al., 2000; Bissonnette et al., 2001; DeEll et al., 2001) were satisfied
with the results obtained.
The S.D. results (Table 1) demonstrate that the system is reasonably stable with
maximal S.D. of 0.051 and 0.077 for the CO2 and O2 concentrations, respectively.
There was no significant difference among S.D. for CO2 (F4,4 /0.46; P /0.6613)
across different MC volumes. However, a significant difference was found in the
S.D. between the treatments for O2 (F4,4 /64.0; P /0.0009), wherein there was
significantly more variation in the 500 l MC, followed by the 110 l and then the 285 l
MCs. This might be partially due to relatively more mechanical problems for the 500
l MCs. Furthermore, their fan capacities were insufficient to properly circulate the
gas mixture, resulting in improper mixing and thus incorrect gas concentration
readings. Gas leaks, mainly from tubing connections and improper cover sealing of
the 110 l rigid MCs caused an increase in O2 and a decrease in CO2. The leaks were
eventually located and resolved. The controller stabilized the gas concentrations
subsequently.
4. Conclusions
the produce type or their respiration rate. By implementing new routines in the
systems software, it can be used to investigate and analyze new control strategies for
CA storage.
In general, performance of the monitoring system was evaluated and judged
satisfactory by end users (researchers). The controller was able to maintain desired
set points with maximum average errors (absolute values) of 0.0035 and 0.0016% for
the CO2 and O2 concentrations, respectively. There was no significant difference in
average O2 errors across different MC volumes and only a slight difference in CO2
errors. The system was reasonably stable with maximum S.D. values of 0.0501 and
0.0701% for the CO2 and O2 concentrations, respectively. There was no significant
difference in CO2 S.D. across the MC volumes, but a significant difference among O2
deviations. The latter difference was due to the large number of mechanical problems
encountered during the storage season for the 500 and 110 l MCs, as opposed to the
285 l MC.
Improvements to a few aspects of MC design could increase system performance
and stability. For example, air inlets and outlets in the MCs could have greater
physical separation to allow for better mixing and thus more exact gas concentration
measurements. Also, fan capacity in the 500 l MCs could be increased to allow better
gas circulation. Finally, CO2 scrubber capacity could be better matched to MC size.
Such mechanical modifications would allow the control devices and software to
operate at their designed capabilities.
Acknowledgements
The authors wish to thank S. Bissonnette for assistance with the experimental
testing and C. Beaudry for her input during the design of the control system.
References
Beaudry, C., Vigneault, C., Landry, J.-A., Raghavan, G.S.V., Gariepy, Y., Goyette, B., 1999. Design of an
automated control system for mini-chamber controlled atmosphere storage. Paper presented at ASAE
No. 99-4114, 10 p.
Bishop, D.J., 1996. Controlled Atmosphere Storage. A Practical Guide. East Sussex, UK, pp. 33 /45.
Bissonnette, S., DeEll, J.R., Levesque G., Vigneault, C., 2001. Evaluation de leffet du delai de
refroidissement et de leffet des concentrations de gaz et de la temperature dans lentreposage sur la
fermete des pommes McIntosh. Report presented to the Federation des Producteurs de Pommes du
Quebec. Horticultural Research and Development Center. Agriculture and Agri-Food Canada. St-
Jean-sur-Richelieu, Quebec, Canada, 37 p.
DeEll, J.R., Toivenen, P.M.A., Vigneault, C., 2001. Postharvest dips to control black speck and scar
discoloration in stored broccoli. J. Food Qual. 24, 575 /584.
Demian, S., DeEll, J.R., Roussel, D., Barrette, M., Goyette, B., Bissonnette, S., Beaudry, C., Gariepy, Y.,
Vigneault, C., 2000. Controlled atmosphere storage of cauliflower. Report presented to Les Jardins
Cousineau et Fils Inc. Horticultural Research and Development Center. Agriculture and Agri-Food
Canada. St-Jean-sur-Richelieu, Quebec, Canada, 23 p.
N.R. Markarian et al. / Computers and Electronics in Agriculture 39 (2003) 23 /37 37
Goyette, B., Vigneault, C., Markarian, N.R., 2002. Design and implementation of an automated
controlled atmosphere storage facility for research. Can. Biosys. Eng. 44 (3), 35 /40.
Hughes, T.H., 1988. Measurement and Control Basics. ISA, Research Triangle Park, NC, 306 p.
Koca, R.W., Hellickson, M.L., 1993. Computer-controlled refrigeration in evaporatory fan-cycled apple
storage. Appl. Eng. Agricult. 9 (6), 561 /565.
Korthals, R.L., Knight, S.L., Crawford, R.R., Christianson, L.L., 1995. Development and testing of a
simple carbon dioxide enrichment controller for growth chambers. Trans. ASAE 38 (1), 207 /211.
Lauro, E.M., Chu, C.L., Wismer, J.D., 1987. An integrated gas control and temperature monitoring
system for controlled atmosphere storage. Paper presented at CSAE No. 87-406, 20 p.
Mittal, G.S., 1997. Computer-based instrumentation: sensors for in-line measurements. In: Computerized
Control Systems in the Food Industry. Dekker Marcel Inc, New York, NY, pp. 13 /53.
Morimoto, T., Hashimoto, Y., 2000. An intelligent control for greenhouse automation, oriented by the
concepts of SPA and SFA */an application to a post-harvest process. Comp. Electron. Agricult. 29 (1),
3 /20.
Plissey, E., 1995. Improve potato storage. Am. Veget. Grow. 43, 41 /43.
SAS Institute, 1988. SAS/STAT Users guide, Release 6.03 ed. SAS Institute, Cary, NC.
Sigrimis, N.A., Arvanitis, K.G., Pasgianos, G.D., 2000. Synergism of high and low level systems for the
efficient management of greenhouses. Comp. Electron. Agricult. 29 (1-2), 21 /39.
Stone, M.L., 1991. Control system applications. In: ASAE Automated Agriculture for the Twenty-first
Century. Proceedings of the 1991 Symposium, Chicago, IL, pp. 163 /166.
Thompson, A.K., 1998. Controlled Atmosphere Storage of Fruits and Vegetables (278 p). CAB
International, New York, NY.
Vigneault, C., Granger, R.L., Raghavan, G.S.V., 1991. Mini-chambers for lab-scale research on controlled
atmosphere storage. Appl. Eng. Agricult. 7 (5), 617 /621.