On-farm implementation of near-optimal control of bulk storage drying for grain

Summary

Sector:
Cereals & Oilseeds
Project code:
PR168
Date:
01 April 1995 - 31 March 1998
Funders:
AHDB Cereals & Oilseeds.
AHDB sector cost:
£95/467 From HGCA (Project No. 0046/1/94)
Project leader:
M Nellist, D Burfoot and W Day Silsoe Research Institute, Silsoe

Downloads

project_report_168

About this project

Abstract

The objective of this research was to test the feasibility of applying, to real stores, theoretical predictions of the optimal control of drying grain in bulk stores. In bulk storage drying, air at or near ambient temperatures is forced through static beds of grain usually between 1.5 and 4 metres deep Airflows 0.05 m3 s-1 tonne-1 are generated by fan pressures between 0.3 and 1.7 kPa. The efficacy of such drying is very dependent on the level, and variation, in temperature and humidity of the ambient air. Control of the drying process is effected by modifying either one, or both, of these factors. Methods of modification vary from simply switching the fan on and off to adding supplementary heaters or dehumidifying units.

The action of bulk storage drying is very dependent on the weather so realistic experimental investigations are restricted to harvest periods and replicated tests are almost impossible. Conventional design criteria have therefore been based over many years on cumulative experience and trial and error. The development of theoretical understanding and mathematical models of drying provide a means of; in effect, carry out many years' of experiments very rapidly on a computer. One of the computer models developed at Silsoe Research Institute, called STOREDRY, can be used to simulate bulk storage drying allowing for hourly changes in weather data and implementing various control policies. Included with the predictions of temperature and moisture content of the grain, the model also predicts the index of grain deterioration.

STOREDRY has been incorporated within an optimising shell which enables improved control strategies to be suggested. The resulting software, called OPTIM-STOREDRY, can be used to estimate ten parameters relating to the drying equipment and control strategy which would minimise the combined cost of energy and overdrying. A conventional optimisation procedure cannot be used due to the unpredictable variations in weather, the number of criteria for successful drying (such as final moisture content), and the constraints on drying time and spoilage. Work prior to the project sought to use the software to seek the optimum size and best control strategy for a fan and heater system. The results always indicated that the most cost effective strategy was initially to overdry the bottom layers of the grain and allow them to rewet during the final stages when the upper layers were tending towards the target moisture content. This required that the set point for the relative humidity was low at the beginning of drying and was increased steadily as drying progressed. This is counter to conventional policies in which the set point relative humidity is reduced during drying, consequently, the new approach would not be recommended until the present project had been carried out to prove its validity at a practical scale. This required the development of:

i. a near optimal control algorithm (based on computer predictions of drying scenarios)
ii. a method of providing signals to a controller (using a model to infer moisture profiles from temperature measurements)
iii. a control system incorporating the algorithm and signal handling (which was to be demonstrated on-farm).

The project period encompassed the three harvest years, 1995, 1996, 1997. In 1995, a suitable store was identified and instrumented but the exceedingly dry harvest precluded any drying of the grain. The 1996 harvest period was also relatively dry although in this year the first tests of the controller were possible by moving to a farm in a more western, and moister, region. This work proved that the controller could infer, from measurements of the grain temperature in the bed, the moisture contents needed to enable control of the dryer. Also, we successfully demonstrated that we could remotely operate, interrogate and, if necessary, modify the controller via a telephone link. However, the drying equipment at the farm was undersize for the required drying duty and the control algorithm was rarely required to fine tune the relative humidity in the way that was intended. There were also problems caused by power failures at the farm.

In the final year, the crucial requirements were to demonstrate that the controller would operate on a reasonably large batch of grain and to ensure that we had wet grain and continuity of power supply. Grain was harvested at over 20% w.b. and was dried successfully over 10 days in a 3 m deep bed. The controller was able to implement the planned control policy but the rate of drying was slightly over-predicted probably because some of the equilibrium moisture data used in the development of the control policy was not exactly appropriate to the grain used in the tests

The control policy was implemented in a data logger, provided by Campbell Scientific, which was easy to use in the development and testing of the control policy but it provides too much flexibility for use in most farm situations. Parts of the logging system could be used to produce a dedicated grain controller. Alternatively, the control policy and the procedure for inferring moisture contents from temperatures in stores could be incorporated into one of the new advanced controller/loggers which are entering the farm industry. The latter could also be useful for proof of traceability. Currently, extensive computer simulations are required to determine the near-optimal parameters for use in the controller for each specific store and conditions. A simplification of this approach would be required to provide parameters for use in a less-optimal, but more generally applicable controller, otherwise the cost of the controller would be prohibitive. The cost of each controller would then be around £2000 irrespective of whether a dedicated controller or controller/logger was used.

×