ADVANCED CONTROLS ENGINEERING - A PATH TO BREWING PROFITABILITY

John T. Kay, President, Universal Dynamics Technologies Inc.
#100-13700 International Place, Richmond, Canada V6V 2X8

Armand Thompson, Head Brewer, Molson Breweries
1550 Burrard Street Vancouver, Canada V6J 3G5

KEYWORDS

Adaptive, Predictive, Auto-Tuning, Dynamic Modeling, Optimization

ABSTRACT

This paper is about the application of an advanced control strategy in brewing. It is written about brewing for 'Practical Brewers' to convey to them that there are opportunities for using new technology to improve the art of brewing. It is not a review of the theory behind the technology which would be more suited to academics. A bibliography is provided for those more interested in theory, however a very brief introduction is provided.

This technology was trialed at a Western Canadian brewery on two process; kettle boil height and warm water mixing. These applications present difficulties to traditional approaches due to their non-linearity, dead time, and changing process characteristics. The trials were conducted to determine the effectiveness for two distinct plant requirements:

• direct control through on-line modeling
• calculating optimal PID settings.

INTRODUCTION

As with virtually all industries today and certainly all Breweries, there is tremendous pressure to reduce operating costs and continually improve product quality in order to remain competitive. The Vancouver plant is no exception. Like all of you, we're trying to do more with less and in order to achieve this we've found it absolutely necessary to take advantage of as much new technology as we can afford and justify.

As the brewing people will know the boiling of wort in the kettle is one of the most critical stages in the brewing process. A vigorous boil is essential to sterilize the wort, to ensure inactivation of enzymes, precipitate which may cause haze in the final product, concentrate the wort, increase the colour, drive off undesirable volatiles and, very importantly, extract the bitterness from the hops.

A good boil, besides the analysis of the wort, is measured by an almost violent rolling action in the kettle itself and more quantitatively by its evaporation rate. At Vancouver we target for a minimum of about 5% evaporation per hour on most of the brands.

The challenge of achieving this kind of a boil is complicated by a number of criteria or a number of objectives we want to attain.

We want to make each brew as large as we possibly can in order to maximize productivity and vessel capacities. However because of the violent foaming that occurs during boiling an adequate head space must be allowed.

We want to maximize the heat transfer and the evaporation rate in order to minimize the boil time and thus put the brews through faster.

We want repeatability form one brew to another so that product consistency is always achieved.

We want to help the steam engineers minimize the fluctuations on the steam draw to reduce peak loading and energy loss through venting.

The Vancouver plant has a capacity of 1.2 million hectoliters annually. It has a 30 foot diameter lauter tun, a cereal cooker, a mash tun and two identical 940 H1 kettles with a common stack and separate dampers. The wort entry is through the bottom of the kettle and hop pellets are added via a chute from a remote hopper. Heat transfer for wort boiling is accomplished by a bottom steam jacket, two side jackets and a central percolator. The steam regulator is under the kettle and the steam system pressure is 65 psi. There is no measurement of the pressure.

The plant control system consists of programmable logic controllers (PLC) and personal computer based man operator interfaces. Kettle #2 was just being put into service and a continuous level capacitance probe was used in addition to the standard 18 inch 3 level discrete probe used in Kettle #1. Kettle #2 was used for the trial and Kettle #1 used a PLC based strategy which had been used for several years. Kettle #1 provided a baseline upon which to compare the results of Kettle #2.

To obtain the necessary head space, Kettle #1 is considered full at 725 HI.

This paper specifically address the use of Dynamic Modeling Technology (DMT) to control the difference between the height of the boiling and non-boiling wort in the kettle during the brewing process and the temperature of water entering the Mash Tun. DMT is another general term which can be used in the same breath as Expert Systems, Fuzzy Logic, Neural Networks, Adaptive and Predictive control. DMT grew out of an area of study which was call "Minimum Effort" control. The mathematics heart of the controller used is Laguerre Orthonormal Series. This approach has only come into commercial use under different names in the last year. The key to its effectiveness is the ability to automatically create a mathematical model of a process which can be continually updated on-line and handle processes which are non-linear, higher order, dead time varying, and have changing time constants and gains without bumping the process.

This model is used to either directly control the process through a PLC/DCS or to calculate the optimal settings for a PID controller. The process selected for direct control was the boiling of wort in the kettle; and, the process selected for calculating the PID settings was the temperature control of water entering the Mash Tun.

BREW KETTLE

The quality of the brewing process is directly affected by the boil characteristics in the brew kettle. If a vigorous, rousing boil can be maintained, a high quality brew can be consistently made by the Brewmaster. It should be emphasized that a good control system only provides repeatability to the process engineer; or, as in a brewery, the Brewmaster. It is his knowledge that produces a "quality" beer. The control engineer just allows him to do it again and again and again with the same settings on the process.

Until now this has not been attainable due to the non-linear characteristics of a boiling liquid This, along with changing formulas for each brew type, meant that only a model based predictive adaptive controller could achieve the goal of maintaining a consistent boil height so that the process of brewing could be optimized. Boil height is the difference in height between a non-boiling liquid and the boiling liquid. The boil height measurement had not been available to the Brewmaster previously. It was only through discussions with the Brewmaster that it was determined that this was a suitable parameter to control. It was hoped that a certain boil height would optimize evaporation for each phase of the boil while maintaining quality.

As can be seen from the following chart the system for Kettle #1 was continually opening and closing the steam valve as the heat caused the liquid to boil until it hit one of the three the high level probes and then the steam valve closed at a rate determined by quite a sophisticated PID PLC strategy based on how many of the three levels probes were covered and for how long. This strategy in the PLC also opened the steam valve in steps. The addition of 1st, 2nd and 3rd hops, the timing of which is at the brewers discretion, allows the steam valve to remain open longer. With this strategy it was found necessary to limit the steam valve to a maximum of 83% open to ensure there was not a boil over which is a serious safety hazard.

As long as the percolators are covered, boiling should start as the beer level changes throughout the filling and continue through the runoff phases of the process. Also the addition of hops dramatically changes the characteristics of the boil as the oils tend to allow the bubbles to burst more easily thus reducing the boil height. Air pressure changes, through climatic changes or as venting fans switch on or off, complicate the control of boil height. It was also found that once the steam is shut off then on repeatedly the rate of rise increases as if the boil height had a 'bounce' characteristic.

Note that "Low Level Probe" refers to the lowest of the three high level discrete probes which are set 9" apart vertically and "Level Probes" refers to the group of three.

KETTLE #1

When we look more closely at Kettle # l's steam valve shortly before the addition of the second hops (shown above) we can see that it is actually going full closed even with the ramp feature in the PLC.

The activity of this cycling changed from batch to batch even for the same brew. This meant that there were inconsistencies between brews of the same brew types.

A Personal Computer platform with an OS/2 operating system was chosen to run the software as PC's are easily upgraded to the latest hardware. This is important with advanced model based control strategies as they are calculation intensive. The software selected does 10,000 floating point calculations per update per loop. A 486/66 CPU can handle 16 to 32 loops with 1 to 2 second updates. A true multitasking operating system such as OS/2 is required when more than one loop is on-line.

The PC was serially linked to the PLC Data Highway network giving it access to all the PLC's in the plant. The Kettle #1 backup strategy using the PLC PID instruction was programmed into the PLC as well as leaving the old high level probe in place and active. The new probe provided continuous level measurement with a 4-20 mA signal to the PLC. The probe was specially manufactured for the application designed through consultation with the manufacturer who visited the site. The Hop Blower and Wort flowmeter were supplied to the system as feedforwards. Had there been a steam pressure sensor, it would have also been used as a feedforward.

The trial was carried out on Kettle #2 in January and February 1994. The limitation of the steam valve maximum opening was removed and the following results were obtained.

KETTLE #2

Kettle #1 full level was limited to 725 H1 to maintain a level of about 3.5 feet below the door opening to ensure that the boil would not go out that opening. For each 6 inches that this head space can be reduced the utilization of kettle capacity is increased by approximately 3%. As can be seen in the above chart the new system maintains the boil height within 6 inches of setpoint. It is felt that this head space can be reduced by 6 to 12 inches thus increasing utilization by 3 to 6%. This cannot be tested until Kettle #1 can be retrofitted with the same control system as both kettles must brew the same size batches in this plant.

You can see that the boil height has a setpoint of 3.7 feet which is being maintained very well. The setting of 3.7 feet was very arbitrary during the test shown and should not be construed as an optimal setting. During the period shown here the Kettle #2 temperature was constant at 100 degrees C. The Hop addition lowered the boil height and even though the steam valve was 100% open it took approximately 15 minutes to return to the same boil height although it never stopped boiling. This leads us to believe that a larger steam source would provide even greater improvements to the system.

It is worth pointing out that a very positive spin-off of doing the process monitoring required by an advanced control trial is that people knowledgeable in both process and control take a close look at the dynamics and often identify opportunities that would otherwise not be noticed.

Over a three month analysis of the new control system the data for the following chart was gathered on evaporation rates for identical brews in Kettle #1 and Kettle #2. The result was an eight percent improvement in the evaporation rate with the new control system. The average evaporation rate for Kettle #1 was 5.17% and for Kettle #2 was 5.59%.

Other measured results over the three month period were that the plant operated with a steadier steam pressures as peak loading was reduced by 6000 to 8000 lbs. Also a 5% reduction in hop utilization was attributed to the new system which results in a saving of $25,000 US annually.

Unmeasured results included a reduction in steam venting due to sudden valve closures which affects the total energy consumption of the plant.

The system cost for the first kettle is approximately $70,000.00 US including hardware, software, installation and setup (exclusive of PLC hardware). To add Kettle #1 to the system will require an additional $30,000 US as only a probe, software and engineering are required. However, as each brewhouse is different, the pay back must be looked at on an individual basis but it is expected to be well under one year.

WARM WATER LOOP

This is an example of a seemingly simple process which appeared to the operator that it was working well because the temperature indicator in the operating room showed minimal variation from setpoint. In this process hot & cold water are mixed to a desired temperature of 50 degrees C which goes to the Mash Tun. The temperature is critical to ensure the proper reaction as quickly as possible.

The valve was hidden from easy view so that any irregularities were not likely to be noticed by maintenance staff. When we conducted a process audit we saw this loop had significant oscillations. Yet it had been tuned several months before. Upon further inspection of the loop we noticed the following difficulties:

• No measurement of cold water temperature, flow or pressure,
• Poor mixing of hot/cold water affects temperature probe,
• Mag flow meter is too close to valve.

It would appear that the loop had been tuned in the summer when the city water was warmer. Now that it was mid-winter the gain had increased causing the loop to oscillate badly as shown below.

The CO line is the movement of the control valve which had about a 45 second oscillation from 0 to 54%. This is not only a problem because of valve wear as it causes the water temperature to cycle with the same period from 45 to 55 degrees C.

Although once mixed into the large Mash Tun tank the average temperature was correct, as seen by the operator, the enzymes were being continuously hit with hot and cold showers. The hot showers were deactivating the enzymes resulting in increased production times and Mash variability. Enzymes have a very narrow optimum temperature range of+- 2.5 to 5.0 degrees C.

It was desired to maintain this loop under PID control and to meet this goal the model was developed on-line and then an iterative approach was used off-line in the personal computer to find the optimal P, I, & D settings for the PID controller used. The follow graph shows how the loop looked after tuning. Notice how quickly it settled into the proper temperature.

This shows the loop is properly responding to an increase in the temperature of either the cold or hot water feed with minimal change (0 to 200 seconds) considering no feed forward signals were available. It is important that a loop properly handle process disturbances as well as process set point changes (changed at 300 and 500 seconds) as this now does. The key point to remember is that no one new nor did existing instrumentation tell anyone that the loop was oscillating and even with the difficulties mentioned above good tuning was provided by the software.

CONCLUSIONS

Advanced control strategies can be successfully used to significantly improve performance of a Brewhouse with minimal cost. This can be done through the use of direct control of loops on a networked PLC/DCS system where PID is not adequate and on-demand control to tune PID loops which are stand alone devices.

Although model based control has been around since the 1950's, the DMT method has only come into commercial products in the past year. The key practical difference to previous ones is the ease of implementation. The automatic building of the process model including feedforward models by monitoring the process on-line in about 5 to 10 updates or process dead times such that it can be put into control mode from the learn mode in less than an hour or two after connection. Keep in mind however that all control strategies must be thoroughly thought out and engineered to ensure all the plant goals are met. It is just that now the model building portion has been significantly reduced. The best advanced control strategy cannot account for erratic sensors, under sized valves and many other physical limitations.

REFERENCES

[1] C.C. Zervos and G.A. Dumont, "Deterministic adaptive control based on Laguerre series representation", Int. J. Control, Vol. 48, No. 6, pp. 2333-2359, 1988.

[2] C.C. Zervos, "An Offiine Method for the Optimal Tuning of a Three Term Controller", Master's Thesis, Dept. of Electrical Engineering, McGill University, Montreal, Canada, 1983.

[3] B. Gough, J. Kay, G Seebach, "Adaptive Control Applications in Pulp and Paper", Canadian Pulp and Paper Association Annual Conference, Montreal, Canada, January 1993.

[4] B. Gough, J. Kay, "Auto-Model Adaptive Process Control", ISA PUPID Annual Conference, Vancouver, Canada, March 1994.

[5] B. Gough, J. Kay, ""Minimum Effort" Advanced Control Solutions", ISA/94 Annual Conference, Chicago, USA, 1993.

[6] Dr. B. Wilson, B. Gough, J. Kay, "Adaptive Control of Sulphur Recovery Units", ISA Calgary '93, Calgary, Canada, May 1993.

[7] B. Gough, "Advanced Adaptive Control Applications", IEEE IAS PPIC, Portland, USA, June 1992.

[8] B. Gough, J. Kay, ""Minimum Effort" Adaptive Control of Effluent Treatment, Kilns and Pulp Brightness", 1993 TAPPI/ISA PUPID Process Control Conference, Nashville, USA, March 1993.

[9] G. Hassell, R. Harper, "Nonintrusive P.I.D. Auto-Tuning Using Dynamic Modeling Technology", IEEE IAS PPIC, Nashville, USA, June 1994.

[ 10] S. Hagemoen, "An Expert System Application for Lime Kiln Automation", National Lime Association Annual Conference, Birmingham, USA, October. 1993.

[11] M. Cameron, D. Lyon, "Implementation of an Innovative Self-Tuning Adaptive Controller for Complex Industrial Processes", Conference Record IEEE/IAS PCIC-90-52, September 1990.