Model predictive/adaptive control reduces the variability of final Brix in orange-juice concentrate

Results of applying predictive/adaptive control to an evaporation process for concentrating orange juice were reported at the Citrus Engineering Conference conducted by the American Society of Mechanical Engineers (ASME) last March by Jose D. "Danny" Milla, control products manager at FMC FoodTech Citrus Systems (Lakeland, Fla.) A QuickStudy Adaptive Process Controller (APC), developed by Adaptive Resources (Pittsburgh, Penn.), was installed for testing on a commercial evaporation system at a Florida orange-juice plant. Food Engineering interviewed Danny Milla to learn how predictive/adaptive control can improve product quality in an evaporation process.

FE: Let's first discuss the evaporation system, then how QuickStudy was applied. By how much is the solids content of orange juice increased through evaporation?

Milla: Natural orange juice is about 11 or 12 degrees Brix, the refractive index used to measure the solids content, which is generally equal to percent sugar. Evaporation concentrates juice typically to 65 degrees Brix for storage and transportation.

FE: What kind of evaporation system was used for the QuickStudy test?

Milla: The evaporator is a Thermally Accelerated Short Time Evaporator (TASTE), developed in the 1960s to better retain the flavor of concentrated orange juice, and is in general use throughout the industry. TASTE systems are manufactured by FMC and by other suppliers as well. Earlier evaporation systems used recirculation or reboiling methods, which resulted in burned flavor and off-flavors as a result of heat treatment. The TASTE evaporator is of multi-stage, falling-film, tube-and-shell design operated under vacuum, and this particular system is of mixed-flow design, which allows the processor to recover volatile compounds which can later be added back to the concentrate.

FE: How does TASTE minimize degradation of juice flavor and color?

Milla: Evaporation occurs in a vacuum, so it can occur at lower temperatures, and the falling film minimizes residence time. So lower temperature and minimum residence time better retain juice flavor.

FE: What was the objective of the QuickStudy study?

Milla: To provide a control system that would more tightly regulate final Brix of the concentrated juice, that would minimize the variability of Brix in the concentration. Processors like to maintain final Brix as constant as possible to minimize post-blending, post-processing and waste.

FE: What is QuickStudy?

Milla: QuickStudy is an adaptive, model-predictive control software that can adapt dynamically to process changes. It allows the engineer to find the variables that affect the process and examine their relationships to each other and to the final product. There are many interrelationships in the evaporation process. You have, for example, two different transport mechanisms: One is the vapor on the outside of the tube, the other is the product inside the tube, and they travel at different speeds. A change on the vapor side can quickly affect the product side. If you can detect that change, you can make the correct adjustment.

FE: What process variables are controlled?

Milla: Feed rate is the ultimate control variable. Typical controllers just look at the discharge concentration and vary the feed rate to control the final concentrate. But there are other things happening here. Feedstock concentration can change, and typically that's not measured. Steam pressure can change and have very profound effects on product concentration. With QuickStudy we're measuring these and minimizing variation in the end product.

FE: What process conditions are predicted?

Milla: By looking at steam pressure, we can predict the effect of pressure change on the final product. And by looking at feedstock concentration, we can predict what effect a change in feed Brix will have on final Brix. So we then know what correction we have to make to the feed rate to account for changes in both of those variables.

FE: How does QuickStudy adapt to a predicted condition?

Milla: That's where the adaptive-control feature comes in. It collects more data on-line and refines the model to give you better control.

FE: What issues or problems were addressed in the study?

Milla: In a typical evaporator, product residence time throughout the system is so long that you get a lot of historisis, or variation, in the output. Dead time in this process was eight to 15 minutes. The process is typically controlled by product feed rate. Suppose you notice the product going to the high side of your concentration limit - up to 66 degrees Brix. By the time you increase your feed rate to bring that Brix down, it may have climbed to 67 or 68 degrees Brix. In other words, by the time you notice a change in the product it's too late to take corrective action.

FE: How is the basic model built?

Milla: You can use existing data, but it's probably better to design a test to collect data specifically for building a QuickStudy model. We took several hundred data points and used Adaptive Resources' off-line program, called QMast, to create the model. Then we installed the model in the system, let it run, evaluated it, then updated it with new data as the process was running. Ideally you want to let the model go through upset conditions -- unplanned events such as a sudden drain on steam pressure caused by starting-up another evaporator, or a change in feed Brix caused by recycling off-spec product into the feed tank. A new load of oranges can cause a natural variation in feed Brix.

We actually developed three models, or PCS blocks in the controller (as shown on the schematic).

FE: What kind of process-control system did QuickStudy integrate with?

Milla: The basic architecture consists of an Allen-Bradley PLC-5 linked via Ethernet to a Wonderware operator interface running on a Windows NT-based PC. We installed QuickStudy on another Windows NT system to collect data directly from the PLC. Then we created a "heartbeat" between the two systems so we could switch between them and thus preserve distributed control.

FE: What were the results of the test, and how did they compare to a conventional control system?

Milla: A conventional process typically controls variability in the final concentration to plus-or-minus 1 degree Brix, which is actually pretty good. For our test, the target Brix was 65, and the plant requested we control variability in the final concentration to plus-or-minus 0.5 degrees Brix. We were able to control variability to within 0.1 degree Brix over any two-hour period. We saw some upsets during that time, and the system handled them. Although we didn't measure it, there is the potential to increase throughput through reduced fouling and thus longer production runs between CIP cycles, and also by minimizing the need for reblending. The system also eliminates the need for operator intervention, and allows analysis of evaporator behavior.

FE: Where do you go from here?

Milla: We applied for a patent on the technology last November. Because of declining consumer demand for concentrated orange juice in favor of not-from-concentrate (NFC) juice, however, we weren't able to commercialize this particular application. But technically we were able to show that we can control variability of an evaporation process much better than any previous method. There are some other applications in citrus, such as pasteurization and aseptic processes, that we'd like to investigate. Beyond citrus, there are many opportunities to apply this technology in production-line applications. In my opinion, this technology will replace PID loops. The potential for bottom-line returns is high, especially in high-value processes which have long lag times and consume a lot of energy.