What's New
Industries
Systems
Services
Applications
Library
Industry Links
Training

  

Home / Library

Copy of article published in Food Engineering, December 1998 issue

"It's only a matter of time before MCP will be as common in the control engineer's toolkit as PID is today"...Ian D. Steel, Engineering Manager, Anchor Products Ltd., Hamilton, New Zealand.

Charles E. Morris, Midwest Editor

Food Engineers Apply Predictive Control

Multivariable Model Predictive Control (MPC) is being applied to a growing number of food manufacturing processes.

MPC is predicting process conditions and product-quality characteristics for dairy ingredients, beer, fruit juices, corn starch, snack foods, bakery products and wastewater. In a batch-continuous retort application for canned foods, MPC predicts microbiological lethality when a process deviation occurs and adjusts process time or temperature to compensate for the deviation.

MPC is defined as a multivariable control strategy which employs a mathematical model, embedded in control-system architecture, to predict the future effects of current control efforts. MPC consists of three key elements: 1.) the predictive model; 2.) optimization in a temporal window; and 3.) feedback correction. The controller predicts process behavior and proactively optimizes control.

BrainWave MPC software boosts brewhouse productivity and quality control while reducing steam costs at the Molson brewery in Vancouver, BC. Repeatable setpoints were formerly not attainable on the brew kettle due to the non-linear characteristics of a boiling liquid. Now, model-based predictive/adaptive control maintains a consistent boil height to optimize the brewing process. (Source: Universal Dynamics Technologies)

Process perfecter

The most advanced applications of multivariable MPC known to date in the food industry are at several plants operated by Anchor Products Ltd., the manufacturing arm of the New Zealand Dairy Group (Hamilton, NZ). Anchor operates nine plants with combined capacity to convert up to 27.5 million liters of milk per day into milk powders, milk proteins, cheese and cream products - all for export. Process control at most Anchor plants is based on PLCs (programmable logic controllers), SCADA (supervisory control and data acquisition) packages and relational databases (site historians), providing the basis for advanced process control (APC).

In 1994, Anchor started-up a new milk powder plant built by McClunie Birch Ltd. (New Zealand) and Stork Friesland B.V. (The Netherlands) at Waitoa, NZ, with process control based on Allen-Bradley PLCs and Realflex SCADA. Anchor and MBL/Stork then investigated MPC software packages to optimize the plant's evaporator and dryer systems, selected the Process Perfecter developed by Pavilion Technologies (Austin, TX) and formed an alliance with Pavilion to further develop APC strategies.

Model-free adaptive control at Tri-Valley Growers

Predictive control systems create and apply process models and are thus known as Model Predictive Control (MPC) systems. But a model-free approach to advanced process control has been applied to improving wastewater-evaporation control by the Tri-Valley Growers plant at Madera, CA. The package: CyboCon, a model-free adaptive (MFA) control system developed by CyboSoft, a unit of General Cybernation Group, inc. (Rancho Cordova, CA). Unlike MPC, CyboCon requires no process identification, quantitative process knowledge or controller design. Unlike PID, it requires no complicated manual toning. According to CyboSoft President Dr. George S. Cheng, CyboCon is a rule-based system with a learning algorithm incorporating some neural-network technology but is not a neural-network based system.

The Tri-Valley application was described by Cheng, Qiang Wang of CyboSoft and Steve Smialkowski of Tri-Valley at a conference on dynamic modeling control earlier this year in Vancouver, BC. Tri-Valley's Madera plant is a zero-discharge facility, where wastewater is processed in a three-effect falling-film evaporator which concentrates the solids (sludge). Two main process variables are controlled on the third evaporator: density of outlet water, and evaporator water level. Two variables can be manipulated: inlet and discharge flow rotes. Both affect density and level.

The original control system, which consisted of three single-loop PID controllers, was difficult to start-up and maintain in automatic control. The relationship between density and flow cannot be described by simple dynamic equations, so it is difficult to built a process model for the density loop. Many operating conditions can change input/output relationships, affecting density and flow - a multivariable control problem.

A CyboCon MFA controller replaced two of the PIDs controlling inlet flow rate (density control) and discharge flow rate (level control), while one PID was retained and cascaded with the MFA to regulate level. The control system consists of PLCs for logic and PID control, and two PCs incorporating HMI (human machine interface) software for data acquisition, networking and monitoring. The MFA is installed in one of the PCs.

The new system was commissioned in one eight-hour business day, the authors report. Results: The evaporator can be started-up and maintained in automatic control; process variables reach setpoints quickly with little overshoot; system stability is improved; variation in product density reduced by more than 50 percent.

As reported earlier by Food Engineering (November 1997), Pavilion's Process Perfecter is a neural-network model-based system which extracts historical data from process controls and solves relationships between complex process variables. This new knowledge is then continuously applied to the model in real time to optimize the process. The model is non-linear - i.e., the learning process is continuous, not intermittent - for continuous steady-state optimization of the process.

The three evaporation systems at Waitoa each consist of a mechanical vapor recompression (MVR) falling-film evaporator followed by a thermal-vapor recompression (TVR) falling-film finisher, which typically concentrate milk at 9 percent solids to 50 percent total solids. The concentrate is then atomized in spray dryers and further dried in vibrating fluid beds.

According to Ian D. Steele, engineering director at Anchor Products, the primary control objectives for the evaporation stage are concentrate density and flow rate. The evaporator project at Waitoa involved applying multi-variable model-based controllers to regulate TRV density and balance-tank level on all three evaporators. Variables manipulated are MVR fan speed, TVR steam pressure and evaporator feed flow, while the only disturbance variable is MVR condenser temperature. To date, says Steele, the three Process Perfecters at Waitoa have reduced variability in concentrate density by up to 70 percent. The evaporators are rapidly approaching target uptimes of greater than 95 percent for density control, and have thus minimized energy consumption, increased throughput and improved product quality.

To date, advanced process control at Waitoa has increased capacity of evaporators one and three by 7.4 percent and of evaporator two by 8.2 percent, Steele reported. To date, the project has achieved a Net Present Value (NPV) of $1.6 million, exceeding target NPV of $1 million, and an Internal Rate of Return (IRR) of 77.3 percent, well beyond target IRR of 77.3 percent.

Anchor has also applied Pavilion's Process Insights package to build "soft sensors" known as Virtual OnLine Analyzers (VOAs), said Steele. Although Process Insights does not provide true MPC by itself, the package can be used to build neural-network non-linear steady-state models relating process inputs and outputs. VOAs can be used to predict quality measurements in real time, especially useful where process outputs have long dead times, are sparsely sampled or where lab analysis is time-consuming and expensive. (One example of a VOA, adds Pavilion: real-time moisture-content prediction of corn starch in a flash dryer.)

According to Ross McCowan, general manager/engineering & projects at Anchor Products, Anchor now has four Process Perfecters and approximately 27 more APC projects including Process Perfecter and VOA applications in various stages of completion. In addition to evaporator and spray-dryer applications, these projects include energy generation, cheese manufacturing, nutritional-products manufacturing and wastewater treatment. By applying such technologies "we can potentially counter [New Zealand's] annual milk growth without building a new plant," said McCowan. Incremental capacity improvements "allow deferment or alternate use of capital." Adds Steele: "It's only a matter of time before MPC will be as common in the control engineer's toolkit as PID [proportional, integral, derivative control] is today!"

AseptiCAL simulated center temperatures of the fastest moving particle and its accumulated lethality values in an aseptic process for multiphase foods (Source: FMC Corp.)

Connoisseur

As reported earlier by Food Engineering (November 1997), the Connoisseur adaptive MPC system installed on an evaporator at a Kiwi Dairy plant in New Zealand controls process temperature and pressure, product flow and product density to save $200,000 per year through improved product quality, reduced rework and lower energy costs. Connoisseur was developed by Predictive Control Ltd. (Norwich, Cheshire, UK), part of the Siebe Group. Additional applications include coffee roasting and sugar refining. Predictive Control was recently integrated into Simulation Sciences, Inc. (SIMSCI), another member of the Siebe Group which along with Siebe companies such as APV, Foxboro and Wonderware creates a potent combination for food manufacturing applications.

In October at ISA Expo `98 in Houston, SIMSCI (Brea, CA) introduced Connoisseur Version 14, which incorporates a graphical user interface closely matching the modeling and control-development process; it is more intuitive for control engineers and is implemented in the Java programming language for platform independence. Connoisseur is currently being applied by a major U.S. starch refiner, where deviation of product-moisture content has reportedly improved by a factor of three.

Predictive lethality

The Food Processing Systems Division of FMC Corp. (Madera, CA) has developed models for predicting and correcting on-line the total microbiological lethality (Fo) delivered to food products when a deviation occurs in a sterilization process.

According to Dr. Zhijun Weng, senior research engineer at FMC, food processors using conventional sterilization systems want to optimize product quality while ensuring product safety and minimizing energy consumption. Process temperature deviations can occur, however, when the heating medium (such as saturated steam) fails to supply adequate energy. In a batch sterilizer, the lethality delivered to each container is not uniform; in a continuous sterilizer, each container (in a rotary pressure-cooker/cooler) or container carrier (in a hydrostatic sterilizer) will experience different time/temperature treatments. The goal of thermal process control is to deliver adequate lethality to the product being least processed in the sterilizer under both normal and deviant process conditions.

In November at IEFP in Chicago, FMC introduced three predictive-control systems:

-- NumeriCAL On-Line predictive modeling software for batch retort control. NumeriCAL runs on FMC's LOG-TEC hardware, which corrects process deviations on-line in real time. The controller adjusts process time or temperature to compensate for the temperature-deviation effect on product lethality in the slowest heating zone and/or the fastest cooling zone in the retort. Accepted by FDA and USDA, the system tracks lethality for each batch through every stage of the retort process, manages variable retort temperatures and provides immediate documentation of the process. The model applies maximum thermal credit for come-up and cooling times in calculating Fo value. This reduces heating times, steam consumption, final product temperatures, cooling times and cooling-water consumption to boost throughput and - through less overprocessing - product quality. The controller precisely controls retort conditions to just above minimum temperature and pressure setpoints for the given process. Any variation which could reduce target lethality is alarmed as a deviation, and an equivalent NumeriCAL model is calculated and applied. According to FMC, the system cuts batch-retort processing time by 15 to 30 percent as compared to the widely-accepted Ball Formula. The first commercial application of NumeriCAL On-Line is on an automated batch-continuous can line consisting of eight retorts.

A Virtual On-Line Analyzer (VOA) samples plant historical data to create a predictive model of a product-quality measurement, such as moisture content of milk powder in the drying application shown above. When integrated with Process Perfecter, the model then controls field devices (transmitters and controllers) to bring setpoints closer to product specification. Abbreviations: Pressure controller (PC); temperature controller (TC); pressure transmitter (PT); temperature transmitter (TT); near infra-red sensor (NIR); vibrating fluid-bed dryers (VFB). (Source: Pavilion Technologies)

-- HydroCAL, for hydrostatic sterilizer control. The dynamic model predicts the coldest spot temperature at any time during the hydrostatic process, identifies the product carrier receiving the least lethality and takes corrective action. (Alternatively, the under-processed containers can be isolated and segregated at the unloading station.) According to FMC, the process reduces cook times by up to 25 percent as compared to the Ball method while improving product quality and reducing cooling water costs. Because modeling software and documentation are accepted by FDA and USDA, the system saves time in process filing. FMC has an off-line HydroCAL model at its food lab in Madera where processors can test their products on the system.

-- AseptiCAL mathematical modeling software for aseptic processing of low-acid and high-acid foods with or without particulates. The program has four modules to accommodate 3-D, 2-D and 1-D particulates, as well as homogenous products. AseptiCAL calculates and predicts the temperature of the fastest-moving particle, immediately analyzes deviations from the aseptic process and optimizes aseptic processing time and temperature. For process development, the model calculates the hold-tube length needed for a scheduled Fo value. Simulation with an accurate model is an FDA requirement for filing an aseptic low-acid particulate process, FMC points out.

BrainWave

As reported earlier by Food Engineering (November 1997), the Molson brewery in Vancouver, BC, boosts brew-kettle productivity and quality control while reducing steam costs by applying BrainWave model predictive control software developed by Universal Dynamics Technologies (Vancouver, BC)). At the ISA Expo `98 Universal Dynamics introduced BrainWave with multi-model capability: The system can now apply up to 10 process models to control the same loop. This means it can automatically select a precise model for each production rate, mode of operation or product type, and tightly control loops with varying process dynamics.

At ISA Expo `98, Universal Dynamics received the ISA President's Award for significant innovation.