16 Instrumentation and Control Systems John R King 1.0 INTRODUCTION The widespread use of advanced control and process automation for biochemical applications has been lagging as compared with industries such as refining and petrochemicals whose feedstocks are relatively easy to characterize and whose chemistry is well understood and whose measure- ments are relatively straightforward. Biological processes are extraordinarily complex and subject to con- siderable variability. The reaction kinetics cannot be completely determined in advance in a fermentation process because of variations in the biological properties of the inoculant. Therefore, information regarding the activity of the process must be gathered as the fermentation progresses. Directly measuring all the necessary variables which characterize and govern the competing biochemical reactions, even under optimum laboratory condi- tions, is not yet achievable. Developing mathematical models which can be utilized to infer the biological processes underway from the measurements available, although useful, is still not sufficiently accurate. Add to this the constraints and compromises imposed by the manufacturing process and the task of accurately predicting and controlling the behavior of biological production processes is formidable indeed. 675 676 Fermentation and Biochemical Engineering Handbook The knowledge base in fermentation and biotechnology has expanded at an explosive rate in the past twenty-five years aided in part by the development of sophisticated measurement, analysis and control technology. Much of this research and technology development has progressed to the point where commercialization of many of these products is currently underway. The intent of this chapter is to survey some of the more innovative measurement and control instrumentation and systems available as well as to review the more traditional measurement, control and information analysis technologies currently in use. 2.0 MEASUREMENT TECHNOLOGY Measurements are the key to understanding and therefore controlling any process. As it relates to biochemical engineering, measurement technol- ogy can be separated into three broad categories. These are biological, such as cell growth rate, florescence, and protein synthesis rate; chemical, such as glucose concentration, dissolved oxygen, pH and offgas concentrations of CO,, O,, N,, ethanol, ammonia and various other organic substances; and physical, such as temperature, level, pressure, flow rate and mass. The most prevalent are the physical sensors while the most promising for the field of biotechnology are the biological sensors. One concern when considering measuring biological processes is the maintenance of a sterile environment. This is necessary to prevent foreign organisms from contaminating the process. In-line measurement devices must conform to the AAA Sanitary Standards specifying the exterior surface and materials of construction for the “wetted parts.” Instruments must also be able to withstand steam sterilization which is needed periodically to prevent bacterial buildup. Devices located in process lines should be fitted with sanitary connections to facilitate their removal during extensive clean- in-place and sterilize-in-place operations, Sample ports, used for the removal of a small portion of the contents from the bioreactor for analysis in a laboratory, must be equipped with sterilization systems to ensure organisms are not inadvertently introduced during the removal of a sample. 3.0 BIOSENSORS Biosensors are literally the fusing of biological substrates onto electric circuits. These have long been envisioned as the next generation of analytical Instrumentation and Control Systems 677 sensors measuring specific biomolecular interactions. The basic principle is first to immobilize one of the interacting molecules, the ligand, onto an inert substrate such as a dextran matrix which is bonded (covalently bound) to a metal surface such as gold or platinum. This reaction must then be converted into a measurable signal typically by taking advantage of some transducing phenomenon. Four popular transducing techniques are: Potentiometric or amperometric, where a chemical or biological reaction produces a potential difference or current flow across a pair of electrodes. Enzyme thermistors, where the thermal effect of the chemical or biological reaction is transduced into an electrical resistance change. Optoelectronic, where a chemical or biological reaction evokes a change in light transmission. Electrochemically sensitive transistors whose signal de- pends upon the chemical reactions underway. One example is the research['] to produce a biomedical device which can be implanted into a diabetic to control the flow of insulin by monitoring the glucose level in the blood via an electrochemical reaction. One implant- able glucose sensor, designed by Leland Clark of the Childrens Hospital Research Center in Cleveland, utilizes a microprobe where the outside wall is constructed of glucose-permeable membrane such as cuprophan. Inside, an enzyme which breaks the glucose down to hydrogen peroxide is affixed to an inert substrate. The hydorgen peroxide then passes through an inner membrane, constructed of amaterial such as cellulose acetate, where it reacts with platinum producing a current which is used to monitor the glucose concentration. A commercial example of a biosensor, introduced by Pharmacia Biosensor AB2, is utilizing a photoelectric principle called sufluceplusmon resonance (SPR) for detection of changes in concentration of macromolecu- lar reactants. This principle relates the energy transferred from photons bombarding a thin gold film at the resonant angle of incidence to electrons in the surface of the gold. This loss of energy results in a loss of reflected light at the resonant angle. The resonant angle is affected by changes in the mass concentration in the vicinity of the metal's surface which is directly correlated to the binding and dissociation of interacting molecules. 678 Fermentation and Biochemical Engineering Handbook Pharmacia claims its BIAcore system can provide information on the affinity, specificity, kinetics, multiple binding patterns, and cooperativity of a biochemical interaction on line without the need ofwashing, sample dilution or labeling of a secondary interactant. Their scientists have mapped the epitope specificity patterns of thirty monoclonal antibodies (Mabs) against recombinant core HIV-1 core protein. 4.0 CELL MASS MEASUREMENT The on-line direct measurement of cell mass concentration by using optical density principles promises to dramatically improve the knowledge ofthe metabolic processes underway within a bioreactor. This measurement is most effective on spherical cells such as E. Coli. The measurement technology is packagedin a sterilizable stainless steel probe which is inserted directly into the bioreactor itself via a flange or quick-disconnect mounting (Fig. 1). By comparing the mass over time, cell growth rate can be determined. This measurement can be used in conjunction with metabolic models which employ such physiological parameters as oxygen uptake rate (OUR), carbon dioxide evolution rate (CER) and respiratory quotient (RQ) along with direct measurements such as dissolved oxygen concentration, pH, temperature, and offgas analysis to more precisely control nutrient addition, aeration rate and agitation. Harvest time can be directly determined as can shifts in metabolic pathways possibly indicating the production of an undesirable by-product. Cell mass concentrations of up to 100 grams per liter are directly measured using the optical density probe. In this probe, light of a specific wavelength, created by laser diode or passing normal light through a sapphire crystal, enters a sample chamber containing a representative sample of the bioreactor broth and then passes to optical detection electronics. The density is determined by measuring the amount of light absorbed, compensating for backscatter. Commercial versions such as those manufactured by Cerex, Wedgewood, and Monitec are packaged as stainless steel probes that can be mounted directly into bioreactors ten liters or greater, and offer features such as sample debubblers to eliminate interference from entrained air. Another technique used to determine cell density is spectrophotometric titration which is a laboratory procedure which employs the same basic principles as the probes discussed above. This requires a sample to be withdrawn from the broth during reaction and therefore exposes the batch to contamination. Instrumentation and Control Systems 679 ~~.s C:- ~ 'I). :::.- c,§ ;::;,~u"5- f ~ ~ ~"'5 r/) "'"' ~-~u~ 0 ]~ ~=~ rz . 680 Fermentation and Biochemical Engineering Handbook 5.0 CHEMICAL COMPOSITION The most widely used method for determining chemical composition is chromatography. Several categories have been developed depending upon the species being separated. These include gas chromatography and several varieties of liquid chromatography including low pressure (gel permeation) and high pressure liquid chromatography and thin layer chromatography. The basic principle behind these is the separation ofthe constituents traveling through a porous, sorptive material such as a silica gel. The degree of retardation of each molecular species is based on its particular affinity for the sorbent. Proper selection of the sorbent is the most critical factor in determining separation. Other environmental factors such as temperature and pressure also play a key role. The chemical basis for separation may include adsorption, covalent bonding or pore size of the material. Gas chromatography is used for gases and for liquids with relatively low boiling points. Since many of the constituents in a biochemical reaction are of considerable molecular weight, high pressure liquid chromatography is the most commonly used. Specialized apparatus is needed for performing this analysis since chromatograph pressures can range as high as 10,000 psi. Thin layer chromatography requires no pressure but instead relies on the capillary action of a solvent through a paper-like sheet of sorbent. Each constituent travels a different distance and the constituents are thus separated. Analysis is done manually, typically using various coloring or fluorescing reagents. Gel permeation chromatography utilizes a sorbent bed and depends on gravity to provide the driving force but usually requires a considerable time to effect a separation. All of these analyses are typically performed in a laboratory; therefore they require the removal of samples. As the reaction is conducted in a sterile environment, special precautions and sample removal procedures must be utilized to prevent contaminating the contents of the reactor. 6.0 DISSOLVED OXYGEN Dissolved oxygen is one of the most important indicators in a fermen- tation or bioreactor process. It determines the potential for growth. The measurement of dissolved oxygen is made by a sterilizable probe inserted directly into the aqueous solution ofthe reactor. Two principles of operation Instrumentation and Control Systems 681 are used for this measurement: the first is an electrochemical reaction while the second employs an amperometric (polarographic) principle. The electrochemical approach uses a sterilizable stainless steel probe with a cell face constructed of a material which will enable oxygen to permeate across it and enter the electrochemical chamber which contains two elec- trodes of dissimilar reactants (forming the anode and cathode) immersed in a basic aqueous solution (Fig. 2). The entering oxygen initiates an oxidation reduction reaction which in turn produces an EMF which is amplified into a signal representing the concentration of oxygen in the solution. Figure 2. Sterilizable polarimetric dissolved oxygen probe. (Courtesy of Ingold Elec- trades, Inc., Wilmington,Mass.) 682 Fermentation and Biochemical Engineering Handbook In the amperometric (polarographic) approach, oxygen again perme- ates a diffusion barrier and encounters an electrochemical cell immersed in basic aqueous solution. A potential difference of approximately 1.3 V is maintained between the anode and cathode. As the oxygen encounters the cathode, an electrochemical reaction occurs: 0, + 2H,O + 4e- + 40H- (at cathode) The hydroxyl ion then travels to the anode where it completes the electro- chemical reaction process: 40H- + 0, + 2H20 + 4e- (at anode) The concentration of oxygen is directly proportional to the amount of current passed through the cell. 7.0 EXHAUST GAS ANALYSIS Much can be learned from the exchange of gases in the metabolic processsuchas O,, CO,,N,,NH,,andethanol. Infact, mostofthepredictive analysis is based upon such calculations as oxygen uptake rate, carbon dioxide exchange rate or respiratory quotient. This information is best obtained by a component material balance across the reactor. A key factor in determining this is the analysis ofthe bioreactor offgas and the best method for measuring this is with a mass spectrometer because of its high resolution. Two methods of operation are utilized. These are magnetic deflection and quadrapole. The quadrapole has become the primary commercial system because of its enhanced sensitivity and its ability to filter out all gases but the one being analyzed. Magnetic deflection mass spectrometers inject a gaseous sample into an inlet port, bombard the sample with an electron beam to ionize the particles and pass the sample through a magnetic separator. The charged particles are deflected by the magnet in accordance with its mass-to-energy (or charge) ratio-the greater this ratio, the less the deflection. Detectors are located on the opposing wall of the chamber and are located to correspond to the trajectory of specific components as shown in Fig. 3. As the ionized particles strike the detectors, they generate avoltage proportional to their charge. This information is used to determine the percent concentration of each of the gasses. Instrumentation and Control Systems 683 Figure 3. Magnetic deflection principle. The quadrapole mass spectrometer also employs an electron beam to ionize the particles using the quadrapole instead of a magnet to deflect the path of the particles and filter out all but the specific component to be analyzed. The quadrapole is a set of four similar and parallel rods (see Fig. 4) with opposite rods electrically connected. A radio frequency and dc charge of equal potential, but opposite charge, is applied to each set of the rods. By varying the absolute potential applied to the rods, it is possible to eliminate all ions except those of a specific mass-to-energy ratio. Those ions which successfidly travel the length ofthe rods strike a Faraday plate which releases electrons to the ions thereby generating a measurable change in EMF. For a given component the strength of the signal can be compared to references to determine the concentration. The quadrapole, when used in conjunction with a gas chromatograph to separate the components, can measure a wide range ofgases, typically from 50 to 1000 atomic mass units (amu). As mass spectrometers are relatively expensive, the exhaust gas of three or more bioreactors is typically directed to a single analyzer. This is possible because the offgas analysis is done outside the bioreactors them- selves. However, the multiplexing ofthe streams results in added complexity with regard to sample handling and routing, particularly if concerns of cross contamination need be addressed. The contamination issue is usually handled by placing ultrafilters in the exhaust lines. Care, however, must be taken to 684 Fermentation and Biochemical Engineering Handbook ensure that these filters don’t plug resulting in excessive backpressure. Periodic measurement calibration utilizing reference standards must be sent to the spectrometer to check its calibration. NONRESONANT ION m W IONIZING ELECTRON BUM dc AND rl VOLTAGES Figure 4. Quadrapole principle 8.0 MEASUREMENT OF pH Metabolic processes are typically highly susceptible to even slight changes in pH, and therefore, proper control of this parameter is critical. Precise manipulation of pH can determine the relative yield of the desired species over competing by-products. Deviations ofas little as 0.2 to 0.3 may adversely affect a batch in some cases. Like the cell mass probe and dissolved oxygen probes described earlier, the pH probe (see Fig. 5) is packaged in a sterilizible inert casing with permeable electrode facings for direct insertion into the bioreactor. The measurement principle is the oxidation reduction potential of the hydrogen ion and the electrode materials are selected for that purpose. Instrumentation and Control Systems 685 Figure 5. Ingold sterilizable pH probe. (Courtesy ofIngo[d E[ectrodes, Inc., Wi[mington, Mass.) 9.0 WATER PURITY Water purity is often very important in biochemical processes. One of the best methods to detect the presence of salts or other electrolytic materials is to measure its resistivity .Conductivity or resistivity probes are capable of measuring conductivities as high as 20,000 microsiemens per centimeter and resistivities as high as 20 megoluns per centimeter . 686 Fermentation and Biochemical Engineering Handbook 10.0 TEMPERATURE Precise temperature control and profiling are key factors in promoting biomass growth and controlling yield. Temperature is one of the more traditional measurements in bioreactors so there is quite a variety of techniques. Filled thermal systems, Fig. 6, are among the more traditional tempera- ture measuring devices. Their operating principle is to take advantage of the coefficient of thermal expansion of a sealed fluid to transduce temperature into pressure or movement. This has the advantage of requiring essentially no power and therefore is very popular in mechanical or pneumatic control loops. Although the trend in control is toward digital electronic, pneumatic and mechanical systems are still very popular in areas where solvent or other combustible gases may be present and therefore represent a potential safety hazard. The primary constraint in these types of systems is that the receiver (indicator, recorder, controller) must be in close proximity to the sensor. Figure 6. Filled thennal system assembly for temperature measurement. (Courtesy of the Foxboro Co., Foxboro, Mass.) Instrumentation and Control Systems 687 Thermocouple assemblies, Fig. 7, are a popular measurement choice in electronic systems or in pneumatics where the sensor must be remote. The thermoelectric principle, referred to as the Seebeck Effect, is that two dissimilar metals, when formed into a closed circuit, generate an electromo- tive force when thejunction points ofthe metals are at different temperatures. This conversion of thermal energy to electric energy generates an electric current. Therefore, if the temperature of one juncture point (the cold junction) is known, the temperature of the hot juncture point is determined by the current flow through the circuit. Depending upon the alloys chosen, thermocouples can measure a wide temperature range (-200 to +350"C for copper, constantan) and are quite fast acting assuming the assembly doesn't contribute too much lag in its absorbance and dissipation ofheat. Its primary disadvantages are its lack of sensitivity (copper, constantan generates only 40.5 microvolts per "C) and requirement for a precise cold junction temperature reading. COYER - HEAD ," I WELL SENSITIVE TIP W -- RTD 3 Figure 7. Cutaway view of thermocouple or resistance temperature detector probe for temperature measurement. (Courtesy of the Foxboro Co., Foxboro, Mass.) 688 Fermentation and Biochemical Engineering Handbook Resistance temperature detectors, RTD’s, are more sensitive than thermocouples especially when measuring small temperature ranges. As a result, they are preferred for accurate and precise measurements. The principle behind these devices is based on the use of materials, such as platinum or nickel, whose resistance to current flow changes with tempera- ture. These materials are used as one leg in a Wheatstone bridge circuit with the other legs being known precision resistors. A voltage is applied across the bridge and the voltage drop midway through each path of the circuit is compared. The potential difference at the midway point is directly related to the ratio ofeach set of resistances in series. Since three ofthese are known, the resistance of the RTD can be calculated and the temperature inferred. If the RTD is remote from the bridge circuit, the resistance of lead wires can affect the measurement. Therefore, for highly precise measurements, compensating circuits are included which require increasing the wiring for this measuring device from two to as many as four leads. Thermistors are a special class of RTD’s and are constructed from semiconductor material. Their primary advantage is their greater sensitivity to changes in temperature, therefore making them a more precise measuring method. Their disadvantage is their nonlinear response to temperature changes. This form of RTD is gaining popularity for narrow range applications, particularly in laboratory environments. 11 .O PRESSURE Pressure is an important controlled variable. The measurement is obtained by exposing a diaphragm surface or seal to the process via a flange or threaded tap through the vessel wall. The signal is translated through a filled capillary to a measurement capsule which will transduce the signal to one measurable by an electronic circuit by one of several methods. One method is to employ a piezoelectric phenomenon whereby the pressure exerted on an asymmetric crystal creates an elastic deformity which in turn causes the flow of an electric charge. A second technology is variable resistance whereby flexure on a semiconductive wafer affects its resistivity which is measured in a similar fashion to RTD’s. The third, shown in Fig. 8, is the use of a vibrating wire where changes in the tension of the wire changes its resonant frequency which is measured as a change in pulse rate. Instrumentation and Control Systems 689 TOOSClLUTOR (ENSOR BODY CIRCUIT / OUTPUT \L FILL FLUID Figure 8. Diagram of resonant wire technology pressure measurement. (Courtesy ofthe Foxboro Co., Foxboro, Mass.) Several types of pressure measurements can be taken. These include absolute pressure, where one side ofthe capsule is exposed to 0 psia in a sealed chamber. Gauge pressure is measured with one side of the capsule vented to atmosphere. Vapor pressure transmitter seals one side of the capsule, filling it with the chemical composition of the vapor to be measured. The vapor pressure in the sealed chamber is compared with the process pressure (at the same temperature). Ifequal, the compositions are inferred to be equal. This technique is used primarily for binary mixtures as multicomponent compo- sitions have too many degrees of freedom. 12.0 MASS Weigh cells or load cells are typically used to measure the mass of the contents of a vessel. These are electromechanical devices which convert force or weight into an electrical signal. The technique is to construct a Wheatstone bridge similar to that used in the RTD circuit with one resistor being a rheostat which changes resistance based on load. Three configurations are popular. These are the column, where the cell is interposed between one leg of the vessel and the ground (see Fig. 9) and is 690 Fermentation and Biochemical Engineering Handbook typically used for weights exceeding 5000 pounds. The second is the cantilever design, where the weight is applied to a bending bar and is used for weights under 500 pounds. The third is the shear design, where the weight is applied to the center of a dual strain gage arrangement. HIGH ACCURACY FULLY SUPPORTED LOAD CELL 8 FIXED MOUNTING \ PLATES Figure 9. Schematic of the installation of a load cell. 13.0 MASS FLOW RATE A Coriolis meter utilizes a measurement technology which is capable of directly measuring mass flow (instead of mferring mass flow from volumetric flow and density). The Coriolis effect is the subtle correction to the path of moving objects to compensate for the rotation of the earth. This appears as a force exerted perpendicular to the direction ofmotion and creates a counterclockwise rotation in the Northern Hemisphere and a clockwise rotation in the Southern Hemisphere. This phenomenon is used by the mass flow meter to create a vibration whose frequency is proportional to the mass of the fluid flowing through the meter. This is accomplished via the geometry of the meter (Fig. IO), specifically the bends to which the fluid is subjected IC it trmdc thrnlloh the meter Instrumentation and Control Systems 691 Figure 10. Schematic of Coriolis Meter flowpath. (Courtesy of the Foxboro Co., Foxboro, Mass.) 14.0 VOLUMETRIC FLOW RATE Quite a number of technologies are available for measuring volumetric flow rates. These include differential pressure transmitters, vortex meters and magnetic flow meters. Each has its advantages and disadvantages. The differential pressure transmitter is the most popular and has been in use the longest. Its measurement principle is quite simple. Create a restriction in the line with an orifice plate and measure the pressure drop across the restriction. The measurement takes advantage of the physical relationship between pressure drop and flow. That is, the fluid velocity is proportional to the square root of the pressure drop, and in turbulent flow, the volumetric flow rate is essentially the velocity of the fluid multiplied by the cross-sectional area of the pipe (Fig. 11). 692 Fermentation and Biochemical Engineering Handbook TRANSMITTER 3-VALVE BYPASS MANIFOLD IOPTIONl \ ORIFICE U-0ENO ASSEM0LY Figure 11. Integral flow orifice assembly, U-bend configuration. (Courtey of Foxboro Co., Foxboro, Mass.) Inaccuracies with regard to transmitting the pressures between the sensor and transducer occur at very low flow rates, therefore closely coupled units have been designed for this purpose. Using this approach and small bore orifice plates, extremely low flows can be measured. A 0.38 millimeter diameter bore can accurately measure flows in the 0.02 liters per minute range for liquids and 0.03 cubic meters per hour for gases. Jeweled orifice plates can have a bore as small as 0.05 millimeters in diameter. The primary disadvantages of the differential pressure producing flow measurements are the permanent pressure drop caused by the restriction in the line; sediment buildup behind the orifice plate (which could be a source of bacterial buildup) and loss of accuracy over time as the edge ofthe plate is worn by passing fluid and sediment. This type of transmitter typically has a limited range (turndown)-usually a 4 to 1 ratio between its maximum and minimum accurate flow rates. Vortex meters utilize a precision constructed bar or bluff through the diameter ofthe flow path to create a disruption in flow which manifests itself as eddy currents or vortices being generated, starting at the downstream side of the bar (Fig. 12). The frequency at which the vortices are created are directly proportional to velocity of the fluid. Although these devices contain a line obstruction, the turbulence created by the vortices make the bluff self- cleaning and they are available for sanitary applications. Also, their linear nature makes them a wide-range device with a ratio of as much as 20: 1 between the maximum and minimum flow rate. Line sizes as small as 1" are available which are capable of reading flow rates as low as 0.135 liters per minute. Instrumentation and Control Systems 693 HIGH VELOCITY FLUID FLUID - LHEAR, LAYER ALTERNATE VOATICES METER BC Figure 12. Vortex creation via shedding bluff, (Courtesy ofFoxboro Co., Foxboro, Mass.) Magnetic flowmeters take advantage of the electrolytes in an aqueous solution to induce a magnetic field in the coils surrounding the meter's flowtube, see Fig. 13. The faster the flow rate, the greater the induced field. Interestingly, the ionic strength of the electrolytes has only negligible effect on the induced field so long as it is above the threshold value of 2 microsiemens per centimeter. Because these meters create no obstructions to the flow path they are the preferred meter for sanitary applications. 15.0 BROTH LEVEL As the broth in a fermenter or bioreactor becomes more viscous and is subjected to agitation from sparging (the introduction of tiny sterilized air bubbles at the bottom of the liquid) and from mixing by the impeller, it has a tendency to foam. This can be a serious problem as the level may rise to the point where it enters the exhaust gas lines clogging the ultrafilters and possibly jeopardizing the sterile environment within the reactor. Various antifoam strategies can be employed to correct this situation, however, detection of the condition is first required. 694 Fermentation and Biochemical Engineering Handbook DIMENSIONS-NOMINAL 8OOH.SCR TO 8OIH.SCR SANITARY, CERAMIC.LINED FLOWTUBES SIGNALCABLE Figure 13. Cutaway schematic of a sanitary magnetic flowmeter. (Courtesy of Foxboro Co., Foxboro. Mass.) Capacitance probes (Fig. 14) are one means to accomplish this. The basic principle is to measure the charge between two conductive surfaces maintained at different voltage potentials and separated by a dielectric material. The construction of the probe provides an electrode in the center surrounded by an insulator, air, and a conductive shell. The length of the probe is from the top of the reactor to the lowest level measuring point. As the level in the reactor rises the broth displaces the air between the capacitance plates and thereby changes the dielectric constant between the plates to the level ofthe broth. The result is a change in the charge on the plate. Ifthe vessel wall can act as a plate (is sufficiently conductive), the preferred approach would be to use an unshielded probe (inner electrode with insulator) to prevent erroneous readings resulting from fouling of the probe. Because of the uncertain dielectric character of the broth, this measurement should only be used as a gross approximation of level for instituting antifoaming strategies. Instrumentation and Control Systems 695 Figure 14. Installation schematic of a capacitance probe in a vessel. Several other forms of level measurement technologies are available. One is the float and cable system, where the buoyancy of the float determines the air-broth interface boundary and the length of the cable determines the level. The density of the broth may render this measurement questionable. A second is hydrostatic tank gauging, where level is inferred from pressure. Again, density, particularly if two phases exist (aqueous and foam), may render this approach questionable. A third is sonic, which computes the distance from the device to the broth surface based on the time it takes for the sound wave initiating from the device to reflect off the surface of the air-liquid boundary and return. Several other ingenious variations of these basic approaches are commercially available as well. 696 Fermentation and Biochemical Engineering Handbook 16.0 REGULATORY CONTROL Automatic regulatory control systems (Fig. 15) have been in use in the process industries for over fifty years. Utilizing simple feedback principles, measurements were driven to- ward their setpoints by manipulating a controlled variable such as flow rate through actuators like throttling control valves. Through successive refine- ments in first mechanical, then pneumatic, then electronic and finally digital electronic systems, control theory and practice has progressed to a highly sophisticated state. r I ltf0 1 AIF L *ClO BAS! IUPUl , Figure 15. Typical instrument configuration around a fermenter. Instrumentation and Control Systems 697 16.1 Single Stage Control The fundamental building block has been the proportional plus integral plus derivative (PID) controller whereby the proportional term would adjust the manipulated variable to correct for a deviation between measurement and target or setpoint; the integral term would continue the action of the proportional term over time until the measurement reached the setpoint and the derivative term would compensate for lags in the action in the measure- ment in responding to actions of the manipulated variable. The classic equation is: m = 1OOIPB (e + 1IR edt - D deldt) Judicious application of this control strategy on essentially linear single variable control systems which don't exhibit a prolonged delay (dead time) between action by the manipulated variable and measured response by the controlled variable has proven quite effective. Fortunately most single loop control systems exhibit this behavior. In highly nonlinear applications such as pH control, or in situations where the dynamics of the process change over time as occurs in many chemical reactions, adjustments to the tuning coefficients are needed to adequately control the modified process dynamics. Self-tuning controllers employing expert rule sets for dynamic retuning the PID settings are available for this class of problem. These are also used by many users to determine the optimum settings for the linear systems described above. One such rule system is the EXACT controller by Foxboro (Fig. 16), which automatically adjusts the controller tuning parameters based on the pattern of the measure- ment signal received. When the process under control exhibits significant dead time, the problem is considerably more difficult. One approach is to use a simple model-based predictor corrector algorithm such as the Smith predictor[l0I which is interposed between the manipulated and controlled variable in parallel with a conventional controller and conditions the measurement signal to the controller based on time conditioned changes to the manipulated variable made by the controller. This works exceedingly well if properly tuned, but is sensitive to changes in process dynamics. Another scheme, introduced by Shinsky["] recently, utilizes a standard PID controller with a dead time function added to the external reset feedback portion of the loop. This appears to be less sensitive to changes in process conditions. 698 Fermentation and Biochemical Engineering Handbook Figure 16. Model 761 Controller with EXACT tWling. (Courtesy of the Foxboro Co. Foxboro, Mass.) 17.0 DYNAMIC MODELING A control system which anticipates adjustments to the manipulated variables based on changes to one or more controlled variables can be constructed by combining single station controllers with signal characterizers, dynamic compensators and computational elements such as summers and multipliers. Simpler implementations, such as cascade control, will minimize the effect of a deviation of a controlled variable from its target value while dynamic models will anticipate changes to process conditions and adjust the control strategy to compensate based on a leading indicator. A simple example would be the effect on the draw rate and energy input to a distillation column based on a change to its feed rate. The dynamic model in this case would be a material and energy balance around the column compensating for the time delays encountered on each tray as the increased flow rate works its way through the column. Instrumentation and Control Systems 699 18.0 MULTIVARIABLE CONTROL Characterizing a process as a set of nonlinear time dependent equations and then developing a strategy which manipulates sets of outputs based on changes to the inputs is another approach gaining momentum in other industries such as petroleum refining. One approach is called Dynamic Matrix Control['2] (DMC) which first automates the process of determining the coefficients for the set of nonlinear equations based on sets of controlled and manipulated variables declared. The method perturbs each of the manipulated variables and determines the corresponding response of the controlled variables. Once the model is constructed, the information is represented in a relative gain matrix to predict the control actions necessary to correct for changing process conditions. Once the DMC is correctly tuned, including dynamic compensations, a predictor corrector algorithm is applied to compensate to changes in the process dynamics over time. This technique has been applied quite successfully to reaction pro- cesses in the petroleum industry including fluid catalytic cracking units and catalytic reformers. 18.1 Batch Control Butch is a general term given to a diverse set of time dependent control State variable control, such as the opening and closing of a solenoid or the starting and stopping of a motor, including the use of any timing circuits which may be used for alarming in the event the action doesn't achieve its specified results in the allotted time. The interlocking, sequencing or coordinating of systems of devices to ensure their proper and coordinated opera- tion. Examples include interlocking a discharge pump to the opening of the discharge valve and the alignment of pumps and valves to transfer materials from one vessel to another. This may include actions such as the resetting and starting of totalizers to ensure the proper amount of material was successfully transferred. strategies including: 700 Fermentation and Biochemical Engineering Handbook The modification of selected process variables in accor- dance with a prespecified time-variable profile. Two examples are the changing ofthe reactor temperature over time to conform with a specified profile or the timed periodic addition of nutrient into the bioreactor. Conducting event driven actions such as adding antifoam upon the detection of excess foam or invoking an emer- gency shut down routine if an exothermic reaction goes beyond controllable limits. Performing a sequence of operations in a coordinated manner to produce the desired changes to the contents of a process unit. This would typically include combinations of the above mentioned activities on various sets of equipment associated with the unit. The Instrument Society of America Committee Group SP88, Batch Control Structure, is drafting a specification which decomposes batch control into a hierarchal set of activities each with their own purview and problem definition. The objective is to define the properties of the control problem at each level and identify conceptually the appropriate control and information management tools needed for each level. Once defined, a building block approach is taken whereby successively higher levels rely on the foundation established by the controls implemented at the lower levels. A strategy directed at the operation of a reflux condenser would rely on the definitions already in place for throttling flow to achieve proper temperature control and would merely direct the devices (such as PID controllers) as to the actions required. This hierarchy is currently depicted[14] as: Loop/Device, Element Level, which deals with the real- time devices which interface directly with the process. Equipment Module Level, which utilizes combinations of loops and devices to manage an equipment fbnction such as a reflux condenser within a reactor. Unit Level, which coordinates the equipment modules to manage the process unit. Train/Line Level, which coordinates a set of units to manufacture a batch of specified product. Instrumentation and Control Systems 701 Area Level, which coordinates the manufacture of sets of products being made at the train/line level so as to ensure adequate availability of resources and the optimum utili- zation of capital equipment. Plant Level is the integration of the manufacturing pro- cess with other plant functions such as accounting, quality control, inventory management, purchasing, etc. Corporate Level is the coordination of various plants to ensure a proper manufacturing balance with market needs and financial goals. 19.0 ARTIFICIAL INTELLIGENCE A considerable amount of attention is being given to the use of various forms of artificial intelligence for the control of bioreactor systems. Two forms of systems are currently being explored. These are expert systems and neural networks. Expert systems combine stored knowledge and rules about a process with inference engines (forward and backward chaining algo- rithms) to choose a best or most reasonable approach among a large number of choices when no correct answer can be deduced and in some situations the information may appear to be contradictory. Neural networks are also being seriously explored for certain classes of optimization applications. These employ parallel solution techniques which are patterned after the way the human brain functions. Statistical routines and back propagation algorithms are used to force closure on a set of cross linked circuits (equations). Weighting functions are applied at each of the intersections. The primary advantage for using neural networks is that no model of the problem is required (some tuning of the weighting functions may facilitate “learning”, however). The user merely furnishes the system with cause and effect data which the program uses to learn the relationships and thereby model the process from the data. Given an objective function, it can assist in the selection of changes to the causes (manipulated variables) to achieve the optimum results or effects (controlled variables). At BPEC, the Engineering Research Center of Excellence at MIT, advanced computer control of bioprocesses is being researched with an eye 702 Fermentation and Biochemical Engineering Handbook toward industrial commercialization. Professor Charles Cooney has directed the effort to develop expert systems and artificial neural networks to achieve this goal. One of the products resulting from this effort is the Bioprocess Expert developed by Dr. Gregory O’Connor, President of Bioprocess Automation, Inc. in Cambridge. This uses an expert system called G2 from Gensym Corporation, also located in Cambridge. 20.0 DISTRIBUTED CONTROL SYSTEMS As the knowledge of the physiology and reaction kinetics of biochemi- cal processes has progressed and the measurement systems for monitoring their activity has improved, the need for sophisticated systems able to execute coordinated control strategies including batch has increased. Fortunately the state of the art of control systems has rapidly evolved to the point where all of the control strategies described above can be embodied in a Distributed Control System (DCS), see Fig. 17. This transformation has been facilitated to a great extent by the technology breakthroughs in computer, communica- tions, and software technology. Distributed control systems are organized into five subsystems. Process interface, which is responsible for the collection of process data from measurement instruments and the issuing of signals to actuating devices such as pumps, motors and valves. Process control, which is responsible for translating the information collected from the process interface subsys- tem and determining the signals to be sent to the process interface subsystem based on preprogrammed algorithms and rules set in its memory. Process operations, which is responsible for communicat- ing with operations personnel at all levels including operator displays, alarms, trends of process variables and activities, summary reports, and operational instructions and guidelines. It also tracks process operations and product batch lots. Instrumentation and Control Systems 703 3 a 3 Lb 704 Fermentation and Biochemical Engineering Handbook Applications engines, which are the repository for all of the programs and packages for the system from control, display and report configuration tools to program lan- guage compilers and program libraries to specialized packages such as database managers, spreadsheets and optimization or expert system packages to repositories for archived process information. Communications subsystems, which enable information flow between the various DCS subsystems as well as to other computerized systems such as laboratory informa- tion management systems (LIMS); plant inventory man- agement and scheduling systems such as MRP 11; plant maintenance systems and business systems. The integration ofthese systems into acohesive whole has dramatically increased the level of automation possible to improve the quality, productivity and economics of manufacturing. REFERENCES 1. de Young, H. G., Biosensors, High Technologv Magazine, pp. 41-49 (November, 1983) 2. Evans, S., Genetic Engineering News, pp. 2, 20 (February, 1991) 3. CEREX Brochure, Cerex announces MAXimum Accuracy in cell mass measurement, CEREX Corporation, Gaithersburg, h4D. 4. Wedgewood Technology Specification Sheet, Model 650 Cell Growth Probe, Wedgewood Technology, San Carlos, CA. 5. Foxboro Bulletin F-20B, Temperature Measurement and Control Systems, (July, 1982) 6. Considine, D., Process Instruments and Control Systems, 3rd. ed., McGraw Hill (1 985) 7. Miller, R. W., Flow Measurement Engineering Handbook, McGraw Hill (1 983) 8. Foxboro Instrument Catalog 583E (October, 1991) 9. Vogel, H. C., Fermentation and Biochemical Engineering Handbook, 1st ed., Noyes Publications, Park Ridge, NJ (1 983) 10. Smith, 0. J. M., A Controller to Overcome Dead Time, ISA Journal (February, 1959) 1 1. Shinsky, F. G., Model predictors: The first smart controllers, Instruments and Control Systems (September, 1991) Instrumentation and Control Systems 705 12. Cutler, C. R. and Ramaker, B. L., Dynamic Matrix Control - A Computer Control Algorithm, Paper #5 1 b, AIChE 86'" Meeting (April, 1979) 13. Meeting Minutes. ISA/SP88/WGl--l990--2. Draft Specification ISA- dS88.01, Attachment 3, Batch Control Systems Models and Terminology, Draft 1 (October, 1990) 14. Gensym Real Times, Vol. 1, Gensymk Joins MIT's Biotechnology Consor- tium, Gensym Corporation (Winter, 1991)