Control of Continuous Flow Chemical Reactors: - CPAC - University ...

Control of ContinuousFlow Chemical Reactors:Using unique approaches to sampling,sensing and data handlingMichael F. Roberto, Thomas I. Dearing, and Brian J. MarquardtApplied Physics LaboratoryUniversity of WashingtonSeattle, WA 98105U.S. Food andDrug Administration

Continuous Flow Chemistry - Approachn Design and develop CF reactor system that includessmart sampling and real-time analytics for thepurpose of quickly and efficiently developing,optimizing and controlling new chemistriesn Focus on taking a “systems approach” to developingand controlling chemistry in continuous flow¨ Define analytical tools – information, speed and cost¨ Define smart sampling and control – smart valves, flowcomponents, reactors, temperature control,….¨ Software and hardware interfaces for automated control ofchemical process/systemn Development – automated design of experimentsn Production – feedback or feed forward control models for QA/QC2

Analytical Control System DesignOptimizationFeedback& ControlProcessModelingData FusionAnalytics and Data HandlingHardware Control and Monitoring(Software)Hardware – NeSSI, Raman,Other PAT, Reactor

Reactor Hardware and Analyticsn Corning CF Reactorn Kaiser 4-channelRaman systemn Mettler Toledo IRn On-line analytics¨ Pressure¨ Temperature¨ Flown NeSSI SamplingSystem

Sampling Systems• NeSSI Components• Industry-driven effort todefine and promote a newstandardized alternative tosample conditioning systemsfor analyzers and sensors Standard fluidic interface formodular surface-mountcomponentsISA SP76 Standard wiring andcommunications interfaces Standard platform formicro analytics

What does NeSSI Providen Simple “Lego-like” assembly¨ Easy to re-configure¨ No special tools or skills requiredn Standardized flow components¨ “Mix-and-match” compatibility between vendors¨ Growing list of componentsn Standardized electrical and communication (Gen II)¨ “Plug-and-play” integration of multiple devices¨ Simplified interface for programmatic I/O and controln Advanced analytics (Gen III)¨ Micro-analyzers¨ Integrated analysis or “smart” systems

NeSSI Sampling System for ReactorRaman Probemonitorcleanbypass

System Control and AnalyticsHPLC Pump(Flow Rate)Back PressureRegulatorFlow MeterPressureGaugeThermocouple Raman Other PATControlAnalytics• Reactant 1HPLC Pump(Flow Rate)Back PressureRegulatorFlow MeterPressureGaugeThermocouple Raman Other PAT• Reactant 2TemperatureControlFlow MeterPressureGaugeThermocoupleRaman Other PAT Needle Valve• Product

NeSSI Reagent Sampling SystemNeSSI Product Sampling Sys.

Reaction Procedure and Setupn Chemistry performed in reactorn On-line Raman monitoring onend of product linen Each variable independentlycontrolled in DoE to enableunderstanding of each criticalprocess parameter¨ Computer control of flow andtemperature¨ Catalyst concentration controlledmanually10

Corning ® Advanced-Flow TM LFLow-Flow Capability (1-10 mL/min)Corning has introduced a reducedflow-rate reactor that retains theoutstanding mixing and heatexchange performance of itsAdvance-Flow TM glass reactorswhile providing:§ Low internal volume (2 mL flow)§ High flexibility§ Metal-free reaction path§ Scalability§ Compatibility with analytics§ T, Flow and Pumping control

NeSSI Raman BallprobeBallprobe Specs.• Hastelloy C-276Ti, SS, Monel• Sapphire optic• Std. temp range:-80 > 400° C• Pressure:0-350 Bar12Matrix Solutions: www.ballprobe.com

Full Reactor System w/Control2.5 ft

Initial System DemonstrationControl of Esterification14

Chemical Reactionn Esterification of benzoic acidn Straight-forward, pharmaceutically relevantn Strong dependence on temperature, flow ratesWiles, C., Watts, P. 2009

Temperature vs. Reaction Progressn Clear correlationbetween temperatureand product formationn Reaction progress notlinearly correlatedwith temperature¨ Expected for thisreactionn Similar results whenflow and catalystconcentration varyTemperature (C)Raman PC1 Scores (AU)16015014013012011010090807060Temperature (C)Raman PC1 Scores60° C 100° C 140° C0 20 40 60 80 100 120 140 160Time (min)

Temperature vs. Reaction Progress

Flow Rate vs. Reaction Progressn Slower flow rate(longer residencetime) correlated withhigher yieldn Too slow results inunsteady yield frominefficient mixing(outside reactorspecifications)Scores on PC 1 (64.43%)Samples/Scores Plot of Raman Data320.2 mL/min8 mL/min4 mL/min 2 mL/min 1 mL/min 0.5 mL/min10-1-2-350 100 150 200 250 300 3504 x 104 Samplen Boundaries defined:8 mL/min & 1 mL/min18

Catalyst Conc. Vs. Reaction Progressn Correlation betweenconcentration andyield of productn Little variation in yieldbelow 0.275Mn 2.2M solution resultedin significant increasein yield, but wasabove reactorspecifications19

Reaction Monitoring with Ramann Spectrum cropped toregion of interestbelow, with a 1 st orderpolyfit appliedn Standards of0% to 100%productn Responsivein RamanRaman Signal (A.U.)54.543.532.521.510.5x 10 4Raman Signal (A.U.)120001000080006000400020000ProductReagent720 740 760 780 800 820 840 860Raman Shift (cm^-^1)0400 600 800 1000 1200 1400 1600 1800Raman Shift (cm^-^1)

PLS Raman Validation ModelYield Predicted by Raman (%)100908070605040PLS Model - Raman Yield Prediction vs. HPLCHigh Catalyst ConcentrationR 2 = 0.9992 Latent VariablesRMSEC = 0.91141RMSEP = 2.1617Calibration Bias = 0Prediction Bias = 0High T, Low R tHigh T, High R t30Y Predicted 120CalibrationTest10Low T, High R t1:10Low T, Low R tfit0 20 40 60 80 100Yield Measured by HPLC (%)

Real-Time Process Understandingn Real-Time monitoring requires new softwaredeployment:¨ Integration of fast on-line spectroscopy¨ Chemical yield information must be within controlsoftwaren Design towards future control implementations¨ Maintain continuous monitoringn Early warning signs for process upsetsn Continuity in modeling univariate parametersn Rapid analysis allows for timely decision making¨ Typical Raman analysis time = 20 seconds22

Real-Time Quantitative Yieldn Able to monitorreaction throughthree designpoints (flow rates)n Identify whensystem isequilibrated1 mL/min2 mL/min4 mL/minManuscript in Preparation23

100PLS Model - Raman Yield Prediction vs. HPLCReal-Time Quantitative AnalysisYield Predicted by Raman (%)908070605040302010Validation of Real-Time Yield Model00 20 40 60 80 100Yield Measured by HPLC (%)1 mL/min2 mL/min4 mL/minManuscript in Preparation24

Real-Time Quantitative Analysisn Rapid online quantitative real-time analysisconclusively demonstrated with Ramanspectroscopy¨ 20 s from start of spectrum acquisition to chemicalconversion informationn Less than 1 second to model spectrumn ~19 seconds Raman spectral acquisition¨ Continuous information output with no moving parts,operators, or supplies25

Current Studiesn Expand Chemistry¨ Working with an industrial collaborator to optimize anew, multistep pharmaceutical reaction for continuousproduction¨ Investigating continuous solvent exchangen Automated design of experiments¨ Automate generation of response surfaces¨ Derive reaction kinetics from spectroscopy¨ Produce generic optimization and control strategyn Implement modeling control strategy through theuse of PID control theory26

Moffat Swern Oxidation of Alcoholn Cryogenic reaction requires -70°C in batch¨ Very exothermic, with unstable intermediatesn Moffat Swern oxidation has been performed incontinuous microliter scale previously¨ All chemistry analyzed offline – no controln Low-temperature reaction in batch, can beperformed at up to room temperature in CFRn 4 feeds in 3 reactors - preparation of oxidizingagent, oxidation, quench

Moffat Swern Oxidation of Alcohol28

Experimental Batch - Developmentn Reactants were precooledusing Corningreactor plate to -20°Cn Continuous addition toensure low tempn Maintaining -70°C bathrequired regular dry iceadditionn Vessel included 3 probesplus port for addition ofchemicals29

Infrared Observationsn Collection timecompared to Raman¨ 4 Raman spectra forevery 1 IR collectionn Temperature exhibitsa large role in thevariation of spectra¨ Preprocessing appliedto spectra to reducethis effectn EMSCSignal Intensity10.80.60.40.201800Raw IR Data1600 1400 1200 1000- 1Wavenumber cm800n Baseline correction

IR Reaction Progressn Reaction progress:start to finishn Clear change spectralprofile as reactionprogresses¨ Formation of peak at1680 wavenumbers¨ Generation ofintermediate at 1750

3D Representation of IR Data0.30.2Intensity0.10-0.11900180017004060801001600Wavenumber cm -11500020TimeSame region in 3D time-space, shows generation of intermediate

Infrared Conclusionsn Spectral changes apparent¨ Clear intermediate formationn Separating rate of formation of product (whichoccurs during warming of reactor) is difficultn Data fusion with Raman spectra (up next) maygive improved resultsn Will not be an issue in continuous flow¨ All acquisitions will be at the same temperatureallowing for direct comparisons33

Raman setup and settingsnRaman¨ 785nm Excitation wavelength¨ Raman 1/8” ball probe¨ 1 s Exposure time, 5 acquisitionsx 10 4108Intensity6420500 1000 1500Raman Shift (cm-1)

Pre-processed Raman Data7000Reaction progress: start to finish60005000Intensity40003000200010000800 900 1000 1100 1200 1300 1400 1500 1600 1700Raman Shift (cm-1)

Region 1560-1720 cm -1 (blue region)SelectivePeak forintermediateSelectivepeaks forProduct

Raman Reaction PlotRAMAN SCORE PLOT (region 1560 to 1720 cm -1 ) and reaction temerature8000406000Score PC1 81.69% (Product)Score PC2 18.07 % (Intermediate)Temperature20Scores400020000Product starts forming quicly whentemperature reaches about -50 CIntermediate formedas Alcohol is added0-20-40Temperature (C)-2000-60TFAA additionDMSO additionAlcohol additionTEA addition-40000 10 20 30 40 50 60 70 80 -80Time (min)

Raman Validated Against HPLC8000RAMAN SCORE PLOT(region 1560 to 1720 cm -1 )and HPLC reference data10090600080Scores40002000PC1PC2AlcoholProductIntermediate7060504030HPLC determined Yield (%)02010-200060 65 70 75 80 0Time (min)

Concluding Remarks onSpectroscopic AnalysisnnnnInfrared shows much promise – CF and Data Fusion¨ More heavily impacted by temperature variation¨ Further work needed to separate temperature effectsnIn continuous flow, this will not be an issue.Raman is very promising for quantitative measurement of productionyield in batch¨ Good selective regions¨ No fluorescence¨ Minimally affected by temperature¨ Capable of monitoring side productsRaman will work well in a continuous setup¨ Multiple probes, multiplexing and very informativeSuccess of PCA modeling of spectral data indicates that HPLC willbe required only for model construction and periodic maintenance39

Further Workn PLS Calibration Model¨ Quantitative informationn Data Fusion¨ Both infrared and Ramanshown to be viable in batch¨ Further analysis of IR –separate temperatureeffects¨ Addition of T, P, RI andflow to fusion modelsn Complete the new reactorsystem build40

Swern Reactor w/sampling + Analytics41

Sampling System and Analytics42

Data Fusionn There is no one perfect process analyzer that canmeasure all samples, in all states under any conditionn Concept:Surround a chemical problem or process withcomplimentary and orthogonal tools to extract a greaterquantity of information than each tool individually providesn In most cases the complimentary and orthogonalmeasurements are already collected throughout aproduction cyclen Use of data fusion can increase the informationextracted from these measurements and thereforeincrease the value of the measurement

Data Fusion ProcedureIRSCALEDIRSCALENMRSCALEDNMRSCALERAMANRAMANPOLYFIT +SCALEPOLYFIT +SCALEIRNMRRAMANFUSED DATADearing, T.; Thompson, W.; Rechsteiner, C.; Marquardt, B.; J. Applied Spectroscopy, February 2011, Volume 65, Issue 2, pp. 181-186.

ResultsComparison of Individual Models to Fused ModelRSEP /%76543210API Weight % H2Raman IR NMR FusedQuantity Raman IR NMR Fused°API 1.96% 1.85% 0.477% 0.237%Weight % 2.50% 6.36% 0.844% 0.451%H 2 Content 0.708% 0.463% 0.063% 0.03%

Acknowledgementsn Marquardt Groupn Applied Physics Lab – Univ. of Washingtonn Center for Process Analysis and Control (CPAC)n FDA, ORISEn Parkern Kaiser Optical Systemsn Mettler-Toledon Corning