Implementation of Near-Infrared Spectroscopy for In-Line Monitoring of

Article pubs.acs.org/OPRD

Implementation of Near-Infrared Spectroscopy for In-Line Monitoring of a Dehydration Reaction in a Tubular Laminar Reactor Aleksandar Mitic,† Albert E. Cervera-Padrell,† Asmus R. Mortensen,†,‡ Tommy Skovby,‡ Kim Dam-Johansen,† Irakli Javakhishvili,† Søren Hvilsted,† and Krist V. Gernaey*,† †

Department of Chemical and Biochemical Engineering, Technical University of Denmark (DTU), Søltofts Plads, Building 229, DK-2800 Kongens Lyngby, Denmark ‡ Chemical Production Development, H. Lundbeck A/S, Oddenvej 182, DK-4500 Nykøbing Sjælland, Denmark S Supporting Information *

ABSTRACT: Production of active pharmaceutical ingredients (APIs), fine chemicals, food products, and so on has in recent years been focused on implementing process intensification and process optimization tools. Lower costs and higher selectivity as well as better sustainability and competitiveness are the main benefits. A good approach to achieve this is to perform continuous manufacturing together with satisfying process analytical technology (PAT) requirements. The example studied here is the dehydration reaction of 9-allyl-2-chlorothioxanthen-9-ol (“N714-allylcarbinol”) to give a mixture of cis- and trans-9Hthioxanthene, 2-chloro-9-(2-propenylidene)-(9CI) (“N746-butadienes”). A simplified procedure for designing mesoscale tubular reactors is demonstrated together with performance outside of the normal operation windows (higher pressures and temperatures above normal boiling points of solvents). Noninvasive in-line real-time monitoring was established by using Fourier transform near-infrared (FT-NIR) spectroscopy and a suitable partial least squares (PLS) model. High prediction accuracy was achieved and additionally validated by using at-line FT-NIR spectroscopy and off-line HPLC analysis. The presence of impurities was noticed and partly identified in the form of polymers. It is important to note that substrates and products in this work are API intermediates in the production of zuclopenthixol, a product of H. Lundbeck A/S.

1. INTRODUCTION The introduction of the Process Analytical Technology (PAT) Initiative1 resulted in significant changes in the modern pharmaceutical industry as well as in the manufacturing of fine chemicals and food.2,3 The implementation of in-/on-line real-time process monitoring, control, and automation together with suitable process intensification/optimization tools has resulted in plenty of economic and environmental benefits in the past decades.4−9 The most common way to achieve advantages in the modern pharmaceutical industry is to establish continuous manufacturing modes. However, not all chemical reactions are suitable for continuous processing according to Roberge et al.10,11 Slow reactions, for example, cause difficulties when continuous production is demanded. It is therefore necessary to accelerate chemical reactions significantly before all of the benefits defined with PAT can be achieved. Acceleration of chemical reactions can be performed by applying different tools and techniques.12 One example is to use downsizing equipment, as was done by Damm et al.,13 who achieved accelerations of many chemical reactions by implementing microreactors instead of batch reactors, for instance. They additionally performed tests by using microwave-assisted organic synthesis (MAOS), leading to the conclusion that equivalence between microreactors and MAOS could be established. This conclusion supports the thermal kinetic theory of MAOS, which claims that the influence of the physical parameters (mass/heat transfers, mixing, temperature gradients, etc.) is the most dominant in the accelerations of chemical reactions.14 © XXXX American Chemical Society

Application of the equivalence established between MAOS and microreactors has received more attention in modern meso-flow chemistry (mesoscale tubular reactors) as well, together with reaction performances outside of the normal operation windowsat high pressures and temperatures above the normal boiling points of solvents.15,16 Mesoscale tubular reactors are defined as tubular reactors with inner diameters in the range from 0.5 or 1 mm up to 3 mm.17 Their design could be based on the use of a new and simplified approach: estimation of the residence times could be performed with respect to the “time prediction chart” that was initially made for MAOS.18 More precisely, chemical reactions that last 4 h in batch processing with conventional heating will be accelerated down to just 4 min if a temperature increase from 60 to 120 °C is introduced together with the assistance of microwave radiation. The same effect is achieved when applications of scalable continuous-flow processes (microreactors or meso-flow chemistry) are applied.13 The chemical reaction studied here is the dehydration of 9allyl-2-chlorothioxanthen-9-ol (“N714-allylcarbinol”) to give a mixture of cis- and trans-9H-thioxanthene, 2-chloro-9-(2propenylidene)-(9CI) (“N746-butadiene”). The chemical reaction is depicted in Scheme 1. All of the constituents of this reaction are intermediate products in the synthesis of Special Issue: Continuous Processing, Microreactors and Flow Chemistry Received: December 23, 2014

A

DOI: 10.1021/op5004055 Org. Process Res. Dev. XXXX, XXX, XXX−XXX

Article

Organic Process Research & Development

Scheme 1. Schematic representation of the dehydration of “N714-allylcarbinol” to give a mixture of cis- and trans-“N746butadiene”

Figure 1. Multivariate calibration procedure involving five basic steps (plus an “extra” step, “remove outliers”) and additional supporting steps (mathematical data pretreatments). Some further explanation is given in the dashed boxes. Abbreviations: PLS, partial least squares regression; RMSEC, root-mean-square error of calibration; LV, latent variable; RMSECV, root-mean-square error of cross-validation; RMSEP, root-mean-square error of prediction.

to establish a step-by-step procedure (Figure 1). The procedure represents a modified version of a previously described approach.19 At the start, the FT-NIR data must be investigated in order to find ranges of wavelengths/wavenumbers suitable for the compounds to be analyzed. After this step, preprocessing of the raw data is usually done in order to retain mostly relevant information in the pretreated spectra. This is a challenging step in the calibration procedure and involves appropriate and careful selection of the mathematical pretreatment methods.30 Mean centering,31 autoscaling,32 smoothing,22 first- and secondorder derivative,33 multiplicative scatter correction (MSC),34 and standard normal variate (SNV)35 are some of the methods that could be used separately or in different combinations in order to pretreat spectral data. Furthermore, several additional methods could be applied, such as a priori variable scaling, normalization, extended MSC (EMSC), Fourier compression, wavelets, and so on.36 The third step involves the development of the partial least squares (PLS) model,37 where the number of latent variables (LVs) is optimized through a cross-validation procedure. The cross-validation is usually based on the leaveone-out variable technique, meaning that calibration is performed with one variable excluded from the calibration

clopenthixol, a product of H. Lundbeck A/S. The chemical reaction itself represents the third synthetic step, after Grignard alkylation and hydrolysis.19,20 It is performed in THF because all of the previous steps are performed in the same solvent. The chemical reaction is performed in a tubular-laminar reactor that was designed by applying the “time prediction chart” and thermal kinetic theory. In-line real-time process monitoring was furthermore established by using Fourier transform nearinfrared (FT-NIR) spectroscopy and validated with at-line FT-NIR and off-line high-performance liquid chromatography (HPLC) analyses. The successful development of a kinetic model was also performed as a good foundation for further work on process control and automation.

2. MATERIALS AND METHODS 2.1. Reactor and Instrumentation. A detailed description of the reactor design, reactor setup, and instrumentation used is provided in the Supporting Information. 2.2. Chemometric FT-NIR Model Development. Calibration of FT-NIR was performed by applying multivariate data calibration.19,21−29 There is no unique approach for implementing multivariate data calibration, and it was therefore important B


Article

Organic Process Research & Development data set. The excluded variable is afterward used in the validation procedure. Therefore, if the root-mean-square error of cross-validation (RMSECV) is minimized, the optimum number of LVs is achieved theoretically. However, over- or underfitting should be avoided, and therefore the root-meansquare error of calibration (RMSEC) should be calculated and compared to the RMSECV. The most suitable criterion for deciding on the optimal number of LVs is to achieve a small difference between the RMSECV and RMSEC, which should be below 20% according to Shenk et al.38 If the obtained value is higher than 20% (overfitting), then a new number of LVs should be applied, and therefore, new values for RMSECV and RMSEC would be obtained (often higher when the number of LVs is decreased). From a practical point of view, the best choice for the number of LVs is the first value that causes a significant decrease in the RMSECV. The consequent step is to calculate the RMSEP, whose value should be in the same range as the RMSECV. If not, the data set should be checked for the presence of practical errors (outliers). In case of obvious outliers, they should be removed from the calibration data set and the overall procedure should be repeated. It is important to note that the calibration procedure should be repeated for several mathematical pretreatments (or combinations of them). The best calibration model is the one yielding the lowest RMSEP value.

Table 1. Data sets for calibration and validation no.

3. RESULTS AND DISCUSSION 3.1. Chemometric Model Development. 3.1.1. Calibration and Validation Samples. The dehydration reaction involved three different constituents: “N714-allylcarbinol”, “N746-butadienes”, and water. The calibration data set included 20 different samples, and the validation set included four samples (Table 1). 3.1.2. Overview of Absorption Spectra and Wavelength Interval Selection. Successful implementation of real-time process monitoring involves good calibration of process analyzers. Focusing on in- and at-line modes of analysis, it is important to note that the first step was to define spectral regions that would give information about the constituents of interest. “N714-allylcarbinol” and “N746-butadienes” could be identified and quantified by analyzing the regions from 6220 to 5925 cm−1 and 4900−4560 cm−1. The first region corresponds to absorptions of allyl, vinyl, and aromatic groups, whereas the second region shows combination bands of allyl, vinyl, and aromatic groups. For practical reasons, the second region was extended to 5370 cm−1 because the third constituent (water) has a strong absorption in that region. It is important to note that THF was used as a background. Its absorption is in the region between 5900 and 5400 cm−1 and therefore does not interfere with the analyzed constituents. Raw FT-NIR spectra including all three constituents are shown in Figure 2 together with the selected regions (indicated by the red vertical dashed lines and the black vertical dashed lines, respectively). 3.1.3. PLS Model Calibration, Cross-Validation, and Validation. After regions with useful information about the analyzed constituents were identified, different mathematical pretreatments of the data were tested. Therefore, removal of scattering, baseline corrections, smoothing, and so on were performed by using four different mathematical techniques: baseline correction (BLC), mean centering (MC), and Savitzky−Golay first derivative (SG1) and Savitzky−Golay second derivative (SG2) derivations with different numbers of points (p). These techniques were emphasized as the most

CN714‑allylcarbinol [M]

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20

0.3000 0 0.6000 0.8000 0.9000 1.1000 1.2000 1.4000 0.1000 0.4000 0.7000 1.000 0 0 0 0 0 0 0 0

1 2 3 4

0.2500 0.7500 0.5500 0

CN746‑butadiene [M]

Calibration Data Set 1.4000 1.1000 0.9000 0 0.6000 0.5000 0 0.2000 0 0 0 0 0.1000 0.4000 1.000 0 0 0 0 0 Validation Data Set 0.3500 0.1500 0 0.7500

Cwater [M] 0 0.5000 0.3000 0.2000 1.3000 0.9000 1.2000 1.1000 0 0 0 0 0 0 0 0.1000 0.4000 0.7000 1.000 1.3000 1.050 0 0.5500 0.1500

comprehensive ones for covering a wide range of potential corrective actions.39 All of these techniques were implemented independently as well as in suitable combinations. The construction of PLS models for both modes of analysis included the leave-one-out cross-validation procedure. The starting number of latent variables was assumed to be 10, which is equal to half of the number of calibration samples. The optimization procedure of the PLS model was based on fulfilling the following three conditions: 1. RMSECV had to reach or be around the minimum value. 2. The difference between the RMSECV and RMSEC had to be less than 20% in order to avoid overfitting.38 3. RMSEP had to have an acceptably low value. On the basis of these criteria, it was decided that the best pretreatment for the in-line analysis of “N714-allylcarbinol”, “N746-butadienes”, and water was BLC with additional MC. The optimal numbers of LVs were 3, 3, and 4, respectively. The same procedure was repeated to evaluate the best calibration models for at-line analysis. As a result, three different mathematical pretreatments were chosen here in order to precisely predict the molar concentrations of the constituents. Raw spectra of “N714-allylcarbinol” showed the best results when SG2 and 15 points were applied as mathematical pretreatment, whereas the spectra for “N746-butadienes” and water were treated with MC and MC with additional BLC, respectively. With respect to the numbers of LVs, they were almost identical to the values for the in-line mode (4 for “N714-allylcarbinol”, 3 for “N746-butadienes” and 4 for water). The last step in successful implementation of the multivariate calibration was to include additional validation of the obtained calibration models. The obtained values of RMSEP were acceptably low (see the Supporting Information for more detailed information). C


Article


Figure 2. Raw FT-NIR spectra of all 20 samples used in the at-line calibration procedure together with the defined spectral regions used for the development of chemometric models.

3.2. Real-Time Process Monitoring. Specifically for “N714-allylcarbinol”, a relationship between in-line data on the one hand and at-line and off-line data on the other hand was investigated. In Figure 3, the X axis shows molar

because of the fast data acquisition and consequently easier process control and automation. Furthermore, real-time process monitoring of the molar concentration of “N746-butadienes” is depicted in Figure 4. It can be seen that quite a good correlation among the different modes of analyses was achieved.

Figure 3. External validation of the experimental data obtained with in-line monitoring during development of the kinetic model for the dehydration reaction (focus on “N714-allylcarbinol”).

Figure 4. External validation of the experimental data obtained with in-line monitoring during development of the kinetic model for the dehydration reaction (focus on “N746-butadienes”).

concentrations obtained when the in-line mode was applied, whereas the primary and secondary Y axes refer to at-line and off-line molar concentrations, respectively. Standard deviations are additionally included as indicators of the error between the different modes of analysis. It can be easily noticed that a very good correlation between in- and at-line experimental data was achieved, resulting in a correlation coefficient (R2) of 0.993. Furthermore, the off- and in-line data showed an excellent correlation as well (R2 = 0.995), meaning that a very high predictive capability of the model was obtained (which is essential for the development of a good kinetic model). It is important to note that the development of the kinetic model was based on only the measurement results obtained from the in-line FT-NIR data

The third constituent, water, showed a great correlation between the at- and in-line modes of analysis. The results achieved during real-time process monitoring (depicted in Figure 5) again demonstrate a very high correlation factor of 0.999. This implies a successful calibration of the process analyzers (FT-NIR), as well as pointing toward the fact that reliable data are collected during the real time process monitoring of the dehydration reaction. 3.3. Development of the Kinetic Model. The development of the kinetic model was based on a set of five different samples for five different molar hydronium ion concentrations. The experiments were carried out at five different temperatures. D


Article


where rr is the reaction rate (in mol dm−3 s−1), [ROH] is the molar concentration of “N714-allylcarbinol” (in M), kr is the reaction rate constant (in s−1), and τr is the mean residence time ( in s). According to previous experiences, the reaction time for complete conversion of the alcohol was 2 h in the batch mode. Following the time prediction chart, it was assumed that a residence time of 256.2 s with a reaction temperature of 120 °C would be enough for the complete conversion in a mesoscale tubular laminar reactor. The kinetic model was therefore developed with respect to these assumptions. Hence, incorporating eq 1 into the material balance for tubular reactors with a constant volume leads to eq 2:

(

ln

[ROH]0 [ROH]

τr

r

(2)

The initial conditions [ROH]0 = 1.06 M and τ0 = 0 s as well as the fixed residence time τr = 256.2 s were used to solve eq 2. This last equation resulted in a set of values for the reaction rate constant that varied as a function of the changes in the molar concentration of sulfuric acid as well as with temperature, of course. Further analysis was therefore performed in order to calculate all of the constituents describing the reaction rate constant, as shown in the Arrhenius equation below:

Figure 5. External validation of the experimental data obtained with in-line monitoring during development of the kinetic model for the dehydration reaction (focus on water).

The starting molar concentration of “N714-allylcarbinol” was 1.06 M, and this value remained the same in all of the experimental runs. Furthermore, the molar concentration of H3O+ ions was varied from 0.02 to 0.1 M, whereas the applied temperatures were in the range from 40 to 120 °C with a step increase of 20 °C. According to Bruckner40 and Carey and Sundberg,41 the initial assumption in the kinetic model development procedure was that the chemical reaction is elementary and that it follows first -order kinetics with respect to the alcohol. Focusing on the conversion of alcohol, the assumed kinetic model is shown in eq 1: d[ROH] = k r[ROH] rr = − dτr

) =k

k r = koe−EA / RT

(3) −1

where ko is the Arrhenius pre-exponential factor (in s ), EA is the energy of activation (in kJ/mol), R is the universal gas constant (in kJ mol−1 K−1), and T is the temperature (in K). The logarithmic version of eq 3 is shown in eq 4. Hence, determining the values of k and EA would include plotting values for ln(k) as a function of inverse temperature. ln(k r) = ln(ko) −

EA 1 R T

(4)

As a result, a straight line is expected with good linear fitting. The results are shown in Figure 6.

(1)

Figure 6. Relationship between ln(k) and 1/T with the main purpose to find values for the energy of activation (EA) and the pre-exponential factor in the Arrhenius equation (ko). E


Article

Organic Process Research & Development It can be seen that the assumed kinetic model was actually correct because a good fit was achieved. Furthermore, five different values of the molar concentration of sulfuric acid resulted in different values for ln(ko), thus demonstrating a pH dependence and thereby influencing further investigations in the kinetic model development studies. However, the value of EA was calculated from Figure 6 and is equal to 67792.86 J/mol. Furthermore, a graph of ko obtained from Figure 6 versus the molar concentration of hydronium ions present in the final solution is shown in Figure 7. A good linear fit was achieved,

Figure 8. Conversion of “N714-allylcarbinol” at different temperatures for a fixed residence time. The lines show the predicted values, and the points show the experimental data. Different molar concentrations of the chemical catalyst are indicated with different colors.

Figure 7. Relationship between the Arrhenius pre-exponential factor (ko) and the molar concentration of hydronium ions [H3O+] present in the reaction medium.

thus confirming the linear dependence of the pre-exponential factor on the molar concentration of hydronium ions. More detailed analysis showed that the pre-exponential factor could be expressed as written below: ko = k[H3O+]

(5) −1 −1

where k = 3 × 10 dm mol s . Therefore, the final version of the kinetic model describing the conversion of “N714allylcarbinol” during the dehydration reaction is shown in eq 6: 8

−

3

d[ROH] = k[H3O+]e−EA / RT [ROH] dτ

(6)

3.4. Predicted Molar Concentrations of Constituents versus Experimental Data. A comparison between the model values for the molar concentration of “N714allylcarbinol” and the experimental data points is shown in Figure 8. A good fit of the alcohol conversion was achieved for all molar concentrations of hydronium ions present in the reaction mixture. It is furthermore important to note that small prediction errors might be present in the areas with very low molar concentrations of alcohol, such as in the concentration range below 0.1 M. Nevertheless, the general conclusion is that a good fit was obtained. Prediction of the molar concentrations of “N746-butadienes” was also done. In the case of perfect conversion, 1 mol of the alcohol would give the same amount of butadienes at the end of the chemical reaction on the basis of the stochiometry (Scheme 1). Hence, the modeled values for “N746-butadienes” were calculated by subtracting the reacted amounts of “N714allylcarbinol” from the initial molar concentration of the same constituent. Figure 9 depicts the predicted and experimental data. It is important to note that standard deviations calculated

Figure 9. Synthesis of “N746-butadienes” during the dehydration reaction. The lines refer to the values expected from mass balance, whereas the points represent the experimental data. Different molar concentrations of the chemical catalyst are indicated with different colors.

between the in-line, at-line, and off-line measurements are also included. The main reason for such an approach was to emphasize the impact of the experimental errors in the calculation of the values for “N746-butadienes”. The plot of the predicted and experimental data for the concentration of “N746-butadienes” hints at several undesired effects in the reaction system. As an initial conclusion, it can be noticed that increasing the loading of the chemical catalyst predicts unexpected formation of the desired product. This, in particular, implies the synthesis of undesired compounds. More precisely, 0.1 M solutions of hydronium ions result in F


Article


improvement was to emphasize the influence of the physical parameters (mass/heat transfers, mixing, avoidance of temperature gradients) and additionally operating outside of the normal operation windows (increasing the reaction temperature from the normal boiling point of THF to 120 °C by applying increased pressure). Applications of FT-NIR spectroscopy for in-line process monitoring were successfully demonstrated. Fast and precise data collection from the in-line mode was validated externally by performing off-line HPLC and at-line FT-NIR analyses. Such accurate data acquisition influenced the development of the appropriate kinetic model, which forms a good starting point for further process control studies and implementing process automation. It is, however, important to note that conversion of “N714allylcarbinol” did not lead deliberately to the desired products. Side reactions were identified, mostly as a consequence of a polymerization reaction. This was caused by the presence of the hydronium ions and THF, a common combination for the formation of poly(THF). It is, however, important to note that additional byproducts could be present in the product mixture, and therefore, future work should include detailed analyses of the formed impurities. Furthermore, quick neutralization of the hydronium ions could cause a decrease in the byproduct formation. Hence, applications of aqueous sodium bicarbonate for the quenching operation should be tested.

significant overpredictions of the amount of synthesized butadiene. This behavior is illustrated within the black dashed square in Figure 9. Furthermore, it is easy to notice that quite high standard deviations were observed in the regions with high molar concentrations of “N746-butadienes”. This is caused by the significantly different results obtained when in-/at- or off-line analyses were applied. The main reason for such behavior might be practical issues during the sampling procedures because inline monitoring was performed almost immediately after the reaction, whereas the at-line and off-line modes included time delays. 3.5. Partial Analysis of Side Reactions. The application of sulfuric acid as a chemical catalyst caused a very fast conversion of “N714-allylcarbinol” during the dehydration reaction. However, it was noticed that a certain amount of the desired product was not obtained and therefore that the predicted and the experimental values did not correspond. It is well-known that cations launch the ring-opening polymerization of THF.42 Thus, any strong Brønsted acid would facilitate the synthesis of poly(THF). Since HPLC and FT-NIR analyses did not indicate the formation of any additional components in the reaction system, it was decided to perform NMR spectroscopy and size-exclusion chromatography (SEC) analyses in order to explore the purity of the components and the presence of polymeric side products, respectively. Qualitative results from SEC are depicted in Figure 10,

■

ASSOCIATED CONTENT

S Supporting Information *

Design of the reactor, sampling procedure, calibration of the NIR spectrophotometer, and NMR investigation of solute purity in the reaction mixture. The Supporting Information is available free of charge on the ACS Publications website at DOI: 10.1021/op5004055.

■

AUTHOR INFORMATION

Corresponding Author

*Address: CAPEC-PROCESS Research Center, Department of Chemical and Biochemical Engineering, Technical University of Denmark (DTU), Søltofts Plads, Building 229, DK-2800 Kgs. Lyngby, Denmark. Telephone: +45 45252970. Fax: +45 45932906. E-mail: [email protected]. Figure 10. SEC trace of the product (refractive index detector response).

Notes

whereas the NMR spectra are given in the Supporting Information. The obtained results corroborated the presence of compounds with very high molecular weight (Mn = 17 000 Da at the elution volume of 14.7 mL), thus confirming the presence of polymers in the reaction medium. Furthermore, it was found that some of the poly(THF) chains incorporated the intermediate of the dehydration reaction (the carbocation).

ACKNOWLEDGMENTS The authors acknowledge the Technical University of Denmark and H. Lundbeck A/S for technical and financial support and thank Emine Yüksel Coskun and Mette Larsen for technical assistance during the experimental work.

4. CONCLUSIONS AND FUTURE PERSPECTIVES Dehydration of “N714-allylcarbinol” to give a mixture of cisand trans-“N746-butadiene” was successfully demonstrated in a mesoscale tubular laminar reactor. Applications of the modern transition approach implied a significant decrease in the reaction time from 2 h in the traditional batch mode with conventional heating to just 4 min in the continuous tubular laminar reactor. The main approach for achieving such an

(1) PAT Guidance for Industry; U.S. Food and Drug Administration: Washington, DC, 2004. (2) Workman, J.; Koch, M.; Lavine, B.; Chrisman, R. Anal. Chem. 2009, 81 (12), 4623. (3) Workman, J.; Lavine, B.; Chrisman, R.; Koch, M. Anal. Chem. 2011, 83, 4557. (4) Hessel, V. Chem. Eng. Technol. 2009, 32 (11), 1655. (5) Moulijn, J. A.; Stankiewicz, A.; Grievink, J.; Górak, A. Comput. Chem. Eng. 2008, 32 (1−2), 3.

The authors declare no competing financial interest.

■ ■

G

REFERENCES


Article

Organic Process Research & Development (6) Moulijn, J. A.; Stankiewicz, A. I. Re-engineering the Chemical Processing Plant: Process Intensification; Marcel Dekker: New York, 2003. (7) Van Gerven, T.; Stankiewicz, A. Ind. Eng. Chem. Res. 2009, 48 (5), 2465. (8) Stankiewicz, A. I.; Moulijn, J. A. Chem. Eng. Prog. 2000, No. January, 22. (9) Nauman, E. B. Chemical Reactor Design, Optimization, and Scaleup; John Wiley & Sons: Hoboken, NJ, 2008. (10) Roberge, D. M.; Gottsponer, M.; Eyholzer, M.; Kockmann, N. Chem. Today 2009, 27 (4), 8. (11) Roberge, D. M.; Ducry, L.; Bieler, N.; Cretton, P.; Zimmermann, B. Chem. Eng. Technol. 2005, 28 (3), 318. (12) Mitic, A.; Gernaey, K. V. Chem. Eng. Technol. 2015, 10.1002/ ceat.201400765. (13) Damm, M.; Glasnov, T. N.; Kappe, C. O. Org. Process Res. Dev. 2010, 14 (1), 215. (14) Kappe, C. O. Angew. Chem., Int. Ed. 2004, 43, 6250. (15) Razzaq, T.; Glasnov, T. N.; Kappe, C. O. Eur. J. Org. Chem. 2009, 2009 (9), 1321. (16) Illg, T.; Löb, P.; Hessel, V. Bioorg. Med. Chem. 2010, 18, 3707. (17) Teng, J.; Liu, C.; Zhang, C.; Chu, J.; Cheng, J.; Lee, M.; Greif, R.; Jin, S.; Huang, S.; Dang, T.; Xu, T.; Yu, X.; Lien, Y. Fluid Dynamics in Microchannels; InTech: Rijeka, Croatia, 2012. (18) Biotage Catalog. http://cnr.ncsu.edu/wpsanalytical/ documents/BiotageCatalogue.pdf (accessed Oct 9, 2013). (19) Cervera-Padrell, A. E.; Nielsen, J. P.; Jønch Pedersen, M.; Müller Christensen, K.; Mortensen, A. R.; Skovby, T.; Dam-Johansen, K.; Kiil, S.; Gernaey, K. V. Org. Process Res. Dev. 2012, 16 (5), 901. (20) Cervera-Padrell, A. E.; Morthensen, S. T.; Lewandowski, D. J.; Skovby, T.; Kiil, S.; Gernaey, K. V. Org. Process Res. Dev. 2012, 16 (5), 888. (21) Gemperline, P. Practical Guide to Chemometrics; CRC Press: Boca Raton, FL, 2006. (22) Brereton, R. G. Chemometrics: Data Analysis for the Laboratory and Chemical Plant; John Wiley & Sons: West Sussex, U.K., 2003. (23) Brereton, R. G. Chemometrics: Applications of Mathematics and Statistics to Laboratory Systems; E. Horwood: New York, 1990. (24) Miller, C. E. J. Chemom. 2000, 14, 513. (25) Beebe, K. R.; Pell, R. J.; Seasholtz, M. B. Chemometrics: A Practical Guide; John Wiley & Sons: New York, 1998. (26) Beebe, K. R.; Blaser, W. W.; Bredeweg, R. A.; Chauvel, J. P., Jr.; Harner, R. S.; LaPack, M.; Leugers, A.; Martin, D. P.; Wright, L. G.; Yalvac, E. D. Anal. Chem. 1993, 65 (12), 199. (27) Wise, B. M.; Gallagher, N. B. J. Process Control 1996, 6 (6), 329. (28) Roggo, Y.; Chalus, P.; Maurer, L.; Lema-Martinez, C.; Edmond, A.; Jent, N. J. Pharm. Biomed. Anal. 2007, 44 (3), 683. (29) Lopes, J. A.; Costa, P. F.; Alves, T. P.; Menezes, J. C. Chemom. Intell. Lab. Syst. 2004, 74 (2), 269. (30) Kramer, R. Chemometric Techniques for Quantitative Analysis; Marcel Dekker: New York, 1998. (31) Gasteiger, J.; Engel, T. The Data. In Chemoinformatics; Gasteiger, J., Engel, T., Eds.; Wiley-VCH: Weinheim, Germany, 2006. (32) Otto, M. Pattern Recognition and Classification. In Chemometrics: Statistics and Computer Applications in Analytical Chemistry; Otto, M., Ed.; Wiley-VCH: Weinheim, Germany, 2007. (33) Savitzky, A.; Golay, M. J. Anal. Chem. 1964, 36, 1627. (34) Martens, H.; Næs, T. Pretreatment and Linearization. In Multivariate Calibration; Martens, H., Næs, T., Eds.; John Wiley & Sons: Hoboken, NJ, 2002. (35) Eriksson, L.; Johansson, E.; Kettaneh-Wold, N.; Trygg, J.; Wikstrom, C.; Wold, S. Part II: Signal Correction and Compression. In Multi- and Megavariate Data Analysis; Eriksson, L., Johansson, E., Kettaneh-Wold, N., Trygg, J., Wikstrom, C., Wold, S., Eds.; Umetrics AB: Umeå, Sweden, 2006. (36) Bakeev, K. A. In Process Analytical Technology: Spectroscopic Tools and Implementation Strategies for the Chemical and Pharmaceutical Industries, 2nd ed.; John Wiley & Sons: West Sussex, U.K., 2010.

(37) Abdi, H. Partial Least Squares Regression (PLS-Regression). In Encyclopedia for Research Methods for the Social Sciences; Lewis-Beck, M., Bryman, A., Futig, T., Eds.; SAGE Publications: Thousand Oaks, CA, 2003. (38) Shenk, J. S.; Workman, J. J.; Westerhaus, M. O. Application of NIR spectroscopy to agricultural products. In Handook of NearInfrared Analysis; Burns, D. A., Ciurczak, E. W., Eds.; Marcel Dekker: New York, 1992. (39) PLSplus IQ User’s Guide; Thermo Galactic: Woburn, MA, 2006. (40) Bruckner, R. Organic Mechanisms: Reactions, Stereochemistry and Synthesis; Springer: Berlin, 2010. (41) Carey, F. A.; Sundberg, R. J. Advanced Organic Chemistry: Part A: Structure and Mechanisms; Springer: New York, 2007. (42) Meerwein, H.; Delfs, D.; Morschel, H. Angew. Chem. 1960, 72 (24), 927.

H


Implementation of Near-Infrared Spectroscopy for In-Line Monitoring of

Recommend Documents