How to Run a PAT Project? A Cooking Recipe!

Transcription

1 How to Run a PAT Project? A Cooking Recipe! Prof. Dr. Rudolf Kessler STZ Prozesskontrolle und Datenanalyse Reutlingen

2 Literature R.W. Kessler, W. Kessler and E. Zikulnig-Rusch: A Critical Summary of Spectroscopic Techniques and their Robustness in Industrial PAT Applications, CIT, 2016, DOI: /cite R. W. Kessler, Perspectives in process analysis. J. Chemometrics, 2013, 27: B. Boldrini, W. Kessler, K. Rebner and R. W. Kessler, Hyperspectral imaging: a review of best practice, performance and pitfalls for inline and online applications, Journal of Near Infrared Spectroscopy 2012, 20, , doi: /jnirs.1003 Kandelbauer A., Rahe M. and Kessler R. W., Process Control and Quality Assurance - Industrial Perspectives, Handbook of Biophotonics, Vol.3: Photonics in Pharmaceutics, Bioanalysis and Environmental Research, First Edition, Edited by Juergen Popp, Valery V. Tuchin, Arthur Chiou, and Stefan Heinemann, 2012 Wiley-VCH Verlag GmbH & Co. KGaA. Published 2012 by Wiley-VCH Verlag GmbH & Co. KGaA, 1-69 D. Oelkrug, E. Ostertag, R. W. Kessler, Quantitative Raman spectroscopy in turbid matter: reflection or transmission mode?, Anal. Bioanal. Chem. (2013) 405, , D. Oelkrug, M. Brun, K. Rebner, B. Boldrini, R. Kessler: Penetration of Light into Multiple Scattering Media: Model Calculations and Reflectance Experiments. Part I: The Axial Transfer, Appl. Spectrosc. 2012, 66 (8): D. Oelkrug, M. Brun, P. Hubner, K. Rebner, B. Boldrini, R. Kessler: Penetration of Light into Multiple Scattering media; Model Calculations and Reflectance Experiments. Part II: The Radial Transfer, Appl. Spectrosc. 2013, 67 (4),

3 Concepts Manufacturing and Processing Industry: Provides 70% of the wealth of the German society although only around 30 % of the population work in the manufacturing industry!!!, 90% of IT research is financed by the manufacturing industry Manufuture of the EU PAT/QbD of the FDA BDI 2006: 32 Thesen Factbook 06 (VCI) Chemie Globalisierung gestalten EU s goals (20% increase in energy efficiency, 20% reduction of CO2 emissions, and 20% renewables by 2020) Namur road map: Prozess-Sensoren World Manufacturing Forum 2012 Industrie 4.0

4 Challenges in the future and the fourth industrial revolution Aging of the population: medical systems Urbanisation and megacities Personalization of products and goods Industry 4.0: Internet of the Things arbitrarily modifiable and expandable (flexible) production, connect arbitrary components of multiple producers (networked), enabling its components to perform tasks related to its context independently (self- organizational) and emphasizes ease of use (useroriented) Smart materials are processed in smart factories with smart sensors to produce smart products : small and decentralized production units for personalized goods

5 Definition: Knowledge Based.. Aoccdrnig to a rscheearch at an Elingsh uinervtisy, it deosn't mttaer in waht oredr the ltteers in a wrod are, the olny iprmoetnt tihng is taht the frist and lsat ltteer is at the rghit pclae. The rset can be a toatl mses and you can sitll raed it wouthit porbelm. Tihs is bcuseae we do not raed ervey lteter by it slef but the wrod as a wlohe.

6 Knowledge Based Ncah eneir Sutide der Cmabridge Uinervtistät, ist es eagl in wlehcer Riehenfloge die Bcuhstbaen in eneim Wrot sethen, Haputschae der esrte und ltzete Bcuhstbae snid an der rhcitgien Setlle. Der Rset knan ttoaels Druchenianedr sien und man knan es torztedm onhe Porbelme lseen, wiel das mneschilhce Gherin nhcit jdeen Bcuhstbaen enizlen leist, snodren das Wrot als Gnazes.

7 Adaptive Processing for Process and Product Optimization Multi Step Process Output parameters of step 1 (intermediates) are input parameters for step 2, etc. Raw Material Step 1 Step 2... Final Product Feed Back Feed Forward Online and Inline Control Reality: Varying input quality Need for Input Control 100 % - use molecular information - build soft sensor models - use process trajectories - survey, control and optimize Customers wish: constant quality Need for output Control 100 %

8 Strategy: Introduce 3 levels of process knowledge Use expert knowledge Experimental Design Use process data: Build Soft Sensors Use molecular markers: Validate quality by spectroscopic measurements on a molecular level

9 however Culture. "Heaven is where the police are British, the chefs French, the mechanics German and the lovers Italian and it is all organised by Swiss Hell is where the chefs are British, the mechanics French, the lovers Swiss, the police German and it is all organised by the Italians."

10 Knowledge Based Production and Causality First Principles & Mechanistic Models Causality Variable Material Input Correlative and Descriptive Models Adaptive Processing Fixed Process Knowledge based Production Consistent Output Variable Output Models: First Principles Design of Experiments Softsensor Models Molecular Markers (Spectroscopy) and Process Trajectories

11 Measurements Data Knowledge: Causality The Japanese eat very little fat and suffer fewer heart attacks than the British or the Americans. The French eat a lot of fat and also suffer fewer heart attacks than the British or the Americans. The Japanese drink very little red wine and suffer fewer heart attacks than the British or the Americans. The Italians drink a lot of red wine and also suffer fewer heart attacks than the British or the Americans. Conclusion: Eat and drink whatever you like. It's speaking English that kills you. What is wrong? To draw conclusions from random or spurious correlations

12 Road Map for a Successful Implementation Step 1: Analysis of historical data Step 2: Selection of critical parameters (CQA, CPP) Step 3: Define knowledge space and variability identified, explained, measured, managed and controlled, online predictable Step 4: Execute experiments and exploit data experimental design model, soft sensor model, molecular markers Step 5: design and implement control

13 Example: Fibre Board Production: Funder Drying Non woven Moulding press Dry Process Wood chips Blow line Defibrator Vat Fibre mat Hot press Wet Process

14 Analysis of historical data Univariate and bivariate analysis of data Normal distribution, Boxplot, Scatterplot, Errors, Trends Reproducibility, process capability etc. Multivariate analysis of data Correlation, Determination of critical parameters by PCA, (PLS) Reduction of the dimension Select key variables for exper. design

15 Biegefestigkeit STZ Process Control and Wasseraufnahme Historical Data: Production b 2c 4a 4b 4c 4d 4e 5b Holzmischung b 2c 4a 4b 4c 4d 4e 5b Holzmischung

16 Biegefestigkeit STZ Process Control and Wasseraufnahme Normal Distribution and Scatter Plots Biegefestigkeit vs Wasseraufnahme Normalverteilung Biegefestigkeit

17 Designed Experiments: CQA, CPP Variables: Raw Material Pressure Retention Time Plate Gap Rotation SFC SFM Feed Back Control Prediction Response Wood, Wood Mixture Lignin, Carbonyl Flexural Strength Bark, Age Reactive Binder Modulus Pressure (Temperature) Morphology Water Uptake Time, Severity C (Moisture) Mouldability Plate Gap, Rotation, Homogeneity (COD) Plate Geometry, Orientation (Press, etc.) Severity M Quality of the Dispersion (Power, Spec. Energy,..) Size of Agglomerates

18 Severity of mechanical impact Design of Experiments for Understanding Raw Material (wood) Severity of Chemistry Design experiments on laboratory scale Find out hierarchical structure of effects Proof of concept on industrial level

19 Wasseraufnahme_Rand STZ Process Control and BB_l_ngs_Mitte Classical of DoE Results Fichte 50% Buch Fichte mit Rind Fichte ohne Rin E_Holzsorte 40.0 Fichte 50% Buch Fichte mit Rind Fichte ohne Rin E_Holzsorte

20 Water Uptake Flexural Strength STZ Process Control and Water Uptake Flexural Strength Response Surface Analysis for Optimization A: Pressure B: Time A: Pressure B: Time Spruce Task: Optimize several response variables Beech Water uptake and flexural strength A: Pressure B: Time A: Pressure B: Time 3.00

21 B: Zeit STZ Process Control and Multi Target Optimization 15 Overlay Plot Druck = 7.30 Zeit = Biege: 24 Wasser: 90 Biege: Wasser: 130 Spalt = 1.00 Umdrehung = Holzart = Biege = Desirability = A: Druck Wasser = 53

22 Fiber Board Production: Process Understanding e.g. Beech (Electron Micrograph) underprocessed optimum processing SFC ~ 2 SFC ~ 3.5

23 Process Control and Optimisation: Soft Modeling Matrixplot of Process Data spruce without bark spruce with bark 450 process data (x-axis) 102 press cycles (y-axis)

24 PCA with Process Data of Spruce without Bark plate gap: coarse plate gap: fine SFC 2.5 SFC 2.1

25 Process Modeling Prediction and Property Design (reverse model, IAF/Siemens) Process Modeling Chip Yard Data Control Parameters Refiner Diagnose Fibre prep. Soft Sensor adaptive Fuzzy-Model Dewatering Press Optimization Quality

26 Refiner Optimization (Soft Sensor)

27 Inverse Model: Design a new Fibreboard

28 High Energy Wavelength 10 STZ Process Control and Optical Spectroscopy Chemistry and Morphology Low Energy [nm] Far Near Near Mid Ultraviolet Radiation Visible Radiation Infrared Radiation Fluorescence Excitation / Absorption: l nm Fluorescence: l nm UV / Vis - snir l = nm (UV) l = nm (Vis) l = nm (NIR) NIR / MIR MIR: l = nm ~ n = cm -1 NIR: l = nm ~ n = cm -1 Raman l 0 = 633 nm l = nm Dn ~ = cm -1 Excitation /Emission of Electrons Excitation of Electrons (UV / Vis) and Overtones (NIR) Excitation of Vibrations, Rotations and Overtones Excitation of Vibrations Complementary to MIR

29 Technology Review (see Kessler (ed.)) UV/VIS/ s-nir NIR MIR Fluorescence Raman Selectivity Sensitivity and Selectivity: NIR: around 1% (for water 0.1% 0.01%) FTIR: at least 0.01%.. standard: 0.001% Raman: at least 1% up to 0.01% (selectivity may +++ beat sensitivity!!!) UV/Vis: ppm or better Fluorescence: single molecule level Sensitivity (+) (+) ++(+) Sampling Working in aqueous media Applicability Process analytical tool Light guide glass (+) Signal Absorption Absorption Absorption Emission Scattering Samling online/inline Techniques s, l, g s, l s, l, g s, l (g) s, l, (g) Transmission Reflectance ATR Transmission Reflectance ATR ATR (Transmission) Reflectance Transmission Reflectance Relative costs

30 Product Functionality Design: Molecular Markers

31 Spectroscopy Calibration Production after Quality Control Sample Set Design Experimental Design - On-Line Capable Methods - Reference Methods - Target Values from Industry Screening - Spectroscopy (specular / diffuse) - Optics & Polarization - Elektrochemistry Detection of Superposition Multi-Linear Starting Point => Exploratory Statistics Selection of Measuring Methods and Optimization Data Reduction: Independent Variables Non-Linear Starting Point =>Neuronal Nets Selective Algorithmical Correlation Measuring System - Optimized According to the Application Construction of Prototypes - Verification by the Application

32 In-Line Control during Manufacturing Diffuse Reflectance Probe and Spectral Imaging

33 Blow Line Spectra

34 PCA of Vis-Spectra of Fibreboards nm reflectance PCA-Analysis of Spectra Clustering in wood mixtures Classification of fineness Classification of Severity factor (SFC)

35 Morphology Spruce Beech Cross-section 20 µm magnification 100 x reminder: sensitivity absorption and scatter

36 PLS of UV/Vis-spectra of Fibremat Water Uptake Correlation nm reflectance VP=45 bar; VT=180 sec PLS-Analysis of Spectra Correlation between technological data and spectral information

37 Wood-Scores STZ Process Control and Prediction and Control Production October 21 till October measured wood-scores 5 WM = wood mixture A B C 4 3 WM 4a less than 1 hour 2 6 hours D 1 much bark WM 2c WM 2c E 0-1 9:20 10:34 12:51 14:48 17:21 19:30 20:17 22:46 0:38 6:57 2:52 12:33 15:28 WM 1 WM 1 WM 2 I 19:15 8:59 11:17 13:30 15:03 17:24 1:16 2:55 5:13 8:43 10:52 12:33 overshot -2 Production over time

38 Iggesund Plant Sweden

39 STZ Process Control and. On-Line Trials at Iggesund Comparison of Spectroscopic Results with Digester Variables RESULT2, (Y-var, PC): (PC1( nm),6) Elements: Slope: Offset: Correlation: RMSEC: SEC: Bias: Measured Y Predicted Y

40 On-Line Trials at Iggesund Comparison of Spectroscopic Results with Digester Variables 1.0 Predicted and Measured _10: _10: _01: _18: _07: _03:50 RESULT2, (Y-var, PC): (PC1( nm),6) Predicted Measured Samples PC 1 score of wood chip spectra versus PLS result of 120 digester variables.

41 Summary (1) PAT/Qbd is a holistic approach: thus people have to talk to each other The manifold of different institutions for standardisation inhibit a broader distribution of PAT systems The potential of innovations of SME cannot fully be exploited because of lack of the critical size To achieve a critical size to integrate PAT into dinosaur companies, system integrators are needed The toolbox for the use of big data, sensor fusion and chemometrics/doe is not yet exploited due to the lack of teaching at universities Intelligent smart sensor need to integrate first principles into the measurement The special problem of the interdisciplinary field of Process Analysis for a sustainable knowledge based production is the lack of skilled personal

42 Summary (2) Deming: Learning is not compulsaryneither is survival!! THANK YOU FOR YOUR ATTENTION