Hyperspectral / Chemical Imaging as Key Technology in Sensor Based Sorting Applications

Transcription

1 Hyperspectral / Chemical Imaging as Key Technology in Sensor Based Sorting Applications Matthias Kerschhaggl BiRT Workshop, 20/03/15 1

2 Outline 2

3 Who we are Smart solutions provider since : HSI camera & classification system introduced to market Industry proven in recycling, mining, food processing and pharmaceutical applications Real time classification and sorting: All in one device 3

4 Applications Household waste Metals from slag E Scrap Potato Sorting & Grading C&D Recovered Paper Wood Sorting Nut Sorting 4

5 References Recycling Glass Sorting Customer: EVK products: EOS Color Imaging ARGOS Conductivity Imaging 5

6 References Recycling Metal Scrap Customer: EVK product: ARGOS Conductivity Imaging 6

7 References Recycling Plastic Sorting Customer: EVK product: HELIOS Chemical Imaging 7

8 Recovered Paper Sorting 8

9 References Recycling Minerals, Metals, Plastic Customer: EVK product: EOS Color Imaging 9

10 References Mining Mineral Sorting Customer: EVK product: HELIOS Chemical Imaging 10

11 References Food Sorting Div. Nut Sorting / Recycling Customer: EVK product: HELIOS Chemical Imaging 11

12 References Food Processing Potato Sorting Customer: EVK products: HELIOS Chemical Imaging EOS Color Imaging 12

13 Hyperspectral Imaging Technology Camera internal classification and RGB visualization of chemical differences for real time analysis 13

14 HELIOS Spectral Ranges 14

15 Outline Recycling Food Pharmaceuticals Mining 15

16 Recycling paper cardboard PE LD 16

17 Recycling PE PVC 17

18 Recycling Paper + Glue, 1st derivative, normalized Quality analysis of paper 18

19 Outline Recycling Food Pharmaceuticals Mining 19

20 Food HSI preview of SE potatoe CCI view of SE potatoe Potatoe after frying Sugar end

21 Food Food sorting Blueberries and stems

22 Food Food sorting Noodles and worms

23 Food Glucose Starch water Simple VIS representation (filter convolution) of HSI data CCI representation of HSI data (PCA: Principal Component Analysis) Quantitative analysis is possible

24 HSI Objectives/Trends Monitoring of input streams for quality and process control > HSI as Process Analysis Technology (PAT) Sorting and monitoring of product flows in one machine using HSI/CIT Quantitative analysis of chemical constituents as add on Removing foreign materials and infer quantitative information on e.g. dry matter and glucose in potatos at the same time 24

25 HSI Quantitative Analysis Predictors (spectra) X Observations (reference analysis) Y X Y wellness.byu.edu Fernadez Novales et al. (2009) Fig 2: PLS schematics (Wold et al. 2001) 25

26 Dry Matter REMINDER 1st inline TEST 2013 R = 0.9 RMS = 1 % HELIOS Camera 26

27 Inline Dry Matter Measurements Dry matter content traced by spectral signatures Inferred FIR filter kernel allows for concentration prediction 27

28 Dry Matter Machine Integrated in sorting machine (INSORT Observer) Valdidated model (R^2=0.9, RMS=5 %) 28

29 Award International FoodTec Award DLG (Deutsche Landwirtschafts Gesellschaft / German Agricultural Society) 29

30 Dry Matter 30

31 Dry Matter water channels 31

32 Dry Matter Statistics DM heat map DM value histogramm Mean DM Standard Deviation of product DM 32

33 Accuracy vs. Precision higher precision lower accuracy lower precision higher accuracy Laboratory precision completed with 100% inspection of entire production stream Plots taken from 33

34 Precision vs. Accuracy A RMS of e.g. 1% for constant monitoring of the product throughout the entire input stream is already way better than what you can do with a precision of 0.1 % every 30 min on a spatially confined sample! The intrinsic variance of the product is greater than the precision of the lab reference, so better have full sampling rather than super precision! Better you measure all the fries all the time with reasonable precision rather than 1 out of with ultra high precision! It is about large scale trends in the whole product stream not about laboratory measurements of unrepresentatively small quantities! Courtesy W. Märzinger 34

35 Outline Recycling Food Pharmaceuticals Mining 35

36 Pharmaceuticals 36

37 Pharmaceuticals 37

38 Pharmaceuticals R^2=0.98 RMSEC = 1.2 mg/cm2 matthias.kerschhaggl@evk.biz 38

39 Pharmaceuticals mg/cm 2 R^2=0.98 RMSEC = 1.2 mg/cm2 matthias.kerschhaggl@evk.biz 39

40 Personalized Medicine Imprinted set of test specimens with ascending API concentration Visible image Intensity unnormalized spectra Pixels representing wavelength µm HSI image 40

41 Drying Processes In situ measurement of moisture content of a sample during drying cycle (left: wet, right: dry) matthias.kerschhaggl@evk.biz 41

42 Product Identification Courtesy of RCPE Hyperspectral Imaging technology analyzes the chemical consistency and checks tablet integrity 42

43 Chemical Composition Chemical mapping of API content Agglomerated API spots 1% 5% 11% 15% Courtesy of RCPE 43

44 Outline Recycling Food Pharmaceuticals Mining 44

45 Sandstone/Slate Classification 45

46 Copper Ores Quantitative Analysis % I [AU] λ [px] matthias.kerschhaggl@evk.biz 46

47 Conclusion HELIOS is a near infrared (NIR) hyperspectral imaging (HSI) smart camera suitable for fast inline applications A broad range of applications in food, recycling, mining and pharmaceuticals have been already realized and are running in the field. HSI has become an established and reliable technology in the field of sensor based sorting. Apart from mere sorting solutions the inference of spatially resolved, non invasive quantitative inline measurements for 100 % product control (PAT) have enormous potential adding to classical, precise but low sampling and invasive laboratory based techniques. 47

48 References M. Kerschhaggl, W. Märzinger, E. Leitner et al. Inline HSI food inspection and concentration measurements of pharmaceuticals a report from an industrial environment. Karlsruhe: KIT Scientific Publishing, Karlsruhe, M. Kerschhaggl Hyperspectral imagery for real time quantitative inline analysis, Sensor Based Sorting 2014, GDMB Verlag GmbH, vol. 135, M. Kerschhaggl Hyperspectral Imaging as Process Analysis Technology for inline Applications. Laboratory Precision meets high Sampling Accuracy Karlsruhe: KIT Scientific Publishing, Karlsruhe,

49 Acknowledgments M. Jeindl W. Märzinger E. Leitner 49

50 Thank you for your attention 50