Horiba, Ltd. · TSE:6856

Horiba — Innopharma Technology — Technical Presentation (2022)

35 slides from Horiba's technical presentation, 2022

Slide 1: The Power of Machine Learning
The Power of Machine Learning Image analysis for particle size and shape ©2022, Innopharma Technology Ltd. 1
Slide 2: Innopharma Labs founded 2009
• Innopharma Labs founded 2009 • Expanded into: • Innopharma Education • Innopharma Technology • Innopharma Technical Services • HQ: Dublin, Ireland • ~120 employees experienced in STEM, pharma, manufacturing, IT, Software Development ©2022 Innopharma Technology Ltd. 4
Slide 3: The Journey
The Journey Solutions for Advanced Manufacturing 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018 2019 2020 2021 2022 Sensors – Process Analytical Technology Innovative sensors for process environments Eyecon - in-line real time particle size and shape Eyecon2 - second generation, smaller GMP desig
Slide 4: Eyecon2 Technical Specifications
Eyecon2 Technical Specifications Size Range 50 to 5500 µm ✓ In-line ✓ At-line Casing materials 304 Stainless Steel, Glass, ✓ Non product contact Silicon (gaskets) ✓ Real time Imaging Area 11.25 x 11.25 mm Output PDF. CSV, JPEG full PSD D5-D95. Instrument GMP compliant design Ratings EyePASS is both
Slide 5: Small particles are Big deal
Small particles are Big deal Configurations Processes ©2022 Innopharma Technology Ltd. 9
Slide 6: Method of Operation: Image Capture
Method of Operation: Image Capture • A flash-imaging technique is used with an extremely short light-pulse to illuminate moving particles for image capture • Red, Green and Blue LEDs illuminating the sample from different angles for detection of particle boundaries ©2022 Innopharma Technology Ltd. 1
Slide 7: Particle size methods
Particle size methods Particle size methods are not created Horiba Particle Guidebook 2022 equally, they all have their advantages and disadvantages From: Particle Size Measurements, Fundamentals, Practice, Quality,Henk G. Merkus From: https://static.horiba.com/fileadmin/Horiba/Products/Scientific/P
Slide 8: Method of Operation: Image Capture
Method of Operation: Image Capture • Each particle initially identified • Best-fit ellipse calculated • Major & minor diameters computed • PSD/D-values determined ©2022 Innopharma Technology Ltd. . 15
Slide 15: What does it involve?
What does it involve? Vision Task Measurement Task Delineate objects of interest Measure necessary attributes of each object e.g., particles e.g., major/minor axes lengths ©2022 Innopharma Technology Ltd. 25
Slide 16: Image Segmentation
Image Segmentation The goal as defined by a human observer ©2022 Innopharma Technology Ltd. 26
Slide 17: Image Segmentation
Image Segmentation Histogram (intensity distribution) Classical Image Processing - Thresholding ©2022 Innopharma Technology Ltd. 27
Slide 18: Image Segmentation
Image Segmentation Classical Image Processing - Thresholding ©2022 Innopharma Technology Ltd. 28
Slide 19: Image Segmentation
Image Segmentation Surface Plot Classical Image Processing - Watershed ©2022 Innopharma Technology Ltd. 29
Slide 20: Image Segmentation
Image Segmentation Classical Image Processing - Watershed ©2022 Innopharma Technology Ltd. 30
Slide 23: Image Segmentation
Image Segmentation Feature Maps … … Input Output CNN Machine Learning - Convolutional Neural Networks (CNNs) ©2022 Innopharma Technology Ltd. 33
Slide 24: Image Segmentation
Image Segmentation Machine Learning - Convolutional Neural Networks (CNNs) ©2022 Innopharma Technology Ltd. 34
Slide 25: Classical Image Processing
Classical Image Processing Pre-processing Segmentation Post-processing • Contrast Enhancement • Thresholding • Size-based Region Selection • Noise Reduction • Watershed • Region Merging • Colour Transformations etc etc etc Images ©2022 Innopharma Technology Ltd. 35
Slide 26: Convolutional Neural Networks
Convolutional Neural Networks MODEL DEVELOPMENT INFERENCE Deployment Labelling/Annotating Training CNN Model Images Dataset Data Augmentation ©2022 Innopharma Technology Ltd. 36
Slide 27: Model Development Process - K-fold cross validation
Model Development Process - K-fold cross validation Dataset Training set Test set Split 1 Fold 1 Fold 2 Fold 3 Fold 4 Fold 5 Split 2 Fold 1 Fold 2 Fold 3 Fold 4 Fold 5 Split 3 Fold 1 Fold 2 Fold 3 Fold 4 Fold 5 Split 4 Fold 1 Fold 2 Fold 3 Fold 4 Fold 5 Split 5 Fold 1 Fold 2 Fold 3 Fold 4 Fold 5 Tra
Slide 29: Image Segmentation compared to EyePASS v3
Image Segmentation …compared to EyePASS v3 ©2022 Innopharma Technology Ltd. 42
Slide 31: EyePASS - Before and After Machine Learning
EyePASS - Before and After Machine Learning Traditional Image analysis Machine Learning Image analysis ©2022 Innopharma Technology Ltd. 52
Slide 32: EyePASS V3.0 - Results
EyePASS V3.0 - Results Glass EyePASS V3.0 1000um Sieve Std um (Sub sample) um um diff % diff D10 882.1 898 15 2% D50 989.6 949 41 4% D90 1145.6 1102 39 3% Correlation of 0.9559 = Strong Positive ©2022 Innopharma Technology Ltd. 53
Slide 34: Inspiring Capability
Inspiring Capability 500um Alu Oxide (Bonded Abrasives) 4 materials ML mitigates issues like false detections 250um Glass Microspheres (Sieve Calibration Standard) 1000um Multiparticulates 1000um Glass Microspheres (Wurster materials, cellets, suglets) (Sieve Calibration Standard) ©2022 Innopharma T
Slide 35: Thank you!
The Power of Machine Learning Eyecon2 Image EyePASS V2.0 EyePASS V3.0 Darren McHugh Girish Mallya Senior Product Manager Senior Software Developer Thank you! Innopharma Technology Ltd Innopharma Technology Ltd mchughd@innopharmalabs.com Mallyag@innopharmalabs.com ©2022 Innopharma Technology Ltd. 83

Innopharma Technology — Technical Presentation (2022) on 6ix

Horiba — Innopharma Technology — Technical Presentation (2022)

Slide 1: The Power of Machine Learning

Slide 2: Innopharma Labs founded 2009

Slide 3: The Journey

Slide 4: Eyecon2 Technical Specifications

Slide 5: Small particles are Big deal

Slide 6: Method of Operation: Image Capture

Slide 7: Particle size methods

Slide 8: Method of Operation: Image Capture

Slide 9: Eyecon2 Data Output

Slide 10: Eyecon2 Data Output

Slide 11: Eyecon2 Data Output

Slide 12: EyePASS Version 3.0

Slide 13: EyePASS

Slide 14: What is Image Analysis?

Slide 15: What does it involve?

Slide 16: Image Segmentation

Slide 17: Image Segmentation

Slide 18: Image Segmentation

Slide 19: Image Segmentation

Slide 20: Image Segmentation

Slide 21: Image Segmentation

Slide 22: Image Segmentation

Slide 23: Image Segmentation

Slide 24: Image Segmentation

Slide 25: Classical Image Processing

Slide 26: Convolutional Neural Networks

Slide 27: Model Development Process - K-fold cross validation

Slide 28: Image Segmentation In EyePASS v2

Slide 29: Image Segmentation compared to EyePASS v3

Slide 30: User Experience

Slide 31: EyePASS - Before and After Machine Learning

Slide 32: EyePASS V3.0 - Results

Slide 33: Untapped Potential

Slide 34: Inspiring Capability

Slide 35: Thank you!