Stefan Bosse - AFEML - Module 0: About the Lecturer -

About the Lecturer

Stefan Bosse - AFEML - Module 0: About the Lecturer - Bremen

Bremen

2016: Habilitation mit der Venia Legendi für die Informatik

Stefan Bosse - AFEML - Module 0: About the Lecturer - Bremen

Bremen

2016: Habilitation mit der Venia Legendi für die Informatik

2016 bis 2024: Privatdozent Fachbereich Informatik, Universität Bremen

Stefan Bosse - AFEML - Module 0: About the Lecturer - Bremen

Bremen

2016: Habilitation mit der Venia Legendi für die Informatik

2016 bis 2024: Privatdozent Fachbereich Informatik, Universität Bremen

2018-2019: Vertretungsprofessur Praktische Informatik und Lehrbeauftragter in der Fakultät Informatik, Universität Koblenz-Landau

Stefan Bosse - AFEML - Module 0: About the Lecturer - Bremen

Bremen

2016: Habilitation mit der Venia Legendi für die Informatik

2016 bis 2024: Privatdozent Fachbereich Informatik, Universität Bremen

2018-2019: Vertretungsprofessur Praktische Informatik und Lehrbeauftragter in der Fakultät Informatik, Universität Koblenz-Landau

2020-heute: Wissenschaftlicher Leiter und Projektleiter in der transregionalen DFG Forschungsgruppe 3022, Automatische Schadensdiagnostik mit geführten Wellen in Laminaten

Stefan Bosse - AFEML - Module 0: About the Lecturer - Bremen

Bremen

2016: Habilitation mit der Venia Legendi für die Informatik

2016 bis 2024: Privatdozent Fachbereich Informatik, Universität Bremen

2018-2019: Vertretungsprofessur Praktische Informatik und Lehrbeauftragter in der Fakultät Informatik, Universität Koblenz-Landau

2020-heute: Wissenschaftlicher Leiter und Projektleiter in der transregionalen DFG Forschungsgruppe 3022, Automatische Schadensdiagnostik mit geführten Wellen in Laminaten

2022-heute: Wissenschaftlicher Leiter und Projektleiter in der U Bremen Research Alliance AI Center for Health Care (Add. Fertigung und Prozessoptimierung)

Stefan Bosse - AFEML - Module 0: About the Lecturer - Koblenz

Koblenz

2025-heute: Professur Praktische Informatik im Fachbereich Informatik, Universität Koblenz

Stefan Bosse - AFEML - Module 0: About the Lecturer - Koblenz

Koblenz

2025-heute: Professur Praktische Informatik im Fachbereich Informatik, Universität Koblenz

Themenschwerpunkte Verteilte Künstliche Intelligenz, Betriebssysteme, Technische Systeme

Arbeitsgebiete

1. Generative Modelle für die synthetische Erzeugung von Sensordaten und synthetische Instrumentenmodelle
2. Analoges Rechnen - Analogrechner kombiniert mit KI Methoden
3. Verteilte Sensornetzwerke, Schadensdiagnostik, SHM, Sensorik
4. Mess- und Prüftechnik (ZfP, NDT, Ultraschall, X-ray)

Stefan Bosse - AFEML - Module 0: About the Lecturer - Siegen

Siegen

Seit 2022: Lehrbeauftragter im Fachbereich Msschinenbau, Lehrstuhl für Materialkunde und Werkstofftechnik, Universität Siegen

Stefan Bosse - AFEML - Module 0: About the Lecturer - Siegen

Siegen

Seit 2022: Lehrbeauftragter im Fachbereich Msschinenbau, Lehrstuhl für Materialkunde und Werkstofftechnik, Universität Siegen

Wissenschaftliche Zusammenarbeit Bremen - Siegen - Koblenz

Stefan Bosse - AFEML - Module 0: About the Lecturer - Teaching and research topics

Teaching and research topics

Stefan Bosse - AFEML - Module 0: About the Lecturer - My Research Topics

My Research Topics

Sensor Data Sciences

Stefan Bosse - AFEML - Module 0: About the Lecturer - Research topics

Research topics

Distributed and Applied AI

Stefan Bosse - AFEML - Module 0: About the Lecturer - Research topics

Research topics

Simulation

Stefan Bosse - AFEML - Module 0: Introduction - Research topics

Introduction

Stefan Bosse - AFEML - Module 0: Introduction - Target Audience

Target Audience

Student curriculum module area: Bachelor and Master Specialization

Stefan Bosse - AFEML - Module 0: Introduction - Main topics in this course

Main topics in this course

Metrics, taxonomies, methods, algorithms of ...

Stefan Bosse - AFEML - Module 0: Introduction - Main topics in this course

Stefan Bosse - AFEML - Module 0: Introduction - Special topics in this course

Special topics in this course

• From physical measurements to material structure analysis

• Image processing (i.e., for 2D and 3D data volumes) is discussed with practical lessons using real-world data

• Tomography fundamentals are introduced with practical applications and lessons using real-world data (including data from Siegen's Zeiss Xradia MicroCT device)

• Microscopy Imaging

Stefan Bosse - AFEML - Module 0: Introduction - Fields of Application

Fields of Application

Stefan Bosse - AFEML - Module 0: Introduction - Organisation

Organisation

1. Introduction to basic methods and algorithms with simple "sandbox" tutorials and exercises (partially home/office work)

2. Discussion of scientific application cases and available data

3. Application of methods and algorithms to different complex application cases

4. Discussions and presentation of results

Stefan Bosse - AFEML - Module 0: Introduction - Data

Data

$${f{{\left(\hat{{D}},\vec{{M}}\right)}}}:\hat{{D}}\times\vec{{M}}\to{\left\lbrace\matrix{\hat{{R}}\\\hat{{F}}}\right.}\\ {g{{\left(\hat{{R}}\right)}}}:\hat{{R}}\to\hat{{F}}$$

F is commonly derived from intermediate feature selected data R, not from raw data D

Stefan Bosse - AFEML - Module 0: Introduction - Data

Data

Input Data

1. Dimension
   • 0D: Scalar data (stationary temperature, voltage, current, field strength)
   • 1D: Time-dependent vibration or guided ultrasonic wave signals measured at a specific geometrical position
   • 2D: Images (single projection, visible light, infrared, X-ray)
   • 3D: Time-dependent 2D images (videos) or 3D spatial data (multi-projection tomography, LIDAR, Coupled Ultrasonic, (Ultra)Sonography, Air-guided Ultrasonic)
   • 4D: Time-dependent 3D spatial data (tomography videos)

Stefan Bosse - AFEML - Module 0: Introduction - Data

Data

2. Variable Class
   • Metric/Numerical, continuous (e.g., temperature)
   • Metric/Numerical, discrete (e.g., number of items)
   • Categorical/Symbolic (e.g., damage class)
   • Categorical/Interval (e.g., range intervals)
3. Size
   • One data value
   • 1000 data values
   • 1M data values
   • ..
4. Temporal Characteristics
   • Periodic
   • Sporadic
   • One-shot

Stefan Bosse - AFEML - Module 0: Introduction - Data

Data

Stefan Bosse - AFEML - Module 0: Introduction - Data

Data

Meta Data

• Any experiment is situated in a context:
  • Date, Place
  • Set-up
  • Operator
  • Environmental conditions (temperature)
  • Measuring technology, device, parameters
• The device under test (DUT)
  • Dimensions
  • Material
  • ..
• Data formats and structure!

Stefan Bosse - AFEML - Module 0: Introduction - Data Sources and Storage

Data Sources and Storage

Stefan Bosse - AFEML - Module 0: Introduction - Data Examples

Data Examples

Anomaly detection in 3D X-ray CT image volumes (composite materials with defects) Materials 2022, 15(13), 4645; https://doi.org/10.3390/ma15134645

Stefan Bosse - AFEML - Module 0: Introduction - Data Examples

Data Examples

Breakage prediction with tensile test data (metal probes) SciForum ECSA 2020, https://doi.org/10.3390/ecsa-7-08279

Stefan Bosse - AFEML - Module 0: Introduction - Data Examples

Data Examples

Damage detection with Guided Ultrasonic Waves (composite material with pseudo defects) SysInt 2022, DOI: 10.1007/978-3-031-16281-7_35

Stefan Bosse - AFEML - Module 0: Introduction - Data Examples

Data Examples

Porosity analysis with metallurgical microphotographs of sliced surfaces (additive manufacturing) Materials 2022, 15(20), 7090; https://doi.org/10.3390/ma15207090

Stefan Bosse - AFEML - Module 0: Introduction - Data Examples

Data Examples

Feature marking of micro cracks and crack propagation of metal probes after applying stress AG Brandt, U Siegen

Stefan Bosse - AFEML - Module 0: Introduction - Features

Features

• Categorical features, e.g., damage classes, machine states, etc.
• Metric/numerical features, e.g., length of a crack, spatial position, time, temperature
• Regions of interest (labelled polygons)
• Anomaly

Stefan Bosse - AFEML - Module 0: Introduction - Features

Features

There are input and output features.

Input signal features

Selected features from the raw signal data that pose a strong correlation with the output target features $\Rightarrow$ Transformation of signal data to Intermediate variables

Examples: Statistical features of time data series (average, deviation, skewness, ..), frequency spectrum, ROI, wavelet transformation coefficients.

Output target features

The final relevant information that is obtained from the data, i.e., the answer to a question.

Stefan Bosse - AFEML - Module 0: Introduction - Features

Features

Stefan Bosse - AFEML - Module 0: Introduction - Methods and Algorithms

Methods and Algorithms

A typical "Data Mining" work flow consists of different stages:

1. Data pre-processing (e.g., filtering)
2. Signal feature selection (what are the best/strongest intermediate variables) and their computation
3. Selecting a predictor and analysis model (e.g., an artificial neural network or a decision tree)
4. Fit the model to training data
5. Apply the model to test data
6. validation and improvement.

Stefan Bosse - AFEML - Module 0: Introduction - Methods and Algorithms

Methods and Algorithms

Statistical Methods

Statistical analysis of signal data and computation of statistical aggregate variables. Mostly applied to data vectors (data series) or matrix data (images) $\Rightarrow$ Horizontal Analysis

Algebraic and geometrical Methods

Analysis of data variables with respect to correlation strength and information content, e.g., Principle Component Analysis, with optional following data transformation $\Rightarrow$ Vertical Analysis

Transformation Methods

Transformation of data spaces, e.g., time-to-frequency (fourier) or time-to-time/frequency (wavelet) spaces. Image processing applies often kernel-based transformations, e.g., contrast amplification, edge detection.

Stefan Bosse - AFEML - Module 0: Introduction - Methods and Algorithms

Methods and Algorithms

Regression Methods

A numerical parametrized model (e.g., polynomial function) is fitted to labelled training data $\Rightarrow$ Supervised Training

Selection Methods

A tree structure is created from labelled training data $\Rightarrow$ Supervised Training

Clustering Methods

A tree or functional graph structure is created from unlabelled training data $\Rightarrow$ Unsupervised Training and grouping of experiments.

Stefan Bosse - AFEML - Module 0: Introduction - Methods and Algorithms

Methods and Algorithms

In general, data processing and machine learning is a functional composition.

• ML searches mapping functions M=F(X): X $\to$ Y.

• ML is commonly an approximation of a hypothesis model H ≈ F of an unknown real (world) model M

• There are infinite approximations (hypothesis models), the so called model space

Stefan Bosse - AFEML - Module 0: Introduction - Methods and Algorithms

Methods and Algorithms

Stefan Bosse - AFEML - Module 0: Introduction - Methods and Algorithms

Methods and Algorithms

Toolbox

Stefan Bosse - AFEML - Module 0: Introduction - Methods and Algorithms

Methods and Algorithms

Stefan Bosse - AFEML - Module 0: Introduction - Software

Software

Stefan Bosse - AFEML - Module 0: Introduction - Software

Software

Stefan Bosse - AFEML - Module 0: Introduction - Software Architecture

Software Architecture

Work flow architecture

Stefan Bosse - AFEML - Module 0: Introduction - Toolbox

Toolbox

• The WorkBook and WorkShell software provide a broad range of methods and algorithms, partially using well-known software packages:
  • Image processing with OpenCV.js (https://docs.opencv.org/3.4/d5/d10/tutorial_js_root.html)
  • Image processing with image.js (https://github.com/image-js/image-js)
  • Artificial Neural networks with neataptic (https://wagenaartje.github.io/neataptic/)
  • Convolutional Neural Networks with ConvNetJS (https://cs.stanford.edu/people/karpathy/convnetjs/)

... andy many more ...

• But most of all: The numerical R+ programming language, an adaptation of R widely used in science and industry (not only for statistical analysis of table data)

Stefan Bosse - AFEML - Module 0: Introduction - R+

R+

• Introduction to R programming
  • Basic data structures
  • Basic computational statements
  • Advanced computational statements
  • Control statements
  • Input and Output operations
  • Image processing

On one hand, R+ extends the core R syntax (e.g., simplified list, vector, and matrix constructors), on the other hand it implements only a sub-set of R (packages, functions).

Stefan Bosse - AFEML - Module 0: Introduction - R+

R+

use math,plot
m = matrix(runif(100),10,10)
plot(m,auto.scale=TRUE)

Stefan Bosse - AFEML - Module 0: Introduction - R+ IDE Live

R+ IDE Live

+

Stefan Bosse - AFEML - Module 0: Introduction - Lessons and Exercises

Lessons and Exercises