**Disclaimer**: The content of this document is public!

Benefit of registration at hackmd.io:
Your editions are automatically connected to [your name](https://hackmd.io/settings#general).

###################################################

### Brain storming session 2

# Application of Machine Learning Methods

## Participants
13.8.20 online via MS-Teams
- Loredana Kehrer (KIT, Host)
- Ludwig Schöttl (C2, KIT, Moderator)
- Luise Kärger (KIT)
- Jennifer Sears (D1, UWindsor)
- Tarkes Dora (PostDoc, ITM, KIT)
- Jennifer Johrendt (UWindsor, Design stream)
- Michael Thompson (McMaster)
- Benedikt Rohrmüller (C1, KIT)
-

## Motivation
- Classification and segmentation of objects within images
- Pattern recognition
- Fast data processing/optimization
- Feature extraction
- Reduce resource-intensive impact/mechanical testing

## Current status

- Standard methods vs. machine learning
- Advantages and Disadvantages of machine learning

#################################################################

## Survey:

Please feel free to extend this list:

### Which tools / algorithms do you use currently?
- Name: Ludwig Schöttl (C2, IAM-WK)
- Keras and TensorFlow (Python)
- Image processing
- Object classification and segmentation
- Name: Jennifer Sears (D1, UWindsor)
- Matlab (GUI/manual algorithm expansion)
- Python (Keras/Tensorflow/PCA/SVR/SVM/Polynomial Regression)
- Pattern Recognition
- Comparison to existing statistical methods (Regression/Spearman reduction)
- Clustering & Curve Fitting
- Name: Michael Thompson
- Quality assurance
- Acoustic Analysis
- Analyzing high frequence signals
- Name: Jennifer Johrendt
- Prediction models based on ML
- Time reduction of simulations
- Name: Luise Kärger
- Process optimization (draping)
- LFT process/geometry optimization (3rd Gen. IRTG)
### What do you need?
- Name: Jennifer Sears (D1)
- expansion of training data set for LFT/HP-RTM/LCM from Canadian or German group (processing parameters/high speed imagery re: crack detection)
- Python or Matlab best practices

### What would you like to have?
- Name: Tarkes Dora (PostDoc, ITM,KIT)
- Introductory seminar: Machine learning/ANN/Deep learning, how it works, state of the art technologies, reliability and robustness, advantages/disadvantages, open source and commerical tools: widely and commonly used tools or considered as standard tool.
- Name: Jennifer Johrendt (UWindsor)
- Knowledge of/access to validation data (measured and simulated) with details of the context in which data was collected/generated
- Data can include:
- material characterization
- processing parameters
- - Name: Jennifer Sears (D1, UWindsor)
- experimental data by process type (LFT/HP-RTM/LCM or UP tape consolidation) including impact and uCT for us in Gen 3 (Canada & Germany)
- manufacturing parameters
- material characterization
- high-speed imagery

### What would you like to know?
- Name: Ludwig Schöttl
- Machine Learning applications of other projects (canada/germany)
- Type of the input data and task of the machine learning method
- Common challenges and helpful solutions
- Applied tools (Python, Matlab, Keras, PyTorch, ...)
- Name: Tarkes Dora(PostDoc, ITM,KIT)
- Possibility of application as fast root-finding solvers (Currently I use standard MATLAB trust-region-dogleg solvers in a viscoelastic homogenization problem. I was wondering if this can be speeded up using ML methods.)
- Material physics driven material modelling (linear/nonlinear)
- Application to mean-field homogenization
- Name: Jennifer Sears (D1, UWindsor)
- Alternate ML tools
- To gain knowledge from the group using draping/FEM-simulation and ML (Dr. Karger) and how this can integrate with LFT for Gen 3.
- Collaboration possibilities between IRTG/ICRC
- Results from other members on the success/failures of ML methods for adjusting design process in its infancy.
- Name: Benedikt Rohrmüller
- General information
- Possibilities to include ML in the research work

### Which tools do you think would be helpful?
- Name: Tarkes Dora (PostDoc, ITM,KIT)
- An introductory seminar will give a general idea. It will be helpful to explore it's utility in different problems of composite materials.
- Name: Jennifer Johrendt (UWindsor)
- Best practices for ML methods application
- Name: Jennifer Sears (D1, UWindsor)
- Monthly meetings with specific ML group (Canada & Germany)

#################################################################

## Recommended literature

- F. Chollet (2017) "Deep Learning with Python"
- S. Samarasinghe (2007) "Neural networks for applied sciences and engineering: from fundamentals to complex pattern recognition"
- J. Moolayil (2019) "Learn Keras for Deep Neural Networks : A Fast-Track Approach to Modern Deep Learning with Python"
- ... please feel free to add literature here ...

#################################################################

## Ideas and suggestions

- Meetings with short presentations focused on ML topics. Exchange of ML experiences (M. Thompson)
- Best practice document (J. Johrendt)
- Inivting experienced ML-user for presentations (A. Langhoff)