# Case Studies

CASE-STUDY ASSIGNMENTS 2015

SPOTSeven team offers three case studies in the Case Study Module: Modeling, Simulation, Control and Optimization (MSCO) for Master AIT students at CUAS. The general topic of these case studies reads:  “Meta-Model based optimization of simulation models for industrial applications”

11. Time Series Methods for HVAC systems (Cooperation with BOSCH Thermotechnology)
Heating, Ventilation and Air Conditioning (HVAC) systems set up a new challenge for up-to-date and comprehensive time series analysis methods. Such systems provide a couple of interesting research questions because different devices in a house are cross- linked and need to be set up with respect to one another. This can only be made by measuring the current state of the house/system and controlling all the devices simultaneously. This case study provides the chance to work with real as well as simulated data from HVAC systems. As a first step, the consumer behavior is observed and recorded to time series, which are analyzed using classical statistical methods in a second step. Here, correlation and sensitivity analysis play a major role in the investigations and real restrictions, constraints have to be considered, respected. After having investigated the recorded data with the classical, conventional data mining methods, several new analysis techniques can be explored during the case study. Different roads might be followed by the case study participant:

• pattern recognition in the costumer data and generalization to cope with new recorded system data,
• development, implementation, and test of new learning strategies with the aim to provide better strategies for HVAC systems,
• definition and realization of system comparisons to develop adequate behavior of simulated data in contrast to real system data.

What do you learn?

• State-of-the-art data-mining techniques
• Application of modern software tools
• Exploratory data analysis
• Communication skills
• Documentation

Requirements:

• Math skills and interest in stats
• Basic programming skills
• Interest in complex real-world problems

Tutors: T. Bartz-Beielstein, J. Stork, M. Zaefferer
Case study can be prepared during the summer months.

12. Visualization for Landscape Analysis of Combinatorial Optimization Problems in R

To understand features and behavior of optimization problems, visualization methods are of major importance. Several methods (e.g., surface plots or contour plots) are available to visualize continuous fitness landscapes. On the other hand, proper visualization for combinatorial or discrete optimization problems is a more difficult problem. Few methods exist, including techniques like Barrier Trees. To review and summarize these existing ideas would be the starting point of the proposed work. Moreover, existing techniques as well as your own ideas should be implemented in (or interfaced with) the R programming language. The resulting approaches and methods are supposed to be tested with some application examples from recent re-search projects.

What do you learn?

• State-of-the-art optimization approaches
• Visualization methods
• Documentation of scientific projects
• Current research problems

Requirements:

• Interest in optimization and visualization
• Scientific working and writing
• Good programming skills

Tutors: M. Zaefferer, T. Bartz-Beielstein
Case study can be prepared during the summer months.

PREVIOUS CASE-STUDIES

2014: Optimization of simulation models: Cyclone Optimization

The reduction of dispersed particles from gas is a demanding task which appears in different devices from coal-fired power plants to hoovers. Cyclone dust separators are frequently used for these tasks. Their advantages are simple structure, low costs and ease of operation. Collection efficiency and pressure loss are the two most important performance parameters. They are heavily influenced by the choice of several geometrical design parameters, like height or diameter.
This case study aims in formalization, implementation, and optimization of the constrained cyclone dust separator optimization problem. Here, constraints may have different natures:

1. An objective constraint is defined on the pressure loss allowing only a certain amount here. This finally drags the multiobjective optimization problem down to an only singleobjective instance.
2. Geometrical constraints are defined on the geometry parameters of the cyclones. They allow only certain values for the parameters or define certain relations between them.

Case study participants are expected to have a background in optimization, typically using standard R implementations of evolutionary algorithms. Furthermore, they should have an interest in extending these algorithms with constraint handling techniques.

Tutors: T. Bartz-Beielstein, B. Breiderhoff, B. Naujoks, and M. Zaefferer

2014: Big Data: Time Series Methods and Data Mining for HVAC systems

Analysis of heating, ventilation, and air conditioning (HVAC) systems
HVAC systems set up a new challenge for up-to-date and comprehensive time series analysis methods. Such systems provide a couple of interesting research questions because different devices in a house are cross-linked and need to be set up with respect to each other. This can only be done by measuring the current state of the house/system and controlling all the devices simultaneously. As a first step, the consumer behavior is observed and recorded to time series, which are analyzed using classical statistical methods in a second step. After having investigated the recorded data with conventional methods, different new analysis techniques can be explored during the case study. Different roads might be followed by the case study participant: pattern recognition in the costumer data, development, implementation, and test of new learning strategies, or definition and realization of system comparisons to develop adequate behavior of simulated time series. This case-study will be done in close collaboration with our industrial partners, see http://www.spotseven.de/partner/

Tutors: Bartz-Beielstein, Moritz, Stork

2014: Damage detection of composites

Composites are high-performance materials, with application in the areas of aerospace, automotive and construction. In recent years, the use of composites has increased due to the superior properties of strength, stiffness, weight, performance, corrosion, etc. However, the degradation process of these materials is complex and there is a lack of a dependable technique to assess the state of a composite structure.  With the use of the NASA dataset with collected information of the fatigue testing on a composite structure, evidence of degradation in the structure is seek through the data analysis of Lamb waves generated and measured by PZT-sensors installed on the area of interest. The main goal is to identify Condition indicators that can be correlated to the occurrence of damage.

Tutors: T. Bartz-Beielstein, M. Rebolledo, M. Zaefferer