Student theses

• “SISProof” Project: Quantified uncertainty of machine learning models: An investigation of utility using the Tennessee Eastman process.

• Functional test of an adaptive safety device for process safety at a laboratory reaction

• “RiIM” Project: Evaluation and testing of classical as well as ML approaches to modeling corrosion on buried high-pressure gas pipelines

• Projekt “sRMC”: Development of an acoustic early detection system for gas leaks in industrial plants

BACHELOR/MASTER THESES | “SISProof” Project

Quantified uncertainty of machine learning models: An investigation of utility using the Tennessee Eastman process

In the context of this work, the potential added value of a quantified uncertainty of machine learning models shall be investigated. Different methods / ML algorithms with this property will be compared and evaluated.

Working steps:

• Literature research: Quantified uncertainty of ML algorithms
• Selection of appropriate ML algorithms/methods to reproduce quantified uncertainty (e.g., Gaussian process regression, neural processes, ensemble models, etc.)
• Creation of ML models for anomaly detection with the ability to quantify their own uncertainty in prediction
• Performance comparison in terms of recognition rate (own models and models from literature)
• Comparison of the created models with respect to the quantified uncertainty as well as the evaluation of their validity
• Evaluation of the usefulness of quantified uncertainty in terms of the suitability of ML models for safety-critical systems.

BACHELOR/MASTER THESES | “SmOP” Project

Functional test of an adaptive safety device for process safety at a laboratory reaction

The aim of this work is to plan, perform and evaluate laboratory experiments to test the adaptive safety device on a laboratory reactor with an exemplary chemical reaction. The experimental results will then be evaluated and the functionality of an adaptive safety device will be assessed.

Working steps:

• Systematic design of an experimental plan taking into account the requirements of the adaptive safety device
• Development and simulation of the proposed test modes with a suitable reactor simulation
• Planning and construction of the laboratory plant (mechanics, control system, software) and startup using water
• Conducting and evaluating the experiments for the developed experimental program with a real chemical reaction
• Analysis of the evaluation results and comparison with the simulation data
• Evaluation of the functionality of the SmOP

MASTER THESES | “RiIM” Project

Evaluation and testing of classical as well as ML approaches to modeling corrosion on buried high-pressure gas pipelines.

Within the scope of this work, physical, empirical and ML models for corrosion modeling are to be researched. The researched models are to be evaluated according to defined criteria. A justified selection of the models shall be investigated by a sensitivity analysis after implementation. Trends, limitations and challenges of corrosion modeling are to be identified through critical questioning.

Working steps:

• Literature research of corrosion processes on buried gas pipeline and modeling by physical, empirical, as well as ML approaches
• Definition of the requirements for corrosion modeling (input variables, data quality, accuracy, etc.)
• Assessment of the researched models against the defined requirements
• Implementation of relevant models based on the previous assessment
• Sensitivity analysis of the models with respect to input variables & influencing parameters.
• Final evaluation of the of corrosion modeling on trends, limitation and challenges & influencing parameters.

MASTER THESES | “sRMC” Project

Development of an acoustic early detection system for gas leaks in industrial plants

Within the scope of this work, acoustic measurement data are to be acquired with the help of a newly developed test system. The measurement data will then be evaluated using suitable signal transformation methods and subsequent pattern recognition (machine learning).

Working steps:

• Literature research on signal transformations and machine learning for acoustic data acquisition.
• Creation of a control and evaluation of the measurement cycle with Python and NI-DAQmx
• Development of a suitable method for data transformation of acoustic measurement data (e.g. Short-Time-Fourier-Transformation, Wavelet-Transformation)
• With the help of various machine learning algorithms, the leakage data is detected in the measurement data and the condition of the plant is evaluated
• Uncertainty analysis of the results

MASTER THESES | “EuroValve” Project

CFD simulation of a safety valve and investigation of the forces acting on the safety valve disc.

Construction of a two-dimensional simulation of a safety valve and investigation of the forces acting on the safety valve disc. It is possible to build on previous simulations and measurements.

Working steps:

• Literature research on available simulations and measurements
• Definition of relevant influential parameters on the distribution of the flow force at the valve disc
• Creation of a two-dimensional simulation of the flow forces
  at the valve disc based on an existing safety valve for single-phase compressible media
• Sensitivity analysis in relation to the relevant influential parameters

MASTER THESES | “RiIM” Project

Identification of risks from NaTech as a result of climate change on gas pipeline.

In this thesis, the risks for gas pipelines due to the consequences of climate change will be investigated. For this purpose, NaTech scenarios, i.e. threat scenarios for technical facilities such as gas pipelines due to natural hazards such as floods, ground movements, lightning or extreme temperatures, are summarized from historical events. The effects on the mechanical integrity of high-pressure gas pipelines are then calculated.

Working steps:

• Literature research on historical events, NaTech, climate models, and mechanical failure of pipelines
• Identification of hazard scenarios using historical examples
• Modeling of the impact on the high-pressure gas pipeline e.g. mechanical stress due to the uplift of the steel pipeline
• AAnalysis of the probability of occurrence due to influencing factors such as environment, line data and climate development
• Evaluation of the risks of the scenarios in relation to the pipeline, climate development and the environment

MASTER THESES | “EuroValve” Project

CFD simulation of real gas effects in a safety relief valve and in its inlet line.

Within the scope of this work, an existing one-dimensional CFD simulation for modeling the safety valve stability during gas flow is to be extended by real gas effects. The inlet line is to be taken into account.

Working steps:

• Literature research on relevant real gas effects in a safety relief valve
• Modeling of the pressure drop in the inlet flow line for real gases
• Extension of the existing simulations with suitable equations of state
  
• Influence analysis of the length of the inlet line on the safety valve stability during flow with real gases
• Evaluation of simulation results based on measurement data

BACHELOR/MASTER THESES | “SISProof” Project

Development of hybrid virtual sensors in safety-related applications: An evaluation of suitable combinatorics and ML methods

Within the scope of this work, virtual sensors with hybrid model structure are to be developed for the verification of safety-relevant sensors. Combinatorics of rigorous and ML models as well as ML methods are to be evaluated.

Working steps:

• Literature research on combinatorics and ML methods
• Definition of requirements for the operation of soft sensors
• Pre-selection of suitable ML methods
• Preselection of suitable combinatorics of rigorous and ML models
• Development of hybrid models for the verification of safety-relevant sensors based on a stirred tank reactor
• Evaluation of the hybrid models with regard to previously defined requirements