Register now After registration you will be able to apply for this opportunity online.
Fully Automated Evaluation of Raman Spectra in a Self-Driven Thermodynamics Lab
This thesis focuses on fully automating the evaluation of Raman spectra in a self-driven thermodynamics lab to accelerate the development of sustainable chemical processes or novel heat pump concepts. By integrating Machine Learning (ML) with advanced spectral evaluation algorithms, the aim is to achieve complete lab autonomy. The methodology combines data-driven and physically-based approaches, including synthetic spectrum generation for ML training.
Vapor-Liquid Equilibrium data of mixtures are an essential part for the design of novel sustainable processes. Despite advancements in predictive models, experimental data quality remains unmatched for dependable process design. Unfortunately, experiments are expensive and time-consuming, often posing a bottleneck in validating models and process designs. The Energy & Process Systems Engineering Lab at ETH Zürich aims to establish a self-driven thermodynamics lab, enhancing experimental throughput to save time and cost. This also includes automating the spectral evaluation necessary for various composition measurements. These evaluations typically rely on expert knowledge and are performed manually, representing one of the final hurdles toward a fully autonomous operation. By employing Machine Learning (ML) techniques alongside advanced spectral evaluation algorithms, we aim to achieve complete lab autonomy to maximize the output of our lab. Consequently, we can provide the essential mixture property data to design novel sustainable processes.
Vapor-Liquid Equilibrium data of mixtures are an essential part for the design of novel sustainable processes. Despite advancements in predictive models, experimental data quality remains unmatched for dependable process design. Unfortunately, experiments are expensive and time-consuming, often posing a bottleneck in validating models and process designs. The Energy & Process Systems Engineering Lab at ETH Zürich aims to establish a self-driven thermodynamics lab, enhancing experimental throughput to save time and cost. This also includes automating the spectral evaluation necessary for various composition measurements. These evaluations typically rely on expert knowledge and are performed manually, representing one of the final hurdles toward a fully autonomous operation. By employing Machine Learning (ML) techniques alongside advanced spectral evaluation algorithms, we aim to achieve complete lab autonomy to maximize the output of our lab. Consequently, we can provide the essential mixture property data to design novel sustainable processes.
This thesis uses data-driven and physically-based approaches to evaluate spectra fully autonomously in a self-driven thermodynamics lab with the option of some experimental work. The main work packages are:
- Familiarize yourself with the topic and conduct literature research.
- Write a script to create synthetic Raman spectra used as training data.
- Training an ML model to perform automatic background removal, initially using synthetically generated spectra and subsequently refining it with real spectra from our setup.
- Implement an automated version of Indirect Hard Modeling (IHM) to evaluate the Raman spectra in combination with the previously developed background removal algorithm. Validate it with actual Raman spectra from our setup.
- Finally, you will document and present the results in a scientific manner.
This thesis uses data-driven and physically-based approaches to evaluate spectra fully autonomously in a self-driven thermodynamics lab with the option of some experimental work. The main work packages are:
- Familiarize yourself with the topic and conduct literature research.
- Write a script to create synthetic Raman spectra used as training data.
- Training an ML model to perform automatic background removal, initially using synthetically generated spectra and subsequently refining it with real spectra from our setup.
- Implement an automated version of Indirect Hard Modeling (IHM) to evaluate the Raman spectra in combination with the previously developed background removal algorithm. Validate it with actual Raman spectra from our setup.
- Finally, you will document and present the results in a scientific manner.