Pretrained Neural Networks in HSC Chemistry Sim
Mansikka-aho, Jarkko (2024)
Mansikka-aho, Jarkko
2024
All rights reserved. This publication is copyrighted. You may download, display and print it for Your own personal use. Commercial use is prohibited.
Julkaisun pysyvä osoite on
https://urn.fi/URN:NBN:fi:amk-202403275242
https://urn.fi/URN:NBN:fi:amk-202403275242
Tiivistelmä
Many researchers are familiar with the Python programming language and can create machine learning models using popular third-party tools such as TensorFlow and PyTorch. Unfortunately, the use of these models in HSC Sim has not previously been possible. The aim of this thesis was to develop a method for integrating Python-based machine learning models into the HSC Sim program, a module of the HSC Chemistry software suite by Metso Oyj. This software is widely used for modelling and simulating various processes in the fields of mineral processing, hydrometallurgy and pyrometallurgy.
HSC Sim faces challenges in processing speed, particularly with large-scale processes, and in modelling phenomena that either lack explicit mathematical formulas or are too complex to be modelled using theoretical first principles methods. In these situations, the incorporation of machine learning models could offer a crucial solution, potentially enhancing computational speed and addressing the complexities of the simulations. For instance, this approach could make it feasible to model the wear and tear of equipment using machine learning models trained with experimental data and then integrate it into the HSC Sim simulations.
This development work was conducted through pragmatic applied research that focused on creating a concrete new solution for HSC Sim based on empirical observations, benchmarking and group discussions. The work was composed of three parts. The first part focused on developing machine learning models and converting them into the ONNX format, which is a standardized format designed to move and use models across different frameworks. In the second part, C# example solutions were developed to find the most appropriate techniques and tools to use these models within the .NET framework. The final part involved the integration of these models into HSC Sim software.
As a result, a successful solution was developed, enabling the straightforward integration of machine learning models within HSC Sim. This solution includes methods for importing, exporting, and using ONNX models and converting them to a more suitable form for HSC Sim. Moreover, an add-in was developed to enable the use of these models in Excel.
HSC Sim faces challenges in processing speed, particularly with large-scale processes, and in modelling phenomena that either lack explicit mathematical formulas or are too complex to be modelled using theoretical first principles methods. In these situations, the incorporation of machine learning models could offer a crucial solution, potentially enhancing computational speed and addressing the complexities of the simulations. For instance, this approach could make it feasible to model the wear and tear of equipment using machine learning models trained with experimental data and then integrate it into the HSC Sim simulations.
This development work was conducted through pragmatic applied research that focused on creating a concrete new solution for HSC Sim based on empirical observations, benchmarking and group discussions. The work was composed of three parts. The first part focused on developing machine learning models and converting them into the ONNX format, which is a standardized format designed to move and use models across different frameworks. In the second part, C# example solutions were developed to find the most appropriate techniques and tools to use these models within the .NET framework. The final part involved the integration of these models into HSC Sim software.
As a result, a successful solution was developed, enabling the straightforward integration of machine learning models within HSC Sim. This solution includes methods for importing, exporting, and using ONNX models and converting them to a more suitable form for HSC Sim. Moreover, an add-in was developed to enable the use of these models in Excel.