Periodic Reporting for period 2 - MatCHMaker (Open data and industry driven environment for multiphase and multiscale Materials Characterization and Modelling combining physics and data-based approaches)
Okres sprawozdawczy: 2024-06-01 do 2025-05-31
The MatCHMaker project is determined to support excellence in research on methods and tools for advanced materials development and has three main objectives:
1. Develop a model-based workflow to accelerate advanced materials design and validation,
2. Reinforce traceability, integrity and interoperability of characterisation and modelling (C&M) data and workflows
3. Make C&M knowledge and data accessible by developing an open data repository enabling the documentation of models and data in an interoperable way.
MatCHMaker aims to validate the project results on three use cases from industries, namely construction (cement), energy (Solid Oxide Fuel Cells/Solid Oxide Electrolysis Cells) and mobility (Proton-Exchange Membrane Fuel Cells).
The main challenge is to accelerate multiphase multiscale materials design. To achieve it, it is necessary first to link macroscopic properties to microstructural features while covering different scales and physics. This link will be provided using advanced physics-based or databased models, coupled with experimental data (for verification or confrontation).
A second challenge is to accelerate materials characterization, through development of efficient and robust machine learning tools.
Finally, a correlation between the material composition and the sustainability and social impacts is provided. These developments are done in an interoperable way through the development of appropriate domain ontologies.
1. Develop a model-based workflow to accelerate advanced materials design and validation,
2. Reinforce traceability, integrity and interoperability of characterisation and modelling (C&M) data and workflows
3. Make C&M knowledge and data accessible by developing an open data repository enabling the documentation of models and data in an interoperable way.
MatCHMaker aims to validate the project results on three use cases from industries, namely construction (cement), energy (Solid Oxide Fuel Cells/Solid Oxide Electrolysis Cells) and mobility (Proton-Exchange Membrane Fuel Cells).
The main challenge is to accelerate multiphase multiscale materials design. To achieve it, it is necessary first to link macroscopic properties to microstructural features while covering different scales and physics. This link will be provided using advanced physics-based or databased models, coupled with experimental data (for verification or confrontation).
A second challenge is to accelerate materials characterization, through development of efficient and robust machine learning tools.
Finally, a correlation between the material composition and the sustainability and social impacts is provided. These developments are done in an interoperable way through the development of appropriate domain ontologies.
Use cases requirements and Key Performance Indicators - The activities focused on the collection of needs and technical requirements from the industrial partners for Characterization and Modelling tools and models that will be developed, verified, and validated later. Moreover, all European and international existing standards initiatives on advanced materials modelling were mapped to identify potential gaps. Finally, the reference Key Performance Indicators framework in terms of technical, economic, environmental, and social Key Performance Indicators list and target values were defined for validation activities.
Semantic data models and ontology development - The activities defined the groundwork for the creation of data models and ontologies in the MatCHMaker project. The data models and ontologies will be used to semantically document the data thatare produced and used within the project, and potentially coming from external sources.
Requirements for Machine Learning - The problems that are to be solved by machine learning (ML) methods for the MatCHMaker project are defined, as precisely as possible. ML tools are planned to be used to analyse data in all three use cases. They must be developed and adapted specifically to the data and the task to address. More specifically, the activity consists of determining the data that will be used as input for the ML models, the output in the supervised cases, and the metric that will measure the error or the performances of the models.
Parametric model for materials sustainability assessment - This activity evaluates the environmental performances of the three use cases (cement, SOEC, PEMFC) using the Life Cycle Assessment (LCA) methodology. Following the results assessment made by hotspot analysis, a parametric model is drawn to identify the correlations between impacts and the variation of selected parameters characterizing the products under analysis. Some assumptions are common within the three assessments: for instance, only the Manufacturing process is included in the system boundaries of the study. Following the impact assessment calculations, the results are evaluated and interpreted, identifying the main hotspots through a contribution analysis. From these results, the LCA parametric model is built. Then advanced study of Life Cycle Assessment (LCA) and Social Life Cycle Assessment (S-LCA) has been applied to three use cases.
Technical specifications of the Open Repository - the technical specifications and software architecture of the MatCHMaker repository and framework are defined according to different predefined “zones” (abstract zones, repositories zones and API service. Following an extensive analysis of the existing repositories for materials and for software systems, a new concept was defined for the MatCHMaker open repository. It especially introduces the concept of Distributed Data and Knowledge Mesh (DDKM), a concept with roots in the Data Mesh concept.
Experimental results - Experiments have been carried out for the three use cases of the project. For cement, SEM, XRF, and XRD measurements of phase fractions, and measurements of mechanical strength and porosity are the main data of interest. Micro-scale 3D imaging was done with FIB-SEM, while the hardening phase (calcium silicate hydrate) was studied by TEM methods. For SOC electrodes, SEM/EDX observations have been made on samples with several oxidation times. The initial microstructures have been characterized by X-ray nanotomography, a non-destructive technique, at the European Synchrotron Radiation Facility (ESRF). For PEMFC electrodes, TEM images of the Pt nanoparticles are the main data of interest. Electron tomography experiments were also carried out on the catalyst powders.
ML models - different ML models have been implemented for the different Use Cases of the project.
Semantic data models and ontology development - The activities defined the groundwork for the creation of data models and ontologies in the MatCHMaker project. The data models and ontologies will be used to semantically document the data thatare produced and used within the project, and potentially coming from external sources.
Requirements for Machine Learning - The problems that are to be solved by machine learning (ML) methods for the MatCHMaker project are defined, as precisely as possible. ML tools are planned to be used to analyse data in all three use cases. They must be developed and adapted specifically to the data and the task to address. More specifically, the activity consists of determining the data that will be used as input for the ML models, the output in the supervised cases, and the metric that will measure the error or the performances of the models.
Parametric model for materials sustainability assessment - This activity evaluates the environmental performances of the three use cases (cement, SOEC, PEMFC) using the Life Cycle Assessment (LCA) methodology. Following the results assessment made by hotspot analysis, a parametric model is drawn to identify the correlations between impacts and the variation of selected parameters characterizing the products under analysis. Some assumptions are common within the three assessments: for instance, only the Manufacturing process is included in the system boundaries of the study. Following the impact assessment calculations, the results are evaluated and interpreted, identifying the main hotspots through a contribution analysis. From these results, the LCA parametric model is built. Then advanced study of Life Cycle Assessment (LCA) and Social Life Cycle Assessment (S-LCA) has been applied to three use cases.
Technical specifications of the Open Repository - the technical specifications and software architecture of the MatCHMaker repository and framework are defined according to different predefined “zones” (abstract zones, repositories zones and API service. Following an extensive analysis of the existing repositories for materials and for software systems, a new concept was defined for the MatCHMaker open repository. It especially introduces the concept of Distributed Data and Knowledge Mesh (DDKM), a concept with roots in the Data Mesh concept.
Experimental results - Experiments have been carried out for the three use cases of the project. For cement, SEM, XRF, and XRD measurements of phase fractions, and measurements of mechanical strength and porosity are the main data of interest. Micro-scale 3D imaging was done with FIB-SEM, while the hardening phase (calcium silicate hydrate) was studied by TEM methods. For SOC electrodes, SEM/EDX observations have been made on samples with several oxidation times. The initial microstructures have been characterized by X-ray nanotomography, a non-destructive technique, at the European Synchrotron Radiation Facility (ESRF). For PEMFC electrodes, TEM images of the Pt nanoparticles are the main data of interest. Electron tomography experiments were also carried out on the catalyst powders.
ML models - different ML models have been implemented for the different Use Cases of the project.
A Life Cycle Assessment (LCA) approach has been implemented, as a parametric model.
Acquisition of experimental data on each use case has been started to feed the data- and physics- based approaches.
ML models have been developed to predict the compressive strength of cement formulations at multiple curing stages (1, 2, 7, and 28 days). SEM and EDX image has been analysed to segment the image according to the expected phases in order to provide the phase assemblage. This segmentation is performed using an unsupervised clustering approach to identify the different cement phases.
A deep learning model (U-Net) was trained to segment scanning electron microscopy (SEM) images of Solid Oxide Cells (SOCs) without relying on time-intensive Energy Dispersive X-ray Spectroscopy (EDS). To improve trust in predictions, uncertainty quantification is implemented using Deep Ensembles.
Acquisition of experimental data on each use case has been started to feed the data- and physics- based approaches.
ML models have been developed to predict the compressive strength of cement formulations at multiple curing stages (1, 2, 7, and 28 days). SEM and EDX image has been analysed to segment the image according to the expected phases in order to provide the phase assemblage. This segmentation is performed using an unsupervised clustering approach to identify the different cement phases.
A deep learning model (U-Net) was trained to segment scanning electron microscopy (SEM) images of Solid Oxide Cells (SOCs) without relying on time-intensive Energy Dispersive X-ray Spectroscopy (EDS). To improve trust in predictions, uncertainty quantification is implemented using Deep Ensembles.