Periodic Reporting for period 2 - BIG-MAP (Battery Interface Genome - Materials Acceleration Platform)
Okres sprawozdawczy: 2022-03-01 do 2024-02-29
Large-scale deployment of intermittent renewable energy and the electrification of the transportation sector critically depend on the availability of low-cost, high-performance, environmentally friendly, and scalable energy storage.
Developing sustainable, ultra-high-performant battery technologies and materials plays a central role in ensuring and accelerating the transition towards net-zero CO2 emission. However, the existing battery innovation and development paradigm is too slow and too costly to address the urgency of the societal challenges resulting from global warming.
The BIG-MAP project developed a first prototype of a fully autonomous battery Materials Acceleration Platform (MAP) capable of accelerating closed-loop materials discovery, cell design, and manufacturing sustainable, ultra-high-performant batteries in Europe. The developed battery “MAP” can acquire and utilize data from different parts of the battery value chain, from raw materials to end-users. The "Battery Interface Genome - Materials Acceleration Platform" project developed the foundation for a novel “chemistry-enabling” methodology and data/research infrastructure.
This was achieved by:
• Providing an accelerated path to disruptive battery technologies with ultra-high performance, full sustainability, and smart operation.
• Developing a modular, closed-loop infrastructure and methodology to bridge physical insights and data-driven approaches to accelerate the inverse design of future battery chemistries and technologies.
Developing sustainable, ultra-high-performant battery technologies and materials plays a central role in ensuring and accelerating the transition towards net-zero CO2 emission. However, the existing battery innovation and development paradigm is too slow and too costly to address the urgency of the societal challenges resulting from global warming.
The BIG-MAP project developed a first prototype of a fully autonomous battery Materials Acceleration Platform (MAP) capable of accelerating closed-loop materials discovery, cell design, and manufacturing sustainable, ultra-high-performant batteries in Europe. The developed battery “MAP” can acquire and utilize data from different parts of the battery value chain, from raw materials to end-users. The "Battery Interface Genome - Materials Acceleration Platform" project developed the foundation for a novel “chemistry-enabling” methodology and data/research infrastructure.
This was achieved by:
• Providing an accelerated path to disruptive battery technologies with ultra-high performance, full sustainability, and smart operation.
• Developing a modular, closed-loop infrastructure and methodology to bridge physical insights and data-driven approaches to accelerate the inverse design of future battery chemistries and technologies.
The BIG-MAP project developed a first proof-of-concept for an autonomous European battery MAP and the infrastructural backbone needed to achieve a 5-10-fold acceleration in the discovery process, as exemplified by a set of 12 Key Demonstrators (www.big-map.eu/key-findings) e.g. a first foundation model for atomistic materials chemistry, and an example of distributed and asynchronous multi-partner materials acceleration platform.
To facilitate an efficient platform for dissemination, a project website (www.big-map.eu) Twitter (@bigmap_eu), and LinkedIn accounts were created.
A unique BIG-MAP data infrastructure was crucial to ensure a seamless flow of data between the partners. The foundation of the data infrastructure was the development of a rigorous data management plan (DMP), which was developed to orchestrate the flow of data. The DMP enabled a simple tracking of the data flows across the project, aligning the expectations to the format and size of the exchanged data and ensuring its operability.
A critical point in making a DMP operational was the development of a shared language or "ontology" to ensure interoperability between simulations and experiments across multiple spatial and temporal scales and different techniques and domains in the battery discovery process. To facilitate battery experts in various fields to convert real-life observations to a standard digital representation of an ontology, BattINFO was developed and made openly available to the community (https://github.com/BIG-MAP/BattINFO(odnośnik otworzy się w nowym oknie)). BattINFO supports the description and characterization of key aspects governing battery interface performance, including the formation of interphase layers, passivating layers, active material dissolution, charge transfer reactions, etc.
The BIG-MAP project developed a tiered and scalable battery archive (https://archive.big-map.eu(odnośnik otworzy się w nowym oknie)) based on the InvenioRDM architecture developed at CERN that underpins Zenodo (see figure 1). Notably, the BIG-MAP Archive serves as a private repository to all the BIG-MAP partners, but, thanks also to the community feature of InvenioRDM, it can now be used to serve independently and privately the other BATTERY 2030+ projects with multiple sharing tiers, allowing either for secure data sharing within a given project or giving access to the data to all BATTERY2030+ projects. The private data can also be pushed publicly to the Materials Cloud Archive, which has a public DOI and guaranteed availability for 10+ years.
Finally, the BIG-MAP GitHub registry and BIG-MAP App Store (https://big-map.github.io/big-map-registry/(odnośnik otworzy się w nowym oknie)) were launched and made openly available to the research community. The BIG-MAP App Store contains 31 apps (to date).
To facilitate an efficient platform for dissemination, a project website (www.big-map.eu) Twitter (@bigmap_eu), and LinkedIn accounts were created.
A unique BIG-MAP data infrastructure was crucial to ensure a seamless flow of data between the partners. The foundation of the data infrastructure was the development of a rigorous data management plan (DMP), which was developed to orchestrate the flow of data. The DMP enabled a simple tracking of the data flows across the project, aligning the expectations to the format and size of the exchanged data and ensuring its operability.
A critical point in making a DMP operational was the development of a shared language or "ontology" to ensure interoperability between simulations and experiments across multiple spatial and temporal scales and different techniques and domains in the battery discovery process. To facilitate battery experts in various fields to convert real-life observations to a standard digital representation of an ontology, BattINFO was developed and made openly available to the community (https://github.com/BIG-MAP/BattINFO(odnośnik otworzy się w nowym oknie)). BattINFO supports the description and characterization of key aspects governing battery interface performance, including the formation of interphase layers, passivating layers, active material dissolution, charge transfer reactions, etc.
The BIG-MAP project developed a tiered and scalable battery archive (https://archive.big-map.eu(odnośnik otworzy się w nowym oknie)) based on the InvenioRDM architecture developed at CERN that underpins Zenodo (see figure 1). Notably, the BIG-MAP Archive serves as a private repository to all the BIG-MAP partners, but, thanks also to the community feature of InvenioRDM, it can now be used to serve independently and privately the other BATTERY 2030+ projects with multiple sharing tiers, allowing either for secure data sharing within a given project or giving access to the data to all BATTERY2030+ projects. The private data can also be pushed publicly to the Materials Cloud Archive, which has a public DOI and guaranteed availability for 10+ years.
Finally, the BIG-MAP GitHub registry and BIG-MAP App Store (https://big-map.github.io/big-map-registry/(odnośnik otworzy się w nowym oknie)) were launched and made openly available to the research community. The BIG-MAP App Store contains 31 apps (to date).
Three of the main outputs of BIG-MAP relate to the “BIG” deep- and machine learning (ML) models, the development of the MAP infrastructure, and the multimodal workflows developed to integrate the two, e.g. a multimodal experimental workflow (see figure 2).
One example in the area of the BIG ML models is the development of an uncertainty-aware deep autoregressive model that, with limited training data and only a few cycles of observations, can predict the capacity degradation over the entire lifetime with intercell variability while maintaining explainability and learning to differentiate degradation mechanisms in a data-driven manner. This model can substantially reduce how long experiments need to run to test new formulations. Another example in this domain is the work on symbolic regression and HTE-acquired datasets on electrolyte conductivity. Here, we discovered a simple, accurate, consistent, and generalizable governing law. Despite emerging from a purely statistical approach, the expression reflects functional aspects from established thermodynamic limiting laws, indicating our model is grounded on the fundamental physical mechanisms underpinning ionic transport.
In the MAP domain, we operate in the paradigm of integrating combinatorial synthesis, high-throughput characterization, automatic analysis, and ML. Within a distributed and asynchronous battery MAP, one or multiple autonomous feedback loops may aim to optimize materials and device-level performance for specific functional properties or generate new insights. Here, the scope of a given experimental campaign is defined by the range of experiment and analysis actions integrated into the experiment framework. FINALES 1 & 2 (https://zenodo.org/records/10987727(odnośnik otworzy się w nowym oknie)) present a transformative methodological development for integrating multiple actions within a hierarchical experimental laboratory automation and orchestration framework. With FINALES, we demonstrate the capability of orchestrating geographically distributed research instruments that incorporate data from experiments, simulations, and databases. FINALES interfaces laboratory hardware and software distributed across several computers and operating systems for executing experiments, data analysis, provenance tracking, and autonomous planning: thus, proving a proof-of-concept for a distributed and asynchronous battery map capable of co-optimization of materials and device-level performance.
Another important aspect of a MAP is the ability to do automated/autonomous analysis of experimental data, which can be fed to the ML models on the fly. We have developed two open-source apps FullProfAPP and PRISMA apps to visualize and process hundreds of spectra from operando experiments. The app implements baseline correction, peak fitting methods, and a friendly graphical user interface. Users load spectra (or diffraction patterns), tune baseline and peak fitting parameters, run a high-throughput processing step, and export the results in a CSV format within minutes. This approach enables extracting spectroscopic trends that characterize the properties and phenomena inherent to the operation of functional materials.
One example in the area of the BIG ML models is the development of an uncertainty-aware deep autoregressive model that, with limited training data and only a few cycles of observations, can predict the capacity degradation over the entire lifetime with intercell variability while maintaining explainability and learning to differentiate degradation mechanisms in a data-driven manner. This model can substantially reduce how long experiments need to run to test new formulations. Another example in this domain is the work on symbolic regression and HTE-acquired datasets on electrolyte conductivity. Here, we discovered a simple, accurate, consistent, and generalizable governing law. Despite emerging from a purely statistical approach, the expression reflects functional aspects from established thermodynamic limiting laws, indicating our model is grounded on the fundamental physical mechanisms underpinning ionic transport.
In the MAP domain, we operate in the paradigm of integrating combinatorial synthesis, high-throughput characterization, automatic analysis, and ML. Within a distributed and asynchronous battery MAP, one or multiple autonomous feedback loops may aim to optimize materials and device-level performance for specific functional properties or generate new insights. Here, the scope of a given experimental campaign is defined by the range of experiment and analysis actions integrated into the experiment framework. FINALES 1 & 2 (https://zenodo.org/records/10987727(odnośnik otworzy się w nowym oknie)) present a transformative methodological development for integrating multiple actions within a hierarchical experimental laboratory automation and orchestration framework. With FINALES, we demonstrate the capability of orchestrating geographically distributed research instruments that incorporate data from experiments, simulations, and databases. FINALES interfaces laboratory hardware and software distributed across several computers and operating systems for executing experiments, data analysis, provenance tracking, and autonomous planning: thus, proving a proof-of-concept for a distributed and asynchronous battery map capable of co-optimization of materials and device-level performance.
Another important aspect of a MAP is the ability to do automated/autonomous analysis of experimental data, which can be fed to the ML models on the fly. We have developed two open-source apps FullProfAPP and PRISMA apps to visualize and process hundreds of spectra from operando experiments. The app implements baseline correction, peak fitting methods, and a friendly graphical user interface. Users load spectra (or diffraction patterns), tune baseline and peak fitting parameters, run a high-throughput processing step, and export the results in a CSV format within minutes. This approach enables extracting spectroscopic trends that characterize the properties and phenomena inherent to the operation of functional materials.