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(öffnet in neuem Fenster)) 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.
