Designing Models for Autonomous Small-Angle Scattering Measurements of Industrial Soft Materials

The Autonomous Formulation Laboratory (AFL) program at the National Institute of Standards and Technology (NIST) develops autonomous platforms that enhance the characterization of complex industrial formulations. A key challenge of our program is building data pipelines and models that can interpret the measurements of messy and non-ideal material systems brought by our stakeholders. This means that simple machine-learning models built from perfectly ideal theories often do not perform well on the real data we encounter in our work. In this talk, we will cover our efforts to develop tools that can correctly label (i.e., identify, classify, label, cluster) small-angle scattering data of soft-materials. We will highlight the unique challenges of working with soft-material systems including dealing with second-order phase transitions and the presence of mixed and non-equilibrium phases. We will show the benefits and drawbacks of various supervised and unsupervised approaches and how they connect into our autonomous agent. Finally, we will demonstrate how these methods work as part of a phase exploration loop during small-angle neutron and X-ray scattering experiments.