Automated data processing and preparation are essential in modern materials research, where large, heterogeneous datasets from microscopy images to extensive simulation outputs must be handled both fast and accurately. High-throughput (HT) experiments [1] and simulations [2, 3, 4, 5] generate massive, time-dependent data reflecting processing parameters, structural features, and the properties of interest. Ensuring data quality through noise filtering, outlier detection, and standardized formatting is vital for maintaining the accuracy of subsequent models, especially those used in data-driven inverse design [2]. Using advanced statistical methods [6], as well as machine learning (ML) [1, 4, 7] and deep learning (DL) techniques [3], researchers can efficiently integrate data from multiple sources [8] with appropriate weighting [9], ultimately preserving the fidelity of processing-structure-property-performance (PSPP) predictions and accelerating the discovery and optimization of new materials.
A compelling illustration of these automated workflows is found in an ML-based algorithm for Tauc plot analysis in HT-optical spectroscopy [1], crucial for rapidly determining band gaps in photovoltaics and photo electrochemistry. This new strategy reduces user intervention, automatically accounts for measurement noise and delivers reproducible results across diverse material libraries including high-entropy oxides. Likewise, the integration of HT-CALPHAD calculations with combinatorial thin-film synthesis provides a powerful pathway for rapid phase screening, as demonstrated in a recent study on several quaternary systems [5]. Moreover, a generalized application of two automated data processing tools — for the selection of key experimental datasets [6] and for phase screening — will be demonstrated using experimental fatigue data from TU Dortmund University and high-throughput investigations of bulk material libraries from OVGU Magdeburg. These examples showcase how streamlined, automated workflows can transform raw experimental data into actionable insights, thereby advancing more reliable data-driven modeling, facilitating inverse design, and ultimately opening new frontiers in materials innovation.
References:
[1] I. Roslyakova, F. Thelen, R. Zehl, T. Piotrowiak, A. Ludwig; In preparation.
[2] A. Müller, I. Roslyakova, et al; MSMSE, 2019, 27, 024001.
[3] U. Nwachukwu, A. Obaied, et al; MSMSE, 30, 025009.
[4] Y. Jiang, M. A. Ali, et al; MSMSE, 2023, 31(3), 035005.
[5] I. Roslyakova, R. Zehl, J. L. Bürgel, A. Ludwig; In preparation.
[6] S. Zomorodpoosh, N. Volz, et al; Journal of Physics Communications, 2020, 4, 075024.
[7] M. Ahmed, A. Obaied, et al.; MSMSE, 2021, 29, 055012.
[8] V. Mohles, Y. Jiang, et al; Materials Science and Engineering: A, 2024, 146780.
[9] N. H. Paulson, S. Zomorodpoosh, et al; CALPHAD, 2020, 68, 101728.