Unsupervised chemical-space organization enables identification of self-healing yet ultra-robust polyurethane elastomers
Although published polyurethane formulations contain detailed structural information, inconsistent measurement protocols limit their utility in data-driven materials discovery. We introduce an unsupervised chemical-space organization strategy that maps 296 literature formulations alongside millions of virtual combinations onto a shared 13-descriptor representation. Uniform manifold approximation and projection (UMAP) and hierarchical density-based spatial clustering of applications with noise (HDBSCAN) partition the space, while a geometry-based ranking anchored to literature formulations with tensile strength > 20 MPa and healing efficiency > 0.90 generates the candidate queue. After screening for formulation plausibility, reagent availability, and synthetic accessibility, six candidates were selected for experimental comparison. The resulting formulation, PUUB (PC2000/HMDI/UPy/PBA), integrates ureidopyrimidinone quadruple hydrogen bonding with phenylboronic-acid-derived dynamic boroxine motifs. PUUB achieves a tensile strength of approximately 98 MPa at 788% elongation, a tensile work-to-fracture of 297 MJ m − 3 , and 86.5% fracture-strength recovery after a four-fifths-width cut and 24 h at 80°C in the presence of water. This workflow converts heterogeneous literature data into prioritized, experimentally validated candidates, providing a generalizable strategy for designing materials with complex, coupled properties.
Authors
- Xiaoyu Ji (ORCID: https://orcid.org/0000-0003-2071-3721)
- Zhigang Xue (ORCID: https://orcid.org/0000-0003-2335-9537)
- Dian Liu (ORCID: https://orcid.org/0000-0001-5472-2832)
- Zhen Shi (ORCID: https://orcid.org/0000-0002-5354-1703)
- Junchen Meng
Institutions
- Naval University of Engineering (CN)
- Huazhong University of Science and Technology (CN)
Publication Details
- Journal
- Research Square
- Published
- 2026-11-19
- DOI
- https://doi.org/10.21203/rs.3.rs-11579608/v1
- Primary Topic
- Machine Learning in Materials Science
- Type
- preprint