Cross-Project Relationships and Preliminary Estimation of Duncan–Chang E-B Parameters for Rockfill Materials
Reliable selection of Duncan–Chang E-B parameters for rockfill materials commonly relies on large-scale triaxial testing, which may be incomplete during preliminary investigations. This study examines cross-project transferability using 191 database test records from 35 dam projects. Material characteristics and engineering classifications were assessed by correlation analysis, mutual information, and held-out project permutation importance. Random forest regression estimated the initial tangent modulus coefficient, bulk modulus coefficient, and reference friction angle; other parameters were represented by shrinkage-adjusted group statistics or empirical ranges. Five-fold validation, with entire projects held out in each fold, and leave-one-project-out validation assessed generalization to unseen projects. With numerical and engineering classification descriptors combined, five-fold coefficients of determination (R2, with one indicating perfect agreement with the reference values) were 0.2992, 0.1907, and 0.3885 for the three directly estimated parameters, respectively. Under five-fold validation, adding engineering classifications to the numerical descriptors reduced pooled normalized mean absolute errors in initial tangent and bulk stiffness responses by 8.59% and 6.84%, respectively. A separate historical statistical analysis yielded stress-dependent response ranges. The records are not assumed to be independent experiments, and improved accuracy was not uniform across projects. These results support preliminary parameter estimation and uncertainty characterization, subject to project-specific experimental verification.
Authors
- Beixiao Shi
- Siyuan Xu
- Cong Chen
Institutions
- Nanjing Hydraulic Research Institute (CN)
Publication Details
- Journal
- Applied Sciences
- Published
- 2026-11-13
- DOI
- https://doi.org/10.3390/app162211255
- Primary Topic
- Dam Engineering and Safety
- Type
- article
- Field-Weighted Citation Impact
- 0.00
Funders
- National Natural Science Foundation of China