Data-driven inverse design of strain-hardening cementitious composites (SHCC): Integrating multi-fiber systems, ANN modeling, and PSO optimization

To address the limited tunability, experience-dependent mix design, and lack of systematic inverse design approaches in multi-performance-oriented SHCC design, this study proposes a target-performance-driven inverse design method based on machine learning and heuristic optimization to achieve wide-range performance design of SHCC. Unlike existing studies focusing on single-performance prediction or localized optimization, this method establishes a performance database covering multiple fiber systems, strength levels, and tensile ductility ranges, enabling the inverse mapping from target performance requirements to material composition parameters. By regulating fiber type, fiber volume fraction, and matrix mix parameters, a series of mechanical performance tests were conducted to construct a comprehensive SHCC database. To improve the coverage and diversity of the database, experimental data with different performance levels were retained, resulting in a database covering compressive strength of approximately 30–115 MPa, flexural strength of 5–36 MPa, axial tensile strength of 2–15 MPa, and axial tensile strain of 0.1–3.5%. This database provides a foundation for data-driven models to capture the complex nonlinear relationships between material compositions and mechanical responses. Based on this database, ANN, SVR, and GPR prediction models were developed, among which the ANN model exhibited the best overall performance, with mean 𝑅 2 values ranging from 0.91 to 0.95 under five-fold cross-validation. A target-performance-driven inverse mix design model was then established by integrating the PSO algorithm, with material cost incorporated into the optimization objective to enable coordinated regulation of performance and economy. The results showed that, at 𝜆 = 0, the predicted performances under all three target-performance scenarios closely matched the prescribed targets. After introducing the cost constraint, the absolute relative errors of all predicted performance indicators remained within 2.0% at 𝜆 = 0.05, while the material costs of the three scenarios were reduced by 30.5%–53.7% relative to the corresponding 𝜆 = 0 solutions. As 𝜆 increased further, material cost continued to decrease, whereas the deviations of some predicted and experimentally measured performance indicators gradually increased. In particular, axial tensile strain under the high-ductility condition exhibited high sensitivity to the cost constraint, demonstrating a clear performance–cost trade-off. The proposed method enables efficient inverse design from target performance to mix proportion parameters and allows the balance between performance requirements and material economy to be adjusted through the cost weighting factor, providing a new technical pathway for the integrated optimization of multiple SHCC performance indicators and economic efficiency.