Sugar-derived carbon fibers (SBCFs) emerge as a promising next-generation sustainable material due to their biomass origin, cost-effectiveness, and superior strength-to-weight ratio. However, industrial adoption remains hindered by inefficient optimization of complex pre-carbonization processes. Here, we present a machine learning (ML)-driven framework to address this challenge, integrating experimental data to establish quantitative correlations between pre-carbonization parameters (temperature, dwell time) and mechanical performance. Gradient Boosted Decision Trees (GBDT) achieved superior predictive accuracy (R2 = 0.857 for tensile strength), enabling efficient identification of optimal conditions: 220 °C pre-carbonization temperature with 100 min dwell time. Experimental validation confirmed a 5.60% tensile strength enhancement over baseline protocols. Optimized protocols yield fibers with 874 MPa tensile strength and 76 GPa modulus, with an average diameter of 28 μm. This machine learning-driven methodology not only advances SBCF manufacturing but also establishes a generalizable paradigm for accelerating functional material development.

Machine learning-optimized pre-carbonization process for sugar-based carbon fibers
    Abstract
    1 Introduction
    2 Materials and methods
        2.1 Experimental
            2.1.1 Materials
            2.1.2 Samples Preparation
            2.1.3 Measurements
    2.2 Machine learning
    3 Result and discussion
        3.1 Mechanical properties of precarbonized fibers
        3.2 Mechanical property prediction and process optimization
        3.3 Microstructural characterization
        3.4 Mechanical properties of carbon fibers
    4 Conclusion
    References

• Wenping Cao(School of Light Industry, Harbin University of Commerce, Harbin 150028, PR China) Corresponding author
• Jiabin Tu(School of Light Industry, Harbin University of Commerce, Harbin 150028, PR China)
• Wanxiaonan Chen(School of Materials Science and Engineering, Harbin Institute of Technology, Harbin 150001, PR China)
• Linsen Zhang(School of Materials Science and Engineering, Harbin Institute of Technology, Harbin 150001, PR China)
• Qianru Lin(Laboratory for Space Environment and Physical Science, Research Center of Basic Space Science, Harbin Institute of Technology, Harbin, PR China)
• Jie Sheng(Laboratory for Space Environment and Physical Science, Research Center of Basic Space Science, Harbin Institute of Technology, Harbin, PR China) Corresponding author