UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

Multi-objective Gaussian process approach for robust optimization and prediction of carbonization process Ramezankhani, Milad


Carbon fibers are high-performance and high-strength reinforcement materials in advanced industries such as aerospace, automotive and energy sector. This class of materials is often derived from polyacrylonitrile (PAN) fiber precursor. The conversion of polyacrylonitrile precursor to carbon fibers is essentially comprised of two major stages; namely, stabilization and carbonization. The carbonization is a very energy-consuming and expensive step due to its high temperature requirements. In order to achieve desirable physical and mechanical properties of carbon fibers through this step, a large amount of energy is required. A cost-effective approach to optimize energy consumption of this process, however, is the use of predictive modelling techniques. In that goal, herein, a Gaussian Process (GP) approach is proposed to firstly, predict multiple mechanical properties of carbon fibers in the presence of manufacturing noise and secondly, optimize them under a minimum energy consumption criterion and a range of process constraints and bounds. The resulting model for each property of carbon fibers consists of two Gaussian Process models. The first model describes the mean value of the property and the second one predicts its standard deviation. The proposed Gaussian Process approach is compared to a traditional regression approach using the measurements obtained from the Carbon Nexus pilot plant in Australia. The Gaussian Process approach clearly proved to be more effective in terms of both prediction accuracy and robustness. Through employing Gaussian Process models, the modulus and tensile strength mean values of carbon fibers along with their standard deviations (STD) were successfully predicted under different process conditions. Squared exponential covariance and linear mean functions were proven to be most suitable in constructing the Gaussian Process in the performed case study. It was found that the modulus and tensile strength responses do not have an evident correlation with respect to each other, hence a multi-objective optimization approach was developed to acquire overall optimum process conditions. To estimate the trade-off between fiber material properties under the multi-objective optimization problem, a standard as well as an adaptive weighted sum method was applied under various constraints and bounds, including the energy consumed during carbonization.

Item Citations and Data


Attribution-NonCommercial-NoDerivatives 4.0 International