Predictive modelling and optimization of carbon fiber mechanical properties through high temperature furnace

Developments in carbon fiber mechanical properties are directly dependent on its manufacturing process, especially the thermal process of carbonization through a High Temperature (HT) furnace. These mechanical properties cannot be straightforwardly predicted experimentally since the thermal process involves complex parameters such as: (i) zones temperature, (ii) applied fiber tension, (iii) the N2 gas flow rate which is influenced by uncertain disturbances. The aim of this paper is to optimize the carbon fiber modulus based on the given range of tensile strength using a predictive modelling technique. The data collection of the carbonization thermal process is significantly costly and time consuming, hence we used a novel DOE method to minimize the number of required experiments. The carbon fiber with the highest and lowest tensile and modulus properties were initially analyzed by X-ray diffraction and Raman Spectroscopy, but no significant differences were observed that could account for the differences in measured mechanical and physical properties, thus a surrogate modelling approach was adopted. Four different mathematical methods were employed for modelling and compared with one another. These are (i) non-linear multivariable regression, (ii) Levenberg-Marquardt neural network algorithm, (iii) thin plate splines, and (iv) a convex hull technique. The results clearly showed that the convex hull technique has less error compared to the other tested methods. This method was selected to optimize the carbon fiber modulus based on the given range of tensile strength. Using the convex hull technique along with an evolutionary search (Genetic Algorithm) method, it is shown that the modulus of carbon fiber can be optimized for any given range of tensile strength, providing an optimum set of control parameters are chosen during the process