Identifying key parameters of electrical curing using GAMs on thermal analysis data

Abstract A procedure based on artificial intelligence statistical techniques for the identification of key design variables and the quantification of their importance for the development of carbon fibre composites by means of electrical current curing is presented. Carbon fibre-reinforced polymer composites are widely used materials in sectors such as aerospace, marine and offshore wind energy, among others, as they offer an excellent balance between high mechanical strength, low mass and resistance to environmental degradation. However, traditional curing methods, such as autoclave processes, are energy and time consuming, require high investments and costly maintenance. Therefore, there is growing interest in alternative curing methods, such as Joule curing, which exploits the electrical conductivity of carbon fibres to generate internal heat, significantly reducing curing time and energy consumption. This research studies the impact of various design parameters on the viscoelastic properties of Joule-cured carbon fibre pre-pregs. The viscoelastic behaviour of the material was evaluated by dynamic mechanical analysis (DMA) and has been modelled by the fitting of semiparametric statistical models. In fact, the main contribution of the present work is the application of generalized additive modelling (GAM) to identify and quantify the influence of design variables such as curing temperature, time and applied voltage, aiding in the development of Joule curing materials with optimized viscoelastic properties. GAM can successfully capture the relationship between these parameters and material properties, obtaining a very high goodness-of-fit, and allowing us for identifying the relative importance of each design variable. It is important to note that this could not be possible by applying common multivariate linear models because the relationships between variables are complex. In addition, GAM is totally interpretable in terms of predictors effects study, advantaging black box machine learning models. Thus, using GAM strongly contributes for increasing the knowledge on complex physical–chemical processes such as Joule curing process. Its application helps to the process optimization, enabling better control of material properties from the control of design variables (curing time, curing temperature) and opening new possibilities for large-scale and cost-effective composite production in demanding marine and offshore applications, as well as other industry sectors. Moreover, this work provides a comprehensive guideline for GAM interpretation results (effects estimation, predictors importance, interaction interpretation, output graphs explanation) specifically applied to the study of Joule-cured materials.