Identification of Critical Process Parameters for Carbon Fiber Precursor Based on XGBoost Algorithm

Abstract

Carbon fiber has excellent properties such as high specific strength and specific modulus, and is widely used in aerospace and civil industries. The quality of carbon fiber precursor (CFP) is one of the important factors affecting the quality of carbon fiber. Identification of critical process parameters plays a vital role in improving the stability of the CFP production and optimization of process parameters. However, the production of the CFP involves nearly 200 process parameters, and there are highly non-linear relationships between quality and process parameters, and interactions between process parameters, which makes the identification of critical process parameters much harder. This paper aims to uncover the critical process parameters based on the extreme gradient boosting (XGBoost) approach because of the great advantages of XGBoost algorithm in solving the problems above. In this paper, a XGBoost-based model is proposed to classify the quality level of the CFP based on actual process monitoring data, and the critical process parameters are identified simultaneously. The performance of XGBoost-based classification model is compared to two traditionally classification model, Support Vector Machine (SVM) and Random Forest (RF), in terms of classification accuracy. In addition, this paper uses the Maximum Information Coefficient to validate the critical process parameters uncovered by the proposed model. It is demonstrated that the proposed method can identify the critical process parameters correctly and intuitively.