Machine-learning based process monitoring for automated composites manufacturing

Automated fibre placement (AFP) is an advanced robotic manufacturing technique which can overcome the challenges of traditional composite manufacturing. The interlaminar strength of AFP-manufactured composites depends on the in-situ thermal history during manufacturing. The thermal history is controlled by the choice of processing conditions and improper interfacial temperatures may result in insufficient bonding. Being able to better predict such maintenance issues in real-time is an important focus of smart manufacturing and Industry 4.0 to improve manufacturing operations. The data analysis of real-time temperature measurements during AFP composites manufacturing requires the temperature profiles from Finite Element Analysis (FEA) based simulations of the AFP process to better predict the quality of layup. However, the FEA simulations of the AFP process are computationally expensive. This study focuses on developing a digital tool enabling real-time process monitoring and predictive maintenance of the AFP process. The digital tool constitutes a machine learning-based surrogate model based on results from Finite Element Analysis (FEA) simulations of the AFP process to predict the in-situ thermal profile during AFP manufacturing. Multivariate Linear Regression, Multivariate Polynomial Regression, Support Vector Machine, Random Forest and Artificial Neural Network (ANN)-based models are compared to conclude that ANN based surrogate model performs best by predicting the important parameters of thermal profiles with a mean absolute percentage error of 1.56% on additional test data and reducing the time by four orders of magnitude as compared to FEA simulations. The predicted thermal profile can be compared with the real-time in-situ temperatures during manufacturing to predict the quality of the layup. A GUI application is developed to provide predicted thermal profiles data for analysis in conjunction with real-time temperatures during manufacturing enabling monitoring and predictive maintenance of the AFP process and paving way for the development of a digital twin of the AFP composites manufacturing process.