Using Machine Learning to Predict Mechanical Properties of Long Fiber Thermoplastics Based on Manufacturing Process Parameters

Abstract

Long Fiber Thermoplastic Direct Extrusion Compression Molding (LFT-D) is an effective manufacturing process for semi-structural composite parts. LFT-D requires the user to know the influence each process parameter (e.g., temperature setpoints, rotational machinery speed, twin screw extruder configurations, matric flow rates, etc.) has on the resulting mechanical property of the manufactured part. This case study highlights the LFT-D manufacturing of carbon and glass fiber reinforced PA6 (CFRP, GFRP) at the Fraunhofer Innovation Platform (FIP) at the University of Western Ontario in London, Ontario, Canada. Data was collected for forty-six trials conducted over three years, varying fiber concentrations, fiber volume weights, screw configurations, temperatures, torques, forces, and pressures. The mechanical properties were measured and recorded in zero- and ninety-degree fiber directions. Machine Learning (ML) models like linear regression, random forest regression, ridge regressor, and lasso regressor were used to predict mechanical properties like Young's modulus and tensile strength of the part based on the process parameters as inputs. Data preprocessing and cleaning techniques were used to reduce the dimensionality of the dataset and to reduce the number of inputs to the model. These models were then evaluated using cross-validation to select the best model. Once the model was selected, the validation and testing datasets were used to determine the accuracy of the model. A sensitivity analysis was also conducted to determine how the variation in the output of each model can be attributed to the variation in input parameters. With this information, models can be further refined to obtain a more accurate prediction of the outputs and determine which inputs are essential when in the design phase of a product. The application of machine learning in the manufacturing process will ultimately accelerate composite product development and reduce the costly and iterative manufacturing process.