This work studies the rate-dependent mechanical behavior of filament-wound fiber-reinforced polymer (FRP) composite pipes. Commercially available tubes with a filament winding angle of ±55° were tested under cyclic axial compression for four loading rates. Stress relaxation under constant strain was observed as well as a dependence of stress on the strain rate. A novel modeling methodology is presented to capture the nonlinear cyclic response, including the viscoelastic behavior of the epoxy matrix and the interaction of axial and hoop strains. This is accomplished by defining an element configuration with separate elements for the epoxy matrix and the glass fibers. The nonlinear and viscoelastic behavior is incorporated using the generalized Maxwell model. A machine learning (ML) calibration framework is adapted for this study and used to calibrate the nonlinear and viscoelastic properties for the analytical model using a convolutional neural network (CNN). The CNN is trained to identify and understand the interdependencies among the model parameters. The calibrated model parameters are used to simulate the experimentally measured response of the FRP tubes and were found to be applicable across the range of strain rates. The proposed modeling methodology accurately predicted the axial stress and hoop strain time histories as well as the rate-dependent stress relaxation during constant axial strains. The accuracy capturing the measured stress-strain responses demonstrated the synthetic dataset was adequate for training the CNN without requiring additional experimental data.
Read full abstract