MODELING CFRP OPEN-HOLE COMPRESSION STRENGTH USING ARTIFICIAL NEURAL-NETWORK WITH MANUALLY DERIVED LAYER

Abstract

This paper models the open-hole compression (OHC) strength of aerospace-grade unidirectional CFRP laminates with standard and non-standard ply angles using deep learning methods. Specifically, an artificial neural-network (ANN) model is developed from experimental test results to predict laminate OHC strength from ply orientation and stacking sequence. This ANN model innovates from conventional models by adding a layer containing an interply compatibility-index that was derived manually from interply failure modes to improve the prediction accuracy. A final root mean square error (RMSE, or standard deviation of prediction errors) of 6.7% of the average failure strength (21.1 MPa) and R2a value of 0.83 is achieved using K-fold cross validation on 16 laminates studied. This paper models all angles from -90° to 90°, not just the standard angles (0°, +/-45°, 90°). This paper shows that interply failure modes are important predictors for laminate OHC strength.