Experimental analysis and prediction of strength of adhesive-bonded single-lap composite joints: Taguchi and artificial neural network approaches

Abstract

Adhesive-bonded joints made up of composite materials offer complex structures with the ease of joining similar or dissimilar materials. The failure behavior of adhesive-bonded joints is influenced largely by geometrical parameters (overlap length and adhesive thickness) and adherend surface preparations. The glass fiber-reinforced epoxy composites are prepared for single-lap adhesive joints to know their strength with different sets of geometrical factors. The adherend surfaces of composites are roughened to 2 ± 0.1 µm, prior to joint preparation. Taguchi L9 experimental matrix representing different combinations of overlap length and adhesive thickness is employed to know the behavior of failure load (FL) and shear strengths (SS) of the adhesive-bonded single-lap composite joints. The results showed that the impact of overlap length of adhesive-bonded joints is more compared to that of adhesive thickness. The interaction effects of geometrical parameters are found negligible toward both the outputs. The optimal factor levels received the highest load to fracture; the joints are found equal to 6096.1 N for FL and 80 MPa for SS, respectively. The empirical relationship based on multiple linear regression (MLR) equations is derived for both the failure load and shear strength. Levenberg–Marquardt algorithm-trained neural networks (NNs) are used for the prediction of both responses. Ten experimental cases are used to check the prediction capabilities of both MLR and NNs. The mean absolute percent error in prediction of both the responses is found equal to 2.27% for NNs and 3.12% for MLR. The NNs and MLR results in accurate prediction might be due to the model development process based on experimental input–output data, rather than assumption-based theoretical and numerical methods.