Natural frequency of laminated composite plates having cutout opening with lining along the cutout edges analysed using six node linear strain triangular elements optimised using aqua search meta-heuristic algorithm and machine learning gradient boos

Abstract. The use of laminates has become an integral unit in every field of Engineering. The laminated plates and shells are a common form of structural shapes which are widely used for various types of Civil Engineering purposes. The use of laminates over the long run sometimes wear out near the ends which are usually not protected. The life span of the laminate can be increased further by strengthening the boundary. Laminates having cutout are widely in use for various purposes of construction. The basic idea behind the paper is to strengthen the boundary of the laminate around the cutout opening. The strengthening can be done by increasing the number of layers, by increasing the thickness, by increasing the modulii of elasticity of the material and few other ways. This lining of laminate can increase the life span and can contribute to the performance of the laminate during its life. The present paper evaluates the fundamental frequencies of laminates having cutouts for both unlined and lined laminates along the cutout edges by increasing the number of layers near the cutout edges. Five different machine learning methods have been used to validate our study and determine the performance metrics. Coding is done in Google Colabs®. The five methods are namely Linear Regression (LR), Ridge Regression (RR), kNearest Neighbour (kNN), Random Forest (RF), Gradient Boosting (GB). The root mean square error and the R2 value are determined to check the accuracy of the model. The best method is found to be gradient boosting with least error nearly equal to zero and an R2 value of 99.9%. The results of non dimensional natural frequency obtained using AIML are cross verified with the results obtained from finite element analysis using six node linear strain triangular elements. The results are encouraging and the error is nearly equal to zero in case of gradient boosting.