Intelligent Modeling and Multi-Response Optimization of AWJC on Fiber Intermetallic Laminates through a Hybrid ANFIS-Salp Swarm Algorithm
Mahalingam Siva Kumar,Devaraj Rajamani,Ahmed M El-Sherbeeny,Esakki Balasubramanian,Krishnasamy Karthik,Hussein Mohamed Abdelmoneam Hussein,Antonello Astarita
DOI: https://doi.org/10.3390/ma15207216
2022-10-16
Abstract:The attainment of intricate part profiles for composite laminates for end-use applications is one of the tedious tasks carried out through conventional machining processes. Therefore, the present work emphasized hybrid intelligent modeling and multi-response optimization of abrasive waterjet cutting (AWJC) of a novel fiber intermetallic laminate (FIL) fabricated through carbon/aramid fiber, reinforced with varying wt% of reduced graphene oxide (r-GO) filled epoxy resin and Nitinol shape memory alloy as the skin material. The AWJC experiments were performed by varying the wt% of r-GO (0, 1, and 2%), traverse speed (400, 500, and 600 mm/min), waterjet pressure (200, 250, and 300 MPa), and stand-off distance (2, 3, and 4 mm) as the input parameters, whereas kerf taper (Kt) and surface roughness (Ra) were considered as the quality responses. A hybrid approach of a parametric optimized adaptive neuro-fuzzy inference system (ANFIS) was adopted through three different metaheuristic algorithms such as particle swarm optimization, moth flame optimization, and dragonfly optimization. The prediction efficiency of the ANFIS network has been found to be significantly improved through the moth flame optimization algorithms in terms of minimized prediction errors, such as mean absolute percentage error and root mean square error. Further, multi-response optimization has been performed for optimized ANFIS response models through the salp swarm optimization (SSO) algorithm to identify the optimal AWJC parameters. The optimal set of parameters, such as 1.004 wt% of r-GO, 600 mm/min of traverse speed, 214 MPa of waterjet pressure, and 4 mm of stand-off distance, were obtained for improved quality characteristics. Moreover, the confirmation experiment results show that an average prediction error of 3.38% for Kt and 3.77% for Ra, respectively, were obtained for SSO, which demonstrates the prediction capability of the proposed optimization algorithm.