Composites Science and Technology 199 (2020) 108251 Available online 3 July 2020 0266-3538/© 2020 Published by Elsevier Ltd. Application of artifcial intelligence models for predicting time-dependent spring-back effect: The L-shape case study Gl� aucio C. Pereira a, * , M.I. Yoshida b , P. LeBoulluec c , Wei-Tsen Lu d , Ana P. Alves e , Ant^ onio F. Avila f a Mechanical Engineering Graduate Studies Program, Universidade Federal de Minas Gerais, 6627 Antonio Carlos Avenue, 31270-901, Belo Horizonte, MG, Brazil b Chemistry Department, Universidade Federal de Minas Gerais, 6627 Antonio Carlos Avenue, 31270 901, Belo Horizonte, MG, Brazil c College of Engineering, Technology, and Computer Science, Purdue University Fort Wayne, Fort Wayne, Indiana, 46805, USA d Mechanical and Aerospace Engineering Graduate Studies Program, University of Texas at Arlington, Arlington, TX, 76019, USA e Physics Department, Universidade Federal de Minas Gerais, 6627 Antonio Carlos Avenue, 31270 901, Belo Horizonte, MG, Brazil f Mechanical Engineering Department, Universidade Federal de Minas Gerais, 6627 Antonio Carlos Avenue, 31270, 901 Belo Horizonte, MG, Brazil A R T I C L E INFO Keywords: Spring-back Artifcial intelligence Statistical analysis Degree of cure Time-dependency ABSTRACT The role of forces and moments in the spring-back effect in L-shaped carbon-epoxy composites is investigated. Statistical models and artifcial intelligence were used to prove the signifcance of these physical quantities in the angu-lar deformation of these composites. We follow the spring-in deformation as a function of time three years span, and recently we reclassify the recovery on angular deformation due to residual cure as spring-back. This angular deforma-tion measured for different confgurations tends to stabilize after approximately three years after the composite fabrication. The variation on the angular de-formation displays direct dependence with the residual curing process for the matrix resin of each specimen. Thirteen angular deformation were measured 3 years span. We calculated the components of forces (N) and moments (M) indirectly through the classical laminate theory (CLT) for each composite con-fguration. The Generalized Additive Models (GAM) evaluate the signifcance of the forces and moments on spring-back effect. Their output results identify the linear and nonlinear cofactors role as spring-back infuencers. The Random Forest (RF) model ranked the infuence of forces and moments in spring-back deformation. Both statistical models are complementary, GAM predicts the impact of cofactors with accuracy close to 90%, whereas Randon Forest model explains the angular deformation in the mean values with accuracy greater than 91%. 1. Introduction In recent years a large effort has been made to predict the behavior of com-posite materials over a long period of time under different me- chanical and chem-istry cure conditions [1]. Carley et al. [2] have shown that the introduction of dispersed carbon nanotubes into the polymer matrix of composite fbers of car-bon signifcantly increase their strength and stiffness. The carbon structures would be responsible for increasing the cohesion of the polymer matrix next to the carbon fber fabric, thereby increasing the strength of the composite. In this work, a large effort join experimental and computational data to investigated the behavior of L-shaped composites with polymer matrix subjected to a residual cure after the standard autoclaving process was made, showing how structural changes occur in their geometry over a period of approximately three years span. It has been shown that spring-in effect is signifcantly infuenced by residual stresses. The residual stresses come from the shrinkage of the polymer resin matrix. In contrast the volume and morphology of the carbon fbers remain virtually constant during the entire curing process [3]. These differences in the volumes of the resin and of the carbon fbers reinforcement before and after the curing process promote the existence of these tensions. Also, the non-uniform distribution of the resin along the laminate is a generator of residual stresses since the curved parts of a laminate structure tend to have a lower amount of resin compared to non-curved parts. Corrado et al. [4] showed that residual stresses can also be induced during the curing process due to anisotropic * Corresponding author. Mechanics of Composites and NanoStructured Materials Laboratory, Universidade Federal de Minas Gerais, 6627 Antonio Carlos Avenue, 31270-901, Belo Horizonte, MG, Brazil. E-mail address: carleyone@gmail.com (G.C. Pereira). Contents lists available at ScienceDirect Composites Science and Technology journal homepage: http://www.elsevier.com/locate/compscitech https://doi.org/10.1016/j.compscitech.2020.108251 Received 20 January 2020; Received in revised form 17 May 2020; Accepted 24 May 2020