Neural Network Based Failure Prediction Model for Composite Hydrogen Storage Cylinders

Abstract

Composite high-pressure cylinders have potential application for hydrogen storage in automotive and transportation systems. Safe installation and operation of these cylinders is of primary concern. A neural network model has been developed for predicting the failure of composite storage cylinders subjected to thermo-mechanical loading. A backpropagation Neural Network model is developed to predict composite cylinder failure. The inputs of the neural network model are the laminate thickness, winding angle, and temperatures. The output of the model is the failure pressure. The finite element model of the cylinder is based on laminated shell theory accounting for transverse shear deformation and geometric nonlinearity. A composite failure model is used to evaluate the failure under various thermo-mechanical loadings. The neural network is trained using failure results of simulation under different thermal loadings and lay-up. The developed neural network model is found to be quite successful in determining the failure of hydrogen storage cylinders. INTRODUCTION The composite high-pressure cylinder is made with a high molecular weight polymer or aluminum liner that serves as a hydrogen gas permeation barrier. A filament-wound, carbon/epoxy composite laminate over-wrapped outside of the liner provides the desired pressure load bearing capacity (Liang, 2002). The cylinder is capable of sustaining pressures of 5000 psi or higher by taking advantage of high modulus, high strength and low specific weight of modern high performance composite (Takeichia, 2003). To design composite highpressure cylinders with the most possible safety, reliability and minimum weight considerations, the failure of composite structures under various mechanical and thermal loadings need to be well understood. To account for complex composite wall structure and environmental temperature influence, a comprehensive finite element model is developed and implemented in commercial finite element code ABAQUS (Mitlitsky, 2000) to analyze the failure of the composite cylinder. Due to a large number of parameters such as varying thermal loads, winding angles, cylinder dimensions and lay up configurations, it is a tremendous task to optimize the cylinder design and predict cylinder failure pressure through case-by-case finite element analysis simulation. A backpropagation Neural Network (NNk) model is employed to predict the failure pressure using the results obtained from a few finite element simulation cases. Three sets of simulation results with various Proceedings of the Artificial Neural Networks in Engineering (ANNIE) Conference, St. Louis, MO, November 11-14, 2007