Refractory linings are vital components of high temperature vessels. Successful lining concept design can avoid the premature wear of refractory linings, allow for more economically efficient configuration of refractories, and improve the efficiency of high-temperature processes and save energy. The thermal and thermomechanical behavior of refractory linings has a significant influence on the lifetime of vessels and is affected by many factors, for instance, the geometry of vessels, the properties of refractory lining, and the process conditions. The direct measurement of stresses at high temperature conditions is nearly impossible. This is unfortunate for the proper design of refractory linings for the specific conditions. The present thesis aims to optimize the lining concept of a steel ladle considering the influence of multiple factors and to predict the thermal and thermomechanical performance of the lining concept. A set of tools in the Taguchi method was used for the lining concept optimization. These tools are orthogonal arrays (OAs), analysis of variance (ANOVA), and signal-to-noise (S/N) ratio. Lining configurations were designed by OAs and finite element simulations were performed with the commercial software ABAQUS to survey the thermal and thermomechanical behavior of the designed lining concepts. The significance of factors was quantitatively ranked by ANOVA and the optimal levels of each factor were evaluated by S/N ratios. Backpropagation artificial neural network (BP-ANN) was applied to predict lining concept performance. The results show that the proposed two lining concepts optimized by the Taguchi method showed a substantial decrease in heat loss through the steel shell and thermomechanical load at the hot face of the working lining. The performance of 128 lining concepts was predicted by BP-ANN models. High prediction accuracy can be achieved by applying suitable BP-ANN models. The coefficients of determination are 0.9970, 0.9950, 0.9364 for maximum compressive stress at the hot face of the working lining, end temperature and the maximum tensile stress at the cold end of the steel shell, respectively. In addition, guidelines to define minimum training dataset size, node number in the hidden layer, and training algorithms were proposed to optimize BP-ANN architectures for a steel ladle system.