This thesis serves as proof of concept for the tensile strength simulation-based nonwoven material design. Objective is the adjustment of the parameters of an underlying production process with regard to a desired tensile strength behavior (optimization). As an example, we focus on the nonwoven airlay production and consider a thermobonding procedure for the consolidation of the nonwoven fabrics. To be able to map production parameters to the associated tensile strength behavior, we present a model-simulation framework composed of a model for the nonwoven fiber structure generation and a model for the nonwovens’ mechanical behavior under vertical load. The model for the fiber structure generation replicates the stochastic fiber lay-down of the airlay production and results in a random three-dimensional fiber web. This web is consolidated using a virtual bonding procedure that mimics the thermobonding of the nonwoven material. The topology of the resulting adhered fiber structure can be described by a graph, which serves as basis for the subsequent tensile strength simulation. The model used for this purpose describes the mechanical behavior of the material at fiber network level. Therefore, the considered fiber structure sample is interpreted as truss and the fiber connections are equipped with a nonlinear material law, which allows to describe the elastic phase of the nonwovens’ tensile strength behavior. The existence and uniqueness of a solution to the model as well as its numerical treatment are discussed. Moreover, we present data reduction strategies that enable more efficient simulations by removing fiber structure parts that do not contribute to the tensile strength behavior. As it becomes evident from the numerical experiments, a single tensile strength simulation for a production-like virtual sample is already computational demanding. Costs accumulate further, since Monte-Carlo simulations are required to account for the randomness in the fiber structure generation. Thus, direct simulations provide an infeasible basis for the nonwoven material design. This motivates the use of a predictive surrogate for optimization. Therefore, we consider regression-based approaches at different levels of information within the simulation framework. It turns out that the coupling of a polynomial model, for the fiber structure feature inference, with a linear one, for the stress-strain curve inference, yields accurate predictions. Once trained, the regression models allow for efficient evaluations and thus represent a suitable surrogate for the nonwoven material design. In this context, we discuss two exemplary problems of interest for the application: First, a tracking-type problem that aims to find the production parameters that result in a desired tensile strength behavior, expressed in terms of stress-strain curves. Second, an in-corridor maximization problem, which aims to identify the production parameters that maximize the probability of ending up in a specified stress-strain corridor.