With the development of science and technology, the industrialization process continues to speed up. The demand for the safety of modern giant systems grows very fast, too. As key research directions to improve the reliability and safety of dynamic systems, fault diagnosis and fault tolerant control (FTC) methods have received more and more attention. Though very rich results have been achieved in these directions during the past decades, fault diagnosis and fault-tolerant control theory still seem to be weak when facing high nonlinear complex industrial systems that iterate even faster. Otherwise, the most advanced creations of science and technology like space shuttles, and civil air-crafts, would not encounter disasters like explosions or crashes. Therefore, fault diagnosis and FTC theories must keep pace with the fast-changing world to ensure safety and stability. In this thesis, we did some extensions to the multiple model approach based fault tolerant control method and proposed several fault diagnosis and FTC algorithms based on model matching strategies. This is under the consideration of designing FTC strategies for complex nonlinear systems using simple ideas and logic. Specifically, the multiple model approach is first used to represent the complex system by a set of simple models. The classical multiple model approach is then extended by increasing its dimensions to create a system that can adapt to different faulty conditions. This results in a multi-dimensional multiple model set that can be adapted to the real system in normal states or under faults. The corresponding multi-dimensional controller bank is designed according to the parameters of the local models. Since the local models are generally set as linear ones, their controllers can be designed easily. When the multi-dimensional multiple model set is constructed, two model matching fault tolerant strategies are proposed. For the traditional system, its output is compared with the model set to determine the interval in which the system is located, and then the controllers, corresponding to the sub-models which are matched to the current system, can be selected to conduct the control calculation. For the controller bank scheduling part, the classical switching or mixing strategy can be used. When using the mixing strategy, the weighted sum of the multi-controller outputs represents the output of the controller bank. The weighting values also have a relatively large research space. They can be linearly weighted, non-linearly weighted, or selectively weighted. Such an integrated output can also be performed in two dimensions, e.g., within one controller set and between fault tolerant controller sets. Inspired by the concept of adaptive model, a two-layer multiple model structure is used as the basic framework to conduct the FTC strategy. In the actual operation, the pre-defined local models set is like a source pool. Parameters of the interval vertex models selected from the multi-dimensional multiple model set that encloses the actual system are used to initiate the adaptive models to approximate the actual system with unknown parameters. In this way, it achieves the purpose of controlling the unknown system while identifying it. A fault diagnosis module monitors the actual system in real time and determines its operating interval within the two-layer multiple model set. When a big change is detected, it indicates a fault. The fault information is transmitted to the fault tolerant module, which initiates an adaptation process to match the actual system according to the new vertex models' parameter. Here, an extra controller is set to the reference model in this process to achieve the control goal by following the original reference while the adaptation process approximating the actual system. Several simulations demonstrate the effectiveness of the proposed model matching FTC strategy for different systems and are worthy of promotions and further study.