The complexity of modern automated and technical processes is continuously growing due to the increasing demands for higher system performance and product quality. Parallel to these demands, the safety and reliability in systems are needed. Therefore, process monitoring and fault diagnosis are becoming an ingredient of automatic modern systems. Model-based fault detection technique is one of the approaches, which has proven its efficiency and now is fully integrated into many successful applications. This approach consists mainly of three steps, which are: residual generation, evaluation process and threshold determination. The purpose of residual generation is to produce the so-called residual signal by comparing the process outputs with their estimates. The deviation of residual signal from zero is an indication of fault occurrence. However, due to the unknown inputs (process disturbances, measurements noises) and model uncertainties the residual signal is nonzero in fault-free operation. Therefore, to extract the occurrence of faults in the existence of disturbances and model uncertainties need an evaluation process for residual signal. The faults in systems can be detected by comparing the evaluated residual signal with a threshold. Determination of the threshold is a critical task in fault detection, which is gene rally determined by the bounds of unknown inputs and model uncertainties.In this thesis, model-based fault detection approach is applied to linear switched systems. A switched linear system is a hybrid system consists of several linear sub-models and a rule which orchestrates the switching between them. This type of system provides a framework which bridges the linear systems and the complex and/or uncertain systems. The profit of utilizing switched linear systems is that, it is relatively easy to be handled as many powerful tools from linear analysis are available to cope with these systems.This thesis presents three novel approaches for fault detection in switched systems. The first approach is a deterministic method for fault detection. The main feature of this method is to utilize the residual signal to estimate the disturbance size and then use it in the evaluation function. The second approach is an integrated fault detection based on weighting factors. In this approach, the effect of disturbance on residual signal is integrated in the residual generation process. The last approach is fault detection based on H-/H8 optimization index. The objectives of these approaches are: (1) enhance the fault detectability in the switched systems, (2) utilize the available information provides by each local sub-model in the fault detection design. The design and implementation forms of these approaches are summarized in implementation algorithms, which are computationally tractable and user oriented. Comparison studies between the proposed FD schemes and the standard ones are established to illustrate the advantages and the efficiency ofthe developed approaches.The fault detection methodologies proposed in this thesis are tested on lateral vehicle dynamic control systems. Lateral vehicle dynamic is a nonlinear complex system, which can be under certain assumptions simplified to linear model called bicycle model (or one-track model). The simulation results demonstrate the feasibility of the proposed approaches and their efficiency in detecting the faults. The results also show a high degree of matching the lateral dynamics when switching strategy is used in the model.