This research mainly deals with fault diagnosis in nuclear power plants (NPP), based on a framework that integrates contributions from fault scope identification, optimal sensor placement, sensor validation, equipment condition monitoring, and diagnostic reasoning based on pattern analysis. The research has a particular focus on applications where data collected from the existing SCADA (supervisory, control, and data acquisition) system is not sufficient for the fault diagnosis system. Specifically, the following methods and systems are developed. A sensor placement model is developed to guide optimal placement of sensors in NPPs. The model includes 1) a method to extract a quantitative fault-sensor incidence matrix for a system; 2) a fault diagnosability criterion based on the degree of singularities of the incidence matrix; and 3) procedures to place additional sensors to meet the diagnosability criterion. Usefulness of the proposed method is demonstrated on a nuclear power plant process control test facility (NPCTF). Experimental results show that three pairs of undiagnosable faults can be effectively distinguished with three additional sensors selected by the proposed model. A wireless sensor network (WSN) is designed and a prototype is implemented on the NPCTF. WSN is an effective tool to collect data for fault diagnosis, especially for systems where additional measurements are needed. The WSN has distributed data processing and information fusion for fault diagnosis. Experimental results on the NPCTF show that the WSN system can be used to diagnose all six fault scenarios considered for the system. A fault diagnosis method based on semi-supervised pattern classification is developed which requires significantly fewer training data than is typically required in existing fault diagnosis models. It is a promising tool for applications in NPPs, where it is usually difficult to obtain training data under fault conditions for a conventional fault diagnosis model. The proposed method has successfully diagnosed nine types of faults physically simulated on the NPCTF. For equipment condition monitoring, a modified S-transform (MST) algorithm is developed by using shaping functions, particularly sigmoid functions, to modify the window width of the existing standard S-transform. The MST can achieve superior time-frequency resolution iii for applications that involves non-stationary multi-modal signals, where classical methods may fail. Effectiveness of the proposed algorithm is demonstrated using a vibration test system as well as applications to detect a collapsed pipe support in the NPCTF. The experimental results show that by observing changes in time-frequency characteristics of vibration signals, one can effectively detect faults occurred in components of an industrial system. To ensure that a fault diagnosis system does not suffer from erroneous data, a fault detection and isolation (FDI) method based on kernel principal component analysis (KPCA) is extended for sensor validations, where sensor faults are detected and isolated from the reconstruction errors of a KPCA model. The method is validated using measurement data from a physical NPP. The NPCTF is designed and constructed in this research for experimental validations of fault diagnosis methods and systems. Faults can be physically simulated on the NPCTF. In addition, the NPCTF is designed to support systems based on different instrumentation and control technologies such as WSN and distributed control systems. The NPCTF has been successfully utilized to validate the algorithms and WSN system developed in this research. In a real world application, it is seldom the case that one single fault diagnostic scheme can meet all the requirements of a fault diagnostic system in a nuclear power. In fact, the values and performance of the diagnosis system can potentially be enhanced if some of the methods developed in this thesis can be integrated into a suite of diagnostic tools. In such an integrated system, WSN nodes can be used to collect additional data deemed necessary by sensor placement models. These data can be integrated with those from existing SCADA systems for more comprehensive fault diagnosis. An online performance monitoring system monitors the conditions of the equipment and provides key information for the tasks of condition-based maintenance. When a fault is detected, the measured data are subsequently acquired and analyzed by pattern classification models to identify the nature of the fault. By analyzing the symptoms of the fault, root causes of the fault can eventually be identified.