Due to the increasing complexity of modern technical processes, the most critical issues in the design of an automated system nowadays are safety/reliability, higher performance and cost efficiency. Therefore, fault diagnosis is becoming an essential part of modern control systems. Contrary to the well-developed fault diagnosis techniques for linear systems, there is still no systematic solution for nonlinear systems. Since most of real systems are nonlinear in nature, the objective of this thesis is to develop nonlinear observer and energy-balance based fault diagnosis approaches, which achieve a high performance and at the same time reduce the difficulties in the design and application.