Fault Diagnosis Techniques for Linear Sampled Data Systems and a Class of Nonlinear Systems

This thesis deals with the fault diagnosis design problem both for dynamical continuous time systems whose output signal are affected by fixed point quantization, referred as sampled-data systems, and for two different applications whose dynamics are inherent high nonlinear: a remotely operated underwater vehicle and a scramjet-powered hypersonic vehicle. Robustness is a crucial issue. In sampled-data systems, full decoupling of disturbance terms from faulty signals becomes more difficult after discretization. In nonlinear processes, due to hard nonlinearity or the inefficiency of linearization, the ''classical'' linear fault detection and isolation and fault tolerant control methods may not be applied. Some observer-based fault detection and fault tolerant control techniques are studied throughout the thesis, and the effectiveness of such methods are validated with simulations. The most challenging trade-off is to increase sensitivity to faults and robustness to other unknown inputs, like disturbances. Broadly speaking, fault detection filters are designed in order to generate analytical diagnosis functions, called residuals, which should be independent with respect to the system operating state and should be decoupled from disturbances. Decisions on the occurrence of a possible fault are therefore taken on the basis such residual signals.