SUPERVISION AND DIAGNOSIS OF INDUSTRIAL SYSTEMS

This Thesis focuses on two specific research areas, both related to the subject of supervision and diagnosis of industrial systems. One topic is the fault detection and diagnosis, a subfield of the control engineering which concerns itself with monitoring a system, identifying when a fault occurred, and pinpointing both the type and location of fault. Within this context, this research addresses the issue of discriminating between different faults which could affect both sensors and actuators of a generic system, by developing a novel proposal based on two main concepts, namely sliding mode observer and residual signature. The validity of the proposed fault diagnosis scheme was tested successfully on a steam separator unit of a thermal power plant. The other topic is the data reconciliation and parameter estimation, a crucial technology which allows for obtaining and validating reliable process models. Within this other context, in this research the joint problem of performing in real-time the reconciliation of measurements and the estimation of the model’s parameters, by exploiting the concept of temporal redundancy, has been examined, and a suitable filtering approach which exploits the concepts of quasi-steady-state and Kalman filter has been developed. An important contribution to the state-of-the-art results in the opportunity of decoupling the two tasks of dynamic data reconciliation and parameter estimation by means of two different schemes, which allow for turning on the parameter estimation filter only when a stable estimate of the state has been achieved. Also in this case, the effectiveness of the proposal has been evaluated on a real industrial application, related to monitoring the healthy conditions of a pyrolysis reactor. A scientific paper about this contribution will be submitted soon.