Simulation Results Sensor Faults

In the simulations, the following classes of faults on the temperature sensors have been considered  

Various simulations with the above kind of faults occurring at different time instants have been done, and the obtained results show that the proposed diagnostic scheme has been able to detect and isolate all the simulated sensor faults. In detail, Figs. 6.5 to 6.12 show the voted measures and the norms of both detection and isolation residuals in the presence of different sensor faults. 

Finally, Figs. 6.11 and 6.12 show the results obtained for an increasing noise added to the output of the sensor ST, i namely, a white noise with zero mean and increasing variance has been added to the sensor output starting at time 12000 s. In this case, the delay for the isolation is about 100 s. 

It can be easily recognized that all the faults have been correctly detected and identified. A wide simulation campaign showed that only when an abrupt freezing on reactor temperature sensors (i.e., 5r i or ST ) occurs during the isothermal phase, the fault can be detected but not isolated. Finally, the measure voted by the DMS results to be always more accurate than the standard duplex measure. 

---

The fault signature matrix excluding sensor faults is given in Table 7.4, where Ui and di (i = 1, 2) are binary and complementary to each other, i.e fl = di, and Rdi (i = 1,..., 4) represent residuals. The fault signature, monitorability, and isolatability of some components change with the operating modes of the system. Simulation results for this system with an adaptive threshold-based decision procedure are presented in the next section. 

A third challenge is simulation fidelity, in particular whether there are faults in the system that the simulation cannot reveal because they depend on details that are not modelled. In this paper, we explore this issue in a simple way - we run simulations with both infallible sensors and sensors subject to random noise, and look at how the results of the latter simulations are richer (in terms of the range of hazardous situations found). 

---