(476b) Kalman-Based Fault-Tolerant Control (FTC) for a Pilot-Scale Cooling Loop

Authors: 
Villez, K., Purdue University
Garcia, H., Idaho National Laboratory
Rieger, C., Idaho National Laboratory


TITLE:

Kalman-based
Fault-Tolerant Control (FTC) for a pilot-scale cooling loop

AUTHORS:

Kris Villez, Humberto Garcia, Craig Rieger and
Venkat Venkatasubramanian,

List
addresses : Purdue and INL details

ABSTRACT:

We report on the real-life evaluation of a
fault-tolerant control strategy for a pilot-scale version of a
cooling loop of a nuclear plant [1]. This cooling loop is equipped
with two valves, a butterfly valve (BFV) and a glove valve (GLV)
which both exhibit substantial hysteresis. This is the result of a
coupling of the motor and vane axles design which allows for leeway
between the valve motors and valve vanes, thereby leading to a safer
valve subsystem. The whole system is used as a study object in the
context of resilient system design and control [2].

In order
to design a supervisory control system allowing for fault-tolerance,
the Fault Detection and Identification (FDI) method from [3] was used
as a first component. This method is based on the deployment of the
Kalman filter for FDI purposes as proposed earlier in literature
[4,5]. The FDI component essentially consists of two steps, namely
fault detection and fault identification. The original method allows
for a single type of fault only, namely the bias type, though in any
actuator or sensor in the studied system. In [3], it was specifically
extended to allow for a wider range of faults, such as drift, stuck
and sticky faults in both actuators and sensors. With this new
method, several types of faults can be identified correctly.

A second
component consists of the supervisory controller which enables
corrective actions based on the diagnostic result. Whenever a fault
is detected, the supervisory controller order the valves to open,
leading to maximum cooling capacity, thus enabling the safest of
possible operations. Following a fixed waiting time, collected data
is used to identify the correct fault (second step of FDI). When that
is done, a final action is taken. For a bias or drift fault, the
regulatory control system can be corrected by means of a parametric
adjustment (e.g. in the corresponding measurement or actuator
signal). For other faults, like a stuck fault, one reconfigures the
control system so that the remaining working valve is used (only) for
flow control. We call this a structural adjustment. In our earlier
work [1], we have applied this method successfully in simulation
showing that the hybrid behavior (discrete and continuous dynamic
behavior) can be handled well in a simulation based study. Indeed, it
was shown that several faults could be detected and diagnosed
correctly. Figure 1 shows one simulated run in which the first valve
gets stuck at 50 seconds in the simulation. Detection follows at 60
seconds and at 80 seconds the fault is identified correctly.
Following that, the supervisory controller decides to use the second
valve instead of the first one.

With this
work, we will report on the real-life implementation of the described
supervisory controller. In particular, we expect to answer whether:

  • The Kalman filter works well for the valves with hysteresis.

  • The FDI strategy allows for successful detection and identification of faults in the real system

  • To which extend the supervisory control system allows to mitigate the effect of introduced faults in the valve subsystems such as faults of the bias, drift and stuck type.

REFERENCES

  1. Villez, K., Venkatasubramanian, V., Garcia, H., Rieger, C.,"Supervisory control of a pilot-scale cooling loop", submitted to the 4th International Symposium on Resilient Control Systems (ISRCS2011), 2011.

  2. Rieger, C. G., Gertman, D. I., and McQueen, M. A., “Resilient control systems: Next generation design research.” In 2nd IEEE Conf. on Human System Interaction, Catania, Italy, May 2009.

  3. Villez, K., Srinivasan, B., Rengaswamy, R. and Narasimhan, S., “Kalman-based strategies for Fault Detection and Diagnosis: Extensions and critical evaluation for a simple benchmark system.” Comput. Chem. Eng., 2011, In press.

  4. Prakash, J., Narasimhan, S., and Patwardhan, S. C., “Integrating model based fault diagnosis with model predictive control.” Ind. Eng. Chem. Res., 2005, 44 , 4344-4360.

  5. Prakash, J., Patwardhan, S. C., and Narasimhan, S., “A supervisory approach to fault-tolerant control of linear multivariable systems.” Ind. Eng. Chem. Res., 2002, 41 , 2270-2281.

Figure 1: Valve 1 stuck scenario. Top: Valve 1
position; Middle: Valve 2 position; Bottom: Flow rate. Valve 1 gets
stuck at 50 seconds. This is detected at 60 seconds and identified at
80 seconds. Fo
llowing
that, the second valve is now used instead of the first.

Topics: