Sensor Fault Tolerance

Introduction

Measurements taken at different points of a system are not random. They are coupled by the dynamics of the process. This is termed analytical redundancy. Analytical redundancy can be employed to provide an estimate for a sensor signal in the presence of a sensor fault or even total loss of the sensor. Albeit the so calculated estimate of the sensor readout may be good enough to control the process for some time after a sensor loss, in the interest of optimal control, it is typically not advisable to abandon the sensor totally in the system design.

Block Diagram

The following block diagram explains the individual stages of a fault management system with sensor fault tolerance.

One keypoint is the process model, which must be of sufficient fidelity to be balte to detect slight deviations in the process behavior and thus be capable of detecting tiny faults. The high model fidelity is also necessary for the application of analytical redundancy, since the estimate of the sensor reading of a faulty sensor must be precise enough to meet the requirements imposed on the control quality of the control loop. The fault management can then employ analytical redundancy as a recovery action. The actual sensor value is no longer supplied by the (faulty) sensor to the controller, but rather calculated by the process model based on other measurements recorded at the testbed.

Video (Length 3:00)

In the video, you can see how the fault detection and diagnosis system switches over from the faulty sensor to the model-sensor employing analytical redundancy to provide an estimate for the measurements of the faulty sensor.

Note: Scripting must be enabled and a current Flash player must be installed in order to see this movie.

Related Publications
18Marco Muenchhof: Fault Management for a Smart Hydraulic Servo Axis. Proceedings of the Actuator 2006, Messe Bremen GmbH, 2006. Language: Englisch
21Marco Muenchhof: Model-Based Fault Detection for a Hydraulic Servo Axis: Fortschrittberichte Reihe 8, Nr. 1105. VDI-Verlag, 2006. Language: Englisch
Note: All downloads are password protected.