Evolutionary detection of sensor discredibility in control loops

When operating control loops, we can encounter hidden malfunctions, usually caused by the sensors used for measuring the controlled variables. These sensors often do not stop operating completely (which would be easy to recognize); they only start to provide slightly wrong measurements of the controlled variables. If the differences between the measured values and real values are not extreme, there is only a very small chance that the operator will recognize any malfunction of the sensor, assuming that there are no additional sensors available in the control loop for performing a hardware check. The problem is as follows: although the control loop seems to work properly, the consequences of such a small sensor malfunction (sensor discredibility) can become substantial and expensive (for example, we can imagine a combustion ratio control where some deviations from an optimal ratio value have no principal influence on the operation of the device, but late discovery of an increase in harmful emissions may be very costly). This paper concentrates on two new ideas: detection of sensor discredibility as a way that replaces usual hardware redundancy and saves the costs of additional measurements; and the use of sensor discredibility detection as a means of improving the function of a controlled system by avoiding hidden impreciseness in the control loop operation.