An application of fuzzy sets in real time filtering problems

In real time updating of a state vector with error corrupted measurement data several types of problems can be encountered. The sensor data may be unusually noisy or have unexpected biases. In the cases of sensors that acquire and track electromagnetic signals the signal acquired may not even be the correct one. Furthermore, a sensor may be tracking the desired signal, lose it, and lock on to an incorrect signal. All of these problems are entirely feasible in space rendezvous navigation where optical and radar sensors are used. They can cause a filter editor to allow erroneous data to be processed, thus destroying the state vector or cause the correct signal to be edited thus losing valuable information. In space rendezvous operations these problems have been avoided by human monitoring of residuals and handling of questions of whether to process or inhibit data. In this paper the human decision making task is modeled with fuzzy sets and the Kalman filter updates to the state vector are weighted using fuzzy functions. Results of the study show that the use of fuzzy set models gives results comparable to those requiring human assistance, e.g. lock on to false targets, and better results when the problem is caused by noise and/or bias, or unexpected errors in the state vector at initial acquisition.