Neural network/expert system-hybrid which identifies untrained damage

A major opportunity for smart structures research is the information processing challenge. Massive amounts of sensor information are required to monitor the health and integrity of any system, mechanical or biological. A prohibitively large computer is needed, if conventional methods are used to monitor a large number of sensors. We have addressed this complex problem by developing a neural network/expert system (NN/ES) hybrid. The NN portion is used for data acquisition and sensor processing. The ES portion monitors the results of the NN to determine if there are anomalies. Anomaly detection feedback is provided to the NN. This unique feedback situation provides our smart structure system with the capability to identify damage which it has never been trained with and which it has never seen. The NN/ES health monitoring capability has been used to detect both temporary and permanent damage introduced into a portable table-top test article. In addition to damage detection and health monitoring, the NN/ES can also provide simulated estimates of maintenance schedules and performance envelopes.