Neural network based signal monitoring in a smart structural system

This paper focuses on the signal processing aspect of a smart structure computational support environment for health monitoring, investigating the use of neural networks to identify and locate structural damage in a steel truss structure instrumented with accelerometers and strain gauges. Cracking damage is simulated by introducing sawcuts into the main members of the struc-ture. Results using accelerometer data alone indicate that Quickprop backpropagation neural net-works constitute a promising tool for these purposes, although network performance in locating damage should be improved by use of strain data as well. 1