AUTOMATED SIGNAL MONITORING USING NEURAL NETWORKS IN A SMART STRUCTURAL SYSTEM

Automated health monitoring of instrumented structures will require an appropriate suite of information processing techniques. One such technique, involving Quickprop neural networks, is developed to identify and locate structural damage in a 3D steel truss-type structure instrumented with accelerometers and strain gauges. In experiments conducted in a structural testing laboratory, transient vibration tests caused by impact hammer strikes were conducted on the instrumented structure which was subjected to various damage scenarios. Results of the investigation indicate that neural networks provide a promising approach as one component of the computational tool kit required for on-line autonomous health monitoring of instrumented structures. Anticipating the need for such a comprehensive tool kit, a computational framework for automated signal monitoring is proposed and introduced as well. This framework incorporates signal processors based on neural networks in an object-oriented model for structural monitoring and diagnosis.