Development of a decentralized artificial intelligence system for damage detection in composite laminates for aerospace structures
Kral, Zachary Tyler
AdvisorHorn, Walter J.; Steck, James E.
MetadataShow full item record
Because of economic impact that results from downtime, aircraft maintenance is an important issue in the aerospace industry. In-service structures will decay over time. Compared to low-cycle loading structures, aerospace structures experience extreme loading conditions, resulting in rapid crack propagation. The research involved in this dissertation concerns development of the initial stages of structural health monitoring (SHM) system that includes a network of ultrasonic testing sensors with artificial intelligence capable of detecting damage before structure failure. A series of experiments examining the feasibility of ultrasonic sensors to detect the initial onset of damage on a composite laminate, similar in structure to that used in aerospace components, was conducted. An artificial neural network (ANN) with the best accuracy was found to be a hybrid of a self-organizing map (SOM) with a feed-forward hidden and output layer. This was used for the single actuator-to-sensor scans on a composite laminate with simulated damage. It was concluded that a decentralized network of sensors was appropriate for such a system. The small four-sensor system was proven to be capable of predicting the presence of damage within a scanning area on a composite laminate, as well as predict the location once damage was detected. The main experimentation for this dissertation involved four ultrasonic sensors operated in a pitch-catch configuration. Simulated damage, verified through experimentation, was placed at various locations in the scanning area of interest. Signals obtained from the ultrasonic sensors were analyzed by a multi-agent system in which each agent describes an ANN. The system was trained to determine damage size. A second multi-agent system was constructed to determine the location of the detected damage. The architecture was similar to the damage-sizing system. Results demonstrated that with the artificial intelligence post-processing of ultrasonic sensors, 95% confidence can be obtained for detecting and locating damage that is 0.375 in. in diameter, which was verified through a bootstrap method. This dissertation validated the initial stages of constructing such a network of ultrasonic sensors. Future research in this area could involve combining the four-sensor network into a larger network of sensors by means of multi-agent processing (i.e., developing scanning regions). The novel method presented here provides the basis for the development of the SHM system for typical aerospace structures.
Thesis (Ph.D.)--Wichita State University, College of Engineering, Dept. of Aerospace Engineering.