A structural neural system for health monitoring of structures

A method for structural health monitoring of large structures based on detecting acoustic emissions produced by damage was developed for this dissertation. The advantage of sensing acoustic emissions is that small damage can be detected in structures built with complex geometry and anisotropic materials. A longstanding limitation of the acoustic emission method is that a large number of bulky sensors are required to monitor cracks that can form at any location on a complex structure. The sensors and data acquisition system are also required to work at a high sampling rate because the frequencies of acoustic waves propagating in the structure due to damage are on the order of hundreds of kHz. To overcome the difficulties with using the acoustic emission method, a very elegant and powerful technique that many researchers have either missed or avoided is presented in this dissertation. The new sensing technique is called a structural neural system. The technique was difficult to develop, and required using electronic circuits to mimic the architecture of the biological neural system. In developing the technique, it was also necessary to recognize the strong linkage between fracture mechanics and fatigue damage detection.; The structural neural system developed uses continuous (multi-node) sensors to mimic dendrites, receptors, and the axon which perform sensing in the biological neural system. Analog electronics were then developed to mimic the thresholding and firing functions of the soma (cell body) in the neural system. The end result is a structural neural system that tremendously reduces the complexity and number of data acquisition channels needed to monitor acoustic emissions and detect damage in structures that have high feature density. Simulation and laboratory testing of a prototype of the structural neural system showed that the structural neural system is sensitive to small damage and practical to use on large structures. A field test was also performed in which a simple two-channel four neuron prototype structural neural system was installed on a 9 meter long wind turbine blade at the National Renewable Energy Laboratory in Golden, Colorado. The blade was loaded to failure in a quasi-static proof test. The structural neural system, using only two channels of data acquisition, identified where damage started during the testing, and monitored the growth of damage at five locations on the blade. The structural neural system detected damage well before final failure of the blade, whereas strain gages on the blade did not indicate damage until just before final failure. A post-failure sectioning and examination of the blade verified the damage locations predicted by the structural neural system, and showed that the structural neural system is a practical technique for health monitoring of large structures. Beyond health monitoring, the structural neural system can tell where damage initiates and how damage propagates in a structure. This information might be useful to improve the design and manufacturing of structures.