1. The main focus of this research was to provide design recommendations and guidelines for the implementation of a structural health monitoring system in a composite structure. A successful design will use several different sensing methods, taking advantage of both the strengths and weaknesses of each. Using the sensor trade spaces shown inFigure 5 and Figure 6 a SHM designer could determine appropriate sensing methods based on the required damage resolution and power budget. It can be seen that they are all generally capable of detecting the same size of damage and can be implemented with similar size and power sensors, however frequency response and Lamb wave techniques are the only ones that can offer full surface coverage for a 1 x 1 m plate. While some other methods, such as eddy currents, can offer better damage resolution, they are only capable of detecting damage directly below the sensor, which would drive the system to use either very large sensors or a large volume of sensors. An estimate could also be made for sensor density based on desired coverage area using the equations presented in previous papers13-17. Thetradebetween redundancy and reliability is essential since missed damage or false-positives could prove financially fatal. Using event-driven processing, such as a passive system triggering a dormant active one could reduce power and complexity, and further gains could be reached by using ambient conditions to provide power or actuation. Lastly, it would be advantageous to design a system that was flexible enoughtoberetrofittedintoexistingagingsystems. A design proposed by the authors would use relatively small (0.25 - 1.0 m2) autonomous sensor patches as its key elements. These patches would include multiple piezoelectric sensors around their perimeter, local wiring between the sensors (longest length of 0.5 m), a data acquisition/processing device (capable of sampling around 1 MHz), a rechargeable polymer battery with an inductive coil for power reception (50 mW required to power all components), and a short range wireless device (10 m transmission range). All of these components would be embedded or deposited onto a conformable insulating polymer sheet with a thermoplastic adhesive backing, so the patch could be removed if it were damaged or if the structure required repair. These patches would be generic so that they could be placed in any region of concern on a vehicle. Other sensor types could possibly be deposited onto the polymer as well in certain regions, such as meandering wires for eddy current tests or differential parallel metal tracks for thermocouple readings. A neuralnetwork algorithm could be used for the sensors to learn the topology of the area of structure they are adhered over, to collect a small database of the undamaged state, and to discern where each patch was in spatial coordinates of the structure. In operation the sensors would passively collect strain and acoustic emission data, passing their data along to their local processing units. When abnormal data is encountered, active transfer function frequency response and Lamb wave methods would be initiated, using the same piezoelectric sensors, to verify the presence of damage. Once damage is located within the patch region, the nearest neighbor patches would be contacted wirelessly to attempt to confirm the damage. This compiled, consolidated and compressed data would then be passed patch to patch to the central processing unit to be interpreted, and the damage type, severity and location would be indicated to the operator and ground crew on a computer terminal along with suggested actions. This system would function continuously during operation, and could also be automatically accessed by the operator or ground crew to perform a mid-air or ground inspection on demand. As a first step towards acceptance of such a system, the operator could rely on it only to speed ground inspections by accessing the in-situ sensor patches via an ethernet connection to replace teardown inspections.