A sensor-based learning approach to prognostics in intelligent vehicle health monitoring

This paper outlines recent progress towards the development of an innovative approach to an intelligent sensing network for airframe maintenance in hidden areas. In the maintenance of aircraft structures, substantial costs are associated with corrosion prevention and control, and a large proportion of these costs are incurred during regular inspections of hard-to-access areas. If the progression of corrosion can be monitored and predicted by a sensor system, then maintenance can be rescheduled on an as-needed basis. The overall framework of our vehicle health monitoring system is based around the principle of network intelligence utilising autonomous sensing agents, and is comprised of sensing, communications, multi-agent data analysis and diagnostics, and prognostics with intelligent objects. Corrosion sensors are used in combination with sensors that monitor the environmental causes of corrosion, e.g. humidity, temperature and conductivity. Relationships between the sensed damage and the causes of the sensed damage are developed during the operation of the system by analysis of the data streams, but are constrained by knowledge of the physical processes of degradation. Of crucial importance to the system is the ability to relate the corrosion sensor readings, both real and predicted, to the actual damage of the underlying aircraft materials. Two complementary approaches to this issue are being developed. In the first, the corrosion damage parameters for a number of aluminium alloys, both bare and primed, have been investigated, and relationships established to correlate test panel damage to the output data of galvanic corrosion sensors. The second approach is to use a purpose-designed ultrasonic Rayleigh wave sensor, currently under development, that can directly detect corrosion defects forming in the structural materials. While this sensor can only be utilised effectively in regions of simple geometry, and the most corrosion-prone areas are often in complex geometrical locations (e.g. within a crevice or fastener hole), it can nevertheless provide two valuable functions: firstly, to provide a 'calibration point' at which the actual sensed damage may be compared with that predicted by the system from the microclimate data; and secondly, to provide a measure of total accumulated damage. These approaches and their utility will be described in more detail in this paper. The predictive capabilities of the system require an estimate of the future microclimates that will be experienced in the aircraft, to allow the progression of corrosion to be calculated using the relationships built up from the existing sensed data streams. Analysis of real flight data has been used to develop a simple model to formulate usage scenarios. The model is based on temperature and absolute humidity, and formulates the in-flight microclimate based on airfield conditions and operational characteristics.