The first decision is what the system is for. A sensor may be intended to replace a scheduled inspection, to extend its interval, or only to gather information alongside it. The three carry very different burdens of proof. The operator should state the objective for each location and open discussion with the authority and the design approval holder before hardware is selected, using the SAE guidance as a common vocabulary (SAE International, 2013).

Candidates are found in the operator's own programme, not in a catalogue of sensors. The useful locations are those where a known damage mode recurs at a defined place, where the present inspection needs substantial access work such as removal of interiors, fuel tank entry or jacking, and where the task repeats often. Typical examples on a pressurised turboprop are fuselage frames and stringer attachments in the propeller plane, pressure bulkhead attachments, wing-to-fuselage fittings, nacelle and engine mount structure, and bonded or riveted repairs under observation. Locations with diffuse or unpredictable damage are poor candidates for point sensors and are better left to conventional inspection.

For each candidate the operator should assemble a short dossier: the task card and its interval, the access steps and their labour, the findings history across the fleet, the damage mode and its expected direction of growth, and any repairs already embodied. The findings history deserves particular care. A location that has never produced a finding offers no evidence about how damage presents there, and a location with frequent findings may be better served by a modification than by a sensor. The dossier also fixes the geometry that the sensor must fit, including sealant fillets, fastener heads and neighbouring systems, which often decides what can be bonded where.

No sensor suits every task. Table 1 compares the main options. Comparative vacuum monitoring detects a surface crack when it breaches a fine evacuated gallery bonded to the structure, and has been the subject of airline validation programmes for specific fittings (Roach, 2009). Piezoelectric wafers excite and receive guided ultrasonic waves that can interrogate an area of plate or a bonded joint from a few fixed points (Giurgiutiu, 2014; Raghavan and Cesnik, 2007), and are attractive for composites, where the damage of concern is delamination (Su et al., 2006). Fibre Bragg gratings measure strain at many points along one fibre and suit load and usage monitoring (Di Sante, 2015). Usage monitoring from strain or flight parameters does not detect damage at all; it refines the fatigue calculation, an approach with long standing in military fleets (Molent and Aktepe, 2000).

A sensor on an aircraft is airborne equipment. Its adhesive, wiring, connectors and any on-board electronics must survive temperature cycling, vibration, fluids and humidity, and must not themselves create a hazard or initiate corrosion or fatigue at the point of attachment. The established environmental test standard for airborne equipment applies (RTCA, 2010). Durability of the bond is critical, because a sensor that detaches gives either a false alarm or, worse, silence. Self-diagnosis of the sensor and its bond should be a design requirement. The architecture should be decided in layers, as shown in Figure 1, and for a small fleet the lightest option is usually sensors and wiring terminated at an accessible connector that is interrogated on the ground with a portable unit.

Credit for an inspection rests on a demonstrated probability of detection for a given crack size. For conventional non-destructive inspection the procedure is codified (DoD, 2009), and it assumes many independent inspections of many flaws by different inspectors. A fixed sensor differs in ways that matter statistically. It occupies one position relative to the flaw, its readings over time are correlated, and variability comes from installation, ageing and environment and not from the human operator. The plan should therefore specify coupon and element tests with representative geometry, sealant, paint and loading, with cracks grown under the sensor, and should state in advance how repeated measurements will be treated. Environmental compensation of the baseline must be shown to work over the temperature and load range expected at the time of reading (Worden et al., 2007).

With capability evidence in hand, the operator and design approval holder propose the change to the maintenance programme. The conservative route is scheduled monitoring at the existing interval, with the sensor reading replacing the access and the inspection. Any indication is confirmed by conventional means before a disposition is made. Interval extension or fully automated monitoring should be deferred until service experience exists. The damage tolerance justification must be revisited, since the detectable crack size of the sensor, not that of the inspection it replaces, now sets the starting point of the crack growth calculation (Schijve, 2009).

Before any task is dropped, the system should be flown on a subset of aircraft while the existing inspections continue unchanged. The trial has three aims: to measure the false-call rate in service, to find installation and durability problems, and to exercise the data path from download to engineering disposition. Aircraft should be chosen to span the fleet's age and operating bases. The length of the trial should be fixed in advance in terms of flight cycles and scheduled inspection opportunities, and the criteria for success written down before it begins.

Every disagreement between sensor and inspection during the trial is valuable and should be investigated to a root cause, whether a sensor indication with no crack found or the reverse. Indications without damage usually trace to bond degradation, connector faults or uncompensated temperature, and each cause has a different remedy. Maintenance staff should be trained on the interrogation procedure before the trial starts, so that handling errors are not mistaken for shortcomings of the system.

The final step compares what the trial showed with the objective of the first step. The technical review asks whether the sensors agreed with the conventional inspections and survived. The economic review sets installation, added weight, data handling and continuing airworthiness of the system against the access hours and downtime saved. The balance depends heavily on how often the replaced task recurs and how much access it needs, and it should not be assumed to be favourable. A location that is inspected rarely, or reached easily, will seldom repay a permanent installation.