Development of a parametric-based indirect aircraft structural usage monitoring system using artificial neural networks

Abstract

development of cost-effective indirect monitoring system because many of the relationships between parameters and strains are often non-linear and difficult to identify with more traditional regression techniques.Researchers in the aeronautical arena have published work using ANNs and similar mathematical techniques to identify loading actions, or to supplement sparse flight test data. Azzam et al (8,9) described the use of a mathematical network to predict loads. Azzam et al (9) and Wallace et al (10) described co-operative work programmes involving Smiths Aerospace, the UK Ministry of Defence and BAE Systems and reported, among other successes, excellent correlation between measured and predicted strains at a number of structural locations for the Tornado combat aircraft, over a large number of test flights. Azzam (8) also described the development of methods to predict damage in helicopter components and the prediction of high frequency events, such as buffet loading on the fin of a fixed-wing aircraft. Jacobs and Perez (11) used a hybrid cascade ANN to augment sparse buffet data by predicting the power spectral densities of surface pressures on a fighter aircraft vertical tail, from parametric data. Kim and Marciniak (12) reported the use of a back-propagation ANN to predict strains in the empennage of a Cessna 172P general aviation aircraft during discrete manoeuvres. Hill et al (13) described the use of a back-propagation ANN to synthesise strain and fatigue damage in several Lynx helicopter components. Furthermore, Manry and co-authors (14) applied ANNs to flight load synthesis, using Bell helicopter data. Levinski (15) used ANN technology to synthesise loads on the vertical tail of a F/A-18 aircraft, using wind tunnel pressure data. Wang (16) used a recurrent ANN to predict loads during discrete manoeuvres for a fighter aircraft. Reed and Cole (17) and Reed (18) reported the development of an ANN-based parametric fatigue monitor for the wing and tailplane of a military trainer aircraft. With the exception of the work reported by Azzam and his associates (8,9,10) and Reed and Cole (17,18) the majority of the methods used were developed to predict discrete loading actions or manoeuvres rather than to predict the usage of the aircraft throughout its operational envelope.Within the following sections, the architecture of the ANN core of the structural health and usage neural network (SHAUNN) is explained. Thereafter, a generic process for developing a SHAUNN system is outlined. Also, results from case-study demonstrator programmes are reported. Finally, conclusions are drawn and recommendations for further work are made. It is noteworthy that software development methods and practices are not discussed within this paper; a range of comprehensive standards and guidance, appropriate to the safety criticality of the software, already exists and is familiar to software developers.

Comparison of real-time flight loads estimation methods

“…In [17] the development of a parametric-based indirect aircraft structural usage monitoring system using artificial neural networks is described. Flight data has been used to predict strains or stresses at key structural locations for several military aircraft types.…”

“…Other publications like [11,17,18] follow the same approach to model the loads at the vertical tail plane of an aircraft in dependency of aircraft system parameters and include additional input parameters such as the deflection of flaps and ailerons. Table 1 summarises the selected parameters used within this paper for each of the considered loads.To simplify, obtaining a mapping function describing the relation between the input and the output parameters basic physical laws are used [17] to combine the selected input parameters to reflect the relation present between the chosen input parameters and the considered output parameters.…”

“…Table 1 summarises the selected parameters used within this paper for each of the considered loads.To simplify, obtaining a mapping function describing the relation between the input and the output parameters basic physical laws are used [17] to combine the selected input parameters to reflect the relation present between the chosen input parameters and the considered output parameters. The lateral acceleration N y is therefore combined with the aircrafts mass W, giving a lateral force…”

Evolutionary computation methods for helicopter loads estimation

“…Much research has been carried out using machine learning methods to model operational loads experienced by fixedwing aircraft structure [19], [7]. In the case of rotary-wing aircraft, the loading spectrum experienced by the airframe structure is significantly more complex since the dynamic rotating components operate at frequencies several orders of magnitude higher than for fixed-wing aircraft.…”

Computational intelligence methods for helicopter loads estimation

“…Much research has been carried out using machine learning methods to model operational loads experienced by fixedwing aircraft structure [19], [6]. In the case of rotary-wing aircraft, the loading spectrum experienced by the airframe structure is significantly more complex since the dynamic rotating components operate at frequencies several orders of magnitude higher than for fixed-wing aircraft.…”