Impact Damage Detection in Aircraft Composites Using Knowledge-based Reasoning

Abstract

A hybrid reasoning methodology is applied to a complex aerospace structure, and its effectiveness is assessed in identifying and locating the position of impacts. Part of a commercial aircraft wing flap is impacted and time-varying strain response data from the structure are sensed using passive piezoceramic sensors. This structure can be regarded as a small scale version of part of a wing span with the corresponding features being a leading edge and trailing edge. The trailing edge is composed of aluminium skins with an aluminium honeycomb core, the leading edge of composite skins with a light weight honeycomb core, and the central section of thin composite material. Nine sensors, to detect time-varying strain response data, are distributed over the surface of the flap; two on the leading edge, two on the trailing edge, and five in the central section. The methodology combines the use of: (i) Case-Based Reasoning; in a `learning mode', an initial casebase is created with the principal features of the impact responses. When the system is working in an `operating mode', the data acquired by sensors are used to perform a diagnosis by analogy with the cases stored in the casebase: reusing and adapting old situations. (ii) The Wavelet Transform is used to extract principal features of a signal providing information about the impact locations. (iii) Self-Organizing Maps are trained as a classification tool in order to organize the old cases in memory with the purpose of speeding up the reasoning process. Finally, when old similar cases are retrieved, the impact location is obtained directly from heuristic considerations.

Multiway Partial Least Square (MPLS) to estimate impact localization in structures

“…Finally, the total average radial error is around 60mm. Results obtained in this work are similar to the ones presented in previous works that used different techniques such as Artificial Neural Networks (LeClerc et al [2007]), Genetic Algorithms (Mahzan et al [2007]) and Case Based Reasoning (Mujica et al [2008]) (Mujica et al [2008]). However, the methodology presented in this work uses the information gathered from the sensors, without computing new or additional features.…”

“…Another option is the utilisation of Genetic Algorithms (Mahzan et al [2007]) or Case Based Reasoning combined with different dimensionality reduction (or features extraction) techniques sucha as: Self Organizing Maps -SOM- (Mujica et al [2008]), Principal Component Analysis (PCA), Partial Least Square (PLS), Curvilinear Distance Analysis (CDA) and some extensions of them (Mujica et al [2008]). …”

Detecting Impacts on a Representative Aerospace Structure: An Implementation with Tests

“…Prime examples include research performed into insulating foam strikes following the loss of the space shuttle Columbia [6], approaches involving the use of embedded optical fibres [7,8], and approaches from the non-aerospace sectors, such as pipeline impact detection and monitoring [9]. These approaches have met with much success and now have a wide range of applications in structural health monitoring, extensively validated by experimental results [10,11,12]. Existing methods can be broadly sorted into two fields: fit-from-new, and retro-fit.…”

“…A logic tree then derived which zone the impact was in, based upon a qualitative comparison of the three dt variables. For instance, a comparison of dt 12 and dt 13 can be used to discern if an impact is on the inboard or outboard half of the wing. Various combinations of zones were trialled, and it was found that six overlapping zones, as shown in figure 4, offered reliable and repeatable detection location.…”

“…Three different densities of foam rubber sheath were used on the impact face for 6 impacts each of 2J conducted at 500mm intervals from the root of the leading edge. The three measured variables were the difference in time of arrival of the wavefront, established by thresholding, at the three sensors, labelled dt 12 , dt 13 and dt 23 , respectively. Figure 6 shows the results for one type of rubber sheath, which also includes linear fits using the least squares method.…”