Joseph Oliver

Development of a composite UAV wing test-bed for structural health monitoring research

In order to facilitate damage detection and structural health monitoring (SHM) research for composite unmanned aerial vehicles (UAV) a specialized test-bed has been developed. This test-bed consists of four 2.61 m all-composite test-pieces emulating composite UAV wings, a series of detailed finite element models of the test-pieces and their components, and a dynamic testing setup including a mount for simulating the cantilevered operation configuration of real wings. Two of the wings will have bondline damage built in; one undamaged and one damaged wing will also be fitted with a range of embedded and attached sensors-piezoelectric patches, fiber-optics, and accelerometers. These sensors will allow collection of realistic data; combined with further modal testing they will allow comparison of the physical impact of the sensors on the structure compared to the damage-induced variation, evaluation of the sensors for implementation in an operational structure, and damage detection algorithm validation. At the present time the pieces for four wings have been fabricated and modally tested and one wing has been fully assembled and re-tested in a cantilever configuration. The component part and assembled wing finite element models, created for MSC.Nastran, have been correlated to their respective structures using the modal information. This paper details the design and manufacturing of the test-pieces, the finite element model construction, and the dynamic testing setup. Measured natural frequencies and mode shapes for the assembled cantilevered wing are reported, along with finite element model undamaged modal response, and response with a small disbond at the root of the top main spar-skin bondline.

A guided-wave system for monitoring the wing skin-to-spar bond in unmanned aerial vehicles

Unmanned Aerial Vehicles (UAVs) are being increasingly used in military as well as civil applications. A critical part of the structure is the adhesive bond between the wing skin and the supporting spar. If not detected early, bond defects originating during manufacturing or in service flight can lead to inefficient flight performance and eventual global failure. This paper will present results from a bond inspection system based on attached piezoelectric disks probing the skin-to-spar bondline with ultrasonic guided waves in the hundreds of kilohertz range. The test components were CFRP composite panels of two different fiber layups bonded to a CFRP composite tube using epoxy adhesive. Three types of bond conditions were simulated, namely regions of poor cohesive strength, regions with localized disbonds and well bonded regions. The root mean square and variance of the received time-domain signals and their discrete wavelet decompositions were computed for the dominant modes propagating through the various bond regions in two different inspection configurations. Semi-analytical finite element analysis of the bonded multilayer joint was also carried out to identify and predict the sensitivity of the predominant carrier modes to the different bond defects. Emphasis of this research is based upon designing a built-in system for monitoring the structural integrity of bonded joints in UAVs and other aerospace structures.

Statistical Damage Identification by Frequency Response Function Based Residual Force Minimization

Systems for structural health monitoring and prognosis systems will increase the performance and safety of future aerospace structures. A key component of SHM is damage identification, defined herein as the process of using dynamic measurements and an analytical structural model to numerically quantify damage and assess the associated estimation uncertainty. In the current work, a new algorithm for damage identification in noisy systems is developed and validated analytically. The algorithm is based on the Bayesian linear optimal estimate in the form of generalized least-squares minimization. The cost-function combines the dynamic residual force vector and parameter regularization, with each term weighted by the inverse of its covariance matrix. An exact sensitivity formulation is incorporated, along with a statistically consistent iterative algorithm which finds an optimally linearized point and then performs the parameter estimation in a single step. The output is an updated set of parameter values and variances. The method is validated on two structurally damped mass-spring systems with additive noise: a grounded 2-degree-of-freedom oscillator and a damped version of the Kabe 8-degree-of-freedom satellite structure system identification problem. In addition, a widely referenced statistical damage detection algorithm from the literature is implemented to provide a comparison. The analytical validations demonstrate excellent performance, in terms of accuracy to known damage cases and resilience against increasing levels of additive noise. In the majority of cases, the new algorithm performs better than the second comparison algorithm. Nomenclature Nω = Number of discrete frequency lines d N = Number of degrees-of-freedom k ω = k

Frequency Domain Statistical Damage Identification Applied to an Experimental Composite Plate

An algorithm for statistical damage identification based on minimization of frequency response function based residual forces is described and experimentally validated on a composite laminate plate with impact damage. The algorithm uses measured frequency response and coherence functions to update a structural finite element model through statistically weighted least-squares minimization. Outputs are location and quantification of the damage, estimation uncertainty, and an updated model. Advantages compared to other approaches include robust applicability to systems which are heavily damped, large, and noisy, with a relatively low number of distributed measurement points compared to the number of analytical degrees-of-freedom of the associated analytical structural model (e.g., modal finite element model). The current work provides a brief overview of the algorithm background and formulation then details experimental implementation, including challenges associated with measured data, preliminary model correlation, and modeling damping in reduced coordinates. The algorithm is then validated experimentally on a graphite/epoxy composite plate structure with enforced impact damage using measured hammer-impact vibration data at 36 measurement points. Results show that the algorithm can locate and identify the damage with acceptable error and that the estimation uncertainty can be used to determine when damage results can be trusted.

Frequency Response Function-Based Statistical Damage Identification Applied to Composite Sandwich Plates

An algorithm for statistical damage identification based on minimization of frequency response function based residual forces, previously described and analytically validated on noisy damped mass-spring systems, is applied to structures modeled with finite elements. The algorithm is based on the Bayesian linear optimal estimate in the form of generalized least-squares minimization, including parameter regularization, an exact sensitivity formulation, and a statistically consistent iterative algorithm. The output is an updated set of parameter values and variances. In the current work, the algorithm has been implemented into a code which retrieves mass and stiffness matrices from an MD.Nastran finite element model of the structure when required, applies dynamic reduction to the measurement degrees-of-freedom, then performs damage identification. In addition, an updated damping formulation is described, along with solutions to some of the challenges presented by using sensitivity based damage identification on larger finite element models. The code is validated analytically on a series of graphite/epoxy/honeycomb sandwich plate models, similar to skin sections of modern all-composite aircraft, with three damage cases and additive noise. The finite element representation of the structures include plate elements for laminates and solid elements for the core, with damage modeled in the core and laminates by decreasing stiffness properties. In addition, several parametric studies further explore the limits of performance capability. The algorithm is shown throughout the validation and studies to perform very well, even with up to a 99.8% decrease in degrees-of-freedom between the analytical model and available measurements, high levels of measurement noise, and up to 36 individual update parameters.

Finite element model correlation of a composite UAV wing using modal frequencies

The current work details the implementation of a meta-model based correlation technique on a composite UAV wing test piece and associated finite element (FE) model. This method involves training polynomial models to emulate the FE input-output behavior and then using numerical optimization to produce a set of correlated parameters which can be returned to the FE model. After discussions about the practical implementation, the technique is validated on a composite plate structure and then applied to the UAV wing structure, where it is furthermore compared to a more traditional Newton-Raphson technique which iteratively uses first-order Taylor-series sensitivity. The experimental testpiece wing comprises two graphite/epoxy prepreg and Nomex honeycomb co-cured skins and two prepreg spars bonded together in a secondary process. MSC.Nastran FE models of the four structural components are correlated independently, using modal frequencies as correlation features, before being joined together into the assembled structure and compared to experimentally measured frequencies from the assembled wing in a cantilever configuration. Results show that significant improvements can be made to the assembled model fidelity, with the meta-model procedure producing slightly superior results to Newton-Raphson iteration. Final evaluation of component correlation using the assembled wing comparison showed worse results for each correlation technique, with the meta-model technique worse overall. This can be most likely be attributed to difficultly in correlating the open-section spars; however, there is also some question about non-unique update variable combinations in the current configuration, which lead correlation away from physically probably values.

Validating finite element models of composite aerospace structures for damage detection applications

Carbon-fiber-reinforced-polymer (CFRP) composites represent the future for advanced lightweight aerospace structures. However, reliable and cost-effective techniques for structural health monitoring (SHM) are needed. Modal and vibration-based analysis, when combined with validated finite element (FE) models, can provide a key tool for SHM. Finite element models, however, can easily give spurious and misleading results if not finely tuned and validated. These problems are amplified in complex structures with numerous joints and interfaces. A small series of all-composite test pieces emulating wings from a lightweight all-composite Unmanned Aerial Vehicle (UAV) have been developed to support damage detection and SHM research. Each wing comprises two CFRP prepreg and Nomex honeycomb co-cured skins and two CFRP prepreg spars bonded together in a secondary process using a structural adhesive to form the complete wings. The first of the set is fully healthy while the rest have damage in the form of disbonds built into the main spar-skin bondline. Detailed FE models were created of the four structural components and the assembled structure. Each wing component piece was subjected to modal characterization via vibration testing using a shaker and scanning laser Doppler vibrometer before assembly. These results were then used to correlate the FE model on a component-basis, through fitting and optimization of polynomial meta-models. Assembling and testing the full wing provided subsequent data that was used to validate the numerical model of the entire structure, assembled from the correlated component models. The correlation process led to the following average percent improvement between experimental and FE frequencies of the first 20 modes for each piece: top skin 10.98%, bottom skin 45.62%, main spar 25.56%, aft spar 10.79%. The assembled wing model with no further correlation showed an improvement of 32.60%.