Structural Health Monitoring in Composites Based on Probabilistic Reconstruction Techniques Procedia Engineering 167 ( 2016 ) 48 – 55 1877 7058 © 2016 The Authors Published by Elsevier Ltd This is an[.]
Available online at www.sciencedirect.com ScienceDirect Procedia Engineering 167 (2016) 48 – 55 Structural Health Monitoring in composites based on probabilistic reconstruction techniques V Memmoloa*, F Riccia, N.D Boffaa, L Maioa, E Monacoa a Univ degli Studi di Napoli Federico II, Dept Of Industrial Engineering, Via Claudio 21, Naples 80125, Italy Abstract Structural Health Monitoring (SHM) based on guided waves (GWs) is responsive to damage occurrences allowing the assessment of composites integrity However, online monitoring is quite complex due to a wide range of parameters affecting wave propagation and consequently diagnostic outputs A useful implementation of a GWbased condition monitoring requires an accurate analysis of reconstruction algorithm and collected data as well In view of a more effective detection of impact induced damages in composites using a sparse array of sensors, several methods are explored in this paper as a first step towards a comprehensive capability assessment of a system permanently installed on aircraft structures Various probabilistic reconstruction methods and signal transformation techniques are developed and used to detect hidden flaws with a multiple analysis From a wide number of simulation carried out it appears that, although a single procedure for the estimation of the structural health is a fast solution for flaw detection, a multiple analysis based on different reconstruction techniques and/or several damage parameters could provide more detailed information about the location and the severity of possible failures if the parameters affecting diagnostics are completely addressed © Authors Published by Elsevier Ltd This ©2016 2016The The Authors Published by Elsevier Ltd.is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/) Peer-review under responsibility ofthe Organizing Committee of DRaF2016 Peer-review under responsibility of the Organizing Committee of DRaF2016 Keywords:Composite structures; delamionation; guided waves; probabilistic damage reconstruction; structural health monitoring Introduction Composite structures are extensively used by aerospace manufacturers with the aim to improve their aircraft performances Unfortunately, impacts occurring during flight likely affect internal layup inducing damages which may be not visible by means of visual inspections The maintenance approaches and the safe life design actually operated often leads to a significant reduction of benefits estimated a priori However, in order to make sure the * Vittorio Memmolo Phone: +39 081 7683576 Email Address: vittorio.mmmolo@unina.it 1877-7058 © 2016 The Authors Published by Elsevier Ltd This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/) Peer-review under responsibility of the Organizing Committee of DRaF2016 doi:10.1016/j.proeng.2016.11.668 V Memmolo et al / Procedia Engineering 167 (2016) 48 – 55 advantages of composites from design without any safety reduction, a system permanently installed on the structure monitoring its health could be a very perspective solution First of all, improving aircraft inspection procedures, such approach could provide a reduction of operative cost, 25 % of which is usually due to maintenance operations [1] Then, integrating SHM within structural design, a relax of allowable stresses and design constraints could be achieved with a condition based maintenance avoiding a safe life approach In view of online monitoring the methodology requires a technique capable to quickly identify emerging defects Ultrasonic approaches respond to such needs, allowing to achieve an efficient solution due to the good compromise between sensitivity to the presence of defects and area inspected Guided waves (GWs) are indeed able to interrogate a wide area from a single location travelling within the structure walls Furthermore, the integration of a cluster of lightweight sensors allows to monitor a prefixed area [2] without compromising the hosting structure [3] However, due to the complexities of GWs propagation in composite components and its sensitivity to the measurement instrumentation and technique [4], a useful implementation of a GW-based monitoring system requires an accurate analysis of collected data and derived information as well as reconstruction approach The advanced signal processing required to extract useful and reliable information on the state of health from wave signal, the analysis of the most probable area of damage and the decision making procedure lead to a complex framework Although a single procedure for the estimation of the structural health is a fast solution for flaw detection, a multiple analysis based on different reconstruction techniques and/or several damage parameters could give more detailed information about the location and the severity of the failure This paper deals with the development of an SHM system for the identification of small emerging impact induced damages in composites using a sparse array of surface ultrasonic transducers Numerical investigations are introduced here in view of a simulated framework for the analysis of damage scenarios in complex structures as well as to achieve a comprehensive point of view on various aspects affecting the diagnostic outputs of an autonomous SHM system Various probabilistic reconstruction methods and signal transformation techniques are developed and used to detect hidden flaws Several wave features affected by damage are extracted from complex signals Moreover, different probabilistic algorithms based on linear and nonlinear reconstruction techniques are developed in view of a multiple analysis capable to increase the effectiveness of the system The ultrasonic-GW tomographic techniques are presented in detail and numerical simulations carried out on a typical composite structure allow to compare different proposed algorithms in terms of effectiveness of damage reconstruction and localization as well Damage scenarios are obtained through a well-validated numerical stage looking to low velocity impact tests assessed by traditional ultrasonic inspection (c-scan) in composites [5] Such part of the works allows to address the effects of some parameters affecting identification and localization of damage The obtained results are indeed analyzed from a theoretical point of view in order to avoid their misleading interpretation Finally, a brief comparison between several reconstruction techniques and other methods developed by authors is presented in order to summarize future perspectives Probabilistic Reconstruction Approach SHM approaches are mostly oriented to identification (detect anomalies present in the structure) and localization of damage (find the most probable position of such anomalies) in view of comprehensive assessment of structural health If the identification is well addressed with a making decision framework providing a confidence in stating that there is a damage [6], the system is indeed able to assess the gravity of damage providing all diagnostic outcomes (presence, location and dimension of damage) However, wide ranges of parameters affecting identification and localization can be identified and they are often quite complex to be addressed in relation to the damage Whereas the detection is related to the useful interrogation and opportune wave feature adopted [7], one of the most crucial aspect for the localization is the reconstruction approach When a pair of sensors is used to interrogate the structure it is indeed quite simple to define a signal response gathering information on the heath condition along the direct line of sight However, a cluster of sensors provides such information along a wide number of paths and the localization is affected by the efficient reconstruction algorithm Every information needs to be opportunely addressed on the structural mesh in order to obtain a probabilistic distribution about the presence of damage as a tomographic representation [8] 49 50 V Memmolo et al / Procedia Engineering 167 (2016) 48 – 55 When a complex waveform is experimentally approached or numerical investigated in view of a future measurement validation, it is useful to assess how signals propagating in the pristine structure can be affected by the hidden flaw The simplest approach consists in the comparison of a wave feature extracted from signal (e.g a parameter representative of the wave behaviour affected by damages) with that evaluated on a known condition reference If no change is currently present in the structure, the current feature should be very close to that evaluated on the pristine structure, resulting in a non-identification response Otherwise, if a change is present, such feature should increasingly change with severity of damage Regardless of the extracted feature, a common metric adopted is the Damage Index (DI) as reported in Eq in which ݂ௌ and ݂ௌ are scalar quantities representing the features extracted respectively from pristine (baseline signal) and current (current signal) structure DI f CS f BS (1) f BS Obviously the damage index defined in Eq provides information about the direct line of sight constituted by the pair of sensors considered When the structure is inspected, each sensor is individually actuated, and the remaining sensors detect guided waves in order to provide the network depicted in Fig (a) The structure is first interrogated in a known undamaged state that represent the reference state (baseline) Then, the structure is monitored through scheduled interrogation and the DI formulation is computed for every pair of sensors providing the comparison between current and baseline condition It is furthermore necessary to define a decreasing probability far from such path with a linear or nonlinear definition of a distance Index (dI) in order to assemble the damage probability index (DPI) in every point of structural mesh as shown in Fig (b) (b) (a) Fig Sparse array of PZT sensors and network on which the damage indices are evaluated (a) and structural mesh on which the damage probability index is calculated with decreasing probability according to specific reconstruction approach (b) The effect which the i-th path indeed induces to the specific point P of the structural mesh is addressed in Eq DPi ( P) DI i dIi ( P) (2) 51 V Memmolo et al / Procedia Engineering 167 (2016) 48 – 55 where dIi defines the decreasing probability from the i-th path In this way the DI accounts the gravity of the damage detected along the direct line of sight and the dI assesses the proximity of such damage Finally, the DPI of the point P is evaluated accounting the effects of every path induced on the point P and normalizing such overall value As discussed and addressed in the literature [7,9], several features have different sensitivity to the damage because different information is extracted from the signal An impact induced damage can be idealized as a complex discontinuity distributed through the thickness A discontinuity in a waveguide is an abrupt change of the guide’s characteristics along its propagation direction The portions of the energy content of the propagating wave transmitted and reflected depend on the impedance variation due to the discontinuity [10] Due to such behaviour, the presence of hidden flaw can be investigated thanks to a transmission factor (TF) parameter Furthermore, the wave velocity can be affected by damage if a dispersive behaviour is tuned, resulting in a different arrival time The time of flight (ToF) can be addressed considering respectively the characteristic time of the detected and excited waves respectively In this work both TF and ToF are computed by means of Short time Fourier transform (STFT) of the digitized signals However, when damage occurs, the signal acquired behind the damage could show a completely different waveform respect to that acquired without damage When a multimodal wave is excited, a mode conversion may occur due to discontinuity interaction resulting in a scattering of the incident guided mode into a finite number of propagating modes and infinite non propagating evanescent modes [11] To completely account the changes in the entire signal, it should be preferable to analyse the entire signal over the time rather than select only a mode or a specific frequency content (i.e carrier frequency from the STFT analysis) To this end, the generalized energy content of the voltage signal can be simply obtained by time integration of its squared amplitude over the signal duration Furthermore, the Pearson correlation coefficient (PCC) ρ is used to directly assess differences between baseline and current signal in the time domain In this case the damage index is evaluated following Eq rather than Eq DI 1 U (3) About the effect of DI on several points of structural mesh, three different distance indices are considered First of all, a linear decreasing probability accounting the distance between paths and structural nodes is computed to assess the dI value Then an elliptical reconstruction is operated considering the distances between the node and both sensors respectively Finally, a modified linear reconstruction is operated accounting the distance between structural node and the path or between the node and the nearest sensor depending upon the position of node The distance index is then evaluated following the Eq (4) where δ(P) is the above mentioned distance and β is the parameter defining the farther influence of the path and its value is usually calibrated on known condition to improve the reconstruction dI E G ( P) , ° ® E 1 °¯ 0, if G ( P) d E if G ( P) t E (4) Obviously each reconstruction has its own capability to assess the position of damage respect to such characteristic inestigated and with several combination of DI and dI many probabilistic reconstructions can be invetigated, like briefly introduced in the next section with few results Numerical investigation The proposed method based on an algorithm capable to elaborate simultaneously several wave features collected by the transducers array and to reconstruct the damage with several techniques is investigated here as a first step towards a more robust and reliable way to identify and locate impact induced damages within composites A simulation based framework is modeled in ABAQUS ® environment following the multi-side validation proposed in [5] in order to idealize a mm thickness composite plate, which has been designed for a lower wing 52 V Memmolo et al / Procedia Engineering 167 (2016) 48 – 55 panel and characterized at a first stage with a cluster of PZT disks [7] In fact, the group velocity of A0 Lamb wave propagating mode used for damage detection purpose is compared with experimental results providing a minimal lack of agreement, actually lower than 5% A 60 KHz frequency is used to interrogate the structure with a low leakage sine-cycle signal opportunely windowed In detail, such excitation frequency is exploited for damage detection purpose due to the optimum balance between noise to signal ratio (tuning) and time detached representation (A0-S0 overlap) of A0 mode According to the multi-side validation, the damage is then idealized using a material degradation approach due the conformation of an impact induced damage trough the thickness Varying the principal damage parameters (size, depth, location, degradation level), their effect on the reconstruction technique and adopted metric are evaluated at a first stage One of the most relevant result regards the position of damage affecting the localization rather than the identification procedure In detail, from several cases investigated it can be point out that the detectability increases when the damage approaches the sensors cluster The Fig (a) shows indeed the effectiveness of reconstruction obtained using elliptical reconstruction with PCC based DI formulation The red region, which represents the likely damaged area assessed, fully detects the degraded region (20x20 mm2 around black cross) and the center of gravity (CG) estimating the most probable location of damage is very close to the center of the degraded region (in view of measurements, CG and impact are respectively estimated impact and real impact location) Furthermore, when the damage is close to the array edges (one of the more challenging position), such reconstruction is able to detect and localize the damage using only few useful path as shown in Fig (b) However, from few reconstruction images not reported here for sake of conciseness, a decreasing level of accuracy can be highlighted when the damage is simulated in the inner sides of the enclosed area (a) (b) Fig Damage reconstruction based on PCC damage index and nonlinear distance index Damage size 20x20 mm2 Distance between damage position and centre of sensors cluster: 100 mm (a) and 150mm (b) Another interesting and reasonable result regards the size of simulated damage From other reconstruction images not depicted here, it can be point out that increasing the in plane size of damage the detectability increases until a specific size is overtaken Such dimension, depending on the parameter and the decreasing probability chosen as well as the position of damage, allows a perfect localization of damage as reported in Fig (a) The same qualitative result can be observed when the depth of damage (depth of degraded region) changes However, the effect of increasing depth can be observed only when a minimum in plane detectable size is considered All results obtained are in line with the Probability of Detection (POD) paradigms [6] demonstrating at a first stage that the distance between impact and estimated impact is a reasonable system response for a reliability approach about damage localization 53 V Memmolo et al / Procedia Engineering 167 (2016) 48 – 55 Other results, not reported here for sake of conciseness, confirm that every parameter is affected by its own sensitivity to the specific damage considered, like addressed in the literature [7, 9] In particular, the dispersive behavior of selected propagating mode affects the localization of damage using ToF parametric representation, as already demonstrated in [7, 9] Such effects, related to the parametric extraction of information from propagating signals, impact the identification firs of all (varying the reconstruction approach respect to [7], the same result is here achieved) as well as the localization of damage Summarizing such investigation, it is possible to point out similar conclusions carried out in [7] from measurements and simulations as well In addition, the investigation on the feasibility of PCC based DI formulation leads to interesting perspective appearing as the most reliable parameter for effective diagnostic However, accounting the time series’ changes correlating two signals, such formulation could be affected by noise level and measurement uncertainties and requires a more detailed investigation As a matter of fact, the optimization of the reconstruction cannot be oriented to find the most suitable approach for the specific case otherwise the methodology effectiveness could be strictly related to such conditions In view of a reliability assessment of the system, several parameters can be weighted and combined in order to reduce wrong effects due to the presence of a parameter which may not be effective Confirmation of such aspect can be addressed looking to the ToF based reconstruction depicted in Fig (a), which shows the lack of agreement between damage position and its estimated location However, the reconstruction obtained with data fusion of ToF and PCC based approach is actually promising, because the wrong effect of the ToF based reconstruction is smoothed by using a multi-parameter approach as depicted in Fig (b) (a) (b) Fig Damage reconstruction based on ToF damage index (a) and ToF-PCC combined damage index (b) with nonlinear reconstruction Damage size 20x20 mm2 Distance between damage position and centre of sensors cluster: 100 mm In the previous pictures are reported only few results obtained with nonlinear reconstruction approach for sake of conciseness However, that obtained using both linear approaches are very close and all considerations made up can be confirmed Only one aspect, which will be addressed with future investigation, can be point out when the linear decreasing probability is used Maintaining information about the projection of the direct line of sight beyond the enclosed area, a damage appearing outside could be detected as well A similar result cannot be approached with elliptical and modified linear reconstruction proposed here Finally, a brief comparison between results obtained with the present methodology and that using the multiparameter approach proposed in [7] is reported in Fig The former approach (a) based on a mesh defined reconstruction is able to identify and localize the damage without uncertainties The same conclusion can be drawn from the latter reconstruction (b) based on a meshless approach Furthermore, the effectiveness of such results can be improved using a multi-parameter approach Although both algorithms lead to a reasonable and similar 54 V Memmolo et al / Procedia Engineering 167 (2016) 48 – 55 diagnostic when damage is inside the enclosed area, the time effort is much more different When all parameters are already extracted, the time required by the meshless approach is markedly lower Such aspect, not emphasized here, could be quite crucial when a wide number of components must be investigated On the other hand, the mesh based approach presented here is capable to identify as well as localize damages very close to the sensors array (Fig (b)) and the possibility to identify anomalies outside the enclosed area is a future perspective using a linear decreasing probability For such reasons, it is really challenging to define at a first stage the more effective approach, but a reliability assessment is needed to analyze the performances of the system (a) (b) Fig Multi-parameter damage reconstruction obtained with mesh based approach (a) and meshless approach [7] (b) Damage size 20x20 mm2 Besides the mixed experimental-numerical approach needed for a well-defined and rigorous procedure for the estimation of such performances [5], a fast management tool is necessary for every damage assessment Thus, a prototype solution is briefly introduced here In order to simplify the management of wide range of settings, options and methods with which the user interfaces, an interactive platform has been developed within MATLAB® Fig GUI Analysis Panel: data and analysis selection, method panel and plot of probabilistic reconstruction V Memmolo et al / Procedia Engineering 167 (2016) 48 – 55 environment The graphic user interface (GUI) is capable to simultaneously analyze several wave features affect by damage and various probabilistic algorithm based on linear and nonlinear reconstruction techniques The designed main panel is depicted in Fig and it is divided in three specific sections The former allows to define all inputs and settings, including an external window to interactively exclude sensors for a robustness analysis of the system Selecting data files, the software is able to elaborate a database with all features extracted by current and baseline configuration useful for data post-processing and damage reconstruction The second section provides the definition of the reconstruction method selecting from the horizontal bar the damage index (single or multiparameter approach), decreasing probability and feature to be considered and/or combined Finally, the central window provides the resulting tomographic image (colormap, contour of isolines or surface of probabilistic function depending upon the visualization mode selected) and the comparison between impact position (if known) and its estimation Concluding Remarks A brief overview of damage diagnostic capabilities of a probabilistic reconstruction method is presented in this work A numerical investigation is carried out in order to validate Structural Health Monitoring systems including technologies and methodologies able to detect and localize barely visible damages due to impact loads on aircraft composite panels The analysis framework includes algorithm capable to extract few metrics from ultrasonic waves propagating in interrogation mode The investigation plan is mostly oriented to detect all parameters affecting such procedures in terms of effective and efficient monitoring in view of the system target definition and assessment Some interesting dependencies are carried out in Section making the present work mostly a first step towards a reliable and quantified SHM system Such methods are indeed able to provide an effective monitoring but are really sensitive to few damage and reconstruction parameters requiring a detail design of experiments and/or simulations to correctly account the wide range of affecting parameters Future researches are mostly oriented to a certification of an efficient multi-parameter SHM system and the relative procedures In view of such perspective, the current work is mainly focused on evaluation of system reliability (Probability of Detection) of SHM systems employing experimental and numerical methodologies (Model Assisted Probability of Detection) which need to account all parameters affecting detection and localization Such quantification will provide a certain metric to evaluate the performance of several techniques available as well as a reliable relation between identification and damage dimension providing a comprehensive diagnostic References [1] B.Y Cohen, Emerging NDT technologies and challenges at the beginning of the third millennium, ndt.net 5(1) (2000) [2] V Memmolo et al., Damage detection tomography based on guided waves in composite structures using a distributed sensor network, Optical Engineering 55(1), 011007 (2016) [3] Y Su et al., Guided Lamb waves for identification of damage in composite structures: a review, Journal of Sound and vibration 295(3-5) 753– 780 (2007) [4] V.Memmolo et al., Experimental Characterization of a Composite Structures Health Monitoring Methodology, Proc of 3rd IEEE International Workshop on Metrology for Aerospace (2015) [5] V Memmolo et al., Model assisted probability of detection for a guided waves based SHM technique, Proc SPIE 9805, 980504 (2016) [6] MIL-HDBK-1823A, Non destructive evaluation system reliability assessment, Department of Defense Handbook, USA (2009) [7] V Memmolo et al., A user interface damage detection technique based on a multi-parameter methodology, Procedia Engineering (2016) [8] X Zhao et al., Ultrasonic Lamb wave tomography in structural health monitoring.” Smart Material and Structures, 20 105002 (2011) [9] V Memmolo e al., Statistical recognition of structural health in composites, ndt.net 21(7), (2016) [10] H Samajder et al., Lamb waves in a Honeycomb Composite Sandwich Plate, Proc SPIE 8695, 869505 (2013) [11] B A Auld, Acoustic Fields and Waves in Solids, R.E Krieger, Malabar, Florida (1990) [12] Monaco et al., Detecting delaminations and disbondings on full-scale wing composite panel by guided waves based SHM system, Proc SPIE 9805, 98050V (2016) 55 ... and baseline configuration useful for data post-processing and damage reconstruction The second section provides the definition of the reconstruction method selecting from the horizontal bar... integrating SHM within structural design, a relax of allowable stresses and design constraints could be achieved with a condition based maintenance avoiding a safe life approach In view of online monitoring. .. capabilities of a probabilistic reconstruction method is presented in this work A numerical investigation is carried out in order to validate Structural Health Monitoring systems including technologies