300 232 Structural integrity of cold spray repaired aerospace components Werner T1 , Madia M1 , Hilgenberg K1 , Klassen T2 , Gärtner F2 1Bundesanstalt für Materialforschung und -prüfung (BAM), 2Helmut-Schmidt-Universität / Universität der Bundeswehr Hamburg Aerospace components are safety critical parts, which are often replaced when found defective after the fabrication process or after periodic inspection in-service. Given that the production of highly-stressed components in the aerospace industry is usually cost and resource intensive, the introduction of alternartive technologies as local repair of defects could lead to large cost savings. Furthermore, repairing represents a resource-efficient and green process in a world targeting at decarbonization of many industrial sectors. In view of that, a promising process is cold spraying, in which metal particles are accelerated to supersonic speed in a gas stream and applied to a surface. High, local plastic deformation removes oxide layers and creates a strong bond between the particle and the surface. This allows new material to be applied layer by layer. For component repair, a defect is first removed by machining, then the removed material is reapplied near-net-shape by means of cold spraying. Eventually, the original component geometry is mechanically restored. The advantage of the cold spraying lies in the low heat input. Conversely, if the process parameters are not optimized, defects, micro-cracks and residual stresses may be generated in the repaired material and at the interface to the substrate, which would lead to an undesidered degradation of the material properties. A further important aspect affecting the mechanical performance of cold spray repaired parts is the adhesion of the repair material to the substrate. This work aims at presenting recent results obtained in the frame of the cluster project CORE (ComputerRefurbishment), which involves diverse number of industrial partners, government agencies, and academic institutions in Germany. The project is devoted to implementing a process chain for the repair procedure in an automated and computer-controlled manner. The investigations are carried out on high-strength aluminum alloys typically used in structural applications in the aerospace industry. The fatigue properties are crucial to the safe application of repaired components. Given the presence of process-induced defects, a damage tolerant framework is peculiarly suitable for judging the possible degradation of fatigue properties in repaired commponents and give an indication of the permissible defects to optimize the parameters of the cold spray process. Therefore, several fatigue and fatigue crack propagation tests have been planned to the aim of comparing the fatigue performance of base and repaired materials. The mechanical tests are corroborated by fractographic and microstructural investigations. Beside pure mechanical cyclic loading, corrosion is expected to contribute to the material degradation. In view of this, slow strain rate tests (SSRT) are conducted
301 233 On the strength of the piezoceramic transducer in the system of structural health monitoring. Pavelko I1 1Riga Technical College The piezoceramic transducer (PZT) is perspective for use in systems of structural health monitoring (SHM) of bearing components of airframe for detecting fatigue cracks and prevention of destruction in operation. As principal element of SHM system the PZT should be embedded in the structure and in this state is subjected by load which is proportional to operational load of monitored component. The long-time operational loading can cause the fatigue destruction piezoceramics and fault of the SHM (Fig.1). Therefore, for successful monitoring mission the strength and the fatigue lifetime of PZT must be provided. In this paper the brief overview problem is done. The analytical model of strength for piezoceramics stripe glued to structural element was modified. The main efforts are focused to improving of the so-called prestressed PZT in which the residual compressive stress is created. The FE analysis of stress state and strength of piezoceramic transducer at static loading is performed. The measurements of electromechanical impedance of the prestressed PZT installed to Al plate at static and cyclic tests show increasing the strength and some decreasing of sensitivity. [1] I.Pavelko. Research on the Protection of Piezoceramics Transducers from the Destruction by Mechanical Loading. Int.Virtual J. for Science, Technics and Innovations for the Industry, Issue 7 (2010), 17-21. The PZT with nine fatigue cracks (detected by penetration) without destruction of host structure[1]
302 234 Universality in magnetically detected residual stresses in steels and method to determine the actual stress level distribution Stamou G1 , Pattakos P1 , Angelopoulos S1 , Ktena A2 , Hristoforou E1 1National Technical University Of Athens, 2National and Kapodistrian University of Athens In this paper, the universality law in the dependence of residual stresses on magnetic properties is presented. Therefore, the method to determine the actual residual stress level distribution dependence on magnetic properties for an unknown type of steel is also presented. The method is based on the determination of the localized stress components in 2 dimensions (surface residual stresses) and in 3 dimensions (bulk residual stresses) and the consequent determination of magnetic properties such as magnetic permeability at the same areas or volumes. Residual stresses are generated by autogenous welding, using TIG, Plasma and RF induction welding process. Thus, the generated residual stresses in the heat affect zones and the fusion zone are in a full agreement with magnetic permeability measurements. Therefore, the Magnetic Stress Calibration (MASC) curve for each type of steel can be determined, thus permitting the residual stress tensor distribution monitoring in steels during production or manufacturing or in periodic tests of steel structures. These MASC curves are different for different types of steel. However, it has been found that the normalization of the stress and permeability axes with the corresponding yield point and the maximum amplitude of permeability respectively, results in collapsing all MASCs in one sigmoid response. Due to this observation, the detection of stresses in an unknown type of steel can be realized by determining the stress-strain curve of this unknown type of steel. During measurement, magnetic permeability is detected in parallel. Thus, the yield point and the maximum amount of permeability is detected.
303 235 On the Barkhausen noise in naval steels Pattakos P1 , Stamou G1 , Angelopoulos S1 , Ktena A2 , Hristoforou E1 1National Technical University Of Athens, 2National and Kapodistrian University of Athens In this paper we depict results on Barkhausen Noise (BHN) measurements in naval steels. BHN can be used to determine anisotropy effects, mean grain size and residual stresses in steels. A multi-pole BHN sensor has been developed to monitor the small Barkhausen Jumps inside the grains, employing minor M-H loops, in order to avoid excitation above the anisotropy field of the steel. Thus, a precise detection of the position and size of the stressed areas inside the steel grains is detected. Furthermore, grain size and anisotropy of grains (either shape or crystalline anisotropy) are determined by means of rotating the BHN sensor on top of the steel. Results illustrate a good agreement with structural characterization. The main handicaps of the method are due to the ultra-high sensitivity of the BHN method, as well as the ability to monitor such noise only on the surface and subsurface of the steel under test.
304 236 AMR sensor for Steel Health Monitoring Stamou G1 , Angelopoulos S1 , Pattakos P1 , Ktena A2 , Hristoforou E1 1National Technical University Of Athens, 2National and Kapodistrian University of Athens Electromagnetic sensors are widely used for safety and maintenance inspection purposes of ferromagnetic materials used in aerospace, construction and transport industry. In this paper, a portable device based on an Anisotropic Magnetoresistance (AMR) sensor for Steel Health Monitoring is presented. The principle of operation of this device is the detection of magnetic anomalies of ferromagnetic materials, such as steel, due to stresses, corrosion, or other kind of defections. For this purpose, a cost-effective, low-consumption and highsensitivity sensor has been developed, consisting of a sensing element, a microcontroller and the required electronics. The sensing element consists of a high-sensitivity, fastresponse AMR sensor, which is placed above the material under test, giving the ability of contactless measurement of the magnetic flux density on 3 axes. Then, a microcontroller with the appropriate electronic components receives the acquired data, enabling their realtime analysis and display of the measurements through a personal computer, smartphone, etc. Moreover, a 3D-printed enclosure has been designed and constructed, which both protects the sensor and maintains the desired position and alignment with the samples. The device can be used to determine the location of the defective spots or zones of the material under inspection, facilitating its treatment through rehabilitation methods that can be applied on the affected area. It can be used for both single-point and scanning mode monitoring, making it suitable for many testing scenarios. Furthermore, its ease of use and portability render it accessible to a wide range of users and appropriate for on-site measurements. The initial measurements using the device on test samples, demonstrate its reliability and accuracy in steel health monitoring.
305 237 Hall Sensors For Steel Health Monitoring Pattakos P1 , Angelopoulos S1 , Stamou G1 , Ktena A2 , Hristoforou E1 1National Technical University Of Athens, 2National and Kapodistrian University of Athens Steel and other ferromagnetic materials are used in a wide range of applications of vital importance. Thus, frequent monitoring of their condition through Non-Destructive Testing techniques is imperative. Steel health monitoring includes a variety of techniques, which can be used to detect the condition of the material and the location of a possible defect. In this work, a portable device is presented, which can be used for Non-destructive testing of ferromagnetic materials, based on Hall effect sensors. More specifically, the device consists of a yoke, formed of a permanent magnet, and two high permeability soft ferromagnetic poles. The yoke is placed above the material under test, forming a magnetic circuit. Hall sensors are placed at the ends of the yoke’s poles, measuring the variations of the detected magnetic flux density, generated due to changes of the magnetic permeability of the material, indicating structural imperfections, stresses, or defects. The device also includes a microcontroller, as well as the appropriate electronics to acquire and process the data. A custom 3d-printed enclosure ensures the optimal lift-off distance between the sensors and the material under test, while maintaining the desired placement of the device above the material. The collected data are transmitted either wired or wirelessly and can be displayed to a receiving device, such as a personal computer or a smartphone.
306 238 Fluxgates for Steel Health Monitoring Stamou G1 , Angelopoulos S1 , Priftis P1 , Ktena A2 , Hristoforou E1 1National Technical University Of Athens, 2National and Kapodistrian University of Athens Magnetic sensors have significant applications in the field of Steel Health Monitoring. Steel structures, used in aerospace, construction and shipbuilding industry can experience fatigue, corrosion, stresses and other forms of damage, that can compromise their structural integrity. By monitoring the health of these structures using magnetic sensors, it is possible to detect and address these issues, preventing severe damage and catastrophic failures. One approach to detect these cracks, dots and impurities can be implemented through the measurement of magnetic flux leakage on the surface of the ferromagnetic material under test, using a calibrated fluxgate sensor. Therefore, a fluxgate magnetometer, designed and developed in our laboratory, has been used, offering high sensitivity and low power consumption, while being small and low-cost. The fluxgate sensor used for this work is based on an appropriate coil topology and the use of a magnetic core made of an amorphous magnetic material. The device includes the above-mentioned sensing element, the required electronics and a microcontroller to generate the excitation signal, as well as receive and analyze the output of the pickup coil. The overall device is portable and can be used for onsite measurements without the need of any equipment other than a personal computer or a smartphone, to display the output of the sensor. Finally, the packaging of the sensor consists of a 3D-printed enclosure, to protect the sensor and ensure its optimal placement.
307 239 Energy Harvester for Remote Sensing Pattakos P1 , Katsoulas A1 , Angelopoulos S1 , Stamou G1 , Ktena A2 , Hristoforou E1 1National Technical University Of Athens, 2National and Kapodistrian University of Athens Modern electronics are commonly optimized for low-energy applications. The integration of energy harvesting systems into such systems can lead to the development of autonomous sensors that can be placed on the field and operate without any external energy source. Energy harvesters are based on various renewable sources, such as light, heat and mechanical or magnetic forces. In this work, an energy harvesting system based on magnetism and the utilization of vibrations, is presented. More specifically, the system consists of coils having movable permanent magnets as a core. The arrangement exploits the vibrations that may exist in a suitable environment, such as a ship or a manufacturing facility, acting as an electromagnetic generator. Thus, the design of the topology can be adjusted to the specific characteristics of each application. Furthermore, suitable electronic circuits have been designed and developed to receive the generated AC voltage, transform it to DC and use supercapacitors as an energy storage. The initial testing has shown that the energy harvesting system was able to supply a sensor and a microcontroller, in order to conduct autonomous measurements. As a result, it can be used in combination with sensors to monitor the condition of structures that are difficult to reach.
308 240 Acoustic Emission Analysis on Mechanical Properties and Damage Evolution of Multiscale Kevlar/Glass Hybrid 3D Orthogonal Woven Composites under Flexural Loading Al-Nadhari A1 , Senol H1 , Ulus H2,3, Topal S1,2, Yildiz M1,2 1 Faculty of Engineering and Natural Sciences, Sabanci University, 2 Sabanci University Integrated Manufacturing Technologies Research and Application Center & Composite Technologies Centre of Excellence, Teknopark Istanbul, 3Huğlu Vocational School, Selcuk University Three-dimensional (3-D) woven fabrics have attracted a lot of attention due to their delamination resistance, great tailorability, and low manufacturing time. In this study, Acoustic Emission (AE) is used to investigate the effect of different fiber hybridization types of 3-D orthogonal fabrics on the flexural properties and damage evolution along the longitudinal and transverse direction of the composite plate. Three different fabric configurations, namely Baseline, Inter-ply and Intra-tow, are studied. All the fabric configurations have the same fabric architecture consisting of six warp and seven weft layers. The warps of all the fabric types are Kevlar whereas the z-binders consist of ultra-high molecular weight Polyethylene fibers. The wefts of the baseline fabric are all Kevlar while Inter-ply fabric has three Kevlar layers flanked by four glass layers. The Intra-tow fabric has three Kevlar layers in the middle with four layers of hybrid glass/Kevlar tows at the top and bottom surfaces. In-situ AE analysis is carried out to understand how damage starts and progresses in the three composite types. AE results are then verified using scanning electron microscopy. Schematic representation of 3D orthogonal woven fabric configurations
309 241 A vibration-based machine learning type Structural Health Monitoring methodology for populations of composite aerostructures under uncertainty Saramantas I1 , Spiliotopoulos P1 , Fera F1 , Bourdalos D1 , Sakellariou J1 , Fassois S1 , Ofir Y2 , Kressel I2 , Tur M3 , Spandonidis C4 1 Stochastic Mechanical Systems And Automation Laboratory, University of Patras, 2Advanced Structural Technologies, Dep. 4445, Israel Aerospace Industries (IAI), Engineering Center, Ben-Gurion International Airport, 70 100, 3 School of Electrical Engineering, Tel Aviv University, Ramat Aviv 699 78, 4Prisma Electronics S.A., 45 Agias Kyriakis str., P. Faliro, 175 64 A robust to uncertainty Machine Learning (ML) methodology for population based Structural Health Monitoring (SHM) in composite aerostructures is postulated. As is well known, SHM methods, which may be highly effective for a single structure operating under constant conditions, may exhibit significantly degraded performance when employed for population based SHM under varying environmental and operating conditions (EOCs). A practically important case relates to structures used in a fleet of flying aircraft. This is due to various types of uncertainty sources that may arise from the varying EOCs, as well as differences among the population of nominally identical structures, often due to assembly, variations in the geometry and materials, and so forth. These uncertainties may significantly affect the structural dynamics, thus leading to either high false alarm rates or to undetected earlystage damages. This problem gets further aggravated for composite aerostructures due to the additional uncertainty in the materials and manufacturing, as well as the sensitivity of their properties to temperature and other conditions. The population based SHM methodology presently postulated is founded upon unsupervised ML type approaches for damage detection and a supervised approach for damage characterization. Damage detection is based on two types of healthy subspace approximations: A Multiple Model Representation (MMR) and a varying radii Hyper-Sphere (HS) type approximation. Both are built upon response-only vibration acceleration or strain measurements of properly selected sensor locations on the composite aerostructure. Based on them, Multiple Input Single Output Transmittance Function AutoRegressive with eXogenous excitation (MISO-TF-ARX) data driven models representing the underlying structural dynamics are estimated. Decision making is then based on the model parameter vector, transformed and reduced via Principal Component Analysis (PCA). Final damage detection is subsequently founded upon information fusion using both acceleration and strain signals and the two healthy subspace approximations. Damage characterization, which refers to damage type, location, and level determination, is achieved in a somewhat similar manner but within a cosine similarity context. The SHM methodology is assessed through numerous numerical Monte Carlo simulations and laboratory experiments. For the numerical Monte Carlo simulations, a population of nominally identical composite beams (Figure 1) is considered based on an Abaqus-based Finite Element Model (FEM). Manufacturing uncertainty is introduced via ±10% variations on the nominal material properties. Uncertainty due to temperature variability within the [- 55...71] ℃ range, and variability due to the force excitation profiles (simulating varying flying conditions) are also considered. The investigated types of damage correspond to early stage debonding and delamination at different locations and of different levels (severities), modeled as stiffness degradation.
310 The experimental assessment includes a population of small-scale composite coupons (Figure 2). The detection and characterization of delamination and impact-induced damage of different levels, under material/manufacturing, temperature, and mounting uncertainty, are investigated. The methodology’s damage detection performance is assessed via Receiver Operating Characteristics (ROC) curves (via correct detection versus false alarms rates), while damage characterization is demonstrated via confusion matrices representing classification accuracy. The obtained results, both numerical and experimental, are very promising, indicating very good overall damage diagnosis. Schematic representation of the composite aerostructure employed in the numerical Monte Carlo simulations (1, 2, 3: sensor locations; E1, E2: excitation locations). Top-down view of the experimental setup (1, 2, 3: sensor locations; E1, E2: excitation locations).
311 242 Development and experimental validation of a Machine Learning based SHM prototype system for composite aerostructures Spiliotopoulos P1 , Fera F1 , Papadopoulos P2 , Giannopoulos F2 , Spandonidis C2 , Tur M3 , Ofir Y4 , Kressel I4 , Saramantas I1 , Sakellariou J1 , Fassois S1 1 Stochastic Mechanical Systems and Automation Laboratory, University Of Patras, 2Prisma Electronics S.A., 45 Agias Kyriakis str., P. Faliro, 175 64, 3 School of Electrical Engineering, Tel Aviv University, Ramat Aviv 69978, 4Advanced Structural Technologies, Dep. 4445, Israel Aerospace Industries (IAI), Engineering Center, Ben-Gurion International Airport, 70 100 The development and experimental validation of a Machine Learning (ML) and vibration based Structural Health Monitoring (SHM) prototype for composite aerostructures under uncertainty is presented. The system is developed under the auspices of the Israel-Greece bilateral collaboration project REALISM. Following initial training (baseline phase) it is intended to be autonomous, operating using a limited number of acceleration and strain measurements from properly selected sensors. SHM is achieved via a methodology utilizing two innovative ML type robust to uncertainty schemes, which employ Multiple Model Representations (MMRs) and varying radii Hyper-Spheres (HSs) schemes, for approximating the healthy structural dynamics under uncertainty. For both schemes, Multiple Input Single Output Transmittance Function AutoRegressive with eXogenous excitation (MISO-TF-ARX) models are employed with their Principal Component Analysis (PCA) based transformed and reduced parameter vectors selected as the sensitive to damage feature. The two schemes are embedded into the system and operate in parallel using acceleration and strain signals, while a decision level fusion is deployed for final decision-making enhancement and false alarm reduction, thus aiming at optimizing SHM performance. The system has been designed in a very compact form for use in small UAVs. It consists of six 24-bit input channels with Integrated Electronics Piezo-Electric (IEPE) signal conditioning and is connected to a minicomputer through USB. Its backend control is based on a .NET Framework 6 operating in C Sharp, while the frontend is a user-friendly Graphical User Interface (GUI) made in JavaScript requiring limited user expertise for the initial set-up (learning phase). The set-up includes options for fundamental signal processing, frequency domain analysis and various other features; it allows for learning options and updating whenever needed, and is also fully upgradable according to developing needs. The system may operate continuously or periodically in time in an automated way, acquiring batches of acceleration and strain responses for performing SHM. The developed prototype system is validated via a series of experiments with full-scale composite aerostructures from actual UAVs under healthy and damaged conditions and in the presence of uncertainty that may be due to variability in materials, manufacturing, mounting, and force excitation profile. The validation results obtained are very encouraging, suggesting that the current progress in hardware and software, along with corresponding progress in robust ML-based SHM algorithms, allow for the development of compact, lowcost, but high-performance SHM units. Due to its capability of coping with various types of uncertainty, the developed unit constitutes a viable option for effective, vibration-based, real time, under-uncertainty SHM for composite aerostructures, such as those employed in current UAVs.
312 Views of the SHM prototype system: Hardware (left) and part of the user interface (right).
313 243 Inverse finite element analysis for delamination detection in composite structures subjected to forced vibration Ganjdoust F1 , Kefal A1 , Tessler A2 1 Sabanci University, 2NASA Langley Research Center The inverse finite element method (iFEM) is a non-destructive structural health monitoring (SHM) technology that operates through processing a set of strain data collected using a network of in-situ sensors and does not require any a priori information regarding the material model of the structure or the loading condition that the structure experiences. Over the past years, this approach has become popular as a damage detection method. This study aims to exploit the damage detection capabilities of the iFEM for the identification of failure in structures operating under complex loading conditions. The damage detection methodology used in this study is developed by adopting the kinematic relations of the refined zigzag theory (RZT) within the iFEM framework. Next, damage identifiers based on the equivalent strain measures are defined, which are used to locate the position and extent of the damage in composite structures. This strategy has been implemented successfully for the detection of delamination damage in composite materials under static loads previously. In this study, the proposed toolbox is utilized to detect delamination damage when the composite structure is experiencing complex loading conditions, such as dynamic or vibrational loads. In this context, two numerical examples have been presented to demonstrate the performance and efficiency of the present approach. In both examples, the failure is modeled as a region with degraded stiffness over the problem domain. In other words, the damage is pre-defined, and the generation and nucleation of the defect are not natural. Using the proposed damage detection approach, in-plane and through-thethickness position and configuration of the structural damage in composite structures undergoing dynamic/vibrational loads are captured and quantified. In addition, it is shown that the present methodology works effectively despite the sparse collection of strain data from the structure. Ultimately, the robustness and superior capabilities of the iFEM as a damage detection approach are demonstrated.
314 244 RANDOM VIBRATION-BASED PROGRESSIVE FATIGUE DAMAGE MONITORING OF THERMOPLASTIC COUPONS UNDER POPULATION AND OPERATIONAL UNCERTAINTY Tserpes K1 , Tsivouraki N1 , Fassois S1 1University Of Patras In the last decade, thermoplastics are increasingly replacing thermosets in aerospace applications mainly due to their weldability and recyclability. Modern aerospace structures are designed using the damage tolerance philosophy. This presupposes very good knowledge of the fatigue behavior of structural materials and the development of reliable fatigue damage diagnosis methods. Yet, as of this date, fatigue behavior of thermoplastics has not been thoroughly studied. The same is true for the use of random vibration-based methods for automated and potentially in-flight condition monitoring and prognosis during the structure’s lifecycle. Random vibration-based technology is in this context of particular interest as it may be based on naturally available, or artificially enhanced, vibration response signals for continuously monitoring developing fatigue damage in-flight, without interrupting regular service. Although work some work in this context has been carried out for composite structures, no studies are reported specifically for thermoplastics. Based on the above, the present work aims at progressive fatigue damage monitoring in thermoplastics using the resulting random vibration signals. A very important and innovative aspect of the work is that it is not based on ideal nominal conditions, but, instead, fully accounts for practically inevitable uncertainty. This uncertainty is of two kinds: Interstructural (population) and operational (for instance excitation profile uncertainty). The experimental process, which integrates mechanical fatigue tests, C-scan tests, and random vibration tests, is applied to a population sample of 12 nominally identical thermoplastic coupons under interrupted fatigue and in the presence of significant population and operational uncertainty. Ten loading states are created by performing fatigue tests at intervals of 10,000 cycles, based on results from preliminary static and fatigue tests. In each interruption, ultrasonic C-Scan tests are conducted to visualize the fatigue damage, which mainly is of the matrix cracking and delamination types, and to correlate the damage states with the number of cycles. After the C-scan tests, the specimens are subjected to acoustically generated white random noise excitation under free-free boundary conditions. A total of 5,040 random vibration tests are conducted, with the laser vibrometer measured random vibration signals undergoing preliminary processing, nonparametric Welch-based spectral estimation, and parametric AR modelling. Fatigue condition monitoring for the available population sample of 12 coupons is based on a single-vibration-response Unsupervised Multiple Model AutoRegressive (U-MM-AR) method. This effectively accounts for the population and operational uncertainty present and utilizes the fatigue damage effects on the AR model parameters as a damage sensitive feature. The results demonstrate very good damage detection performance, achieving a correct damage detection rate of over 95% under population and experimental uncertainty. Further efforts are directed towards the precise characterization of the level of damage through the measured random vibration signals. The available results suggest that the method has high potential for fully automated, in-operation fatigue monitoring and remaining useful life prediction for thermoplastic structures without interrupting regular service and, most important, under population and operational uncertainty.
315 245 Condition Monitoring Framework for Damage Identification in CFRP Rotating Shafts using Model-Driven Machine Learning Techniques Karyofyllas G1 , Koutsoupakis J1 , Seventekidis P1 , Giagopoulos D1 1Aristotle University of Thessaloniki The advancements in key industry sectors such as automotive, wind turbines, oil, gas etc., have led to an increased usage of Carbon-Fiber Reinforced Polymer (CFRP) filament wound tubes with metal connectors, for transmission shafts applications. Monitoring the health conditions of these parts can prevent failures and arrange for effective maintenance schedules, to meet the current industrial requirements regarding availability, reliability, and maintainability. Real-time condition monitoring (CM) of such systems via vibration measurements allows for detection of faults in them and facilitates their predictive maintenance. The accuracy and robustness of a CM application depends among others on the availability of data for different health states which typically requires complete experimental measurements. In this work, a novel CM framework for damage detection and identification is presented and applied to a lab-scale CFRP transmission shaft, using Convolutional Neural Networks (CNNs) trained by data generated through numerical simulations of an Optimal Rotordynamics (RD) model of the physical system. First, the RD model corresponding to the real shaft structure, including support bearings, is developed and optimized using a small number of initial healthy state measurements throughout the operating speed range. The goal is to perform damage identification on different health states by exploiting simulated instead or experimental responses for supervised health state classification. The damage scenarios considered concern cracks and delamination of the composite material at specific locations on the shaft. Damaged states of the system are then simulated as a local reduction of stiffness on the optimal RD model, allowing for generation of labeled simulated training datasets for the CNNs. Uncertainty is introduced to the models by sampling their key parameters from a Gaussian distribution, thus considering the real system’s inherent uncertainty. The trained CNNs’ robustness and accuracy are then validated by accurately classifying faults on the physical system, proving the proposed damage detection method’s generalization capabilities and highlighting its potential.
316 246 An SHM architecture for indirect load estimation in wind turbine rotor blades through strain sensing Liangou T1 , Zilakos I2 , Anyfantis K1 1NTUA, National Technical University of Athens, 2 Light Structures Objective: The current industry practice on the inspection, monitoring and maintenance actions for wind turbines blades is based on preventive maintenance schemes. Potential damages and their corresponding locations (hot spots) are nominally identified during the design phase. Having available a theoretical load spectrum, designers estimate the inspection intervals and blade’s life based on these nominal hot spots. A major field of research and application in Structural Health Monitoring is related to load identification, where we may solicit for methods that may be used for recognizing the exact loads applied to a blade during its lifetime and reduce hence the uncertainty associated with damage accumulation indices. The problem of load identification is an inverse problem per se, since we can seldom find cases where the load may be directly measured. Indirect load identification, has always been problematic due to the ill-conditioning exhibited in the inverse problem. Within the inverse estimation of an ill-conditioned problem, a small error or perturbation in the input data can easily scale up in a significant error or perturbation in the load that is of estimation interest. The objective of this work is to demonstrate a sensor placement design process according to which the variance in the load estimates remains the minimum possible and alleviating hence the ill-conditioning problem. Methods: We employ methods sourcing from statistical Design of Experiments and in particular the D-optimal criterion as the means for arriving at the optimal sensor placement. A typical 7 m long wind turbine blade geometry has been selected for demonstrative purposes. The problem involves a massive design space, i.e. numerous potential locations (see Figure 1) and hence combinations for sensor placement and as such an automated selection process based on Genetic Algorithms is employed. Sensor topologies based on engineering judgment are considered as well for demonstration purposes. A pool of synthetic data has been constructed through Finite Element simulations. The pressure distributions were considered as lumped equivalent beam loads applied at several stations length-wise and are the quantities of estimation interest. Results: A limit state FE analysis (design loads), reveals that the aspect ratio of the maximum tip deflection of 110 mm compared to the blade’s length (~7m) suggests that linearity in the response holds (see Figure 2). The problem is framed in a static load identification setting where inertia effects are considered negligible. Conclusion It is concluded that the D-optimal criterion used as the fitness function within the GA optimization process arrives at an optimal sensor grid that promises the maximum possible variance reduction.
317 Figure 1 Finite Element mesh of the considered wind turbine rotor blade. Each element corresponds to a potential sensor placement location. Figure 2 Deflected shape at the limit state condition.
318 Under the auspices of With the support of 3