Application of Structural Health Monitoring over a Critical Helicopter Fuselage Component  

 

C. SBARUFATTI1, A. MANES1, M. GIGLIO1,  U. MARIANI2, R. MOLINARO2, W. MATTA3, I. DI LUZIO3, D. TOSCANI4, F. ARCHETTI4, L. BJERKAN5, R. HUDEC6, V. WIESER6, P. MAKYS6, J. GOZDECKI7, C. KATSIKEROS8 

1 Politecnico di Milano, Dipartimento di Meccanica, Via  la Masa 1, 20156 Milano, Italy 2 AgustaWestland, Via Giovanni Agusta 520, 21017 Samarate (VA), Italy 

3 Vitrociset S.p.A., Via Tiburtina 1020, 00156 Roma, Italy 4 Consorzio Milano Ricerche, Via Cicognara 7, 20129 Milano, Italy 

5 SINTEF, Box 124 Blindern, NO‐0314, Oslo, Norway 6 UNIVERSITY of ZILINA, Faculty of Electrical Engineering, Univerzitna 8215/1, 010 26 Zilina SK‐Slovakia 

7 AGH University of Science and Technology, Department of Telecommunications, al. Mickiewicza 30 30‐059, Krakow, Poland 

8 Laboratory of Technology & Strength of Materials, Department of Mechanical and Aeronautics Engineering, University of Patras, Panepistimioupolis Rion, 26500 Patras, Greece 

 

Contact e‐mail address: claudio.sbarufatti@mail.polimi.it 

  

ABSTRACT

The helicopter design is a challenging experience for fatigue concern as it is subjected to a very wide range of low- and high-frequency load cycles per flight, very much more than a fixed wing aircraft. Moreover, thinking of the various and harsh environments where the helicopter could operate, also corrosion and low velocity impacts could generate further crack nucleation and propagation into the fuselage. Health and Usage Monitoring Systems (HUMS) has received considerable attention from the helicopter community in recent years with the declared aim to increase flight safety and mission reliability, extend duration of life-limited components and of course reduce inspection and maintenance activities. In particular, Structural Health Monitoring (SHM) seems capable to help in reducing the maintenance and operational costs, which is about 25 per cent of the direct operating cost of the helicopter, thus playing an important role especially in the case of the ageing helicopters. In fact, the damage tolerant design approach makes the fatigue resistance evaluation not only a safety issue but also a maintenance related concern. In effect, thanks to the continuous evaluation of the current structural health of the helicopter through a SHM system, it could be possible to set a Condition Based Maintenance, which means substituting a component according to its current structural condition instead of relying just on the design assumptions. The approach could bring to a maximization of both the machine availability and reliability, thus conjugating safety with economy. Strictly connected to a damage tolerant design, a sensor network is thus needed in order to monitor the structural health of the machine and the recent improvements in non-destructive techniques for crack detection are making the concept more affordable from both the technological and the economical points of view. The aim of the present work is to explain a novel method to apply the SHM concepts on a critical zone of a helicopter fuselage, passing through the creation of a complete FE Model of the fuselage, either in healthy and damaged situations, and considering the different stress distributions caused by a progressive crack in the most critical areas. This would represent the key step for the extraction of information from the sensor data, thus allowing to distinguish between the undamaged and damaged structures. The helicopter tail structure is presented herein as a good candidate for the application and testing of the SHM system. The main reason is the criticality of the region, where the torque generated by tail rotor to balance the rotation induced by the main rotor is undergone. In particular, the attention will be focused on some simplified reinforced panels, well suited to indicate the general behaviour of the entire structure and particularly adapt for the safe and early application on board of the machine.

1. Preface The work presented in this article is part of a wider project named HECTOR (HElicopter fuselage Crack moniToring and prognosis through On-board sensoR network), coordinated by the European Defense Agency (EDA) and funded by 10 EDA Member States ( Cyprus, France, Germany, Greece, Hungary, Italy, Poland, Slovakia, Slovenia, Spain) and Norway in the framework of the Joint Investment Program on Innovative Concepts and Emerging Technologies (JIP-ICET). The project consortium, coordinated by Politecnico di Milano, comprehends universities (Politecnico di Milano, University of Zilina, AGH University of Science and Technology), non-governmental laboratories (Laboratory of Technology & Strength of Materials – University of Patras) and research entities (Consorzio Milano Ricerche), as well as industrial partners (AgustaWestland, Vitrociset) and small-medium enterprises (Stiftelsen SINTEF), thus representing five important European Countries that are regarded as leader for the project purposes (Greece, Italy, Norway, Poland, Slovakia). 2. Introduction to condition based maintenance for helicopters Nowadays, the structural safety of helicopters is guaranteed with a deep fatigue analysis in the design phase [1,2,3] and a clear schedule of inspection during life. However the design and maintenance of helicopters are particularly important and complex compared to the aircrafts. The peculiarity is manifested in two ways: from one hand, the load spectrum, which is composed by a high number of low-amplitude cycles resulting from the mechanical rotation of the rotor blades (severe vibratory loads); from the other hand, the low velocity impact damage, coming from the harsh environments where these machines have to operate. This type of loads can lead to high fatigue damage accrued in short time or rapid crack propagation from accidental flaws or damages. For that reason, up to now the condition based maintenance has not been the most typical approach for helicopter fatigue tackling, since it frequently requires short inspection intervals. The FAR (Federal Aviation Regulations) standard indeed points out how the determination of the real operative usage is a fundamental issue, as for the design of every aircraft, but even more critical in the case of helicopters. Actually, the adoption of redundancy and low stress level is recommended but cannot always be adopted. However, because of the difficulties in predicting loads and accidental failures, a reliable method to detect any damage and, at the same time, to predict the residual life of a component, a part or the entire machine would be very helpful to prevent failures and increase machine availability and reliability [4]. Considering these issues the development of Health and Usage Monitoring Systems (HUMS) has received considerable attention from the helicopter community in recent years [5,6] with the declared aim to increase the flight safety and the mission reliability, to extend the duration of life limited components and of course to reduce the maintenance costs [7]. Structural Health Monitoring (SHM) seems capable to reduce the maintenance cost, which is about 25% of the direct operating cost of the helicopter [5] and plays an important role especially in the case of ageing helicopters. The idea is to update the scheduled maintenance intervals according to the actual condition of the structures. However this is not an easy task, as it is governed and influenced by many variables, each one characterized by a stochastic distribution. In particular, the key factor is the disposal of a detection and monitoring system as reliable as possible, on the basis of which all the machine stops can be optimized in order to maximize the machine availability with the minimum loss of reliability, thus conjugating safety with economics.

3. HECTOR: the progress beyond the state of the art The actual state of the art is mainly represented only by devices that monitor vibrations generated in components critical for the flight performance. The purpose of HECTOR is the creation of a demonstrator for helicopters which permits to perform the Embedded Structural Health Monitoring and the on-line evaluation of the degradation mechanism, based on advanced damage criteria. A complete knowledge of the stress and strain state of the fuselage (advanced FE model) and the data collected by a low power sensor network positioned in the most critical areas, should act simultaneously for the evaluation of the structural health. This package could thus be mounted on new helicopters to define a more reliable, efficient and economic program of inspections, or on ageing airframes to perform a life extension of the machines. Traditionally, cracks are monitored by conducting visual and non-destructive testing on a large area of the aircraft during the operating life. Often this can require significant downtime to get access to such an area with a handheld testing machine. The advantage of HECTOR is that once the sensors network is installed and set up, inspection is possible not only without disassembling parts but also in continuum using a real time strategy to detect failures (existence, location, type, extent) and for the prognosis of their effect on the overall reliability of the structure. The application of low-power sensors will bring these benefits with low costs in terms of the overall energy consumption of the helicopter. The base of this advanced prognostic program is the availability of finite element models (FEM) of the structure with and without damage. This is the key step for the extraction of information from sensor data, i.e. the identification of the damage-sensitive properties, derived from the measured dynamic response, which allows to distinguish between the undamaged and damaged structures. Different types of sensors will be evaluated during the project, selecting the most suitable to perform a structural monitoring for the helicopter fuselage. The goal would be to make non-destructive testing technology an integral part of the aircraft structure itself (Embedded Structural Health Monitoring – ESHM). In case of damage, the system directly identifies the location and suggests the actions that can be taken. The ESHM approach on helicopter structure can have a strong impact as a means of possibly revolutionizing the current structural maintenance and the design process. Thus the application of ESHM on helicopter fuselage will develop the following objectives, in which an innovation in methods, systems and/or hardware is considered:

• the creation of a complete FE model of the fuselage, both in healthy and damaged situation, considering the different stress state caused by a progressive crack in the most critical areas;

• the definition of an integrated network of different sensor types (optical fiber sensors, strain gauges, piezoelectric sensors, crack gages, comparative vacuum monitoring, etc.) chosen on the basis of their suitability for crack detection and compatibility with the helicopter instrumentation;

• the construction of an advanced communication system, choosing the most reliable between the wireless and traditional ones;

• the utilization of an automated advanced signal processing with filtering techniques applied so that the signal resulting from damage can be clearly identified from any other noisy signals being around.

4. The Finite Element Model The Helicopter Rear Fuselage (see Figure 1) is presented as the best candidate for the application and testing of the SHM technologies, mainly because of the following reasons:

• Highly loaded structure, with stresses coming from the reaction to the torque induced by the main and tail rotors;

• High criticality of failure; • Typical aeronautical Semi-Monocoque structure consisting of a skin stiffened with frames

and stringers; • Damage tolerant design approach; • Part which is difficult to access internally but available externally for visual or conventional

inspection; • Deep presence of riveted connections, which act as crack nucleation points.

Figure 1: Semi-Monocoque structure for the rear fuselage A reliable and accurate FE model of the helicopter fuselage has been realized with ABAQUS, with the aim to individuate the most important zones for the sensor positions and to perform a prognostic model about residual life evaluation. For this purpose, the critical zones will be defined on the basis of the most stressed zones considering also the presence of natural crack arresters or accelerators which respectively can provoke an acceptable growth of the crack or generating a worsening of the structural integrity. Figure 2a shows calculated maximum principal strain distributions across the entire rear fuselage. Under a typical mission load spectrum, the extent of microstrains ranges from around zero to about 1500 , which can be well recorded with a sensor network, for instance including Fiber Bragg Gratings, for which the strain resolution is down to a few microstrains, however depending on the interrogating equipment. This example shows that, even when looking at the entire rear fuselage, some areas can be highlighted where the maximum stresses and strains are concentrated, thus where the sensor network should be installed. A submodel (Figure 2b) should thus be created in order to appreciate more precisely the strain distribution in those zones as well as to allow for a very fine mesh for the crack modelling (Figure 2c) without a significant worsening of the model computational complexity. Comparing the healthy and cracked FEM results, the main objective should be the creation of a data matrix that contains the damage effect accumulated in the helicopter fuselage in terms of crack length and position, allowing for the best interpretation of the information obtained from the sensor network; this matrix of influence will thus be treated from a statistical point of view, integrated to the advanced algorithms for data processing and feature extraction to obtain a reliable evaluation of the crack length and position (detection and monitoring) and residual life (prognosis) starting from a sensor network data storage. In [17] a feed forward Artificial Neural Network has been used to

recognise some damage configurations, focusing on Multiple Site Damage (MSD), in riveted lap joints. A FEM based training of the network was implemented, proving a good performance in the test phase in terms of achieved resolution for damage identification.

Figure 2: Calculated strain distribution across an entire fuselage from finite element methods

a) General model of the entire rear fuselage b) Submodel for the most stressed zone, with crack modelled inside c) Crack model.

The strain difference between the presence of a crack and the healthy state is shown in Figure 3 along the crack length direction (vertical direction in Figure 2) for two crack lengths (55mm and 10mm) and along two paths, one (blue line) passing straight through the crack and the other (red line) along a parallel line with a 16cm offset from the damage. For the 55 mm crack the strain difference is significant and easily measurable through the crack, but along the 16 cm offset path the strain difference is much smaller, but still measurable in a somewhat smaller range. For the 10 mm crack the measurable range is sensibly narrowed for the path through the crack, while at the 16 cm offset path it becomes immeasurable. From these examples it is obvious that in order to determine the optimal sensor placement which requires the minimum acquisition points, a reference damage level (crack length) should be reasonably defined.

Figure 3: Strain differences with presence of a crack vs. no crack along the vertical direction of the fuselage obtained from finite element calculations. Blue curve: Through the centre of

the crack; Red curve: Offset 16 cm from the crack.

5. The Sensor Network definition The concept of Structural Health Monitoring (SHM) involves in general the observation of a system over time using periodically sampled dynamic response measurements from an array of sensors, the extraction of damage-sensitive features from these measurements, and the statistical analysis of these features to determine the current state of the system’s health. For long-term SHM, the output of this process is periodically updated information regarding the ability of the structure to perform its intended function in light of the inevitable aging and degradation resulting from the operational environments. In the most general terms, structural damage can be defined as changes introduced into a system that adversely affects its performance. Implicit in this definition is the concept that damage is not meaningful without a comparison between two different states of the system, one of which is assumed to represent the initial, and often undamaged, state. A crack that forms in a mechanical part produces a change in geometry that alters the stiffness characteristics of that part. Depending on the size and location of the crack and the loads applied to the system, the adverse effects of this damage can be either immediate or may take some time before they alter the system’s performance. The basic premise of most damage detection methods is that damage modifies the stiffness, mass, or energy dissipation properties of a system, which in turn alter the measured dynamic response of the system. Although the basis for damage detection appears intuitive, its actual application poses many significant technical challenges. The crucial challenge is the fact that damage is typically a local phenomenon and may not significantly influence the global response of a structure that is normally measured during vibration tests. Another fundamental challenge is that in many situations damage detection must be performed in an unsupervised learning mode, thus implying that data from damaged systems are not available for all the possible faults. Environmental and operational variations, such as varying temperature, moisture, and loading conditions affecting the dynamic response of the structures again will complicate the interpretation of possible damages. The ideal sensor for the purpose of damage detection should have the following properties:

• Be sensitive to the measured property • Be insensitive to any other property • Does not influence the measured property • Be linear over the expected range of the measured property

There are several available sensing principles for mechanical sensing. These range from the well-established strain gauges to innovative and so far unproven technologies like for example sensing devices based on nano-materials. Based on available knowledge, literature and information from vendors the following sensor candidates had been evaluated in this study: strain gauges, crack gauges, eddy current sensors, fiber-optic sensors, potential drop, MEMS devices and mechanical wave-based sensors including general piezoelectric sensor systems, ultrasonic, Lamb waves and acoustic emission. Two among the most innovative and relevant for the current purposes are described hereafter. The goal of an extended sensor system should be to provide an early warning system that material fatigue is under development. The focus is thus on strain sensors since formation of cracks usually follows from excess strains. While there are several methods for measuring strains, the most common is with a strain gauge, a device whose electrical resistance varies in proportion to the amount of strain in the device. The most widely used gauge is the bonded metallic strain gauge. A system based on fiber Bragg gratings (FBGs) is again a strain sensing system which can be made capable of displaying a strain map of the structure. These sensors, easily and economically embeddable into the laminates without significantly affecting the mechanical properties of the hosting material, have several advantages such as: light weight, low power consumption (less than

1W is required to power the filter and the SLED optical source), immunity to electromagnetic interference, long lifetimes and high sensitivity. They don’t need initial and in-service calibrations and are affected by very low signal drop. From an economic point of view, because of their diffusion and industrialization, also the costs are by far reduced. Last but not least, comes the multiplexing option, or the possibility to photowrite more FBGs in series inside one optical fiber, thus becoming particularly attractive for damage location identification through the analysis of the whole FBG network reflected spectrum. The continuous improvements concerning the operating range of such sensors [8], would make possible their utilization also in the harsh environments typical of the helicopters. It is so possible to correlate the effects of various damages (mainly cracks and delaminations) on the reflection spectrum of the embedded FBG sensor [9]. However, there are some challenges that must be addressed in order to construct a system based on FBG sensors from a practical point of view. First of all, an optical fiber for sensor purpose in which a Bragg grating is inscribed consists of the circular glass fiber surrounded by a thin and hard protective coating like for instance polyimide. When this is integrated with a surface which is flat (or at least nearly so) with an adhesive like for example epoxy, the contact area is very small and the gluing process is critical in order to secure a proper strain transfer. Variations in fixing the sensor to the surface can easily lead to variations in sensor responses and difficulties in interpretation of the measured results. For a practical application, a repeatable and reproducible attachment of the optical fiber is needed so that mounting related effects are reduced to a minimum. A practical method for use in aeronautical applications has been addressed in [10-11]. The installation procedure is claimed to be easy, fast and reliable and happens by means of a specially designed mounting tool called a sensor pad. It is used in combination with a UV-curable adhesive. The pad allows easy installation of the optical fiber and provides for easy removal of the pad after mounting of the fiber to the structure surface. This is claimed to be useful for mounting fibers for use as strain gauges when the fiber is in direct contact with the surface to be monitored and preferably without interfering contact to other rigid material which may exert additional strain to the fiber. Finally, this system cannot detect cracks directly (unless a fiber breaks due to a crack), but crack formations are usually the result of build-up of excessive strains and may provide an early warning system that a crack may form in some part of the structure. From this point of view, comes out the importance for the introduction of a Finite Element Model able to predict the extent of the strain while different damages are occurring over a loaded structure. The same model can be used in the learning phase to calibrate the detection algorithms allowing for a crack localization as well as damage extent estimation.

The SMART Layer® developed by Acellent Technologies, Inc. is also presented as a valid alternative for monitoring the structural integrity of composite and metal structures. It consists of an array of networked piezoelectric sensors embedded in a thin dielectric film, eliminating the need for each sensor to be installed individually. A pre-defined diagnostic signal can be transmitted by one of these sensors. It then travels through the structure under investigation as surface acoustic waves and is picked up by the neighbouring sensors, see Figure 4. Each sensor can function both as an actuator and as a detector, creating a multitude of actuator-sensor pairs. By looking at the modulation of the transmitted signal, information about the structural health of the object can be extracted. Information about parameters such as loading, delamination, crack initiation and growth as well as corrosion can be deduced. The SMART Layer technology has been tested in monitoring the health and condition of diverse structures ranging from aircraft and rotorcraft to pipelines, bridges, wind turbines, automobiles etc. This technology seems to be a very promising candidate for Helicopter SHM, but further tests are needed in order to prove the long term stability and airworthiness of the sensors.

Sensor

Sensor Sensor

Scattered waves

Actuator

Figure 4: The actuation and sensing principle of the SMART Layer®. Each node can function both as an actuator and as a detector, creating a multitude of actuator-sensor pairs.

6. The communication system The main objective of the current SHM system is to develop the hardware and software solutions for the real-time monitoring of the structural integrity of helicopter fuselage. In general, the overall concept of the system shall consist of two main hardware parts, namely:

• Helicopter side o Sensor communication network placed in the helicopter’s fuselage. o Main control unit including the communication link HW between the helicopter and

the ground operator’s tablet. • Service side

o Serviceman’s tablet or other kinds of handheld devices. And two main software parts:

• GUI (Graphic User Interface) at Helicopter side. o Signal pre-processing, enhancement and feature extraction block. o Simple crack prognosis block.

• GUI at the service side. o Feature extraction and signal classification block. o Detailed crack prognosis block. o Database block.

The communication system should be capable to deliver measured data from the sensors to the access points for further processing. The problem connected with communication systems should be classified as follows:

• Which technology (wired or wireless) will be convenient for data transport? Wired technology seems to be a good choice, but there are problems to be solved (e.g. reliability of physical connection among sensors, lots of wires, etc.). Besides, there is a possibility to enhance the connectivity reliability by using both technologies (wired – main, wireless – secondary).

• There are problems in the wireless solution to be solved: o Radio wave propagation in closed metal environment (signal attenuation, time

dispersion of the signal).

o Signal frequency selection (to avoid mutual interference to/from other communication systems).

o Communication technology selection (wireless standard for indoor and outdoor communication).

o Network topology selection, etc.

HELICOPTER SIDE SERVICE SIDE

STATUS OVERVIEW

CRACK PROGNOSISsection F – unable to fly section J – crack to 30 minutes

PILOT’S TOUCH LCD PANEL – main screen

SECTION F DETAILS“sensor description” – status“sensor description” – status “sensor description” – status“sensor description” – status“sensor description” – status“sensor description” – status ...PILOT’S TOUCH LCD PANEL – section screen

PILOT’S TOUCH LCD PANEL

SERVICE MAN’S TABLET

IEEE 802.15.1

DATABASE

A – SECTION SENSOR NETWORK

B – SECTION SENSOR NETWORK

IEEE 802.15.4

SENSOR Ax

PRIMARY SENSOR μPC

SENSOR A1

PRIMARY SENSOR μPC

INDUSTRIAL PC

SENSOR Bx

PRIMARY SENSOR μPC

SENSOR B1

PRIMARY SENSOR μPC

CAN BUS

X – SECTION SENSOR NETWORK

μPC

SENSOR Xx

PRIMARY SENSOR

SENSOR X1

PRIMARY SENSOR

STATUS HISTORY

SERVICE MAN’S TABLET – main menu

TCP/IP-------------------

HTTP

Figure 5: The principal block diagram of communication concept.

In Figure 5 the communication technology and the standards selection for communication infrastructure are pointed out. There are two main technologies to be used for indoor communication – CAN bus for wired solution and IEEE 802.15.4 (ZigBee) for wireless communication. To deliver collected data to service side, the IEEE 802.15.1 (Bluetooth), IEEE 802.15.4 (ZigBee) or IEEE 802.11a,g,b (WiFi) are taken as possible candidates. Problems connected with radio wave propagation in closed metal environment is due to the almost ideal wave reflection from a conductive plate. The result is the signal propagation only through holes or apertures in metal plates (Figure 6).  

Figure 6: Signal propagation through aperture in a conductive material.

Signal propagation in such an environment may be roughly described through the following formula:

( ) ( ) ww LnddLdL ⋅+⋅⋅+= )log(10 00 α

where L(d) is the signal attenuation (in dB) in distance d, L(d0) is signal attenuation in a reference distance (1 m), α0 is the path-loss exponent, nw represents the number of obstacles, Lw is the obstacle attenuation. This expression doesn’t take into account the signal reflection from walls. For precise evaluation of signal attenuation the ray tracing method is commonly used [12], which was also used for the example simulation (Fig. 6). From Figure 6 (red ball is transmitter, red squares and green square indicate receivers) it is obvious, that the signal transmitted by transmitter is able to propagate outside the cube only following one path through the aperture (blue line), while other paths (yellow and green lines) are blocked. Signal attenuation becomes high outside the cube (changing squares colour from red to green). In conclusion, to deliver radio signals in metal conducting materials with barriers is possible only by convenient arrangement of transmitters and receivers. Indoor wireless communication system will be based on IEEE 802.15.4 standard. Two different device types can participate in an IEEE 802.15.4 network; a full-function device (FFD) and a reduced-function device (RFD). The FFD can operate in three modes serving as a personal area network (PAN) coordinator, a coordinator, or a device. An FFD can talk to RFDs or other FFDs, while an RFD can talk only to an FFD. An RFD is intended for applications that are extremely simple, such as a light switch or a passive infrared sensor; they do not have the need to send large amounts of data and may only associate with a single FFD at a time. Consequently, the RFD can be implemented using minimal resources and memory capacity. Depending on the application requirements, an IEEE 802.15.4 network may operate in either two topologies: the star topology or the peer-to-peer topology. Both are shown in Figure 7.

Figure 7: Basic topologies in IEEE 802.15.4 network In the star topology the communication is established between some devices and a single central controller, called the PAN coordinator. A device typically has some associated application and is either the initiation point or the termination point for network communications. A PAN coordinator may also have a specific application, but it can be used to initiate, terminate, or route communication around the network. The PAN coordinator is the primary controller of the PAN. All devices operating on a network of either topology shall have unique 64- bit addresses. The PAN coordinator might often be mains powered, while the devices will most likely be battery powered.

Applications that benefit from a star topology include home automation, personal computer (PC) peripherals, toys and games, and personal health care. The peer-to-peer topology also has a PAN coordinator; however, it differs from the star topology in that any device may communicate with any other device as long as they are in range of one another. Peer-to-peer topology allows more complex network formations to be implemented, such as mesh networking topology. Applications such as industrial control and monitoring, wireless sensor networks, asset and inventory tracking, intelligent agriculture, and security would benefit from such a network topology. A peer-to-peer network can be ad hoc, self-organizing, and self-healing. It may also allow multiple hops to route messages from any device to any other device on the network.

a) c)

b)

Figure 8: Sensor network topology scenarios: Star (a), Clustered star (b), Clustered star bus (c)

Up to now some proposals and simulations were prepared with IEEE 802.15.4 (ZigBee) standard as solution for indoor wireless propagation. Several topologies for wireless networks have been compared – Star, Clustered star and Clustered star bus (Figure 8), discarding Peer-to-Peer topology because of problems with radio signal transmission through apertures in structures. Instead, star topology was simulated with hybrid link connection – wired solution for connection among coordinators and PAN coordinator and wireless connection for individual networks belonged to coordinators. The division of the whole network to several sub-networks helped to avoid problems with propagation of radio signal through metal apertures in indoor environment. 7. Algorithms for crack detection and residual life prognosis One of the key features of the proposed SHM system is the possibility to automatically detect cracks during the helicopter flight and perform a prognosis on the future health status. The signal recorded by the different sensors during the flight can be easily stored in low cost, low weight solid state memories, which are less sensible to vibrations than magnetic hard drives. These are traditionally analysed offline by service men, which can run complex analysis that can involve data collected in different flights. As an additional feature, HECTOR leverages the availability of storage memory, together with the presence on-board of low weight computational devices, to run on line data analysis algorithms for crack detection and prognosis. Such fast and computationally not expensive prediction can be shown through the pilot’s console. The pilot’s attentions should be

directed to the cracked sector through, for instance, a “red blinking” message; the alarm can also indicate which sensor(s) produced the alarm and eventually the kind of cracks. The design of the user interactions has to consider as primary constraint the complex tasks that the pilot has to continuously perform; thus, the interface has not to disturb the pilot’s concentration during the flight and it should be as simple as a panel which shows messages like:

• Critical crack – “immediate landing”, red signal. • Potentially critical crack – “time prediction of crack progress – time estimation to safety

landing”, orange signal. • Crack not affecting the helicopter’s flight – “service man attention to the crack”, blue signal. • Sector without crack – “all is OK”, green signal.

The software architecture shown in figure 5 depicts the high level view of the system components:

• A MySQL database, running on the on board memory and synchronized with the Service man’s Tablet, to store sensor data.

• A JAVA infrastructure for data integration and querying, to provide an abstract layer for easy management of heterogeneous sensor data.

• A set of software components for on line data analysis, which run on the on board computer. • A set of software components, deployed on the service man tablet, to show the life cycle of

the helicopter’s components. • A JAVA GUI (graphical user interface), embedded in the pilot’s console, to show alarms,

and information messages. • A JAVA GUI running in the service tablet to show statistics and forecasts on structure

health. The development of software components for online and offline analysis has to face several complex objectives, including:

• the development of algorithms for suppression of noisy signals at the sensors side. • The application of prognostics algorithms to make a basic prediction on the health situation

at the monitoring locations. • The development of a software architecture and a framework for the integration of sensors

data and algorithms for diagnosis and prognosis. • The development of the main control unit.

The data collection, data analysis and data exchange between on board devices and off board maintenance units is supported by the software architecture for integration of sensors data and algorithms. Its objective is to create a data access layer on top of the database. It is deployed in an industrial PC, since data access and analysis has computational needs which cannot be satisfied by the small CPUs embedded in sensors. The proposed architecture allows to directly integrate the data analysis algorithms in this infrastructure, to run queries on sensor data following the approach of [13][14]. The main features provided are, from one hand, the explicit representation of domain concept (sensor data), to support easy comparison and analysis; from the other hand, the data analysis algorithms (soft sensors) provide the possibility to formulate queries over multiple sensors. Once the data are collected and made accessible to the main control unit, they can be elaborated. The basic elaboration is through algorithms for suppression of noisy signals, for signal enhancement and filtration, de-noising and feature extraction for correct signal interpretation and evaluation. The second step is the application of prognostics algorithms to make a basic prediction on the helicopter state. An innovative idea of HECTOR is to provide online data analysis to perform basic prediction and determine if it is possible to continue in flying ; such analysis, which is actually performed only offline, during maintenance phase, and using statistics on vehicle usage, can leverage the availability of on board sensors and improve the safety of the operations.

Several techniques can be applied for this purpose; a common requirement is the need of sensor data collected during a high number of tests, both simulated or during flight, to understand the behaviour of structures and set up/tune the models for damage (crack) diagnosis and prognosis. The basic approach consists in applying heuristics and ad hoc rules to combine multi-sensor data (e.g., obtaining an improved estimate of a physical phenomenon via redundant observations), thus setting a basic knowledge for SHM purposes. This is best performed through human supervision; however an alternative approach is to learn rules from examples. It requires a common representation language to model all the system’s entities, as well as a relevant human expertise and domain knowledge to define rules for the integration and for the calibration of inference algorithms, adapting them to specific sensors and application cases. Furthermore, human evaluation is also applied to evaluate the performances of the rules, which makes this approach expensive in terms of development and maintenance cost. Automatic methods, like Kalman filters, can be applied to the automatic processing of sequences of data readings (e.g. vibrations, stress...) using multi-sensor data to determine faults in the structure. Kalman filters (and in general state-vector fusion methods) are effective when the underlying processes are linear: the data should be first analyzed by a human expert, to evaluate the applicability of such a model. Kalman filtering is essentially data driven and requires training data for the creation of the model of the observed phenomena. These data must have statistical relevance and hence be in a quantity sufficient for achieving acceptable performance in recognizing the classes of interest, defined by the user. If one class has not enough samples, the algorithm will not be able to distinguish it. Data must be provided in structured format, pre-processed and aggregated in order to be machine-readable (data should be stored e.g. in relational databases, csv tables...). The algorithms are not able by themselves to deal with technological heterogeneity of the data collection devices; data collection and feature selection procedures must then be performed by the data integration infrastructure. Other automatic techniques that can be applied are the supervised classifiers for pattern recognition like, for instance, neural networks and statistical classifiers. While there are numerous techniques available, the ultimate success of these methods depends on the ability to select good features, i.e. those which provide good class recognition in the feature space. Neural networks are the most applied supervised classifier, whose major advantage over Bayesian methods is the capability to perform data fusion processing without the need for a priori information. Neural networks always require a “learning” period in order to fully establish and test the specific patterns or rules that will guide the system. During this process, the network must be run so that each neuron can be “taught” the proper association between diverse data inputs and assimilated output. This knowledge can be obtained through the annotations produced by a human expert, which has also to define relevant features or by running automatic tests for feature extraction and feature selection. Then, training data for model learning must be labelled, i.e. a set of observed situations for which a human expert has provided an identifier (e.g. the adjustments needed by rotor blades as consequence of vibrations). A step further in the Artificial Neural Network performance could be achieved by means of a training phase based upon the use of Finite Element Models as well as experimental data. High level data analysis can be performed also through Bayesian methods [15], used to generate a probabilistic model of uncertain systems by consolidating and interpreting data provided by several sensors. The advantage of the Bayesian methods is the possibility to model hidden variables, i.e. perform inference over sensor data and health conditions that are not directly observable. One of the innovative approaches proposed in HECTOR could be the application of Finite Element Models (sect. 4) to refine the probabilities that link the model’s variables and are estimated from the data. Markov Chain Monte Carlo sampling is also applied to perform inference over the status of a structure and to determine anomalous situations, following the approach of [16].

8. Conclusions The SHM is the next future key factor of all the cutting edge structures, in particular concerning helicopters and (going ahead) tilt rotors. The raw instruments (sensors, algorithms and stress analysis) are quite ready for this task, as evaluated and described in the present document. They form three basic elements for the SHM that should be integrated into one expert system. The sensor network, together with data acquisition, processing and fusion units, are key elements to obtain information about the current status of the system. Numerical models based on the FE analysis represent a powerful tool to develop virtual images of the real system. They are currently used to study various scenarios that might occur under different conditions. Moreover, the numerical models show the possibility to be used in-situ to identify a degradation mechanism based on the current status of the system as well as its history, to evaluate the impact of the degradation on the overall structure and, consequently, to precisely predict its future development. The results, processed by the adaptive prognostic assessment, are the basis for a maintenance advisory generation. This process is partially applied and in progress only for typical aerospace applications, like fixed wing aircrafts. The next step is therefore to improve analytical and laboratory based studies to set the available technology for helicopter applications, considering the main characteristics (high frequency vibratory loads, different load manoeuvres, missing of pressurization in the fuselage, etc.) of this very complex machine. Bibliography [1] Lazzeri L, Mariani U, “Application of Damage Tolerance principles to the design of helicopters”, Int J Fatigue,

2008, in press.

[2] Paul W. Tan, Catherine A. Bigelow and John G. Bakuckas, Jr. “Widespread Fatigue Damage Assessment Approach” presented at the RTO AVT Specialists’ Meeting on “Life Management Techniques for Ageing Air Vehicles”, held in Manchester, United Kingdom, 8-11 October 2001

[3] Abubaker Ahmed, John G. Bakuckas, Jr, Jonathan Awerbuch, Alan C. Lau, and Tein-Min Tan, “Fatigue Testing of a Stiffened Lap Joint Curved Fuselage Structure”, 46th AIAA/ASME/ASCE/AHS/ASC Structures, Structural Dynamics & Materials Conference 18 - 21 April 2005, Austin, Texas

[4] Thomas E. Munns and Renee M. Kent, “Structural Health monitoring degradation mechanism and system requirements”, 19th proc. of Digital Avionics Systems Conf. DASC, 2000, Philadelphia, USA

[5] P.M.Pawar and R. Ganguli, “Helicopter rotor health monitoring – a review” Proc. IMechE Vol. 221 Part G: J. Aerospace Engineering, 2007

[6] “A Review of Structural Health Monitoring Literature: 1996-2001”, Los Alamos National Laboratory Report, LA-13976-MS, 2003

[7] C. Boller, M. Buderath, “Fatigue in aero structures—where structural health monitoring can contribute to complex subject”, Phil. Trans. R. Soc. A, (2007) 365, 561–587

[8] http://www.micronoptics.com/

[9] Yashiro S, Takeda N, Okabe T, Sekine H, “A new ap. to pr. Mult. Dam. states in comp. Lam. with emb. FBG sens.” Composites Science and Technology 65 (2005) 659–667

[10] Van Roosbroeck J., Jacobs E., Voet M. & Vlekken J.: “Installation and Test Procedures of Optical Strain Gauges for Aeronautical Applications”, 20th Int. Conf. on Optical Fibre Sensors, Proc. of SPIE Vol. 7503, 75037Q1-4, (2009)

[11] Vlekken J, Voet M. & van Roosbroeck J.: ”Method and means for monitoring a Bragg grating fibre on a surface”, Int. Patent Application, WO2009/106576 A1

[12] Rappaport, T.S.: Wireless Communications. Principles and Practice. Prentice Hall, New Jersey, USA, 1996

[13] D. Toscani, L. Quarenghi, F. Bargna, F. Archetti, E. Messina, "A DSS for Assessing the Impact of Environmental Quality on Emergency Hospital Admissions", In proceedings of the WHCM 2010 - IEEE Workshop on Health Care Management, February 18-20, 2010 - Venice, Italy.

[14] D. Toscani, I. Giordani, F. Bargna, L. Quarenghi, F. Archetti. A software System for Data Integration and Decision Support for Evaluation of Air Pollution Health Impact. In Proc. ICEIS 2010 - 12th International Conference on Enterprise Information Systems. Funchal, Madeira – Portugal, 2010

[15] Biao Chen and Pramod K. Varshney. A Bayesian Sampling Approach to Decision Fusion Using Hierarchical Models. IEEE TRANSACTIONS ON SIGNAL PROCESSING, VOL. 50, NO. 8, AUGUST 2002

[16] F. Archetti, E. Messina, D. Toscani and M. Frigerio - IKNOS – Inference and Knowledge in Networks Of Sensors. International Journal of Sensor Networks (IJSNet), Vol.8 No. 3, 2010

[17] Ch.E. Katsikeros and G.N. Labeas, Development and validation of a strain-based Structural Health Monitoring system, Mechanical Systems and Signal Processing, Volume 23, Issue 2, Feb 2009