Skip to main content

Neural Net based defect detection system using LRU technology for aircraft structure Monitoring

Final Report Summary - SELF-SCAN (Neural net based defect detection system using LRU technology for aircraft structure monitoring)

Executive summary:

The periodic maintenance and inspection of aircraft is both costly and time-consuming. The critical structural components of an aircraft need careful examination as small cracks can appear in large and complex structures. In some cases, the defects are hidden beneath layers of other attachments, so accessing these components involves dismantling the assembly and long downtime hours for airline operators.

Guided wave technology is a promising technique for structural monitoring. It provides large area coverage from a limited number of sensors, combined with potentially high defect detection sensitivity. However, due to the complex nature of ultrasonic guided waves, the often complex geometry of the components requiring monitoring and the variable environmental conditions they exist in, the interpretation of the captured signals can be very challenging. Within SELF-SCAN, it has been shown that such complex signals can be used as input for a neural network system to facilitate in situ defect detection. The main goal of the project was to develop a guided-wave and neural networks based monitoring system for critical aircraft components using permanently installed sensors to allow the testing of critical and inaccessible areas without the need to dismantle during every inspection. The detectability of defects is increased when historical data is used to identify changes over time.

The aim for the legacy of the project is to enable a fundamental realignment of inspection / maintenance strategies, which can then be based on the actual momentary condition of the aircraft structure. The capability of the developed system has been demonstrated in laboratory tests on representative aircraft components.

SELF-SCAN is a collaboration between the following organisations from 6 different European Union (EU) countries: TWI LIMITED, Przedsiebiorstwo Badawczo-Produkcyjne Optel sp.Z o.o. Phillips Consultants, Isotest Engineering SRL, Smart Material GMBH, Cereteth, NDT Expert.

Project context and objectives:

The main aim of the project was to develop a monitoring system for critical aircraft components using permanently installed sensors to allow the testing of critical and inaccessible areas without the need to dismantle during every inspection using guided wave and neural networks. The capability of the developed system has been demonstrated in laboratory tests on representative aircraft components.

To accomplish the project objectives, the work activities have been organised into a number of discrete work packages (WPs). Those were divided into research (WPs 1 to 6), demonstration (WP7), dissemination and exploitation (WP8) and management (WP9) and the description for those WPs are shown below:

WP1: Project specification
WP2: Theoretical study and modelling
WP3: Development and evaluation of transducers and arrays
WP4: Experimental programme design
WP5: Experimental programme
WP6: Development of advanced signal processing and analysis tools
WP7: Demonstration and trials
WP8: Exploitation and dissemination
WP9: Project management

At the beginning of the project, with the advice of the end users, the consortium has identified the cases where the application of guided wave technology would be most beneficial to resolve the most critical problems within the aircraft industry.

End user has provided the research and technological development (RTD) partners with a representative sample to allow the design arrays and analysis of large sections with optimised number of sensors. Sensor arrays have been designed along with finite element (FE) models.

Generally the approach used for inspection or monitoring with ultrasonic guided waves is to establish a means of transmitting waves through a structure and receiving them after they have interacted with potential defects. The signals are interpreted to identify features of the signals that correspond to defects, and often the emphasis of the technique development is on producing a system that yields simple signals so that a human can easily see from the signals where defects occur. This often introduces a requirement for many sensors and elaborate electronics to simultaneously operate the sensors. It is also limited to structures that do not lead to very complicated signals due to their geometric complexity. In the case of many aircraft components the geometries are complex and it is not feasible for an operator to classify the signals. This has been shown through experiments at TWI and corresponding modelling work by Cereteth.

The consortium has performed fatigue tests on the representative samples to understand the effect of crack growth on the guided wave response. This data has been useful for further training of the neural network algorithm. The SELF-SCAN project has included experiments to monitor the life of aircraft components over long periods. Since, the monitoring technique is based on analysing the changes in the signals over a period of time, it is essential that the sensors themselves stay stable over that period. The use of macrofibre composite (MFC) transducers, monolithic shear piezoceramic transducers and 1-3 piezocomposite transducers was evaluated for use in aircraft monitoring. FE analysis was used to evaluate the excitation of guided waves in the target structures using idealised sources, which was compared to equivalent experiments using the transducers evaluated.

Laser vibrometry and direct measurement was used and compared with the modelling work and it was found that transducers could be used to transmit and receive the ultrasonic guided waves in a suitable manner for monitoring.

A neural network classification system has been developed and demonstrated for distinguishing between healthy and defective aircraft components using data gathered from an ultrasonic guided wave based monitoring system. Experiments were conducted to produce sets of data from components that were used to evaluate the performance of a number of neural network approaches. The performance of the neural networks were evaluated both on sets of data for fixed temperature and for mixed sets where temperature was used as a feature for input into the neural network. High accuracy of defect detection was demonstrated (100 % of test signals were correctly classified) using both isolation of similar temperature data and through using temperature measurement as an input feature. The project achieved its goal of demonstrating the application of ultrasonic guided wave based monitoring and defect detection through neural networks. Such condition monitoring has great potential to increase understanding of the structural integrity of aircraft components, increasing their service life and greatly reducing the risk of catastrophic failures.

It was presented that the developed neural network system was able to differentiate between data from defective and non-defective cases with high accuracy. It was shown that the most optimal neural network developed was able to distinguish between data from non-defective cases and those with a fatigue crack of surface length equal to or greater than 2.5 mm with high accuracy in laboratory conditions. This was achieved despite the defective region being more than a few hundred millimetres from the sensor location and in a location considered inaccessible to other means of inspection.

The SELF-SCAN project has developed an advanced integrated system for structural health monitoring (SHM) and impending failure detection for aircraft components. The project was very successful and achieved all intended objectives and it was demonstrated that defect detection could be achieved with high accuracy using the guided waves and Neural Networks.

Project Results:

The SELF-SCAN project has developed an advanced integrated system for structural health monitoring (SHM) and impending failure detection for aircraft components. Such condition monitoring has great potential to increase understanding of the structural integrity of aircraft components, increasing their service life and greatly reducing the risk of catastrophic failures.

It has been demonstrated at the final demo that the developed neural network system was able to differentiate between data from defective and non-defective samples with high accuracy. It was shown at the final meeting that the most optimal neural network developed was able to distinguish between data from non-defective samples and those with a fatigue crack of surface length equal to or greater than 2.5 mm with high accuracy in laboratory conditions. This was achieved despite the defective region being more than a few hundred millimetres from the sensor location and in a location considered inaccessible to other means of inspection.

The details of technical and scientific results for each WP are presented as follows:

WP1: Project specification

WP1 was 100 % complete in month 5 and is reported separately as D1.1.
A specification document was produced and submitted to the European Commission (EC) as part of D1.1. The document included the requirements of the condition monitoring system and also describes the critical components that have been selected for investigation. The small and medium-sized enterprises (SMEs), RTDs and the end-user discussed the main aims of the project at the kick-off meeting (3 March 2010, Cambridge, United Kingdom (UK)), and identified critical components for monitoring following a brainstorming session. During the discussion, the capabilities of existing guided waves system were explained to the partners. In the past, long-range and ultrasonic testing (LRUT) technologies have been applied to a range of different metal / composite guides with simple as well as complex geometries. Based on this knowledge, a number of possible components were discussed. These included flat metal sheet components, metal rib components and wing flap control drives. The end user mentioned that while all of the possible application areas may be highly suitable for the application of LRUT technology, they may not be commercially feasible. The reasoning for this argument was based on the fact that cheaper/simpler technologies already exist for such applications.

Based on these discussions, two representative aircraft component(s) were selected:
- Case 1: The first scenario was related to the use of LRUT to replace scheduled maintenance inspection to detect cracks on fittings and frames radius. Current maintenance on these parts can require heavy disassembly due to accessibility issues.
- Case 2: The second scenario related to the unscheduled maintenance of structures subjected to impact damage. Here the problem was also the accessibility but the monitoring of a large area (different frames and stringers) was the real issue.

LRUT can bring benefit due to its capability to detect damage from 'faraway'. Both components were procured in the first quarter and LRUT experiments have been performed on both. For case 1, the end user has set the requirement for crack detection on the curved surface to at least 5 mm, even though, a defect detection resolution of up to 2 mm would be highly desirable. In case 2, first step consists in choosing a specific area sensitive to crack, a fuselage area with a stringer or a frame, and implement LRUT to assess its capability to detect structural element rupture. This could be done through test monitoring on ground. The damages to be monitored were impact damages.

In all monitoring problems, temperature was a key factor, which affects the quality of the signals. Any changes in temperature have a direct impact on the velocity of different wave-modes in a sample. Since, for an aircraft in service, monitoring data was collected in varying temperatures and their influence on guided wave signals were examined.

WP2: Theoretical study and modelling

WP2 was 100 % complete in month 12 and is reported separately as D2.1. However due to the nature of the work, Task 2.2 Task 2.3 and Task 2.4 continued during the second half of the project as well.

FE models were used to demonstrate that ultrasonic guided waves could be excited in the locations deemed accessible on the target structures such that these waves would propagate to all regions of the structures, and that at suitable frequencies these waves would interact with defects of the required kind such that the returned signals may be altered in a way that facilities the detection of the defects. They showed that frequencies as high as 500 kHz may be needed, which prompted the development of further data collection systems to enable experiments at such frequencies.

Modelling results were able to demonstrate that a narrow frequency band, short tone burst disturbance occurred in the axis parallel to the surface on which the defect resides, then ultrasonic guided waves with desirable directionality would propagate into the structure, filling the region of interest. This showed that it could be possible to collect signals with sensitivity to the desired defects using a single transducer for a relatively large component. It also highlighted that these signals would be complicated and difficult for human interpretation, although possible sufficient for using with statistical tools, such as the neural network approached that was later developed under SELF-SCAN.

WP3: Development and evaluation of transducers and arrays

WP3 was 100 % complete in month 12 and is reported separately as D3.1.
The MFC transducers for exciting ultrasonic guided waves in thin, riveted panels was assessed. Data was captured over time using a number of these permanently installed on a panel sample. It was shown that ultrasonic waves could be excited over a range of frequencies and were stable over time. A means of transducer encapsulation was designed and adhesives were evaluated.

1-3 composite and monolithic in-plane shear transducers provided by the partner Smart materials have been evaluated and demonstrated both in transducing the desired ultrasonic guided waves and being used to find defects through the neural network technique. Both transducer types could be used successfully over a wide range of frequencies (50 to 700 kHz).

The 1-3 composites demonstrated the advantage that they can be easily cut to fit specific locations. In particular the trials found that when two receivers were situated near opposing edges (the top and bottom sides of the sample), the receiver on the edge on which the defect occurs (which was the top side in the fatigue experiment) will detect the defect with greater accuracy. This could be used to determine on which edge a defect occurs. Software to automate an existing low frequency system (20 to 400 kHz) was produced to facilitate early monitoring investigations. A national instruments based system was developed to perform the signal control aspects of a higher frequency system (500 to 700 kHz), whilst amplifier and power supply boards were developed to provide high voltage capabilities to this higher frequency system. This system was also automated for regular data collection. This system was developed in time to allow all the high frequency data collection needed for developing the technique, but represents hardware far from deployment in the field because of the high cost and weight. Further research was conducted to develop and demonstrate a system with the same capability that was lighter, smaller and could be produced more cheaply, this is reported in Appendix A of D7.1.

WP4: Experimental programme design

WP4 was 100 % complete in month 21 and is reported separately as D4.1. A sample that was both representative of a component with a realistic structural health risk and suitable for use in an accelerated fatigue experiment was designed by NDT Experts and fabricated by TWI. Fixings were developed that would allow it to be subjected to three point bending whilst ultrasonic transducers were in place was designed. A 1 mm notch was designed into the side of the sample to provide stress concentration so that a crack would initiate in a timely fashion.

Fracture mechanics modelling software (CrackWise) was used to predict the crack growth rate and used to determine the fatigue load and cycling rate. The load was converted to a strain value and the fatigue cycling machine operated on feedback from a strain gauge bonded to the sample. A crack was grown from 0 mm to 5 mm surface length (the size specified by NDT experts are desired detectable size) over 20 increments, which were also interspersed with rounds of manual NDT and ultrasonic guided wave data collection. It was planned that the data gathered from the long term fatigue tests were going to be collected as a defect was grown so that the accuracy of the system could be assessed for different defect sizes. To accompany the data gathered using the prototype equipment and technique, it was deemed necessary to use conventional NDT methods (eddy current, surface shear wave ultrasonic testing, alternating currents potential drop and die penetrant) to confirm at regular intervals throughout the testing what size the defect had reached. Since it was important to accurately learn when defect initiation occurred the use of NDT was planned conservatively. The experiment was planned so that the fatigue cycling would be interrupted approximately every 35 000 cycles for ultrasonic data collection and NDT. A mixture of eddy current testing, surface shear waves, alternative current potential drop (ACPD) and die penetrant visual inspection was chosen for defect finding for the fatigue sample, which was to be used at each interval.

Experiments were designed were data could be collected over time from a number of samples so that external influences were present in all data sets, which would be included in the system performance evaluation. Since the main external factor likely to affect performance was temperature, temperature controlled trials were planned for samples placed inside a thermal cycling chamber, for the temperatures 20, 3C and 40 degrees of Celsius, whilst ultrasonic guided wave data could be gathered. This data was then planned to be later used for evaluating the use of neural networks with both mixed data such that the influence of temperature could be part of the performance trials.

WP5: Experimental Programme

WP5 was 100% complete in month 22 and is reported separately as Deliverable D5.1 Transducers selected (during the work carried out between WP3 and WP5) were bonded to the samples in regions deemed by the end user as accessible in the target scenario, which was some hundreds of millimetres from the region deemed inaccessible and prone to fatigue cracking. A combination of direct ultrasonic measurements using the transducers and vibrometry was used to confirm that the transducers were operating in a suitably similar fashion to that required. It was observed that the output ultrasonic guided waves were similar to that predicted by the earlier FE models.

A mounting rig was developed for a fatigue cycling machine such that the fatigue sample could be subjected to three point bending at the same time has having ultrasonic transducers attached. This rig was developed to cause the sample to cyclicly deflect between 1and 3 mm, at a rate of 10 Hz.

Fatigue cycling was conducted and successfully introduced the target sized defect in the target location, whilst LRUT was used to collect monitoring data. The crack size was grown at a rate that allowed many iterations of long range ultrasonic data to be collected and for other methods of NDT to accurately monitor the crack growth. It took 640 000 cycles to initiate a defect, which then grew at an approximately linear rate of 0.37 mm per 100 000 fatigue cycles. After a total of 2 000 000 cycles the crack had been grown to the target length of 5mm. This process was done over 56 iterations, which each involved the collection of sample ultrasonic data and the use of multiple manual NDT techniques. The sample temperature was also measured at each iteration. For each iteration of measurement, the developed high frequency pulser receiver was used to excite tone burst pulses in each transducer sequenctially, whilst simultaneously reception was carried out on co-located transducers. Collection repetition and signal averaging was used to reduce noise to a signficantly low level.

Eddy current testing, surface shear waves testing, and ACPD were used together to detect that only one defect occurred and when it occurred during the fatigue test. Die penetrant was then used in combination with visual inspection through a microscope to measure the size of the defect as it grew. This ultimately led to a high confidence that the defect length was known for all 56 collections of ultrasonic guided wave data. There were 36 iterations before a defect was initiated and 20 more whilst the defect was grown to 5 mm.

Experiments were also conducted on two pairs of identical components in a thermal cycling chamber, such that data could be gathered from each sample over a range of controlled temperatures. As the fatigue specimen was too large for use in a thermal cycling chamber, these tests were done on similar, but alternative shaped specimens in parallel to the fatigue testing. A 5 mm by 5 mm triangular saw-cut defect was introduced into one of each of the sample pairs.

These experiments yielded large data sets that were later used for evaluating the accuracy of using ultrasonic guided wave data for use with various neural network designs whilst taking into account defect size and temperature as varying factors.

WP6: Development of advanced signal processing and analysis tools

WP6 was 100 % complete in month 24 and is reported separately as D6.1
For SELF-SCAN, two basic neural networks were developed. One neural network relies on the Bayesian classifier and has a single layer of nodes, and the other implementation is based on the multi-layered perceptron with two hidden layers. Both networks were trained with all acquired data and several feature combinations were evaluated against their accuracy in defect detection.

Methods of pre-processing signals were evaluated for their ability to reduce random noise and a band-pass filter with limits of +/- 20 % was found to be sufficient. A number of methods of processing signals to produce features that could be used in combination as input for neural network systems were chosen. Four set of these features were then chosen for evaluation and tested as neural network input. Of these feature sets, the combination of central frequency deviation and correlation with reference signal were found to be effective at differentiating data from defective and non-defective data. However, since some positive results were found for other feature sets too, the assessment of various feature sets was extended into the system evaluation stage where more data was going to be available. A pattern input to the neural network is considered to be a vector of the described features. For defective and non-defective plates, 330 in total measurements were acquired. The training set is considered to be those 330 measurements with the a-priory knowledge of classification. The results provided here were obtained by the following the approach below:

(a) remove a pattern from the training set;
(b) train the neural network;
(c) input the omitted pattern for classification;
(d) acquire the output of the neural network and store;
(e) return to the first step until all patterns in the training set have been used as input.

A large data set comprising of LRU signal measurements acquired by incorporating the proposed inspection technique documented in D4.1 and D5.1 was initially pre-conditioned using noise filtering and by applying feature extraction routines to derive metrics in the spatial, temporal, statistical and spectral domains, candidate feature vectors were formed. The acquired measurements and thus feature vectors, were categorised with respect to the temperature during the acquisition time, the pulse excitation frequency and the condition of the inspected specimen (defective vs. defect-free). In turn an artificial neural network, more specifically a 3-layer perceptron with 2 inputs, was designed and trained against the candidate feature vector combinations. Using a feed-forward propagation technique we accumulated results from the output of each trained neural network and evaluated against the classification accuracy with respect to the condition of the specimen sample. The results presented in Section 3 of Deliverable 6.1 demonstrate the successful outcome of the proposed technique and will served as the foundation for the final stage of system evaluation.

WP7: Demonstration and trials

100 % complete in month 27 and is reported separately as D7.1
A video discussing both the developed monitoring systems and the project in general can be found on the project website or can be accessed on the web service YouTube following this address: http://www.youtube.com/watch?v=Lxdp19K-lSE

The SELF-SCAN consortium witnessed a successful demonstration of an NDT monitoring system applicable to detecting cracks as specified by end user on 20 April 2012 at TWI, Cambridge. The demonstration was on how each stage of the inspection system is processing. This was completed through a mixture of a live demo and presentations. A large volume of data has been collected from a number of experiments throughout the project and collated for evaluation of the neural networks. Once the performance of the neural networks was complete, it was shown that new signals could be captured using the developed hardware, processed using our signal processing and the health of the sample under test could be correctly identified using the neural network that was developed and trained under SELF-SCAN.

The system was very well received by the entire project Consortium. It was reviewed and agreed by all partners that technology readiness level (TRL)-4 or TRL-5 was reached with this development work. However it was also raised that it would be beneficial if this concept could be tried in real environments, before being implemented in the aerospace market that has strict regulations. This was foreseen as a potential opportunity for a demonstration project by the SMEs.

In deliverable D7.1 the system development was reviewed and the evaluation of the final results was given.

In earlier WPs data was collected from the long term fatigue experiment where a sample had a fatigue crack induced incrementally whilst data was captured and from the temperature control experiments where two pairs of defective and non-defective samples were tested at different temperatures. Also, the hardware system developed for high frequency testing was made to allow the transmission of aribitrary signals with frequencies in the range of 200 kHz to 1MHz and used for testing at 500 and 700 kHz. These national instruments based system allowed automatic data collection over time for the lab trials needed to develop the technique. This system was relatively high weight and cost, but hardware developed under SELF-SCAN also demonstrated that a similar approach to transmission and reception could be achieved with much lighter, smaller and cheaper hardware, as discussed in Appendix A of D7.1.

The single layer Bayesian neural network performed better than the multilayer perceptron. Highest accuracy was achieved. The set of the following features were identified as generally resulting in the most accurate classification across all data evaluated:

(a) estimated central frequency;
(b) power content of estimate central frequency;
(c) correlation with reference non-defective signal;
(d) covariance with reference non-defective signal.

The neural network results from the fatigue experiment data was assessed with the data split in two ways. Firstly, the data was split into data from when the sample had no defect and data from when the defect was any size above 0 mm. Then the data was split again into data from when the sample had no defect and when the defect size was above 2 mm (excluding the intermediate data when a defect was present but very small). The accuracy of the neural network was as follows:

- 0 - 2 mm cracks with 93 % accuracy,
- cracks over 2.5mm with 100 % accuracy.

WP8: Exploitation and dissemination

WP8 was complete by the project end and is reported separately as D8.1 D8.2 and D8.3. (D8.1 and D8.2 were submitted in 1st Reporting Period). This WP is 100 % complete in month 27.
Due to the nature of the project, the exploitation strategy took the form of public dissemination and demonstration events towards the end of the project. Therefore, most of the dissemination activities have taken place in the second half of the project. The successful technical achievements have provided a strong foundation for effective dissemination during the second year. The plan for those activities have been aligned with the project goals and in accordance with the requirements of EU contract Grant Agreement No 232212. The SMEs, as beneficiaries, has identified relevant customers and industries, and will keep them informed of developments with short emails, newsletters or conversations after the completion of the project.

Within WP8 marketing material was developed and provided for the purposes of dissemination at various events. Details of the events are reported in D8.3 (Final PUDK). The project beneficiaries have decided to limit the number of publications in order to prevent breaches to commercially sensitive information (IP) until the results reach to the desired TRL level. The main dissemination is planned to continue with the SELF-SCAN project website beyond the project term, which was launched in the first half of the project. Public area of the website has been enriched with the presentations from SELF-SCAN dissemination events.

The project flyer was designed and printed in the first reporting period to initiate the promotion of the project technical developments. It was also translated into German and Greek in order to increase public awareness in different markets in Europe.

The following can be highlighted to summarise the WP achievements for WP8:

- The ownership and licensing status of each exploitable product has been confirmed.
- The Foreground IP arisen from the successful exploitation of the background IP has been established.
- The benefits of the exploitable IP and the successful implementation of technological integration for the successful detection of defects on aluminum components used within the aerospace market has been established.
- The approach for exploitation of the integrated SELF-SCAN technology beyond the project term and establish a path to permit SMEs to serve new markets has been agreed between the project beneficiaries.
- The activities carried out for further dissemination of the project achievements in the second half of the project via publications, panel, video, webinar and the SELF-SCAN website was presented.

WP9: Project Management

All the management tasks indicated in the Grant Agreement were followed and achieved in the second half of the project and the activities have been reported separately as D9.1 D9.2 and Deliverable D9.3. The comments on the first periodic report were also considered when project management activities were planned for the second half. Kamer Tuncbilek (TWI) has taken over the Project coordinator role from Devashish Fuloria in August 2011. The change of Coordinator has not affected the overall progress as smooth transition was achieved.

The day to day management activities consist of:

- Reporting: In the second half of the project, there were four deliverable reports (D4.1 D5.1 D6.1 and D7.1) final PUDK, final exploitation agreement, 2nd periodic report and final periodic report to be submitted. The project coordinator has been responsible for the coordination of these reports: Technical reports were coordinated by TWI, after completion of the task undertaken. Any relevant reports and documentation were published on the SELF-SCAN website (members area). All the reports submitted to the EU were reviewed, approved by the consortium and formatted in the project template prior to being sent to the EU by TWI.

- Organising communication: TWI set up a website where information can be uploaded/downloaded by each member of the consortium. Moreover, a list of each company representatives with their respective contact details was circulated at the beginning of the project. TWI as a project coordinator encouraged communication and discussion on the technical work using different communication tools. The SELF-SCAN consortium met every six months to review the technical work undertaken and plan the activities for the forthcoming months. Between meetings communications were maintained between the partners via the SELF-SCAN website, emails, partner visits and phone calls, conference calls (WebEx, Skype) as well as organising online and interim meetings.

- Secretarial support: TWI coordinated the preparation of progress meetings to ensure the attendance of and input from all the partners. TWI prepared the agendas and chaired the progress/interim meetings and circulated the minutes with the agreed actions after the meetings. Each agreed action was allocated to one or more partners responsible for its completion with a fixed deadline. TWI ensured that those actions were completed on time.

- Supervision of progress: Progress was reviewed by TWI and discussed with the consortium at progress meetings. Any deviation from the time schedule was discussed and an action plan was prepared to recover any delay and to prevent further deviation from plan. The technical progress made against the work programme and the results obtained from the research activities were monitored on a regular basis as a day to day management activity. These issues were also discussed and assessed within the consortium. Various methods including the traffic light method was used by the Project Coordinator to highlight the status of each task in WPs to create a universal language in a multi-lingual partnership. According to this, the green showed the tasks completed, the yellow showed on-going tasks and the red highlighted any delayed task, which deviated from the original schedule. This was presented to the partnership at every progress meeting to keep the partners updated on the status. The presentations were made available on the website in Project Files / Meetings / Relevant Meeting Folder e.g. 18M Meeting / Plan for next 6 months. Moreover, the costs incurred were also regularly assessed against the technical work carried out.

Any discrepancies from the original work programme were raised with the EC Scientific Officer and prompt actions to recover were taken in accordance with the agreement with project partners and the EC Scientific Officer where and when necessary. Unexpected research outcomes in systems development and defect creation caused some delays. The major issue tackled during the project was the delay in technical progress. Contingencies have been put in place to ensure that significant benefit is gained from the research in the allotted time. To achieve all the aims of all the tasks and to get maximum benefit from the project a three month extension was requested from the Commission. Together with this request, the Commission was also informed of a modification in the distribution of the financial contribution. This will be explained further in the next section. Despite the delay, with an appropriate time-frame extension and the technical successes to date together with effective cooperation between the project partners (using the remaining budget to complete the outstanding tasks) it is considered by the consortium that a complete and beneficial project has been delivered. The project objectives were successfully achieved and the milestones were all reached. All deliverable reports were completed, submitted to the commission and uploaded onto the project website (see www.self-scanproject.eu online). TWI continued to support the data update on the project website and encouraged partners to use it efficiently in the second half of the project. The filing in the member area was reorganised and the website is now more user friendly than the previous version. Presentations, agendas, meeting minutes and reports as well as auxiliary documents produced at each meeting can be found in the member area. A copy of presentations from the webinar and the panel and translations of flyers in partner languages could be seen on the public domain of the website.

Finally the TRL and a road map, which shows the exploitable results from the project and route to the market was created by TWI, with the feedbacks from partners and references from the grant agreement and approved by project partners at the final meeting. These were presented and discussed at the final meeting and can be found on the project website in the Members Area.

Potential impact:

The expected impacts of the SELF-SCAN technology will be also the key selling points for the system eventually and these can be listed as follows:

- Cost efficient monitoring / increased operational availability: Reduces the risk of incurring costs due to dismantling of the airplane parts for periodic inspections, stoppage and downtime.
- Advanced failure warning: Integrated LRU and neural network capabilities that are able to detect flaws on the represented structures thus reducing risk and cost using traditional inspection techniques. Enabling operators to decide whether to carry out further inspection, and plan to replace the parts. This will contribute to reduce the number of catastrophic accidents and aircraft losses due to structural failure.
- Life extension of aircraft: This project has contributed to develop technology that facilitates an increase in the operational life of aircraft. leading to better returns on investment and profitability. The countries of Eastern Europe have ageing fleets that include 30 - 35 year old aircraft. In the absence of inspection data to support life extension programs, these ageing aircraft will be decommissioned in the near future to meet EU safety standards. This will seriously affect East European airlines that are currently struggling to be competitive and survive in the global market. SELF-SCAN will contribute to European wide sustainability and growth, particularly enhancing employment prospects in the new Member States.
- Reduced in-service maintenance cost- Reducing the frequency of labour intensive inspection and unnecessary maintenance will provide more efficient service and cut cost.
- Reduced insurance premiums: Detection of growing defects at earliest possible stage limits damage, enhances robustness and extends airplane's life-expectancy. The exploitable SELF-SCAN deliverables are expected to reduce the number of aircraft crashes and loss of life due to structural failure by detecting defects before they grow to a critical size The transport policy of the EU aims to deliver an integrated and efficient pan-European transport network meeting the demands of a cleaner environment and of safety and reliability. The EU has stated that major programmes should focus on making travel safer, easier and less polluting. In the Seventh Framework Programme (FP7) key action relating to aeronautics, the EU has identified enhanced inspection and analysis for improving operational capability and safety of aircraft as an important area. The field of aircraft safety is such an important issue to the Commission that a directive was approved on 15 September 1998 requiring Member States to carry out inspection of any aircraft from third countries that show signs of poor maintenance.

Structural failure due to fatigue cracking in the fuselage and engines is a major cause of accidents involving airframe failure or loss and as a consequence injuries and fatalities. The moving 10-year average trends show a decrease in the average number of fatal accidents for Asia, North-, South-, and Central America over the past six to seven years. The progress in preventive maintenance plays an critical role to reduce the number of accidents. SELF-SCAN technology adds value to the current state of art with significantly higher reliability and rate of flaw detection, leading to a major improvement in safety of air transport. Hence, the project contributes to the implementation of EU transport policy as well as quality of life, health and safety.

Within WP8, many marketing materials were developed and provided for the purposes of dissemination at various events. The details of the events attended and held by consortium are reported in D8.3 (Final PUDK).

The main dissemination is planned to continue with the SELF-SCAN project website beyond the project term, which was launched in the first half of the project. Public area of the website has been also enriched with the presentations from SELF-SCAN dissemination events. The project flyer was designed and printed in the first reporting period to initiate the promotion of the project technical developments. It was also translated into German and Greek in order to increase public awareness in different markets in Europe. Flyers were the main marketing tools for the project and were handed out at various events by partners. All translations are available on the project website.

A video discussing both the developed monitoring systems and the project in general can be found on the project website or can be accessed on the web service YouTube following this address http://www.youtube.com/watch?v=Lxdp19K-lSE. This has been published to ensure the continuity of dissemination even after the project is completed.

The panel and webinar provided a great platform to access an appropriate audience and to disseminate the project achievements. There was a good response from some of the industry players and these interests will be pursued after the completion of the project.

Contact details: Project Coordinator: Kamer Tuncbilek
kamer.tuncbilek@twi.co.uk
Granta Park, Great Abington
Cambridge CB21 6AL, UK

List of websites: http://www.self-scanproject.eu/

Related documents