In-field CFRP surfaces Contamination Assessment by aRtificial Olfaction tool

Project Information

ICARO

Grant agreement ID: 286786

Closed project

Start date 1 May 2011

End date 28 February 2013

Funded under

• Specific Programme "Cooperation": Joint Technology Initiatives

Total cost € 237 037,00

EU contribution € 177 777,75

Coordinated by

AGENZIA NAZIONALE PER LE NUOVE TECNOLOGIE, L'ENERGIA E LO SVILUPPO ECONOMICO SOSTENIBILE
Italy

Final Report Summary - ICARO (In-field CFRP surfaces Contamination Assessment by aRtificial Olfaction tool)

Executive Summary:

Transport industry accounts for a significant percentage of CO2 emissions. Obtaining a lower fuel per mile ratio will not only help to reduce pollution and greenhouse gas emissions, it will also contribute to the competitiveness of aerospace industries and airlines. As such, Europe is significantly investing in research that pursues the achievement of the so called Green Aircraft industry with the CleanSky joint technical initiative targeting, among others technologies, increased reliance on laminar flows and improved usage of composite material for structural components in aircrafts.
CFRP components assembly is not achieved through riveting, it actually relies on adhesive bonding. A significant increase in the adoption of CFRP technology in primary structures will only be achieved if a reliable quality assurance technology will ensure the mechanical properties of the bonded panels. Up to now, no objective technology has been validated to ensure the quality of bonds and that is having a detrimental effect on the quest for the Green Aircraft. With the ICARO proposal (Infield CFRP surfaces Contamination Assessment by aRtificial Olfaction tool, JTI SP1-JTI-CS-2010-05, GA Number 286786) , ENEA was tackling the challenge primarily targeting pre-bond inspection and specifically the search for surface contaminants in maintenance and repair scenarios. Contamination in fact strongly affects the mechanical properties of adhesive joints assembly lowering the amount of force needed to cause joint failures. In facts, more than 20% (and up to 70%) reduction of interlaminar fracture toughness is expected to occur based on exposure to hydraulic fluids and release agents. . A number of different fluids can act as surface contaminant both in the assembly and operative life of the aircraft: hydraulic fluids, moisture, release agents (assembly phase) and deicing fluids being among the most dangerous.
During the ICARO project, benefitting from the CleanSky contribution, the ENEA electronic nose technology has been studied and adapted for the use as a highly portable, fast contaminant detection and quantification tool and eventually developed in the ICARO e-nose prototype. The combined use of different solid state sensing technologies as well as pattern recognition methodologies have given birth to a tool that is able to detect surface cleanliness state and purposely quantify the amount of contamination especially in the maintenance scenario. After the scarfing process and before the patch adhesion and curing process, the ICARO e-nose can be used to detect contaminations ensuring the quality of the maintenance patch adhesive bond.
Thanks to the JTI efforts the ICARO prototype has been tested versus the capability to detect hydraulic fluids, moisture and anti-ice agents. Cooperation with SFWA partners and specifically with Fraunhofer IFAM institute (as a task responsible) has allowed developing the tool in a realistic framework thanks to appropriate sample preparation methodologies and focusing versus relevant contaminant and relevant contaminants concentrations in the maintenance scenario. During the project some limitations of artificial olfaction technology have been tackled and successfully solved while a long term test period will allow to finally fueling the validation of the use of e-nose technology as a CFRP fast pre-bond inspection tool.
After the execution of the test procedures the ICARO prototype has shown capable to obtain performance in excess of 96% correct classification when dealing with a test scenario including the detection of Skydrol 500-B hydraulic fluid and Kilfrost as surface contaminants in the range of 2 grams per m^2.

Project Context and Objectives:

In the quest for the so called green aircraft technologies, aircraft components weight is a primary concern. Actually, within the framework of research framework program 7, EU is investing in the development of adequate technologies for achieving weight reduction and the use of composite materials such as CFRP goes in this direction. The assembly of CFRP components is carried out by adhesive bonding technologies which require appropriate quality assurance techniques and tools. Novel NDT and quality assurance tools for composite materials adhesive bonding are among the most promising. The availability of such technologies will, in fact, contribute to secure the adoption of lightweight CFRP technology in aircraft assembly and maintenance. As a primary consequence, aircraft and transport industry competitiveness will benefit for a reduction of fuel per mileage, greenhouse gases and pollutant emissions. This will in turn allow for an enhanced sustainability and cost reduction of transport and people mobility. Sustainable development at its broadest meaning is thus concerned.
In this framework, ICARO (In Field CFRP surfaces Contamination Assessment by aRtificial Olfaction tool) is a project arising from an ITD belonging to the CleanSky SFWA demonstrator effort dealing with NDT for quality assurance of adhesive bonding and specifically with pre-bond surface cleanliness state in a maintenance and repair framework.
It has been recently shown how limited contamination by different aerospace industry related fluids severely affects mechanical properties of CFRP bonding. As shown by Markatos et al. in their paper 20% to 70% reduction of forces needed to obtain cohesive failures or delamination of bonded CFRP samples are expected to occur when CFRP panels are subjected to contamination by those fluids. Among these, hydraulic fluids (e.g. Skydrol 500-B), moisture and de-icing fluids are the most relevant in a maintenance and repair scenarios because of the high probability of contamination occurrence during aircraft operative life.
Detecting contaminations on surfaces of CFRP structures before bonded repair take place is thus an important requirement in order to apply an efficient surface treatment. Up to now no technology has been fully validated or qualified as an inspection technology for CFRP bonding quality assurance and this pose a limit to the diffusion of CFRP especially when primary structures are concerned. Actually they are not widely in use for load critical structures unless mechanical fasteners are used for bonding. The latter pose significant weight penalties arising for the need to deal with stress concentrations around the fasteners.
Several requirements exist for a suitable contamination detection technique:

 Measurements must be carried out rapidly (ideally, less than 5 minutes should be required to assess the cleanliness of a 1m^2 panel area)
 Measurement should be carried out in a and fully automated fashion without needing a specific knowledge in the particular adopted technology.
 A single detection techniques must be sensitive for all the relevant contaminants
 Measurement reproducibility must be high

Finally, the technique must be suitable for use in technical environments where bonded repair takes place.
JTI CfP officers have selected artificial olfaction as a possible technology to be investigated and adapted for its possible use in surface cleanliness assessment. However artificial olfaction state of the art, at that moment, was poor of relevant previous experimentation with specific relevance to the CFRP surface contamination detection scenario. As a matter of facts, only a few paper contributions dealt with surface contamination analysis and relevant contaminants were completely new in the artificial olfaction scenario. As such, the objectives of ICARO project included the investigation, the development and adaptation of artificial olfaction technologies to the above described framework. Different gas sensing technologies have been investigated for their suitability and responsiveness to the complex mixtures of chemicals that made up the different relevant contaminants. Surface volatile uptake enhancing techniques have been selected for their compatibility with CFRP handling in a maintenance and repair scenario. An ad-hoc electronic nose prototype has been designed and developed by using the gained knowledge. The prototype, named ICARO after the ancient greek myth, has been than trained by using and ad-hoc developed calibration methodology, focused on obtaining a fast response by point by point classification and quantification of volatiles coming off the surface. Pattern recognition sw components have been produced by implementing the investigated computational intelligence methodologies as selected for their suitability to the project requirements. Trainings of pattern recognition sw have been carried out using appropriate measurements campaigns allowing the validation of the tool by following a developed test procedures.

Project Results:

The ICARO project, as depicted in the DoW, has successfully followed an incremental approach that encompassed:

 Preliminary suitability analysis of artificial olfaction technologies, conducted with two different commercial electronic nose platform (polymeric sensors based and hybrid sensor array based)
 Volatile sampling and measurement methodology adaptations
 Ad-hoc ICARO prototype design and development
 ICARO prototype calibration and testing

During all phases, the adoption or development of multivariate data analysis tools has been pf primary importance to grab a firm knowledge about the behaviour of sensor array when dealing with the harsh scenario.

The results of the preliminary suitability analysis: Two different commercial platforms have been adopted to obtain a wide, as far as it was possible, knowledge about the responsiveness of different solid state sensor technologies to the contaminants selected in the relevant scenario. Conductive Polymers sensors technology was screening for sensitivity and stability towards:

 A hydraulic fluid dilution in water (0.6% v/v)
 A diesel oil solution dilution in water (0.6% v/v)
 Hydraulic fluid contaminated CFRP samples immersed in water
 Hydraulic fluid contaminated CFRP samples immersed in water and sonicated (to enhance volatile stripping)

The commercial platform hosting the sensor was chosen to be the Bloodhound BT116 electronic nose. Results accounted for sufficient sensitivity and specificity both for the watery dilution and for sonicated water immersed CFRP contaminated samples, but stability was largely unsuitable since repeatability of the measure faded in just a few hours due to poisoning effects. The aggressive nature of the hydraulic fluid under investigation was significantly detrimental to the sensors working principles and its structure. These results ruled out the possibility to use conductive polymer technologies. In order to enlarge the diversity of sensor technologies to be investigated an Airsense GDA2 hybrid sensor array commercial platform was acquired. The electronic nose is actually equipped with 2 metal oxide (MOX) and 1 electrochemical (EC) sensors, a photo ionization detector (PID) and an ion mobility spectrometer (IMS). The IMS sensor allowed for the strengthening of true solid state sensor technology with mass spectrometry enhancing both sensitivity and discrimination capabilities. First results were very encouraging and in fact, the new platform allowed us to tackle discrimination of samples prepared in full accordance with scenario requirements as selected by SFWA officers (5x5 cm^2 M21 scarved CFRP samples).
The analysis of the dynamic responses of the sensors shows that the system had the capability to detect small traces of hydraulic fluid contamination. The same sensor array showed a significantly lower sensitivity to runway deicing agent contamination (Kilfrost) as it could be expected since this fluid is basically composed by a water mixture of salts.
A final suitability analysis, conducted with pattern recognition techniques accounted for a percentage of total correct classification in excess of 90% with 4.5% false positive rate and 2.6% of false negative rate. Those figures, although preliminary and conducted during a single run with a limited set of samples account for a positive result of suitability check as regards as specificity and sensitivity of the electronic nose technology towards the specific scenario.
Several attempts have been conducted to define a suitable measurement methodology that could deal with the requirement of the MRO scenario (especially speed of assessment) simultaneously maintaining sufficient detection and discrimination capabilities. In order to obtain fast and reliable assessment of the cleanliness state of the surface, the electronic nose should be capable to achieve a very fast stable response, this, given the typical sensors dynamic with T90 in the range of 60 to 100 seconds, rule out the possibility to check for the entire sensor response during the exposure cycle, the typical measurement and assessment procedure used in artificial olfaction. A possibility is offered by our previous studies on point by point estimations both using dynamic features extraction approaches or simpler, single point approaches when possible. The second approach was chosen in order to keep high the readability and human accountability of the extracted knowledge since it is well known that complex computational intelligence approaches tends to be refused in high critical applications such as aerospace applications due to the difficulty to assess the reliability of cryptic knowledge representations in multivariate space.
According to these requirements, we have selected to design a measure cycle with the following characteristics:

• 15sec baseline acquisition
• 60 sec sampling time
• 60 sec recovery time

However the electronic nose sw components should be able to express its first estimation about the presence, the nature and the amount of contaminant, if present on the surface, as soon as the sampling phase starts. In this way we should be able to express first estimation on a 10cmx10cm area only after 15 seconds baseline acquisition time. Every results expressed in this document is related to this methodology and account for the capability of the electronic nose to express detection and discrimination capability in a sample by sample (each second) fashion.
Several tests have also been conducted to enhance the capability of the electronic nose to extract relevant volatile from the surface of contaminated samples as documented in the relevant deliverables. The final decision was made to employ an IR low power emitter that resulted to be compatible with the requirements about samples handling. The IR emitter is switched on during the first 30 seconds of sampling phase.
ICARO Prototype design and development
After having achieved a grasp on the possibility to use electronic nose technology for contamination assessment of CFRP, ENEA has proceeded to use the acquired knowledge for the design of an ad-hoc electronic nose. The basic concept was to allow for the integration of different sensors technology although continuing to basically relying on the sensors technology that has shown the best responsiveness and can guarantee a significant diversification in specificity i.e. MOX. Their easy to integrate architecture also allows for the fabrication of different sensing layer showing different specificity, furthermore the possibility to use thermal modulation profiles to tune specificity will allow for further specialization in following development of the ICARO e-nose design. The other basic concept was to integrate a thermal desorption aid in the form of a halogen lamp whose IR emission are particularly high. Those two concepts also allowed for reaching a compact, lightweight and low cost design that could be beneficial in a market phase. The design was also leaded by a strong commitment to modularization and use of open source hw designs (Arduino MEGA + WiFi shield) so to guarantee the needed degree of freedom for optimization without sacrificing integration aspects and an adequately large community of developers to easy further developments.
The architectural design is based on two interacting subsystems:
The sampling system was, as previously mentioned, equipped with an IR emitter (20w Halogen lamp), a plastic support has been devised so to help orientation of the lamp. While in the previous design, the inlet was very close to the halogen lamp, in the final design inlet has been moved away to help maintenance (see following figures). Details on specific models of the integrated precision pump and valves as well as electronic boards are available in the specific deliverable 2.1.1.

• A precision pump based pneumatic s/s responsible for volatiles uptake, and transport towards sensors chamber and finally towards gas exhausts; a filtering unit allow for the establishment of a baseline response useful for the following signal processing.
• An electronic control unit that provide interface services towards PC based control sw, sensor polarization, signal conditioning and data sampling, pump and valves on-time control and wireless telecommunication capabilities as well as local storage of captured data.

The e-nose communication capability is achieved by the integration of an additional ad-hoc shield, the wireless Shield based on IEEE 802.11 standard, however sensor data may be also saved locally (on board) to files on a removable SD memory card. This completes the interfacing possibility of the developed design and led it to the most update commercial standard for electronic noses design. A number of different parameters regarding data acquisition (structure of the measurement cycle, sampling frequency, amplification ranges, etc.) can be tuned via the GUI in order to let the researchers to experiment with different setups for measurement optimization in the relevant scenario. In a second version, the parameters set could be fixed in several different configuration files for the different scenarios to be tackled. At the moment, just as an example, sampling frequencies will be configurable in the range 1÷10 Hz.

ICARO Prototype calibration and testing
Once developed, the ICARO e-nose prototype has undergone a calibration phase (see Deliverable 3.3.1) followed by multiple testing phases performed according a developed testing procedures document, the last of them was depicted in the relevant deliverables 3.3.2 and 3.3.3.
Calibration of the electronic nose was performed within a state of the art framework based on statistical pattern recognition methodologies developing ad-hoc feature extraction, classification and regression tools. The state of the art framework foresee, for the electronic noses, a calibration phase in terms of a machine learning process in which the e-nose is exposed to set of relevant samples in order to build a suitable knowledge base made up of response pattern to be used for classification and regression purposes. According to this and exploiting cooperation with the Fraunhofer IFAM institute, a set of relevant samples have been defined and measured by the electronic nose. Following the developed measurement methodologies, the so called training set has been produced.
A model selection has been performed accordingly. In particular, model selection is the design task that encompasses the choice of a pattern recognition model capable to deal with the requirements of the specific application. Different pre-processing techniques have been adopted to generate descriptive parameters from the sensor array responses. One of these techniques was the well-established data analysis by PCA (Principal Component Analysis). For the electronic nose application, where the analysis space is constructed on basis of multi-parametric sensor array responses, PCA is the primary method of choice, to extract relevant information from the data, preserving as much data variance as possible. Linear Discriminant Analysis (LDA) has been also used to extract relevant projection in the data exploration phase. Finally the use of Neural Networks (NNs) has been tested in order to establish the final classification framework.
Finally, neural networks are universal functional approximators that can be used both for classification and regression problems; they have been also used for the estimation of pollutant concentrations. NN knowledge representation and operative functional is extremely compact and so the computational and space footprint. It remains the best candidate for the implementation of the next generation ICARO prototype that will be equipped with on board sensor pattern recognition component.
Training data with adequate use of measurement cycles scale cross validation techniques allowed for the evaluation of the e-nose detection, classification and quantification capabilities. The finalized pattern recognition architecture foresees a two stage system. The first is responsible for point by point classification of the sampled sensor array responses. One discrimination have been performed, i.e. the contaminant have been identified, a second layer built up by different regressors is in charge for estimating the contamination amount in terms of g/m^2. Preliminary data analysis as well as final results indicate that the ICARO e-nose suffer from environmental influences in the measurement phase, however t has shown capable, thanks to the developed calibration methodology, to distinguish with very interesting performance figures among the target analytes of this project. The percentage of classification error made us confident that the operator, by reading, the on line estimation performed by the e-nose software, will be able to correctly classify the samples under analysis. Quantification capability has also been shown allowing reaching milligrams per square meter absolute error (8x10-3 for Skydrol, 1x10-2for Kilfrost) with a detection threshold estimated to be under 2 g/m2. In order to rule out environmental influences, the e-nose should be operated in controlled environmental conditions such to remain within 18÷25 °C and 30÷50 RH intervals. Further development will allow relaxing these requirements.

Potential Impact:

The ICARO project, consistently with its applied research nature, has followed dissemination and exploitation strategy mainly based on scientific outcomes publication and participation to JTI SFWA 1.1.3 foreseen dissemination activities. In addition the primary foreground outcome, that is the development of the first version of the ICARO prototype will be further developed in a more advanced prototype and could be the subject of a patent proposal filing. ENEA will activate its technological transfer unit in order to propose the developed technology to relevant actors in the Italian and international stakeholder scenario. In this framework the participation of ENEA in the Campania Aeronautic district will foster the possible transfer of the generated foreground in further research or technological transfer projects. Contacts with MROs, the primary target group for the ICARO outcomes) are already in an active stage to propose the development of a second version of the ICARO prototype in a commercial product.
Cooperation with research group throughout the EU has been sought and established in order to sustain the research and development of such technology for quality assurance of surfaces adhesive bonding. Although tackling different scenarios the project outcomes will also be beneficial for the development of the FP7-ENCOMB project.

List of Websites:

isense.portici.enea.it/isense/index.php/projects/icaro-project

Contacts:
Dr. Ing. Saverio De Vito e-mail:[e-mail] ------ Dr. Ing. Mara Miglietta e-mail:[e-mail]