Final Report Summary - ICARUS (An ICT Enabled Approach to Optimising the Reliability of Manual Ultrasonic Non-destructive Testing) Executive Summary:Manual Ultrasonic Testing (MUT) which has been a well-established and widely adopted Non-Destructive Testing (NDT) technique continues to play a key role among the wide range of NDT inspection methods as no automated system exists, as yet, which is as dextrous as the human hand for moving a probe over an area while simultaneously adjusting the probe to achieve the maximum amplitude from a reflector. However, method must be found to reduce the dependency on the actual performance of operators in the testing site from calibration of ultrasonic equipment through adhering to inspection procedures and interpretation of results. This report describes the work undertaken and the results obtained throughout the entire duration of the ICARUS project. An integrated UT probe which exhibits the capability of tracking the translational position (x, y) of the probe during the scan operation based on a wireless optical sensor. This dead reckoning approach not only offers an advantage of being light weight and less cumbersome in setting up but more importantly provides the quality of the scan during the scan operation. In addition, the total system (ICARUS) consist an Bayesian inference engine formulated using a binary Guassian Process classification process to monitor the quality of the acoustic coupling. For the evaluation of the ICARUS system, trials will be performed with the ICARUS working prototype and conventional UT flaw detectors using a pool of 3 qualified/experienced and 3 unqualified participants on a set amount of test pieces with artificial flaws. According to the trial results, ICARUS provides a 38% improvement in the inspection efficiency and a 24.52 % improvement in the Probability of Detection (POD) in comparison to a conventional MUT system. Another common issue for MUT however is the length and cost of the training required to bring operators to an adequate standard of inspection for service industry and end users. Integrating virtual simulation training with current procedures would reduce costs by providing virtual substitutes for expensive test pieces, remove the need for supervision under qualified personnel and reduce time wasted due to unavailability of test pieces. The probe described above can readily be applied to a virtual training environment tailored for NDT where the working prototype employs the use of a smart UT probe with an inherent position tracking capability and plot the c-scan with the known pre-stored data as the trainee moves the integrated probe along the surface. Related test bed trials are analysed with data collected from participants using 2 virtual training blocks of similar dimensions but with different defects (2 flat bottom holes on each block) locations. The distance error between actual and measured defect locations is contained within 10 mm. To the best knowledge of the consortium, this virtual training approach for MUT is novel in the industry.Project Context and Objectives:Manual Ultrasonic Testing (MUT) which has been a well-established and widely adopted Non-Destructive Testing (NDT) technique is used in various fields of applications. Even to date, MUT continues to play a key role among the wide range of NDT inspection methods as it is capable to delivers higher attainable flaw sensitivity among all volumetric NDT system. As the name suggests, it requires the presence of an operator to accomplish the task. A simple illustration of the MUT operation involves the operator to scan the surface of a component by means of holding and moving a single transducer over the region where he would like to test. The transducer acts as a transmitter and receiver which propagate the ultrasonic pulse into the material, subsequently the pulse reflects from discontinuities (such as defects) in the specimen, which are received by the hardware and interpreted by skilled operators. Owing to the simplicity of MUT, it often meets the required performance at a reduced cost. Since the initial usage of Carbon Fibre Reinforced Polymer (CFRP) in secondary aircraft structure, knowledge and development of the materials has improved. This led to the increasing usage of primary structures such as wings and fuselages as seen in the case for Boeing Dreamliner 787 where up to 50% of the mass is manufacture from composite materials and the rest made up of conventional material (i.e steel). This situation gives rise to renew interest in portable devices and extensive NDT system which are often more expensive for in-service inspection particularly in the aerospace sector. It is also widely documented in the literature for MUT to be dependent on the actual performance of operators in the testing site from calibration of ultrasonic equipment through adhering to inspection procedures to interpretation of results. Extensive inspection programmes such as Programme for Inspection for Steel Components (PISC), NORDTEST, carried out in Scandinavia; NIL of the Dutch Welding Institute (Nederlands Instituut voor Lasterchniek); ICON (Inter Calibration of Offshore Non-destructive examination) and TIP (Topside Inspection Project) were launch, predominantly to evaluate the reliability of different NDT techniques where MUT occupied a central part. Form these studies, it is shown that the MUT system generally comprises the sum of its equipment, procedures and personnel. Given the essential need for robust inspection systems, and the known variability of man-machine interfaces, the reliability of MUT has been comprehensively investigated in recent years such as the first three workshops (1997 – 2002) from the series of European-American Workshops with a particular focus upon the influence of Human Factors upon the Probability of defect Detection (PoD). Without exception, these trials have demonstrated that the reliability of MUT is suboptimal and human factors have been identified as one of the principle element affecting the reliability of NDT. Despite the major technology and advanced computing developments (most notably in the development of digital technology and advanced computing) that took place over the past few decades, the human element remains an essential component in MUT. Studies have shown that the contributing factors resulted from the long working hours, the local environment, inherent capability of the inspectors (training, qualification and experience), applied inspection systems and procedures. As a result, MUT is still view to be insufficiently reliable to assure the safety and integrity of components such as the case for complex geometry. This has led to the development of complex automated and semi-automated ultrasonic scanners to overcome the variability introduced by the human factor. Although the human factor often results in inefficient work, resistance to change by using a machine to do what a human has always done is still a challenge faced in the industry. Possible reasons are that inefficient work often brings overtime which act as a motivation for such resistance and many people would like to be in control. Since damages may not be visible on the surface, it is also foreseen that less skilled NDT personnel will need to perform more frequent unscheduled inspections. No automated system exists, as yet, which is as dextrous as the human hand for moving a probe over an area while simultaneously adjusting the probe to achieve the maximum amplitude from a reflector. A human hand can also cope with complex shapes and curve surfaces easily, where machine will find it difficult. The prospects of a cost effective robotic arm which can manipulate the probe in such situation are still remote. So, rather than developing mechanised and automated alternatives to MUT as others have, the present research aims to combine aspects of automatic systems with MUT methods which can be performed in situ.One of the main concerns of manual ultrasonic testing is to ensure that all the region of interest has been scanned (full coverage) with full operator attention. More often than not, the ultrasonic operator is faced with the difficult task of discerning genuine flaws from geometrical artefacts. Moreover, there is the problem of characterising defects which may entail detailed investigation of A-scan signals as well as the geometry of the situation. As a consequence, the interpretation of the results is often deemed inadequate when reporting on possible flaws. It is the aim of the ICARUS project to address the known variability of man-machine interfaces and bring an appreciated degree of reliability to achieving its purpose regarding the detection and the characterisation of component defects while retaining those attributes of cost, simplicity and portability. The design of Information and Communication Technology (ICT) tools for MUT are aimed at assisting the operators in those areas where they are least adapted to controlling the quality of the test, whilst leaving open the possibilities for exploiting those areas of particular strength of skilled inspectors. Such aids are most useful in configuring and storing calibration parameters and plotting profiles and positions of reflectors (i.e. C-scan map). The advancement of artificially intelligence and expert systems now offer more opportunities to the manual operator to perform more sophisticated tasks like interpreting ultrasonic results. It is also hope that not only the ICARUS system will be used to improve the reliability of manual testing but also to be used as a virtual training environment for novice and experience operators alike.Having this objective in mind, a number of tools aimed to assist a manual operator will have to be developed. A tracking system to record the positional information of the ultrasonic testing probe with respect to pre-defined world coordinate system (defined point of origin or datum) is essential. In most cases, it might suffice for an operator to record the Cartesian coordinate (x, y) of the probe position and its skew assuming that it is possible to monitor the back-wall echo in case the device is lifted-off the surface. While there are a number of possible ways to achieve this, they remain limited in the amount of information when used to train manual inspector or to provide suitable data sets for inference engines. Alternative methods sought to overcome these shortcomings will be able to determine the x, y, z, roll, pitch and yaw (i.e. 6 degrees of freedom) of the probe to within 2 mm. There are basically three categories of positional system that may be used. First, mechanical linkage or coordinate measurement arms can provide highly accurate positioning information (less than 100 m translational accuracy and sub-arc second rotational accuracy) but they are often unwieldy and prohibitively expensive. Second Microelectromechanical Systems (MEMS) based Attitude and Heading Reference Systems (AHRS) are inexpensive, small in size and require no measurement universe. Such dead-reckoning positioning system is subjected to cumulative errors in translational information over a time period and therefore presents a large obstacle for this technology to take hold. Last, one or more receivers that are sensitive to sources can be attached to the probe. Obviously, a clear line of sight is required for all receivers, nonetheless, it is one practical option. One of the most subtle parameters to ensure a successfully scanned area when manoeuvring the probe is monitoring the acoustic coupling of the device to the test piece. This often manifests itself into attenuation of the ultrasonic amplitude signal. The reasons for loosing the acoustic coupling are manifold: lack of couplant, inadequate pressure on the probe, device lifted off surface, fast hand movement, etc. Skilled operators are, however, taught diminishing backwall echo (if present) as a sign of poor coupling regardless of the causes. There are even more challenging tests where the back wall echo does not exist, leading manual inspectors to rely on the weak amplitude grain noise to determine whether sufficient pulse energy has entered the component under test. It seems that, so far, operators have relied on their experience of empirical measurements to evaluate the validity of their A-scans. This is often subjective and prone to error. To this end, a checker routine for acoustic coupling for acoustic coupling is implemented through Bayesian methods in the ICARUS inference engine. This routine will extract from the A-scan signal relevant features for classification and build robust inference rules to keep or reject recorded signals. The inference engine will also be invoked to aid with final interpretation of the results. This is anticipated to be an even more complex task. The objective here is to obtain 90% probability of defect detection at 95% confidence level rendering the engine highly reliable and precise for online quality monitoring. The inference engine will need to be trained using a suitable data set and then tested for validation. The considered test cases and the types of flaws will be of paramount importance in achieving the desired output. As part of the system integration in ICARUS, ultrasonic data will also need to be acquired through suitable instruments. These will need to be small, portable and capable of transmitting an ultrasonic pulse and receiving a reflector signal. This will not be discussed here in any detail as the technology is assumed to be mature and basic calibration as well as the test procedure parameters can be easily defined: probe angle, beam spread, delay, gain etc.It is anticipated that some of this ideas suggested in the course of this project have already been implemented on the latest instruments for MUT, others may be novel.Project Results:The initial phase of ICARUS project focussed on defining those potential shortcomings of MUT in the aerospace industry. A questionnaire was sent to industrial partners within the consortium to provide the specifications of their MUT applications. This has help to set the scene and focus on the objectives of the project. A literature survey in the subject of MUT reliability has been conducted. The main findings were summarised in a recent report of Programme for the Assessment of NDT Industry (PANI 3) conducted for the Heath and Safety Executive in the UK. It was established through round robin-robin exercises that the probability of a given inspector finding a given fault is only 52% 17% under typical test conditions. The UK Health and Safety Executive attributed this poor performance to the complex nature of ultrasonic inspection task. They explained that human errors arise from poor technique producing scanning errors and during interpretation of the echoes on the flaw detector; random measurement errors for example in the measurement of range and probe position; lapses in vigilance; isolated inexplicable errors or blunders. All these sources of error are affected by the motivation of the operator, the difficulty of the inspection and external factors such as time pressure or uncomfortable environments. In addition, the inspection may impose technical demands which are beyond those for which the operator has been trained or qualified. Likewise, another review of research in this area presented a qualitative evaluation of human reliability on MUT, and identified the complex interaction between 59 Factors that shape Human Performance in MUT. Of these, 12 key ‘Performance Shaping Factors’ represent 80% of the human performance weighting. Consequently, the simple tasks of manual inspection, as it may appear, encompasses many complex issues affecting the performance of operators during in-service inspection and may become am overwhelmingly daunting task to analyse and redress rationally. Moreover, simple real time systems become tedious when utlilsed to inspect large pieces such as in the aerospace industry. Perhaps in some ways this explains why the application of MUT is confined and more economically viable in small space applications (less than 1m). Proposed solutions from different technology readiness levels including the latest commercialised instruments have equally been surveyed. The ideas described are to help the user with sensors and monitors that are capable of ensuring a reliable test. It is arguable possible to classify these aids into two categories; tools that are capable of checking a successfully scanned area and tools which are able to automate inspection results’ interpretation. Most if not all commercial systems fall in the first category. There is, however, some research work in the field of artificial intelligence and expert systems on characterising flaws which has not been implemented yet in an integrated inspection system.Based on this work and discussions with the consortium, a draft specification document was produced. This describes the devices required for a system that can offer a reliable MUT inspection. It also contains some targets which different sub-components of the system need to meet. Given that the project will involve validation through test cases which are selected and manufacture with artificial flaws to generate suitable data sets.The proposed ICARUS operating system architecture takes two main sensor inputs namely the positional information and ultrasonic data and filters only the reliable ultrasonic amplitude scans for storage, visualisation and automated interpretation of the results. The rest of the scans are used to advise manual operators during the inspection and for training purposes. The positional system plays a central part and has been extensively researched. Initially, a low cost infra-red (IR) cameras capture IR light emanating from LEDs affixed to the probe, which may be processed to determine pose and position information robustly. Preliminary results have shown that 2mm accuracy can be achieved. Considerable effort has been put towards building a prototype camera holder with design flexibility to allow horizontal, vertical and skewing movements. Likewise, a robust and functional fit for unit has been designed and manufactured. The camera unit requires a calibration step to extract its intrinsic and extrinsic parameters. These are fed into a pose estimation algorithm which calculates the translation vector and orientation matrix. The low cost camera has also been characterised in terms of its field of view and dynamics frame rate.Although the IR cameras system above could serve the purpose well however a less cumbersome (in term of setting up) system will be more desired as this would be able to draw greater interests from the industry. This arise a new development work for an alternative Spatial Positioning System for ICARUS based on a dead reckoning approach. This new development work should primarily fulfil the objectives stated in WP2 which is the ability to track, in real time, an ultrasonic probe relative to the component under test. The fundamental technique involved the use of a low cost optical sensor (similar as a computer optical mouse) to function as a two-axes (x, y) displacement sensor (i.e. odometer). The optical sensor module which consist of a single light source (i.e Light-Emitting Diode which is used in this development), lens for focusing, Complementary Metal-Oxide-Semiconductor (CMOS) sensor in the integrated chip and battery pack (3.7 V) made up a complete optical imagine system. An optical navigation engine (basically feature extraction software) is then used to process the optical data (i.e. identify the microtexture in the images to track the motion) collected by the sensor (ADNS - 5090) at very high frame rates (2250 frames per second) and resolution (1000 Count Per Inch). In order to shorten the development time, the optical sensor was extracted from a commercially available wireless optical mouse (Logitech model M210). The miniature size of these components provides the possibility of attaining a light weight design for the ICARUS system. The optical sensor requires a calibration step to determine its scaling factor associated with certain surface material, reducing non-systematic error such as homogeneous surface feature. According to the laboratory results (see table 3, D2.2) it shows an achievable error of 0.1% (i.e. 0.2 mm over 200 mm) using this current system on two different types of material (steel and Carbon Fibre Reinforce Polymer). It is also worth mentioning at this stage that the pose will require the use of an Inertial Measuring Unit (IMU) as described in the deliverable (section 2B, D2.1). The interface electronics between the spatial positioning systems (translational information) to the ICARUS operating system is necessary and also have been developed using a multi-channels data acquisition system (i.e. Arduino BT). This established a bidirectional communication between different sensors and the ICARUS operating system implemented through a common platform (Matlab). Numerous factors have been considered to select this interface circuit board such as the high performance, fast operating frequency (maximum 16 MHz), low operating voltage (which can be simultaneously powered using the same battery source for the optical sensor by means of a parallel connection), open source, small in size and inexpensive. The Lithium-ion battery pack can be automatically charged through the IMU via a standard USB port connection. The translational data (relative positions) are transmitted via the Bluetooth link from the Ardunio BT to the computer for post processing to yield the movement of the probe in a body frame. Although the Arduino BT is using a Class 1 Bluetooth communication module which typically able to transmit data at maximum range of 100 m however the ICARUS system operating distance is still limited by the cable between the ultrasonic transducer and the pulser-receiver. Great effort has been put towards the design and manufacture of the integrated probe prototype by housing all the components in the casing. This prototype is manufactured from a combination of materials such as aluminium and nylon taking the account of functionality, safety and ergonomic factors. The prototype design drawings have been circulated to all the partners and consensus have been reached before commencing the actual manufacturing of the probe. The system overview of the ICARUS integrated operating system is conceptually defined by means of a system flow diagram (see figure 9, D3.1). This concept diagram gives a breakdown on the various modules such as the main operating system, spatial positioning system, ultrasonic system, inference engine, Graphical User Interface (GUI) and storage implemented on a common platform (Matlab). This system architecture seeks to intelligently monitor and performs a decision support during the inspections done by human inspector. The ultrasonic equipment selected for use in the operating system is the Le Coeur Electronique US-key which is a single channel transmitter/receiver (Tx/Rx). This compact and lightweight device has all the functions of a digital ultrasonic flaw detector which can be connected to a computer (usually a laptop) via a standard USB connector. In addition, the DLL provided by the manufacturer is fully compatible with Matlab. During the course of research, another type of pulser-receiver (Optel OPBox ver 2.0) has also been explored.The ICARUS integrated virtual training environment has been developed in the current ICARUS system. It provides digital substitutes for real specimens. The system includes a library of virtual test pieces in a form of raw ultrasonic data indexed by location across the entire region of interest. Efforts also have gone into the development of the ICARUS inference engine, based on Bayesian statistical approach, which is an embedded system within the ICARUS operating system. It will be used to monitor the acoustic coupling and to perform the interpretation of the inspection results. It is necessary to utilised a knowledge base to perform inferences about a given problem, therefore data sets for training the inference engine have been collected by means of manual inspection, automated immersion C-scan and simulated defect response as reported in deliverable D3.2. Due to redundancy and memory size, it is neither necessary nor possible to save all A-scans during inspection. Therefore a conceptual study of using a storage policy for the raw data based on a probabilistic model is performed and highlighted in the same deliverable. A generic binary Bayesian classification model using Gaussian Process (GP) is developed for the current ICARUS system. However, the described pseudo marginal approach for GP classification is an active area of research thus consider to be more favourable to be implemented for the ICARUS system. Since underlying problems were formulated in low dimensions, the MCMC method could play out their strengths – they are more useful in high-dimensional data cases. The inference engine is successfully able to 1) Detect coupling problems,2) Detect flaws in the material3) Provide a probabilistic prediction with uncertainty estimates, which is more informative than using just a simple threshold for generating binary decisions.4) Perform inference in the spatial distribution of flaws which is in particular useful in case when the material was not fully scanned or coupling problems made certain scan invalid.The overall ICARUS system will comprise of a calibrated optical sensor capable of functioning as a two-axes displacement sensor to track the translational movement of the probe. The integrated probe should eliminate the cumbersome steps of setting up additional device to obtain the position so as to satisfy a fully dead-reckoning system approach. It should allow transmitting of the optics data wirelessly and also take account of the size, weight and ergonomics for the packaging design. In addition, the operator will be able to look at information (i.e. A-scan and C-scan map) simultaneously on the display to aid his scan operation. Clear and concise procedure instructions which can be stepped through in the ICARUS Graphical User Interface will serve as guidance to inexperience operator. The Bayesian inference engine shall also be integrated fully into the ICARUS system to aid the operator in determining the flaws and acoustic coupling during the scan operation. The availability of the virtual training environment is also foreseen in the ICARUS system prototype. Evaluation of the ICARUS prototype by means of trial will reveal its performance and short coming in comparison with a convention MUT flaw detector. We believe that this work represents a major improvement in the automated processing of ultra-sonic scanning for non-destructive testing.A detailed breakdown of the work performed and main results for each WP is shown below:WP1Overall specification of the project detailing industry requirements and the ICARUS operating system are documented. The system architecture has five main modules: 1) the operating system to handle the task scheduler routine and manages the system’s resources.2) The spatial positioning unit to locate the probe. It was agreed during one consortium meeting that the probe can be tracking with an accuracy of 2 mm without affecting inspection reliability. 3) The inference engine to check the integrity of the ultrasonic amplitude signals and generate an automated interpretation of inspection results. This is not been fully specified at the stage of the reporting period. This is due to the late accession of UCL into the consortium. UCL’s technical leader Prof. Mark Girolami believes that it should not affect the outcome of the project, as work on this aspect is not to take place during the early stages. 4) The storage module where data can be saved and loaded. Its specifications are partly linked to the inference engine, which is still a work in progress.5) The visualisation module and the graphical user interface where data is displayed in an intuitive and accessible manner for the manual operators.16 blocks of steel are manufactured for test piece I and II, An equal number of curved blocks of steel are manufactured for Test piece III. The three test pieces contain a mixture of artificial defects and geometrical features. Two plates of composite material used as test pieces IV are made available for ICARUS project. There is still work in progress with regards to inspection procedure and modelling. Test piece V is a wing spar also made of steel. The inspection procedures for all steel test pieces are written. These were also modelled using CIVA version 10. These can later be compared to inspection results acquired on similar structures using both manual scan and automatic immersion C-scans. The Inspection simulation parameters were extracted from the inspection procedure for each test piece. Immersion C-Scan is carried out on all test pieces. Specifications described for tracking the ultrasonic probe are given below:• Able to monitor probe’s position in a particular coordinate frame.• Able to locate and size flaws to a high degree of accuracy than the one would achieve with previous scans. • Able to provide 6 degrees of freedom information.Specifications described for the graphical user interface is given below: • Enables a display of a two or three dimensional map showing detected flaws. Features in form of signals/patterns in the inspected region of a component. The complete hardware system comprises an ICARUS spatial positional unit, a receiver and a laptop computer.WP2The objective in WP2 is to develop a spatial positioning unit capable of tracking, in real time, an ultrasonic probe relative to a test component. The requirements and specifications for the most appropriate technology to determine the x, y, z coordinates of the tip of an ultrasound probe relative to a test object, its orientation and the contact force applied by the probe are reviewed. There are a number of possible technologies that can be applied to MUT applications to obtain positional information of UT probes but a simple, low cost and accurate positioning system. Initially, optical tracking for the positioning system is implemented using infra-red camera from Nintendo Wii Remote controls. The use of these devices in conjunction with four markers, make ICARUS 6-DOF tracker. The selection of the Nintendo Wii Remote fulfils all the requirements for ICARUS project’s spatial positioning unit i.e. system complexity, accuracy/precision, response time and usability. The camera signal is transmitted through a wireless link by means of Bluetooth connectivity. It also combines an infrared sensor with three accelerometers, vibration feedback, a speaker and a variety of buttons within a single device. There is another appealing alternative to optical tracking which is even smaller, more portable, more lightweight and integrated, namely the attitude and heading reference system (AHRS) devices. However, obtained results can only provide accurate orientation information and more research is required to build a drift-free translation information signal. The optical tracking is demonstrated to achieve 0.5 mm spatial resolution in x, y and z at less than 600 mm. Experiments are conducted to measure the effect of contact force on the integrity of the ultrasound signal. It is found that it is not a strong feature and the measurement of the acoustic coupling of the ultrasound probe to the test piece is more complicated than simply measuring the pressure on the probe. It is clear that a more intelligent approach is required where the ultrasonic signal is processed by the inference engine to classify its quality. An intuitive graphical user interface (GUI) is developed to handle the data coming in and provide feedback on the measurement data in the grid. It provides a step-by-step guide for the user to follow. The GUI allows you to select the desired procedure and test piece. It is also possible to edit the ultrasonic settings before starting to scan the test piece. During the scanning stage the live A-scan are shown and saved if the inspection conditions are good. The corresponding value is also plotted in the c-scan display at the correct position.Subsequently, large amount of time has been utilised to develop an alternative Spatial Positioning System for ICARUS based on a dead reckoning approach. This new development work should fulfil the objectives stated in WP2 which is the ability to track, in real time, an ultrasonic probe relative to the component under test. The fundamental technique involved the use of a low cost optical sensor (similar as a computer optical mouse) to function as a two-axes (x, y) displacement sensor (i.e. odometer). The optical sensor module which consist of a single light source (i.e Light-Emitting Diode which is used in this development), lens for focusing, Complementary Metal-Oxide-Semiconductor (CMOS) sensor in the integrated chip and battery pack (3.7 V) made up a complete optical imagine system. An optical navigation engine (basically feature extraction software) is then used to process the optical data (i.e. identify the microtexture in the images to track the motion) collected by the sensor (ADNS - 5090) at very high frame rates (2250 frames per second) and resolution (1000 Count Per Inch). In order to shorten the development time, the optical sensor was extracted from a commercially available wireless optical mouse (Logitech model M210). The miniature size of these components provides the possibility of attaining a light weight design for the ICARUS system. The optical sensor requires a calibration step to determine its scaling factor associated with certain surface material, reducing non-systematic error such as homogeneous surface feature. According to the laboratory results (see table 3, D2.2) it shows an achievable error of 0.1% (i.e. 0.2 mm over 200 mm) using this current system on two different types of material (steel and Carbon Fibre Reinforce Polymer). It is also worth mentioning at this stage that the pose will require the use of an Inertial Measuring Unit (IMU) as described in the deliverable (section 2B, D2.1). The interface electronics between the spatial positioning systems (translational information) to the ICARUS operating system is necessary and also have been developed using a multi-channels data acquisition system (i.e. Arduino BT). This established a bidirectional communication between different sensors and the ICARUS operating system implemented through a common platform (Matlab). Numerous factors have been considered to select this interface circuit board such as the high performance, fast operating frequency (maximum 16 MHz), low operating voltage (which can be simultaneously powered using the same battery source for the optical sensor by means of a parallel connection), open source, small in size and inexpensive. The Lithium-ion battery pack can be automatically charged through the IMU via a standard USB port connection. The translational data (relative positions) are transmitted via the Bluetooth link from the Ardunio BT to the computer for post processing to yield the movement of the probe in a body frame. Although the Arduino BT is using a Class 1 Bluetooth communication module which typically able to transmit data at maximum range of 100 m however the ICARUS system operating distance is still limited by the cable between the ultrasonic transducer and the pulser-receiver. Great effort has been put towards the design and manufacture of the integrated probe prototype by housing all the components in the casing. This prototype is manufactured from a combination of materials such as aluminium and nylon taking the account of functionality, safety and ergonomic factors. The prototype design drawings have been circulated to all the partners and consensus have been reached before commencing the actual manufacturing of the probe.WP3The dead reckoning positioning system unit has been successfully integrated with operating system using a common platform (MATLAB), as such the ultrasonic probe is being tracked by the system in real time. The operating system has been designed with a modular approach in order to provide flexibility whereby features can be easily added or removed. The work for the interface with the ultrasonic device to read ultrasonic pulse dynamics in real time has also been successful. Preliminary experiments are being conducted to obtain data sets for training and testing of the acoustic coupling monitor module. Several samples which includes flat steel block, tabular steel sections and step wedge steel block with different surface roughness (quantitatively incremented) have been manufactured. Experimental data were collected by performing a manual scan on each band of surface roughness using FORCE Technology P-scan system. This consists of 8400 A-scans from 7 different levels of surface roughness. Currently, the acoustic coupling quality is simulated through different roughness levels which have been created on one of ICARUS test pieces. Specifications for the classifier are being researched. Initially, a simple Gaussian Process classifier will be attempted building mainly the features from the energy of the full wave. It is anticipated that the user interface will be further developed to take into account these new features. Specifications for the acquisition of suitable data sets to build the inference engine are being drafted. This will help guide the manual operator during the inspection and provide aid in the interpretation of results after the inspection. The inference engine work does not officially start during this reporting period and comes under work package 4.WP4Training data taken with the ICARUS prototype which consists of a collection of A-scans (measured at area with and without defect, poor and good coupling conditions) are labelled along with an annotation of the class. Two pair of binary cases is considered which are damaged and undamaged classes as well as correctly coupled and coupling error. Data cleaning and data normalisation are being performed. A-scan data are truncated to remove signal which contains no or little knowledge regarding whether there is a flaw in the material. Since thickness of scanned material might be variable which consequently affect the time to travel the material so in order to overcome this time dependence problem, data are normalised for each scan such that it starts with the first reflection and more importantly ends with the backwall echo. Median filter is implemented via a sliding window to reduce noise (sharp spikes) in the data. To achieve better classification performance for the inference engine, a dimension reduction technique called principal component analysis (PCA) is implemented to allow mapping of A-scans to a representation of lower dimensionality. To overcome computational costs and runtime due to the large amount of data obtained, spatial distribution (act as a smoothing filter) was used to lower the number of training points. In this manner, data were projected to a lower dimensional subspace while keeping most relevant information. Care has been made to select an optimised window size which is set according to the resolution of the ICARUS system hardware in order not to lose important information needed for classification purpose. Binary Gaussian Process (GP) is formulated which involves a popular approximation method, Laplace approximation and efficient algorithms to perform fast inference over the latent variables of the model. Two approaches were studied for hyperparameter training of the used squared exponential kernel which are maximum likelihood related approach that produces point-estimates of hyperparameters and Markov Chain Monte Carlo (MCMC) approach. It was found that MCMC may play a more useful role in high-dimensional data cases and the pseudo marginal approach for GP classification is an active area of research and could well serve the purpose of ICARUS. Promising results were obtained for experimental results both on ideal (i.e. data taken from a very clean and noise free environment) and on real life data taken with the current ICARUS probe. WP5Success development of the virtual training environment which includes the integration of the software, positioning system and the library consisting of the ultrasonic datasets for the virtual test pieces. An operational basic mode is available for assessing the operator ability to fully cover the inspection surface and their proficiency in interpreting the displayed signals. 2 virtual test pieces (Steel block with artificial flaws) are in place where the data was collected during WP3 and the system allows more training test pieces to be added with ease. Advanced modes such as the ability to monitor the acoustic coupling and memory efficient system are proposed for future development. The analysis methodology of the candidate’s performance measurements has also been devised. The virtual training environment prototype has been demonstrated to the partners during the meeting at Warsaw. This working prototype has been tested on the ICARUS final trial with success (see WP6) and some analysis of the results were obtained through the methodology devised here.WP6The outline of the qualifications trial have been defined which includes the selection of test pieces, selection of personnel, procedure for tests, reporting of test results and analysis of results. Evaluation of the ICARUS system trials will be performed by a number of persons on a finite amount of test pieces with artificial flaws using both the ICARUS system and conventional flaw detectors. The test program has been prepared after consultation with the industrial partners participating in the consortium. The ICARUS system trial was conducted to evaluate its performance against a conventional MUT flaw detector. The trial was conducted by a participant pool consisting of 3 qualified/experienced and 3 unqualified Manual Ultrasonic (MUT) inspectors on industry standard test pieces containing known flaws (Flat bottom holes), with locations unknown to the participant (blind test). As a proof of the ICARUS concept, the test material used in this trial will consist of flat steel block, as further modifications of the current system are necessary for composites (i.e. Carbon fibre reinforced polymer) and also the probe casing will required modification to conform to curved surfaces. Typical performance variables found in NDT (Probability of Detection (POD), Probability of False Alarm (POFA), coverage, inspection efficiency etc.) are calculated for each system for the real inspection and Root Mean Square (RMS) error is used to evaluate the accuracy of the ICARUS’s inference engine. According to the results of the trial perform on real specimens, ICARUS provides a 38% improvement in the inspection efficiency and a 24.5 % improvement in the POD in comparison to a conventional MUT system Results indicate that whilst ICARUS achieves a significant improvement in comparison to traditional MUT techniques in both qualified and unqualified groups, improvements are required to achieve targets of 90% POD and 95% certainty as defined.WP7A final Plan for use and Dissemination of the Foreground (PUDF) is produced (see D7.2). It covers two aspects namely the management of intellectual property and exploitation of results. Details on the This PUDF has been issued at the end of the project after a prototype has been demonstrated in a relevant environment for training purposes (industrial research and consultancy) and SME end user (service). This PUDF details the business plans for progressing ICARUS and its components through the stages from prototype to production systems after the project life. Management procedures and structures have been put in place to enable the partners to communicate their business plans to the exploitation Manager and for the Manager to coordinate the partners as well as external agencies and bodies engaged in the technology transfer activities. The E&D committee envisages the roadmap for technological development of the ICARUS system and an estimates of investment required have also been detailed. 2 papers have been presented at the BINDT 2012 Aerospace NDT symposium and BINDT 2012 Annual conference. Further funding application have been apply through the Research European Agency (REA) DEMO project call in order to develop the prototype into a commercial product.Patent search has been carried out (see D7.3) and according to the search results, no relevant patent which suggest a computer-like mouse system to track manual ultrasonic probe, inference engine or virtual training environment for MUT can be found. However patent application haven not been proceeded at this stage, while partners consider the financial implications. Some considerations for the possible effect on standards and norms in Don-destructive testing (NDT) of the ICARUS system and consequence socio-economic effects within the European Union have been studied (see D7.5). The website for ICARUS has been updated throughout the lifetime of the project.WP8A review was done after half way through its 2 years programme and the findings were reported in deliverable D8.2. Deliverables and achievements made after one year are evaluated and plans to attain the milestones remaining are highlighted. The plan to exploit and protect the intellectual property resulting from this project is outlined and this leads to the detailed descriptions of the idea and plan of the road map beyond the end of the project life. It is worth noting that a progress report including management, finance and technical details has been submitted for review at the ninth month stage. Despite the difficulty and high development risks in the ICARUS project, the consortium is making a good progress towards meeting the milestones set out in the project Description of work.A video (>5min video) which showcase a demonstration of the use for the current ICARUS prototype have been recorded for both the virtual training environment and real inspection trial. The screen shots were captured in deliverable D8.1. Potential Impact:It is anticipated that the ICARUS system has the potential to significantly change the nature of manual ultrasonic NDT, both for inspection by aiding the collection and interpretation of test data and for training by providing simulations of a much wider range of defect conditions that could be accommodated by using physical test pieces. This inspection system will not only be applied to the aerospace sector but can also be applied to other safety critical industries such as power generation, chemical process plants, oil and gas industries. A hierarchy of products and services spinning out of the ICARUS project has been identified and members of the ICARUS consortium could possibly form an industry supply chain with interdependencies between inspection companies, sensor companies, electronics companies, NDT equipment companies, NDT training institutions, software modelling companies and end users in predominantly the aerospace and oil & gas industries. Test samples are expensive to manufacture and very limited in the scope of geometries and defect types, it is envisage that the ICARUS virtual training environment will remove this limitation and provide test situations that will help improve UT procedure and UT operator performance. This is identify as a novel way of providing UT training and can potentially be an attractive business opportunity in the NDT training establishments.List of Websites:http://www.icarus-project.euRelevant contacts details:Tat-Hean Gan, Brunel UniversityTel: +44 (0) 1223 899455Fax: +44 (0) 1223 890952E-mail: tat-hean.gan@brunel.ac.uk
