Uwe Zscherpel

A comparison between neural networks and decision trees based on data from industrial radiographic testing

Abstract In this paper, we are empirically comparing the performance of neural nets and decision trees based on a data set for the detection of defects in welding seams. This data set was created by image feature extraction procedures working on digitized X-ray films. We introduce a framework for distinguishing classification methods. We found that more detailed analysis of the error rate is necessary in order to judge the performance of the learning and classification method. However, the error rate cannot be the only criterion for comparing between the different learning methods. This is a more complex selection process that involves more criteria that we are describing in this paper.

A Comparison between Neural Networks and Decision Trees

In the paper, we empirical compare the performance of neural nets and decision trees based on a data set for the detection of defects in welding seams. The data set was created by image feature extraction procedures working on x-ray images. We consider our data set as highly complex and containing imprecise and uncertain datas. We explain how the data set was created and what kinds of features were extracted from the images. Then, we explain what kind of neural nets and induction of decision trees were used for classification. We introduce a framework for distinguishing classification methods. We observed that the performance of neural nets is not significant better than the performance of decision trees if we are only looking for the overall error rate. We found that more detailed analysis of the error rate is necessary in order to judge the performance of the learning and classification method. However, the error rate can not be the only criteria for the comparison between the different learning methods. It is a more complex selection process that involves more criterias that we describe in the paper.

OPTIMUM EXPOSURE CONDITIONS FOR COMPUTED RADIOGRAPHY DEPENDING ON FIXED PATTERN NOISE AND EFFICIENCY OF IMAGING PLATE‐SCANNER SYSTEMS

The presently active standards on Computed Radiography (CR) need a major revision. It was observed by many users that the image quality for class B of EN 14784‐2 is not achievable under the same exposure conditions as used for film exposure. A mathematical model was developed and tested, which allows the calculation of the image quality, proven by image quality indicators (IQI), depending on the fixed pattern noise and the efficiency of the used imaging plate (IP) scanner system. All tested IP scanner systems provide a fixed correlation between the measured signal/noise ratio (SNR) and the grey values in the digital images. The maximum achievable SNRmax depends on the fixed pattern noise of the plate (high dose limit). Depending on the exposure dose an optimum visibility of IQIs can be predicted by calculation and measured considering the attenuation coefficient for calculation of the contrast/noise ratio (CNR). The diameter of the just visible wire is proportional to 1/sqrt(CNR). The optimum tube voltage...

Modelling based radiography for NDE of subsea pipelines

This work presents the use of limited experimental measurements to develop a set of calibrated simulation parameters that can then be used for reliable simulation of subsea pipeline inspections. The modelling software aRTist is used as the simulation tool, and the calibration is through comparison with experimental images of a well characterised sample in a water tank. Image quality parameters such as signal-to-noise ratio, contrast and basic spatial resolution are compared with the aim of matching simulated values to experimental results. Currently the model is partially calibrated, with signal-to-noise ratio successfully matched while differences are still found in contrast-to-noise ratio comparisons. This means that measurements depending on absolute intensity are not accurate enough in the simulation at this stage. However, the simulation is found to be accurate for wall thickness measurements in tangential images, which are not based on absolute intensity, with simulated and experimental cases produc...

Performance Control: Nondestructive Testing and Reliability Evaluation

The performance of materials – as constituents of the components of engineering systems – is essential for the functionality of engineering systems in all branches of technology and industry. Instrumental for characterizing the performance of materials are 1. methods to study and assess the basic damage mechanisms that detrimentally influence the proper functioning of materials, such as materials fatigue and fracture (Chap. 7), corrosion (Chap. 12), friction and wear (Chap. 13), biogenic impact (Chap. 14), materials–environment interactions (Chap. 15) 2. methods to study and assess the performance of materials in engineering applications and to support condition monitoring of materialsʼ functional behavior. In this chapter the following experimental and theoretical methods for performance control and condition monitoring are compiled. Nondestructive evaluation (NDE) methods Methods of industrial radiology Methods of computed tomography (CT) Embedded sensors techniques to monitor structural health and to assess materials performance in situ under application conditions Methods to characterize the reliability of materials with statistical tools and test strategies for structural components and complex engineering systems.

Image recovery by removing stochastic artefacts identified as local asymmetries

Stochastic artefacts are frequently encountered in digital radiography and tomography with neutrons. Most obviously, they are caused by ubiquitous scattered radiation hitting the CCD-sensor. They appear as scattered dots and, at higher frequency of occurrence, they may obscure the image. Some of these dotted interferences vary with time, however, a large portion of them remains persistent so the problem cannot be resolved by collecting stacks of images and to merge them to a median image. The situation becomes even worse in computed tomography (CT) where each artefact causes a circular pattern in the reconstructed plane. Therefore, these stochastic artefacts have to be removed completely and automatically while leaving the original image content untouched. A simplified image acquisition and artefact removal tool was developed at BAM and is available to interested users. Furthermore, an algorithm complying with all the requirements mentioned above was developed that reliably removes artefacts that could even exceed the size of a single pixel without affecting other parts of the image. It consists of an iterative two-step algorithm adjusting pixel values within a 3 × 3 matrix inside of a 5 × 5 kernel and the centre pixel only within a 3 × 3 kernel, resp. It has been applied to thousands of images obtained from the NECTAR facility at the FRM II in Garching, Germany, without any need of a visual control. In essence, the procedure consists of identifying and tackling asymmetric intensity distributions locally with recording each treatment of a pixel. Searching for the local asymmetry with subsequent correction rather than replacing individually identified pixels constitutes the basic idea of the algorithm. The efficiency of the proposed algorithm is demonstrated with a severely spoiled example of neutron radiography and tomography as compared with median filtering, the most convenient alternative approach by visual check, histogram and power spectra analysis.

Reliability Evaluation of NDT Techniques for Cu-Welds for Risk Assessment of Nuclear Waste Encapsulation

Abstract In order to handle the long living radioactive waste (spent nuclear fuel) SKB is planning to build a deep repository that requires no monitoring by future generations. The spent nuclear fuel will be encapsulated in copper canisters consisting of a graphite cast iron insert shielded by an outer 30–50 mm thick copper cylinder for corrosion protection. The most critical part of the encapsulation process is the sealing of the canister, which is done by welding the copper lid to the cylindrical part of the copper shell using friction stir welding and electron beam welding. The quality of the welding process and the reliability of the NDT system – radiographic and ultrasonic testing – must be satisfactorily determined and combined to derive assumptions regarding the frequency of undetected welding defects for a set of canisters as input for the risk assessment. This is done using the POD method according to the “Reliability Handbook MIL 1823” and its generalization.

Simulation supported POD for RT test case-concept and modeling

Within the framework of the European project PICASSO, the radiographic simulator aRTist (analytical Radiographic Testing inspection simulation tool) developed by BAM has been extended for reliability assessment of film and digital radiography. NDT of safety relevant components of aerospace industry requires the proof of probability of detection (POD) of the inspection. Modeling tools can reduce the expense of such extended, time consuming NDT trials, if the result of simulation fits to the experiment. Our analytic simulation tool consists of three modules for the description of the radiation source, the interaction of radiation with test pieces and flaws, and the detection process with special focus on film and digital industrial radiography. It features high processing speed with near-interactive frame rates and a high level of realism. A concept has been developed as well as a software extension for reliability investigations, completed by a user interface for planning automatic simulations with varying...

Digital radiography using digital detector arrays fulfills critical applications for offshore pipelines

Digital radiography in the inspection of welded pipes to be installed under deep water offshore gas and oil pipelines, like a presalt in Brazil, in the paper has been investigated. The aim is to use digital radiography for nondestructive testing of welds as it is already in use in the medical, aerospace, security, automotive, and petrochemical sectors. Among the current options, the DDA (Digital Detector Array) is considered as one of the best solutions to replace industrial films, as well as to increase the sensitivity to reduce the inspection cycle time. This paper shows the results of this new technique, comparing it to radiography with industrial films systems. In this paper, 20 test specimens of longitudinal welded pipe joints, specially prepared with artificial defects like cracks, lack of fusion, lack of penetration, and porosities and slag inclusions with varying dimensions and in 06 different base metal wall thicknesses, were tested and a comparison of the techniques was made. These experiments verified the purposed rules for parameter definitions and selections to control the required digital radiographic image quality as described in the draft international standard ISO/DIS 10893-7. This draft is first standard establishing the parameters for digital radiography on weld seam of welded steel pipes for pressure purposes to be used on gas and oil pipelines.

Untersuchungen zur Optimierung der Bildqualität in der Computer-Radiografie für die Schweißnahtprüfung

Kurzfassung In der Filmradiografie kann durch den Einsatz von Metallfolien und Metallfolienkombinationen der Bildkontrast verbessert werden. Die Mechanismen zur Erzeugung der latenten Röntgenbilder und deren Sichtbarmachung für das menschliche Auge sind in Speicherfolien nicht mit denen in einem Röntgenfilm identisch. Demzufolge lassen sich die in den Normen zur Filmradiografie veröffentlichten Empfehlungen zum Einsatz von Metallfolien und zur Auswahl der Röhrenspannung nicht unmittelbar auf die Computer-Radiografie zur Schweißnahtprüfung übertragen. Die Qualität digitaler Röntgenbilder wird von der Bildunschärfe und dem Kontrast-zu-Rausch-Verhältnis (CNR) bestimmt. Während die Norm EN 14784-1 handhabbare Verfahren zur Bestimmung der inneren Unschärfe und des Signal-zu-Rausch-Verhältnisses (SNR) beschreibt, sind die in EN 14784-2 vorgegebenen Vorgehensweisen zur Kontrastverbesserung durch die Vorgaben zur geeigneten Röhrenspannung und Metallfoliendicke für die Schweißnahtprüfung nicht ausreichend dargelegt. Um die Parameter zur Verbesserung der Bildqualität von CR-Systemen für den Fall der radiografischen Durchstrahlungsprüfung von Schweißnähten zu optimieren, wurden systematische Untersuchungen zum Einfluss eines Kupfer-Vorfilters, von Bleifolien-Speicherfolien-Kombinationen und der Röhrenspannung auf den Bildkontrast von Stahlstufenaufnahmen durchgeführt. Auf der Grundlage dieser Ergebnisse wurden Empfehlungen zum Einsatz der Metallfilter und -folienkombinationen sowie zur Auswahl geeigneter Röhrengrenzspannungen für die Speicherfolientechnik zur Optimierung der Durchstrahlungsprüfung von Schweißnähten abgeleitet.