achieved using commercially available, products. The system was evaluated on field-removed samples with ~rvice-induced IGSCC and is currently being evaluated by utilities. The paper describes the development of the expert system. 48091 Bridgeman, J.; Theisen. K.; Whang, B.; Shankar, R. An expert system for erosion/corrosion inspections Proceedings of the lOth International Conference on NDE in the Nuclear and Pressure Vessel Industries, Glasgow (Scotland), i 1-14 Jun. 1990. pp. 569-572. Edited by M.J. Whittle, J.E. Doherty and K, lida. ASM International, (1990) Carbon and low alloy steel power piping may experience Erosion/Corrosion (E/C). E/C is an accelerated form of corrosion which causes pipe thinning. NSP is developing an Expert System to ~lect E/C inspection locations and to manage E/C inspection results. The Expert System uses rules of thumb developed by experienced engineers over the cour~ of a number of inspections. Desirable aspects of an E/C Inspection Expert System are: Identification of E/C sumeptible components (avoid piping failure): Eliminate non-susceptible components (avoid unneeded: Identify thinned components from inspection data: Provide reinspection interval guidance: Allow others to function as experts: Maintain expertise during job turnover.
Sasahara, T.; Yoneyama, H.; Arakawa, T. Applications of new NDE techniques for the nuclear industry
48059
Proceedings of the 10th International Conference on NDE in the Nuclear and Pressure Vessel Industries, Glasgow (Scotland), 11-14 Jun. 1990. pp. 759-763. Edited by M.J. Whittle, J.E. Doherty and K. lids, ASM International, (1990) This paper summarizes NDE techniques developed for new applications in the nuclear industry. The new techniques discussed are, ultrasonic transducer and couplant to monitor components under extremely high and low temperature conditions, ultrasonic technique applied for weld examination of thin structure ultrasonic technique for crack discrimination using frequency analysis and neural networks and eddy current technique for deeper current penetration and examination of weld.
Barga, R.S.; Friesel, M.A.; Melton, R.B. Classification of acoustic emission waveforms for nondestructive 48042
evaluation using neural n e t w o r k s Proceedings of the Conference on Applications of Artificial Neural Networks, Orlando, Florida (United States), 18-20 Apr. 1990. pp.545- 556. Edited by S.K. Rogers. SPIE, Vol. 1294 (1990) Neural networks were applied to the classification of two types of acoustic emission (AE) events, crack growth and fretting, from a simulated airframe joint specimen. Signals were obtained from four .sensors at different locations on the test specimen. Multilayered neural networks were trained to clagsify the signals using the error backpropagation learning algorithm, enabling AE events arising from crack growth to be distinguished from thorn caused by fretting. In this paper we evaluate the neural network classification performance for sensor location dependent and ~nsor location independent training and testing sets. Further, we present a new training strategy which significantly reduces the time required to learn large training sets using the error backpropagation learning algorithm, and improves the generalization performance of the network.
Pellionisz, P.; Szendro. S. Development of an expert system for acoustic emission testing
48041
Central Research Institute for Physics, Budapest (Hungary), KFKI-199131/L, 20 pp. (1991) Application of acoustic emission techniques and the data evaluation in nondestructive testing requires special expertise. For facilitating this task, the development of an expert system named AE DATA EXPERT has been commenced. The system is supported by a mt of knowledge base modules - called ~parately or in a chained mode - and cooperates with the AE mflware package Defproc. After a brief summary of the principles and discussing recent publications, system considerations are presented and the first knowledge base module is introduced. 47909 Windsor, C.G. The classification of defects from ultrasonic measurements United Kingdom Atomic Energy Authority, Harwell, AERE-R-13306 10 pp. (Nov. 1988) The Rumelbart back propagation and Hopfield algorithms have been used to classify the nature of the defects found in steel pressure vessels. Ultrasonic data from some g3 defects of known type had previously been processed into six feature parameters demribing the ultrasonic reflections. Both neural network methods performed as well as a weighted minimum distance algorithm in classifying the defect type. 47857 Andersen, A.; Kragh, E. An intelligent input to inspection planning on offshore platforms taking gross h u m a n e r r o r s into account Proceedings of the 10th International Offshore Mechanics and Arctic Engineering Conference, Stavanger (Norway), 23 - 28 Jun. 1991, Vbl. 2, Safety and Reliability, pp. 297-303. The American Society of Mechanical Engineers ( 1991 ) Major defects in a structure are often caused by human errors committed by designers, contractors, and operators. In this paper, a new method of coping with human errors is introduced. The method is based on the following idea. During the design, construction, operation, and maintenance pba~s, gross human errors may give rim to various types of serious defects in the structures. A correlation matrix may
40
therefore be defined relating human errors to structural defects. Based on an updated matrix including experience in terms of reports, a number of potential defects and their location may be suggested as a response tO a detected defect. A prototype stand alone expert system, GERM, has been developed to illustrate and verify the method. Experience with the prototype has so far been promising. The aim of this paper is to present the philosophy behind the method and the experience obtained from the prototype. 47854 Galliard, A.; Wunsch, D.C.; Escobedo, R.A. Neural hypercolumn architecture for the preprocessing in radiographic weld images Proceedings of the Conference on Applications of Artificial Neural Networks, Orlando, Florida (United States), 18-20 Apr. 1990. pp.378- 388, Edited by S.K. Rogers. SPIE, Vol. 1294 (1990) A general neural hypercolumn architecture is applied to radiographic weld images to locate regions of strong spatial intensity gradients. The hypereolumn output provides information on both the direction and the orientation of local spatial intensity gradients. These outputs can also be used to form an enhanced decimated image which can be proces~d for feature recognition. Parametric tuning of the architecture is discussed, with particular emphasis on the requirements of the application. The performance of this architecture is compared with that of Sobel filters and other edge-detecting convolution masks. The possible representation of these various discrete convolution masks - including hypercolumns - as generalized non-adaptive neurons is also dimussed.
Royer, J.C.; Merle, A.; de Sainte Marie, C. An application of machine learning to the problem of p a r a m e t e r setting in non-destrnctive testing
47847
Proceedings of the 3rd Conference on Industrial and Engineering Application of Artificial Intelligence and Expert Systems, Charleston, Tennessee (United Stales), 15-18 Jul. 1990, pp. 972-980. Edited by M.M. Mathews. Association for Computing Machinery (i 990), ISBN 0897913728 This article presents an aid system for the setting of non-destructive testing instruments. Some problems inherent in this field are briefly discussed, before showing how they led us to introduce machine learning techniques into the system. The approach u~s learning from examples. The goal of the learning module is to determine dependencies between parameters of different experiments in order to automatically generate a set of rules. A prototype, called MANDRIN, has been implemented and is being evaluated on a real application: an X-ray tomograph. The first results are presented in the last ruction.
Ahraad, K.; Langdon, A.; Frieze, P.A. An expert system for offshore structure Inspection and maintenance
47846
Computers & Structures, VoI. 40, No. 1, pp. 143-159 (1991) The full extent of knowledge required to completely rationally execute inspection and maintenance of offshore structures is beyond the scope of one individual without the assistance of a comprehensive system with access to the necessery databases and algorithms. The PLAIM (Platform Lifetime assessment through Analysis, Inspection and Maintenance) project has been established to provide such a system. 'KnowledgeBased' Expert Systems methods and techniqugs, particularly those related to the acquisition of experiential knowledge and representation of the knowledge for solving problems, have been used in the PLAIM project. The major deliverables of this project include the documentation of informal and qualitative knowledge together with a prototype expert system for eliciting details of defects in structures, empirically assessing the severity of the defect and recommending a series of remedial actions.
McNab. A.; Dunlop, I. Information technology - IT in NDT
47843
British Journal of Non-Destructive Testing, Voi. 33, No. 12, pp. 611- 615 (Dec. 1991) An assessment of the impact of Information Technology on NDT is given. Particular areas which are highlighted include the use of ultrasonic arrays for flaw imaging. An NDT workbench based on a aandard workstation is proposed for performing a number of tasks including the imaging of flaws in three dimensions. Artificial Intelligence techniques which include Knowledge-Based Sy~ems and Neural Nets are described in the context of flaw cheracterisation. Finally, the importance of standard, NDT-data formats for information exchange is discussed in the context of interfacing NDT systems to global industrial networks. 47792 Szyszko. S.; Payne, P,A. Artificial neural networks for feature extraction from acoustic emission signals Colloquium on Measurements, Modelling and Imaging for Non-Destructive Testing. London (United Kingdom), 27 Mar. 1991. pp. 6/1-6/6. lEE (1991) Digest No. 1991/054 Production proces~s can be probed non-inv&sively using acoustic emission (AE) nondestructive testing. Complex pattern recognition techniques are neaxled to classify AE features and correlate them with changes in the proce.-,abeing monitored. This can be done using artificial neural networks (ANN) which have proved to he economic in terms of computer facilities and effective at classification of defects and AE features extraction. How these ANN were developed is the subject of this article. 47776 Kaswasaki Jukogyo KK Automatic weld defect image d a t a extraction f r o m r a d i o g r a p h i c film
images of pipe welds utillsing computer data base based on inspector experience European Patent No. 437,280 (17 Jul. 1991)
NDT&E International Volume 25 Number 1 1992