Bundesanstalt für Materialforschung und -prüfung

International Symposium (NDT-CE 2003)

Non-Destructive Testing in Civil Engineering 2003
Start > Contributions >Lectures > Acoustic Emission: Print

Automatic analysis of acoustic emission measurements on concrete

Jochen H. Kurz, Florian Finck, Christian U. Grosse,
Hans-Wolf Reinhardt
University Stuttgart, Institute of Construction Materials, Pfaffenwaldring 4, 70550
Stuttgart, Germany, Phone: +49 - 711 - 6856792, Fax: +49 - 711 - 6856796
Email address: kurz@iwb.uni-stuttgart.de (Jochen H. Kurz,).

Abstract

Acoustic emission data contains a lot of information about the fracture process it was emitted from. A quantitative acoustic emission analysis permits detailed investigations on the fracture mechanics and the failure process. But often there are several thousand events from one fracture area. Doing the data analysis by hand is a very time consuming process and therefore often only a rudimentary analysis is performed. Furthermore, the data's signal to noise ratio is often very low. Therefore, the development of an automatic analysis method containing data conversion, de-noising, localization, moment tensor inversion and other features from the field of statistical mathematics is indispensable. The general analysis concept is based on the model-view-controller (mvc) paradigm which is realized by using different development environments. Therefore, the software development packages MATLAB and LabVIEW were used. The connection of these two programmes via so called 'Formula Nodes' makes a fast development possible because the advantages of both packages can be used. Furthermore, external code (source code or dynamic link library) can be linked to the analysis environment. A good balance between transparency and automation of the processing steps themselves and the entire acoustic emission analysis should be reached.

Key words:
acoustic emission, automation, mvc paradigm, data processing

1 Introduction

Acoustic emissions are defined as the spontaneous release of localized strain energy in stressed material. Due to micro cracking in the material this energy release can be recorded by transducers on the material's surface (Grosse, 2002). Acoustic emission analysis is capable of revealing damage processes in materials during the entire load history. It is obvious that the recording of damage processes from the microscopic to the macroscopic scale produces large data sets even during relatively short time spans. The number of events can be about several thousand during one test. The high number of acoustic emissions is one fact that needs to be considered during the analysis procedure, due to the signal's often low signal to noise ratio, several further data processing steps are needed before interpretable results are gained (Fig. 1).

Fig 1: Principle sketch of the signal based acoustic emission analysis. The experiment provides raw data of the fracture process which needs to be converted and conditioned before a 3-dimensional localization or a statistical analysis is possible. Finally a moment tensor inversion can be calculated which leads to the fracture mechanism that initially generated the acoustic emissions.

The automation of the data processing steps is a highly preferable status. Especially signal conditioning and onset detection are very time consuming. However, acoustic emission analysis means obtaining information about the fracture process from the recorded signals. But there are different stages of information content like localization which says where the fracture occurred, statistical analysis which permits several statements about the development of the fracture and moment tensor inversion which gives an image of the fracture process itself. Each stage needs different numerical algorithms. Often the stages synthesize one another e. g. the moment tensor inversion needs the localization results, the sensor coordinates and the onset amplitudes. The applied numerics are generally non-trivial, that means error search within a fully automatic process is really difficult. Therefore, a good balance between transparency and automation of the processing steps themselves and the entire acoustic emission analysis is needed.

A conceptual way of realization is the application of the model-view-controller paradigm (mvc paradigm) which is originally a powerful architecture for graphicaluser- interfaces (GUIs) (Buschmann et al., 1996). But the mvc paradigm is also in a general sense helpful for structural concepts regarding the interaction of programs or program parts. This paradigm is realized in the self written software and scripts which themselves represent automatized parts of the whole acoustic emission analysis and in the interaction schemes of these programs. A detailed description of the mvc paradigm and its application in the above described sense will be given in the following.

2 Composition of the automation

The mvc paradigm is in the GUI programming sense a way of breaking an application, or even just a piece of an application's interface, into three parts: the model, the view and the controller. The view manages the graphical and/or textual output to the portion of the bitmapped display allocated to its application. The controller interprets the mouse and keyboard inputs from the user, commanding the model and/or the view to change as appropriate. The model manages the behavior and data of the application domain, responds to request for information about its state, and responds to instructions to change state (Buschmann et al., 1996). Using this strict separation a high flexibility is gained because the application of the mvc paradigm cares for decoupling model, view and controller. This leads to a high reusability, scalability and maintainability of the application.

The mvc paradigm is used for the structure of the self developed software using LabVIEW (all2sdf) or wx Windows (WinPecker) which will be described in the next paragraph. Furthermore, the mvc structure is also found in the Matlab scripting procedure. This is the common usage of the mvc concept. But due to the mentioned aim that the best balance between transparency and automation of the processing steps themselves and the entire acoustic emission analysis should be reached the mvc paradigm is also applied to the interaction structure of the different programs. That means the user is able to control and to interfere the interfaces of the different programs. The complexity of the acoustic emission analysis and the variety of applicable mathematical concepts lead to such a proceeding because a coherent interaction with the user is needed.

2.1 Software development packages
The used development environments can be summarized in the following form:

  • LabVIEW is a graphical programming language that uses icons instead of text.Furthermore, LabVIEW uses dataflow programming where data determine execution.The programming structure of building virtual instruments VIs (subprograms) makes complex but still comfortable programs possible (Jamal and Hagestedt, 2001).
  • wx Windows is a C++ library which allows software development for several platforms. It's main target is the programming of the Graphical User Interface (GUI) (WXWINDOWS: http.//www.wxwindows.org, 2003).
  • Matlab is a high-performance matrix/array language for technical computing.Especially the Matlab functions (M-files) which are of script structure make applications on data sets very comfortable (The Math Works, Inc., 2000).
  • The principal example for used numerical code written in fortran is a relative moment tensor inversion approach by Dahm (1996).

The software and development packages are used to write stand alone programs which are applied to parts of the data processing and signal conditioning of the acoustic emissions. These parts are processed automatically. That means the steps work in an automatic way and are set together to form the complete analysis. The developed software itself is built up with the mvc structure and will be discussed in the next section in detail as well as the interaction schemes.

3 Implementation of the processing steps

The automation of the working steps like data conversion and signal conditioning is one important fact. But the interaction of these steps is also grave because it finally defines the degree of automation of the whole analysis (Fig. 2). The pre-condition concerning the acoustic emission analysis is that a relative high degree of transparency is needed, on the one hand for error detection during the complex single processes and on the other hand to read out parts of the results or to implement new or modified algorithms quickly during data processing.As mentioned above the best balance between transparency and automation of the processing steps themselves and the entire acoustic emission analysis is needed. As stated in the last paragraphs a fully automatic acoustic emission analysis is not preferable because the entire analysis is not reliable using a black box process. The application of the chosen mvc paradigm from the software development level to the overall interaction structure is a good way to combine transparency and automation. Concerning the principal interaction scheme the mvc paradigm can be understood in the following way:

m: Different models i. e. Matlab scripts for statistical analysis, onset detection algorithms or wavelet de-noising/filtering algorithms can be chosen. That means they are available in some kind of modular form.

v: The results of modifications must be uncovered from the stage where they were applied. The possibility regarding the visualization of the different analysis procedure levels is maintained.

c: The user is able to reprocess parts of the results. That means not the whole analysis must be repeated, only that part starting from the point where the modifications were done.

Fig 2: Flowchart of the data processing steps. The interaction of the different programs/ development environments is described through arrows. Especially the good interaction skills of the Matlab - LabVIEW formula node allows a high degree of automation of the routines handled by these two environments.

This is the superior structure of the different routine's interactions. The developed software and their contribution to the reached degree of automation in detail will be discussed in the next subsections.

3.1 Data maintenance and signal conditioning
Non destructive testing using acoustic emissions produces large data sets. Several analysis procedures need to be applied to the raw data before onset detection and localization using the self developed program WinPecker is possible. Therefore, a program based on the development environment LabVIEW called all2sdf (Fig.3) was evolved. Often the raw data is in a binary format which cannot be read by the onset detection software or which is wanted in ascii format. Data conversion is the first step which must be processed. This is done automatically by all2sdf. Within the conversion process signal conditioning is implemented. The so called formula node makes a direct interaction of LabView and Matlab possible. That means the whole Matlab functionality regarding signal conditioning (filtering and de-noising) is used. The signal conditioning can be added to the conversion procedure via a simple switch (Fig. 3). Due to the often low signal to noise ratio and disturbances (low frequent) of the testing apparatus this procedure is generally applied (Kurz et al., 2003). The Matlab script parameters for the formula node must be implemented directly in the source code. An adaptive algorithm for filtering and de-noising is under development but is also a topic that will still be prevailing on the field of data processing in the future.

Fig 3: Screen shot of the all2sdf program which makes the data conversion for the further processing steps with an included signal conditioning possibility.

Furthermore, the conditioned or the raw data can be written into a Matlab array database. This is especially interesting for statistical analysis techniques using the whole Matlab functionality. Using another simple switch the user can choose if the creation of a data base is wanted or not. This data base is of matrix format and is finally stored in the same directory as the other data. The access on the database is arranged by the use of Matlab scripts which contain the access, the mathematical and the visualization features. This is done with a high degree of automation.Data maintenance and signal conditioning form the base for further analysis procedures.In the present shape fast and efficient access on large data sets is possible and transparency of the different steps is maintained.

3.2 Onset detection, localization and external code
The existence of a database is important for statistical analysis. Another not minor important analysis feature is the localization of the acoustic emission events. The recently rewritten programWinPecker (now version 2.0) which is also based on the mvc paradigm is used for automatic onset detection and 3-dimensional localization (Fig. 4). Arrivals in acoustic emission signals are often characterized only by increasing energy and not spectral changes. This effect is mostly related to the sensor characteristic of the used acoustic emission transducers because signal and noise portions of the data are of the same frequency range and difficult to discriminate from one another. The chosen automatic onset detection algorithm is using the partial energy as a statistic (Hinkley criterion) for detection of first arrivals (Grosse,2000). After a successful signal conditioning automatic onset detection is possible but nevertheless further algorithm development is practiced.

Fig 4: Screen shot of WinPecker 2.0. The new GUI facilitates the software handling and is more stable than the older versions.

Within the WinPecker software a 3-dimensional localization routine called HYPO AE (Oncescu and Grosse, 1998) is implemented. The results are visualized using Matlab. The localization is often essential for the further proceeding. The determination of the fracture mechanism using the relative moment tensor inversion requires cluster and amplitude determination (Dahm, 1996). The moment tensor inversion itself is numerically elaborate and therefore, this process is sourced out in some way from the rest of the routine. The cluster determination of the localized events is done within Matlab. From some events of the clusters the amplitude of the onsets first half wavelet is needed. This is again done using the WinPecker software.

4 Discussion and Conclusion

Finding the best balance between transparency and automation of the acoustic emission analysis is a non-trivial problem. A fully automatic processing concept for the acoustic emission analysis of the presented form is in our opinion not reasonable at the moment. Signal based acoustic emission analysis is too complex for the use of a black box routine because many different processing steps are involved. Error detection in such a process is unpromising. But the reached degree of automation covers all steps which would be very time consuming during manual data processing like data conversion, signal conditioning, onset detection and statistical analysis.The power of our concept is that the whole analysis procedure is adjustable to the experiment. The application of the mvc paradigm guarantees transparency of and user interaction with the analysis routines. Furthermore, the conceptual structure enables further development and completion of the automatic acoustic emission analysis in the present form. A fully automatic analysis would require highly adaptive algorithms at the majority of the different processing step but they are currently not available. The flexibility of the mvc approach regarding the automatic analysis of acoustic emissions is especially useful for concrete which represents due to aggregates, pores and reinforcement a uniformly distributed heterogeneous material.The automatic acoustic emission analysis shown in this article is optimized for the application on concrete where the specimens allow 3-dimensional localization of the recorded events and thus the moment tensor inversion. But due to the mvc conceptual structure practice on other materials should be a relatively easy tackling problem.

5 Acknowledgements

The funding of this work by the Deutsche Forschungsgemeinschaft is gratefully acknowledged. The helpful discussions with the members of the SFB 381 and of our whole working group must be mentioned here, too.

References

  1. Buschmann, F., Meunier, R., Rohnert, H., Sommerlad, P., Stal, M., 1996. Pattern- Oriented Software Architecture, Volume 1: A System of Patterns. Vol. 1. John Wiley & Son Ltd., New York, USA.
  2. Dahm, T., 1996. Relative moment tensor inversion based on ray theory: theory and synthetic tests. Geophysical Journal International 124, 245 - 257.
  3. Grosse, C., 2000. Winpecker version 1.2. Instruction manual, University Stuttgart,Stuttgart.
  4. Grosse, C. U., 2002. Basics of acoustic emisson measurement techniques. Kluwer Academic Publishers, Hingham MA, USA, Ch. 9, p. 45, in print.
  5. Jamal, R., Hagestedt, A., 2001. LabView: Das Grundlagenbuch, 3rd Edition.Addison-Wesley,M¨unchen, Germany.
  6. Kurz, J., Ruck, H.-J., Finck, F., Grosse, C., Reinhardt, H.-W., 2003. Wavelet algorithms for non destructive testing. NDT-CE, 2003 Berlin (see same proceedings).
  7. Oncescu, L., Grosse, C., 1998. HYPO AE - A program for the localization of hypocenters of acoustic emissions. Dokumentation zur Software f¨ur PC und Workstation Rev. 2.1, University Stuttgart, Stuttgart.
  8. The Math Works, Inc., 2000. Matlab: The Language of Technical Computing, 6th Edition. Vol. 12. The Math Works, Inc., Natic MA, USA. WXWINDOWS: http.//www.wxwindows.org, 2003. Tech. rep.
STARTPublisher: DGfZPPrograming: NDT.net