If most of the examples are reserved for the

This preview shows page 22 - 23 out of 39 pages.

If most of the examples are reserved for the definition of a good test set, there will not be enough examples left for the training set and vice versa. Therefore, random splits of the available data in training and testing sets and multiple trainings and testing of different size sets gives a better estimation of the classification error. If a simple linear classifier or a minimum distance classifier based on two-dimensional space (that is, using only two features) performs well, there is no reason to use anything more complicated. However, it frequently happens that such a classifier makes too many errors. Possible reasons include the following. 1. The features may be inadequate to distinguish the different classes. 2. The features may be highly correlated. 3. The decision boundary may be complex or there may be distinct subclasses in the data. In any case, proper estimation of classification error rate and classifier stability should be evaluated. Demonstration of Classifier Three 150 kHz resonant frequency transducers were mounted in a triangular pattern on a thick metallic plate measuring 320 × 540 mm (12.6 × 21.3 in.). A four-channel system was used for real time data acquisition and pattern recognition software was used for analysis and pattern recognition. Acoustic emission signals were simulated by 0.3 mm (0.012 in.) diameter pencil graphite breaks at various positions on the plate. Sliding a small metal piece across the surface of the plate created mechanical friction signals. Finally, electromagnetic interference signals were generated by unplugging the transducer cable during acquisition. The plot of amplitude versus rise time of Fig. 17 presents the sequence of experimentation. Because the experiment was performed in a controlled and repeatable manner, representative acoustic emission hits – examples — from each signal class were available and the associated pattern matrix could be saved for supervised pattern recognition. Because two different sources, hsu nielsen and electromagnetic interference, resulted in high amplitude hits (Fig. 17), a classification scheme based on amplitude alone would not perform. Similarly, class overlapping can be seen once rise time is considered alone. Because a single feature does not provide sufficient discrimination power for classifier development, the dimensionality of the feature space must be increased by at least one. Two-dimensional scatter plots were investigated such as the signal strength versus rise time and the energy versus amplitude (Fig. 18). Close observation of the scatter plot of signal strength versus rise time indicates overlapping of feature 163 Acoustic Emission Signal Processing F IGURE 17. Simulated acoustic emission used for examples of supervised pattern recognition.
Image of page 22

Subscribe to view the full document.

Image of page 23

{[ snackBarMessage ]}

What students are saying

  • Left Quote Icon

    As a current student on this bumpy collegiate pathway, I stumbled upon Course Hero, where I can find study resources for nearly all my courses, get online help from tutors 24/7, and even share my old projects, papers, and lecture notes with other students.

    Student Picture

    Kiran Temple University Fox School of Business ‘17, Course Hero Intern

  • Left Quote Icon

    I cannot even describe how much Course Hero helped me this summer. It’s truly become something I can always rely on and help me. In the end, I was not only able to survive summer classes, but I was able to thrive thanks to Course Hero.

    Student Picture

    Dana University of Pennsylvania ‘17, Course Hero Intern

  • Left Quote Icon

    The ability to access any university’s resources through Course Hero proved invaluable in my case. I was behind on Tulane coursework and actually used UCLA’s materials to help me move forward and get everything together on time.

    Student Picture

    Jill Tulane University ‘16, Course Hero Intern