If most of the examples are reserved forthe definition of a good test set, there willnot be enough examples left for thetraining set and vice versa. Therefore,random splits of the available data intraining and testing sets and multipletrainings and testing of different size setsgives a better estimation of theclassification error.If a simple linear classifier or aminimum distance classifier based ontwo-dimensional space (that is, using onlytwo features) performs well, there is noreason to use anything more complicated.However, it frequently happens that sucha classifier makes too many errors.Possible reasons include the following.1. The features may be inadequate todistinguish the different classes.2. The features may be highly correlated.3. The decision boundary may becomplex or there may be distinctsubclasses in the data.In any case, proper estimation ofclassification error rate and classifierstability should be evaluated.Demonstration of ClassifierThree 150 kHz resonant frequencytransducers were mounted in a triangularpattern on a thick metallic platemeasuring 320 ×540 mm(12.6×21.3 in.). A four-channel systemwas used for real time data acquisitionand pattern recognition software was usedfor analysis and pattern recognition.Acoustic emission signals were simulatedby 0.3 mm (0.012 in.) diameter pencilgraphite breaks at various positions on theplate. Sliding a small metal piece acrossthe surface of the plate createdmechanical friction signals. Finally,electromagnetic interference signals weregenerated by unplugging the transducercable during acquisition.The plot of amplitude versus rise timeof Fig. 17 presents the sequence ofexperimentation. Because the experimentwas performed in a controlled andrepeatable manner, representative acousticemission hits – examples — from eachsignal class were available and theassociated pattern matrix could be savedfor supervised pattern recognition.Because two different sources, hsunielsen and electromagnetic interference,resulted in high amplitude hits (Fig. 17), aclassification scheme based on amplitudealone would not perform. Similarly, classoverlapping can be seen once rise time isconsidered alone. Because a single featuredoes not provide sufficient discriminationpower for classifier development, thedimensionality of the feature space mustbe increased by at least one.Two-dimensional scatter plots wereinvestigated such as the signal strengthversus rise time and the energy versusamplitude (Fig. 18). Close observation ofthe scatter plot of signal strength versusrise time indicates overlapping of feature163Acoustic Emission Signal ProcessingFIGURE17.Simulated acoustic emission used for examples ofsupervised pattern recognition.
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.
Temple University Fox School of Business ‘17, Course Hero Intern
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.
University of Pennsylvania ‘17, Course Hero Intern
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.