Bearing damage detection based on statistical discrimination of stator current

Feature representation and discrimination based on Gaussian mixture model probability densities—Practices and algorithms

“…The same approach can be used in two class problems where training examples are available only for one class and observations which do not pass a predefined confidence level are assigned to the unknown class (e.g. in detection of motor failure conditions [12]). In that sense the confidence does not refer to a single value limit but to a certain allowed region in the feature space.…”

“…[11,12]), the assumption of single component leads to strict requirements for characteristics of the phenomenon: a single basic class which smoothly varies around the class mean. The most significant problem is not typically the smooth behavior but the assumption of unimodality.…”

Recognition of Human Iris Using Accurate Iris Map

“…GMMs have been widely used in all kinds of pattern recognition problems, including speech processing [21], human skin detection [22], realtime tracking [23], hazardous chemical agents detection [24], and bearing damage detection for induction motors [25]. The advantage of GMM is its modeling ability.…”

Toward Automatic Motor Condition Diagnosis

“…The first-order statistics approach is not sufficient since it simply selects the frequency band where the distance between the expectations is largest, but neglects the variance information, and thus, a significant overlap of the class probabilities may exist [5,4]. In the second-order statistics typically an assumption must be made about forms of probability distributions p x (n) and p y (n) which describe the spread of features of the both classes.…”