A Method for Minimum Sample Size Calculation for a Classification Problem in Fault Diagnosis of Centrifugal Pump

Monoblock centrifugal pump plays a key role in various applications. Any deviation in the functions of centrifugal pump would lead to a monetary loss. Hence a condition monitoring and fault diagnosis has become very essential for centrifugal pump maintenance. Of late, vibration signals based machine learning approach to fault diagnosis is gaining momentum. The important two activities involved in machine learning approach are training and testing the classifier. Choosing number of samples to train the classifier in order to get good classification accuracy is still a challenging task. Engineers do this activity heuristically or arbitrarily. This paper proposes a systematic method to determine the minimum sample size using power analysis and the results are validated using a decision tree algorithm namely J48.