An Intelligent Machine Condition Monitoring System Using Time-Based Analysis: Neuro-Fuzzy Versus Neural Network

Monitoring and predicting machine components' faults play an important role in maintenance actions. Developing an intelligent system is a good way to overcome the problems of maintenance management. In fact, several methods of fault diagnostics have been developed and applied effectively to identify the machine faults at an early stage using different quantities (Measures or Readings) such as current, voltage, speed, temperature, and vibrations. In this paper, an intelligent machine condition monitoring and diagnostic system is introduced with experimental verification. An adaptive neuro-fuzzy inference system (ANFIS) and a neural network system (NN) are used to monitor and predict the fault types of a critical mechanical element in the Potash industry (namely; a Carnallite surge tank pump). The system uses a piezoelectric accelerometer to generate a signal related to machine condition and fault type. Combinations of the vibration time signal features (i.e., root mean square, variance, skewness, kurtosis, and normalized sixth central moment) are used as inputs to both ANFIS and neural nets, which in turn output a value for predicted fault type. Experimental validation runs were conducted to compare the actual fault types with the predicted ones. The comparison shows that the adoption of the time root mean square and variance features achieved the minimum fault prediction errors for both ANFIS and neural nets. In addition, trapezoidal membership function in ANFIS achieved a fault prediction accuracy of 95%, whereas, a cascade forward back-propagation neural network achieved a better fault prediction accuracy of 99%. © 2009 Jordan Journal of Mechanical and Industrial Engineering. All rights reserved