Abstract:
Fault prediction and maintenance of equipment plays a critical role in today's industry. Failures
should be identified as soon as possible because if the equipment continues to run under
abnormal conditions, it could damage equipment and the employees. Many of the previous
studies were focused on developing mathematical model predicting the machine's operating
conditions. During past years, machine learning techniques have been widely applied in
machine fault diagnoses. Many studies have demonstrated that analyzing the vibration
spectrums of machines can be a useful tool for detecting faults at early stage. Moreover, analysis
of acoustic waves generated from machines could also be a reliable methods in detecting faults.
This study is focused on developing an external data acquisition unit and fault detection system
that utilize acoustic waves produce by the machine in identifying the potential faults. Further,
the proposed unit is a portable unit that can be used in wide range of machines. The primary
objectives of this research are to acquire acoustic signals at higher sampling rate and analyze
the data using machine learning algorithm. In this study acoustic data is collected using a sound
sensor module and Arduino development board. However, the vibration signal could not be
detected accurately at higher sampling rates with the KY-037 acoustic sensor module and
Arduino Thus, Case Western University bearing data are used in machine learning algorithm
to study the accuracy of the fault prediction. The data is converted to frequency domain and it
is extracted the frequencies and amplitudes of dominant peaks as the main features for the
machine learning algorithm. Those selected features are analysed using K-Nearest Neighbor
and Support Vector Machine classification algorithms. The overall accuracy of 90.1% and
81.8% were obtained using KNN and SVM algorithms.