A machine component monitoring system using audio acoustic signals.

The main objective of this study is to develop a new type of machine-component monitoring system which is non-intrusive and non-contact in nature. Moreover, the design of the system to be developed must be robust enough for it to be implemented in an industrial environment. Therefore, this study was initiated to overcome some of the problems that were encountered using the well-established vibration method. For instance, vibration measurement of a machine component is dependent on the quality of contact between an accelerometer with a vibrating surface. Vibration measurement of a machine component is also affected by the vibration of other machine components near the vicinity, in addition to the presence of power-supply-line frequency and its harmonics. On the other hand, the application of a desirable non-intrusive and a non-contact nature of sound pressure measurement method is difficult to carry out if the background sound level is high. This is because sound pressure measurement is dependent on the characteristics of a sound field where a measurement is carried out. For these reasons, air-particle acceleration signals were utilised in the study. Air-particle acceleration is a vector quantity and measurement of vector property can improve the signal-to-noise ratio of the measured signal, even in a noisy environment.A dedicated test rig was constructed to carry out the experiments and to test the hypothesis. Rolling element bearings were used for the experiment because of the many different types of defect that can develop in them, such as inner race, rolling element and outer race defects. Moreover, the dynamic behaviour of bearings are well understood and can be compared with experimental results obtained from the study. Several different methods of analysis were used in the study including statistical, spectral, cepstral and wavelet transform methods. The results from using air-particle acceleration signals were compared with results obtained from utilising sound pressure and vibration signals. These results showed that the performance from using air-particle acceleration signals were superior to the performance from using sound pressure signals. Results from the analysis of air-particle acceleration signals can clearly indicate the presence of a defective component in the test-bearing. This is so even when the overall background noise was 14dB higher than the overall noise level emitted by the test-bearing. Moreover, the sensitivity of the measurement of air-particle acceleration signal to indicate the presence of a defective bearing was similar to the sensitivity when using conventional vibration equipment.Applications of artificial neural networks were also included for automatic identification of defect signals. The multilayer perceptron network was chosen and tested to classify the bearing signals because of the suitability of this type of network to be used for pattern recognition. Finally, a new type of machine-component monitoring system using air-particle acceleration signal was successfully developed and tested in industry.