Bearing fault detection and oil debris monitoring by adaptive noise cancellation

Bearing fault detection is critical in preventing machine failure and ensuring machine is operating in optimal condition. However, the monitored bearing vibration signal is often corrupted by interference signals from other sources in the machine. Adaptive noise canceller can be used to extract the bearing signal from the corrupting interference signals, thus enable bearing fault detection. Adaptive noise canceller utilizing least mean square algorithm is also suitable for on-line monitoring because of its computational efficiency.;The effectiveness of the adaptive noise canceller is controlled by its operating parameters: the transversal filter length and the step-size parameter. This thesis establishes the criteria in determining the proper operating parameters. The Akaike information criterion is used to obtain the transversal filter length because of the adaptive noise canceller's autoregressive structure. The adaptive filter weight fluctuations affect the performance of the adaptive noise canceller. The misadjustment positively correlates to the filter weight fluctuations. A small misadjustment value can be used to obtain the required step-size parameter to ensure the satisfactory performance of the adaptive noise canceller. A procedure to achieve zero settling time in on-line monitoring under the stationary environment is also illustrated in this thesis. Simulation and experiments are performed to demonstrate the effectiveness of the adaptive noise canceller in bearing signal extraction.;Inductive oil debris monitoring sensor is used to detect metal particles in the lubricating oil. However, the sensor is affected by the interfering vibration signal and the metal particle signatures can no longer be correctly detected from the resulting mixed signal. In order to obtain reliable metal particle counts and sizes, it is critical that the interfering vibration signal is removed.;The metal particle signal in the oil debris monitoring signal is non-periodic and can be considered a broadband signal. Thus a special form of the adaptive noise canceller which does not require a separate reference input source can be used. A delay is incorporated to the collected signal to derive the reference input. A delay value longer than the length of the characteristic output signal sampling units is needed to achieve the decorrelation between the metal particle signal in the primary input and the metal particle signal in the reference input. The criteria to determine the operating parameters established in bearing fault detection application can also be implemented in oil debris monitoring application. Simulation and experiments are performed to verify that the adaptive noise canceller can effectively remove the interfering periodic signals and reveal the metal particle signal from the corrupted signal.