Bearing fault diagnosis based on information fusion

Bearing failure is one of the primary causes of breakdown in rotating machinery. Such failure can result in costly downtime and catastrophic consequence, especially for aircraft propulsion systems. Therefore bearing health estimation based on vibration signals is critical to increased safety and reductions in life-cycle cost. Despite the success in indicating the incipient bearing fault and inferring the bearing fault type, current fault diagnosis algorithms fail to effectively estimate the bearing fault severity due to complex failure modes and changing operating conditions. The bearing dynamic response to damage determines the resulting vibration pattern, and consequently the bearing vibration pattern change exhibits the bearing fault progression. Therefore incorporating the temporal information of the bearing vibration pattern is vitally important to indicate the bearing damage level. In order to achieve improved accuracy for bearing health estimation, the paper presents a procedure for bearing fault diagnosis through incorporating the dimension-temporal information of multiple vibration features extracted from vibration signals by both time and frequency domain techniques. The procedure effectively combines the mutually complementary dimension information of multiple vibration features with the temporal information obtained by trending analysis from these features. The bearing fault progression data from bearing experiments validates that the proposed procedure is effective and robust to indicate the bearing health condition.