News Center

Home

Current Status and Development Trends of Research on Condition Monitoring of Aeroengine Spindle Bearings (III)

Release time:

2023-01-13

Source:

Journal of Aerodynamics

The information fusion method is actually a simulation of the comprehensive processing of complex problems by the human brain. In a multi-sensor system, the information provided by various sensors may have different characteristics. By fully utilizing multiple sensor resources in space and time, various observation information is reasonably controlled and used, and redundant and complementary information is combined according to certain criteria, resulting in a consistent description of the observation object and a new fusion effect. Based on the independent observation information of each sensor, by optimizing the combination of information, more effective information is derived, ultimately improving the effectiveness of the entire system. The signal fault characteristics of spindle bearings are weak and the background noise is strong, and information fusion methods have good advantages in processing this type of signal.

3. Multi sensor signal analysis method based on information fusion

3.1 Characteristics of Information Fusion Methods

The collection of information by a single type of sensor is limited, and the use of multiple sensors can complement information and improve the accuracy of health status monitoring. Duan et al. [77] analyzed the advantages and disadvantages of different types of condition monitoring technologies and believed that multi-sensor information fusion is the future development trend of mechanical equipment condition monitoring; Lin Tong et al. [78] proposed a multi feature fusion evaluation method based on standardized Euclidean distance, which was proven to be superior to principal component analysis (PCA) and support vector data description (SVDD) methods through experiments.

According to the level of data abstraction, fusion can be divided into three levels: data layer fusion, feature layer fusion, and decision layer fusion.

The advantage of data layer fusion is that it directly fuses the observed values of sensors, which has accuracy that other hierarchical methods cannot achieve. The disadvantage is that it requires a large amount of computation and cannot perform heterogeneous data fusion. The main algorithms include linear weighted algorithms, Kalman filtering [79] methods, etc.

The feature layer fusion calculates a feature vector that represents the observed values of each sensor and performs fusion processing on this vector. The advantage is that this type of method achieves considerable data compression; The disadvantage is that subtle information in the original data may be lost. The main algorithms include kernel principal component analysis (KPCA), support vector machine (SVM), neural network, k-nearest neighbor (kNN) classification algorithm, etc.

Decision layer fusion obtains system decisions by fusing the decisions of each sensor. The advantage is that it requires low computational performance and does not require the collection device to be a sensor of the same type. The disadvantage is that the loss of original data is large and the performance of minor factors is not obvious. The main algorithms include expert systems, Dempster Shafer (D ⁃ S) evidence theory [80 ⁃ 81], etc.

3.2 Main algorithms

The Development of Information Fusion Methods and the Continuation of Mathematical Theory