A New Multi-Sensor Information Fusion Technique Using Processed Images: Algorithms and Application on Hydraulic Components

Multi-sensor fusion technique is used to combine the complementary information source from the multiple sensors. However, the multi-sensor data are obviously different with the characteristics of complex types, different dimensions or different weights, which is easy to cause the difficulty of the fusion and the decline of the ability of information representation although the fault information is enriched. Therefore, a new multi-sensor information fusion technique using processed images is proposed. The core of this technique is to convert the information from different sensors (especially for heterogeneous sensors) into images for weighting feature matrix and constructing image fusion to realize fault diagnosis. In the technique, the processed images can enhance the weak signal in a complex environment and avoid the weak applicability caused by multi-sensor sampling differences. The proposed algorithm is based on improved data-enhanced Gramian Angular Sum Field (DE-GASF) and multi-channel dual attention convolutional neural network (MC-DA-CNN). And the performance of the algorithm is validated by experiments on basic hydraulic components, taken axial piston pump and hydraulic reversing valve as an example. The experimental results show that the average fault diagnosis accuracy of axial piston pump and hydraulic reversing valve is 97.6% and 99.4% respectively, but the traditional monitoring method and single-sensor intelligent method are difficult to detect their faults due to their bad working environment. In addition, a comparative analysis of the image processing method and the time domain signal processing method confirms the effectiveness of the proposed technique.