Bidimensional local characteristic-scale decomposition and its application in gear surface defect detection

Abstract

Visual image-based inspection methods can directly reflect the type of defects on the surface of gears. However, these methods have many problems: firstly, as a two-dimensional signal, the data volume of images is large and the processing is relatively time-consuming. Although some existing image signal processing methods (e.g. bidimensional empirical mode decomposition (BEMD)) have good decomposition results, their decomposition speed is slow. The bidimensional local characteristic-scale decomposition (BLCD) method is proposed in this paper, which adaptively decomposes an image from high to low frequencies into several bidimensional intrinsic scale components. It is demonstrated that the BLCD method maintains the advantages of the BEMD method in terms of good decomposition ability and adaptive capability while significantly reducing the processing time and improving the computational efficiency. Secondly, in the running state of the gears, the obtained images sometimes contain noise, which is not convenient for detecting surface defect types. A gear surface defect detection method based on BLCD image denoising is proposed in this paper. Firstly, it uses the BLCD denoising module for preprocessing to provide high signal-to-noise ratio images for the subsequent detection module, and then uses the detection module for defect identification and classification. Experiments prove that the BLCD denoising module has excellent performance and it is well coupled with the detection module, giving the whole method higher accuracy and stability than other classification methods.