Research On Image Detection And Recognition Method For Defects On The Surface Of The Steel Plate Based On Magnetic Flux Leakage Signals

Steel has become indispensable raw materials of machinery manufacturing,automobile manufacturing,chemical industry,ship building,aerospace and other industries,which occupies the important position in the national economy.However,there are often different types of quality defects on the surface of the steel plate,due to rolling equipment,continuous casting billets,processing and many other reasons.As a result of the existence of these defects,productions produced by the steel plate will exist safe hidden trouble,which not only affects the use of the products,but also even can cause huge losses to the national economy and the harm to safety of personal property.Therefore,the introduction of effective detection system to evaluate the quality of the surface is of profound significance to control and improve the quality,in the process of steel production.In the first,this thesis analyzes the existing technique for detecting and recognizing defect on steel plate surface and summarizes the characteristics and applications of these methods.In the next,we design the plate defect magnetic flux leakage detection system,including acquisition module design and control module design.We design the main programming of FPGA and CPLD based on the Verilog language.After that,we correct base value and interpolate to the collected magnetic flux leakage data.We convert it to two-dimensional grayscale images by using image conversion technology.We propose an image denoising method based on prethresholding wiener filtering and fusion of multi-wavelets and use it to denoise the image.At the same time,we propose an improved image edge detection algorithm based on wavelet multi-scale registration and wavelet multi-scale modulus maxima,and use it to locate the edge of the defect boundary,and extract geometric characteristics of the defect such as area,circumference,eccentricity,long axis and short axis.Finally,we build up a BP neural network identification model to identify the size of defects.Take geometric characteristics of the defect and the corresponding length,width,and depth of defects as training samples to train the model.More importantly,we validate the length,width and depth from the aspect of mean error rate,which can demonstrate the effectiveness of defect identification model for the recognition of length,width and depth through simulation experiments.At last,we obtain identification results of unknown defects by using this model based on geometric characteristics.