Search

CN-121563985-B - Method, device and program product for discovering periodic defects on surface of strip steel

CN121563985BCN 121563985 BCN121563985 BCN 121563985BCN-121563985-B

Abstract

The invention provides a method, a device and a program product for discovering periodic defects on the surface of strip steel, and relates to the technical field of machine vision. The method for finding periodic defects on the surface of the strip steel comprises the steps of training a target detection model, reasoning the surface image of the strip steel of the production line acquired in real time to obtain a defect detection result, carrying out grid division on the whole strip steel in a self-defined size, constructing defect distribution coordinates by taking grids as units, marking the defect detection result in the grid to obtain a two-dimensional defect distribution map, establishing a defect counting sequence, carrying out short-time Fourier transformation to obtain a time-frequency map of the periodic defects, analyzing the time-frequency map of the periodic defects, and determining the period and the duration of the defects. The invention accurately detects the periodic defects of the hot rolled strip steel through the application of machine vision, and discovers the possible stability process problems in the hot rolled strip steel as soon as possible.

Inventors

  • ZHANG HANBO
  • ZHANG ZHE
  • WANG YUJIN

Assignees

  • 华院计算技术(上海)股份有限公司

Dates

Publication Date
20260508
Application Date
20260121

Claims (12)

  1. 1. The method for discovering the periodic defects on the surface of the strip steel is characterized by comprising the following steps: S1, defect identification is carried out based on a target detection model, and the method specifically comprises the following steps: s11, constructing a training data set; s12, training a target detection model, reasoning the product line steel surface image acquired in real time, and determining a candidate target frame of the periodic defect; S13, carrying out high-precision identification and screening of various morphological features on defects in the candidate target frames to obtain defect detection results; S2, discretizing defect signals, wherein the specific steps comprise: carrying out grid division of the self-defined size on the whole coiled steel, constructing defect distribution coordinates by taking grids as units, marking the occurrence and the category of defects in the defect distribution coordinates, obtaining a two-dimensional defect distribution map, and further obtaining a discretization signal; s3, defect period detection, which comprises the following specific steps: S31, counting discretization signals in the length direction of the strip steel, and establishing a defect counting sequence; S32, sliding the window column by column, performing dense analysis on the defect counting sequence, and performing discrete short-time Fourier transform on the partial sequence subjected to windowing to obtain a time-frequency chart of periodic defects, wherein periodic defect frequency corresponding to a specific position of the strip steel and periodic stability intensity of the frequency are recorded; s33, analyzing a time-frequency diagram of the periodic defect, and determining the period and the duration influence length of the defect; S34, checking the defect period by taking the circumference of the roller as a reference; S4, post-processing and outputting results, wherein the method comprises the following specific steps: outputting the period and the duration influence length of the defect, positioning the position of the fault roller, and giving a treatment suggestion.
  2. 2. The method according to claim 1, wherein the specific step of step S11 comprises: collecting historical image data of the surface defects of the strip steel as a training data set; Adding a training data set by a data labeling, data enhancement and data generation method; the data enhancement method comprises rotation and scaling.
  3. 3. The method according to claim 1, wherein in the step S12, the object detection model is obtained by training YOLO model.
  4. 4. The method according to claim 1, wherein in the step S13, the screening includes: Excluding non-periodic defect data; a specific type of periodic defect data is selected.
  5. 5. The method according to claim 1, wherein in the step S2, the method of marking the defect detection result is to construct mesh data corresponding to each mesh, and mark whether defects occur in the corresponding mesh area and the types of defects by the mesh data.
  6. 6. The method according to claim 1, wherein the specific step of step S32 includes: inputting a defect counting sequence, and setting a window length, a window moving step length and a window function; Performing short-time Fourier transform on defect count data segments which are sequentially intercepted according to the window length from the head of the strip steel, and calculating a frequency spectrum result corresponding to each window, namely frequency components and intensity of occurrence of defects on the section of strip steel; And combining the frequency spectrum results to form a steel strip time-frequency diagram, wherein the horizontal axis represents the length position of the steel strip, the vertical axis represents the frequency, and the brightness or color of each point in the diagram represents the intensity of the periodical defect signal at the position and the frequency.
  7. 7. The method according to claim 6, wherein the specific method of the short-time fourier transform is: Let the defect count sequence of the ith row of the grid be Wherein Represents the grid column index along the length of the strip, , Selecting a window function of finite length for the total length The window length is The short-time Fourier transform is realized by performing discrete Fourier transform on the local sequence after windowing, and the mathematical expression is as follows: ; Wherein: Is the first The starting column index of the analysis window, The window is sequenced along a fixed step1 The sliding motion of the slide-on plate is performed, As the number of the total window(s), As a function of the frequency component, Is a window function, selects a hamming window or a hanning window to realize, Is a temporary variable, is a cyclic index of the summation operation; Is shown in the first Window position, frequency Spectral amplitude at all window positions by Sum frequency Calculating to obtain a two-dimensional time-frequency diagram The amplitude is 。
  8. 8. The method according to claim 1, wherein in the step S33, the method for analyzing the time-frequency chart of the periodic defect includes extracting a significant light-band frequency peak value of the time-frequency chart, calculating a corresponding period, and combining the period with the unit length of the grid to obtain the period of the defect.
  9. 9. The method according to claim 1, wherein the specific method of step S34 is to set an allowable deviation threshold on the basis of an integer multiple of the circumference of the roll, and to verify the defect cycle.
  10. 10. A computer device comprising a memory, a processor and a computer program stored on the memory, characterized in that the processor executes the computer program to carry out the steps of the method according to any one of claims 1-9.
  11. 11. A computer readable storage medium, on which a computer program is stored, characterized in that the computer program, when being executed by a processor, carries out the steps of the method according to any one of claims 1-9.
  12. 12. A computer program product, characterized in that the computer program, when being executed by a processor, implements the steps of the method according to any one of claims 1-9.

Description

Method, device and program product for discovering periodic defects on surface of strip steel Technical Field The invention relates to the technical field of machine vision, in particular to a method, a device and a program product for discovering periodic defects on the surface of strip steel. Background Along with the construction of industrialization 4.0, the surface defect detection technology is widely applied to quality inspection of steel strips (hereinafter referred to as strip steel). The periodic defects are taken as special defects, the generation of the periodic defects often means that the stability of the rolling process of the strip steel is insufficient or abnormal, and the periodic defects can be found in time to help production technicians to adjust production equipment and process parameters quickly, so that the mass production of defective products is avoided. The existing periodic defect detection technology mainly comprises 2 modes, namely 1) periodically finding an original picture of the strip steel directly through Fourier transformation, and 2) discretizing a continuous signal and mining a periodic mode through a tree model. However, the former has high complexity and poor robustness, and the latter cannot accurately extract the period and reflect the trend of the change. Specific: Chinese patent publication No. CN107475509a discloses a method for detecting a pair of roller marks. Roll marks are a relatively common periodic defect in strip rolling. The detection method of the patent is to filter and judge by presetting a series of fixed rules. The rules are typically defined based on defect topographical features along with a range of roller circumferences. In the scheme disclosed in the chinese patent publication No. CN116952968a, the periodic start point is first determined by extracting the periodic defect, then the periodic feature is extracted by fast fourier transform, and the identified periodic picture is superimposed and compared to determine the defect. Prior art literature (Wang Sheng. Research on time series based utility cycle pattern mining algorithms [ D ]. University of electronics and science, hangzhou, 2024.doi: 10.27075/D) investigated time series signal cycle pattern discovery, which mentions that the common practice of time series cycle pattern mining is to first discretize and symbolize a time series, and then mine cycle patterns based on a tree model. The prior art method described above has the following disadvantages: (1) The defect morphology is various, the periodic defect morphology is similar, but the area and the gray scale of the periodic defect morphology can still be various, for example, the periodic roll marks caused by rolling roll attachments into the strip steel can be generated, the periodic roll marks on the surface of the strip steel generally have the area reduction trend along with the consumption of the surface attachments of the rolls, and the defect comparison and the regular matching can not lead to missed detection in the situation. (2) The period identification time of the traditional periodic defect signal is longer, and the strip steel rolling generally has higher requirements on real-time property. The method adopts the fast Fourier transform to identify the period of the original image signal or mode, so that on one hand, the calculation complexity is high, the real-time requirement of the production line is difficult to meet, and on the other hand, the accuracy of period identification is easily reduced due to the influence of tiny noise. (3) The defect picture detected by the meter can be missed, the periodicity of the periodic defect is destroyed, the data required by the fast Fourier transform for the period identification meet the strict periodicity, and the condition that the period changes along with the time cannot be processed. In addition, periodic defects are not always distributed over the whole strip steel, only occur periodically in local areas, and the fast Fourier transform models the whole area to obtain the periodicity, so that the periodic defects only occur in the local areas can not be identified. The existing algorithm has insufficient robustness. (4) The method of discretizing a continuous value into a finite symbol based on the symbolization method loses a large amount of amplitude and accurate frequency information. The method mainly reveals the mode repeatability, on one hand, the accurate frequency or intensity of the mode cannot be directly obtained, and on the other hand, the periodic frequency change trend cannot be described. Disclosure of Invention Aiming at the defects existing in the prior art, the invention provides a method, a device and a program product for discovering periodic defects on the surface of strip steel, which are used for accurately detecting the periodic defects of hot rolled strip steel through the application of machine vision, discovering possible stability process problems i