As a raw material, steel pipes are widely used in industries such as petroleum, chemical industry, electric power, shipbuilding, and automobiles. In recent years, the development of economic globalization has made enterprises put forward higher requirements for product quality. Defects on the surface of steel pipes will seriously affect their service life. Enterprises cause property losses. Therefore, in order to control the quality of steel pipes, relevant enterprises will conduct quality inspections on them, but the inspection measures are usually implemented manually, and it is impossible to quickly and accurately detect defects.
The process of steel pipe production, due to various reasons such as raw materials, rolling equipment and processing technology, will lead to different types of defects such as scratches, roll marks, scale, surface inclusions, holes, cracks, pitting, etc.
These defects not only seriously affect the appearance of the product, but also reduce the performance of the product such as corrosion resistance, wear resistance and fatigue strength, which have a negative impact on the development of the enterprise. Security risks. The surface defect area has the characteristics of stress concentration and weak force. At the same time, performance mutation, fatigue damage and corrosion often concentrate in this area, which greatly reduces the working performance of the steel pipe in complex and harsh environments.
It is of great significance to detect defects in time by detecting the defect areas on the surface of steel pipes and to provide a basis for the adjustment of production processes and the improvement of equipment status. At present, the detection of surface defects in steel pipes is mostly done manually. Manual methods rely on on-site experience and are inefficient. Affected by the on-site environment, the labour intensity is high, and it is easy to cause missed and false detections, and cannot fully reflect the quality of the steel pipe surface. , poor real-time detection, few types of detection, low detection efficiency, and lack of comprehensive evaluation of the surface quality of the product. With the development of the computer level and the rise of the field of artificial intelligence, machine vision technology has been widely used. The use of machine vision methods can effectively make up for the lack of manual detection, and the detection accuracy is high, which can further provide a data platform for intelligent manufacturing.
1. Characteristics of steel pipe surface defect detection
Metallurgical products using machine vision methods at home and abroad mainly detect metallurgical products such as plates, steel strips, and steel bars. These products have relatively flat surfaces, low roughness, and consistent material reflectivity. Both array or line array cameras can obtain ideal images of surface defects of the inspected material, which also effectively reduces the complexity of subsequent image processing algorithms; The ideal lighting result can be obtained with an area array camera; it is easier to achieve with a linear array light source because the distance from each point in the illuminated area to the centre of the light source is equal.
Schematic diagram of light distribution for surface defect detection of planar materials For steel pipes, due to their geometric structure, when using an area array light source, the arc-shaped outer surface makes the distance between the centre of the light source and the irradiated area too large, and the shortest distance from the light source The distance position shows higher brightness, while at the shortest distance position, the light brightness distribution is weakened, and the image imaging results also show similar characteristics. The grey level distribution of pixels in the middle area is higher and the grey level values of pixels on both sides are smaller.
In another case, when using a line array camera and a line array light source to realize the dynamic detection of steel pipe surface defects, due to factors such as vibration and assembly errors, the centre line of the line array light source and the length direction of the line array camera’s field of view is not on the same line, and the direction of the camera’s field of view is usually the same as that of the line array camera. The centre of rotation of the steel pipe is on the same straight line; this situation will reduce the coincidence of the irradiation area and the field of view area, resulting in serious uneven illumination of the imaging results, and further increasing the difficulty of image processing.
2. Key technical difficulties
Due to its geometric structure, steel pipes are prone to uneven illumination; in order to realize the dynamic real-time detection of the steel pipe arc surface, it will inevitably affect the coincidence of the light source illumination area and the camera field of view, which will easily cause uneven illumination distribution, which will cover the defects. characteristics of the region. When image acquisition is not ideal, it will increase the difficulty of image processing. Although relevant scholars have done a lot of work in the field of machine vision detection, there are few domestic studies on the detection of surface defects in steel pipes.
There are mainly the following difficulties:
(1) Hot-rolled seamless steel pipes are similar to hot-rolled strip steel and heavy rails, and the surface is covered with a large amount of oxide scale, which will lead to various false defects;
(2) The arc-shaped outer surface of the steel pipe is prone to uneven illumination;
(3) Due to the influence of uneven illumination, the grey level of defects is quite different, which makes the missed detection serious;
(4) Affected by curvature, out-of-roundness and surface protrusion defects, steel pipes vibrate during the detection process, which makes image acquisition errors and features not obvious;
(5) When dynamic detection is realized, the coincidence of the light source irradiation area and the camera field of view area will be reduced, resulting in uneven illumination distribution.
3. Imaging optical path design
The lighting system includes the selection of lighting methods and the determination of the positional relationship between the camera and the light source. The lighting methods on the surface of the steel pipe can be divided into bright field lighting and dark field lighting. In this paper, the bright field lighting method is selected, which is beneficial to the high contrast between the surface defects of the steel pipe and the background. Due to the use of a single line array camera and line light source, its effective working area is a narrow strip. Different from the surface of other metallurgical products, the surface of the hot-rolled seamless steel pipe is not polished during the forming process due to the characteristics of the process. The light reflection type is mainly diffuse reflection.
According to the principle of image acquisition, it is necessary to determine the position of the line array camera, line light source, etc., which is beneficial to the determination of parameters in the subsequent hardware selection. In order to ensure that the field of view can cover the surface of steel pipes of different lengths, in the optical path design, the width of the field of view is greater than the length of the steel pipe.
4. Defect detection of steel pipes
The surface defects of steel pipes are pits, scratches, warped skins and roll marks.
Pit defects, which are characterized by point-like or massive depressions, are formed by embedding on the surface of the steel pipe during the rolling process and falling off due to scale or foreign matter not being removed;
Warped skin defect is the metal layer attached to the outer surface of the steel pipe. During the deep processing process of the pipe, the accumulated inclusions leak out due to the thinning of the pipe wall, and the cracks form and extend to make the skin warped;
Scratch defects, the surface of the steel pipe is scratched by external metal or hard objects, usually in the form of slender and sharp grooves or shallow pits;
Roll mark defects, are caused by improper adjustment of rolls or surface damage, and are distributed periodically or continuously.
Before extracting the surface defect features of steel pipes, it is necessary to determine which features are effective. For pits, warped skins, scratches and roll marks, it is necessary to select features with good discrimination to form feature vectors; feature vectors are numerical representations of defect features, feature extraction usually follows the following principles:
1) The features in the image should be easy to obtain;
2) The selected features are numerically free from noise and irrelevant factors;
3) The characteristics of the same defect have compactness, and the characteristics of different defects have good discrimination.
The distribution and size of defects on the steel pipe surface are not regular, and the shape is complex, so it is necessary to select the features that can accurately describe the defects. Machine vision technology converts the target image captured by the CCD camera into an image signal in real time, and then inputs the image signal into the embedded visual image processing system. According to image saturation, pixel distribution, target image edge, brightness and other information, it is converted into digital signals recognized by the computer, and advanced algorithms are used to identify features of the image, evaluate the results of feature identification, and output the final defect results, including defects, size, angle, number, qualified and unqualified, presence or absence, etc., to realize the automatic identification function.
All in all, the requirements for automatic recognition of machine vision adopted by the steel pipe defect detection system must solve the following main problems:
1. It must be able to detect the flaws on the surface of the steel pipe online, such as scratches, scratches, holes, scars, pad pits and other surface abnormalities.
2. It can cope with the interference caused by the change in the width and length of the steel pipe, the distortion or inclination of the steel pipe during the moving process, oil stains or water droplets on the surface.
3. Defect detection has self-learning and self-adaptive functions, which are suitable for different widths, different colours, and different speeds. Functions such as pattern recognition, automatic exposure, anti-shake, and defect alarm must also be applied. The defect detection and defect alarm are dynamic and real-time.
4. It must have the characteristics of high precision and few fault points, and it needs to combine industrial-grade digital cameras and industrial-grade PCs to complete system tasks.