A system and algorithm capable of intelligent detection of surface defects on a curved surface with a reinforcing rib

Abstract

The application is based on the demand for surface defect detection of curved surface members with reinforcing rib structure, designs a brand-new mechanical device, builds a multi-degree-of-freedom cooperative motion platform, designs a detection algorithm based on a lightweight convolution network, constructs an intelligent detection system, and provides a system and algorithm capable of realizing intelligent detection of surface defects of curved surfaces with reinforcing ribs. The application can be applied to the fields of infrared thermal imaging nondestructive testing technology, magnetic powder detection technology, fluorescence detection technology and the like, aims to realize automatic detection of surface defects of complex members, improve detection precision and detection efficiency, and has important practical value and scientific significance.

Description

Technical Field

[0001] This invention relates to the field of nondestructive testing, specifically to a system and algorithm for intelligent detection of surface defects on ribbed curved surfaces. Background Technology

[0002] In recent years, research on surface defect detection of structural components has received widespread attention. In aerospace, high-speed rail transportation, and other fields, reinforced curved load-bearing components are subjected to long-term cyclic loads and are prone to fatigue crack initiation in the rib-matrix transition zone. Current defect detection methods have low automation levels, rely heavily on human experience, and suffer from low detection efficiency and low accuracy in detecting minute defects. Existing automated defect detection systems mostly can only detect a simple plane of the component, making it difficult to achieve complete surface coverage detection for complex components with reinforcement and curved surfaces. These problems restrict the promotion and application of defect detection technology. This invention aims to improve the level of automated detection of defects in complex components and meet national strategic needs. Against this background, researching a system and algorithm capable of intelligent surface defect detection for reinforced curved surfaces is of great significance.

[0003] Existing surface defect detection technologies mainly include infrared thermal imaging non-destructive testing (NDT), magnetic particle testing, and fluorescence detection. Infrared thermal imaging NDT heats the surface of an object and records the temperature using a thermal imager, analyzing the temperature distribution to determine the surface condition. Magnetic particle testing sprays magnetic powder onto the surface and applies a magnetic field, analyzing the powder distribution to determine the surface condition. Fluorescence detection sprays a fluorescent agent onto the surface and detects the fluorescence, analyzing its distribution to determine the surface condition. Most existing detection technologies rely on manual processing of components before surface analysis and evaluation, resulting in low efficiency. Defect identification primarily depends on visual inspection or traditional algorithms, leading to low accuracy. Existing automated inspection solutions can only inspect planar or simple cylindrical surfaces, failing to automate the inspection of complex components with reinforced curved surfaces.

[0004] To address these shortcomings, the present invention aims to provide a system and algorithm for intelligent surface defect detection of ribbed curved surfaces. This system can achieve comprehensive detection of curved surfaces containing ribbed structures and, combined with artificial intelligence technologies, improves the efficiency and accuracy of defect detection. This solution can be applied to fields such as infrared thermal imaging nondestructive testing, magnetic particle testing, and fluorescence detection, aiming to achieve automated detection of surface defects in complex components, improving detection accuracy and efficiency, and possessing significant practical value and scientific significance. Summary of the Invention

[0005] This invention designs a novel mechanical device, builds a multi-degree-of-freedom cooperative motion platform, designs a detection algorithm based on a lightweight convolutional network, and constructs an intelligent detection system, providing a system and algorithm capable of intelligently detecting surface defects on ribbed curved surfaces.

[0006] This invention provides a system for intelligent detection of surface defects on reinforced curved surfaces. The system mainly includes an imaging device, an angle adjustment platform, a rotary lifting platform, and a computer.

[0007] The imaging device provided by this invention can be an infrared thermal imager or a visible light imaging device such as a common camera.

[0008] The component with a reinforced curved surface provided by the present invention has a main body of the surface to be tested, which is an outer cylindrical surface with a central angle of 120°. A horizontal reinforcing rib plate with a thickness of 5mm and a vertical reinforcing rib plate with a thickness of 5mm are distributed on the outer cylindrical surface. The component to be tested is then subjected to heat treatment, fluorescent treatment, or magnetic powder spraying before being photographed and recorded by the imaging device.

[0009] This invention provides an angle adjustment platform, including a fixed platform, a movable platform, a quick-release clamp, a servo motor, a servo motor base, a drive gear, and a driven gear. The fixed platform has a U-shaped structure. Its base plate has bolt holes for fixing to the ground. The upper left and right sides of the fixed platform each have concentric shaft holes. A boss and bolt holes are located in the center of the left side of the fixed platform. The movable platform also has a U-shaped structure. Its base plate has bolt holes. The upper left and right sides of the movable platform each have concentric shaft holes. The shaft holes on the left and right sides of the fixed and movable platforms are connected by two shafts. The fixed platform is fixed to the ground, and the movable platform can rotate around the shafts. The quick-release clamp is used to connect and fix the shooting device. The bottom of the quick-release clamp has bolt holes. Bolts can be used to fix the quick-release clamp to the base plate of the movable platform. Adjusting the length of the bolts can adjust the distance between the quick-release clamp and the base plate of the movable platform. The driven gear is bolted to a boss in the middle of the left side of the stationary platform. The servo motor is fixed to the lower surface of the base plate of the moving platform via the servo motor base. The servo motor is connected to the driving gear, and the driving gear meshes with the driven gear.

[0010] This invention provides a rotating lifting platform, including a lifting platform and a rotating platform, such as... Figure 4As shown. The lifting platform is an electric lifting platform capable of lifting, and the rotating platform is an electric rotating platform capable of horizontal rotation. The lifting platform is fixed to the rotating platform with bolts to form the rotating lifting platform. The rotating lifting platform can adjust both the horizontal rotation angle and the height.

[0011] This invention provides a relative positional relationship between the various parts: the shooting device is disposed on the angle adjustment platform, the angle adjustment platform is fixed on the ground, and the angle adjustment platform can change the pitch angle relative to the ground; the rotating lifting platform can rotate horizontally relative to the ground and change its height; the component to be tested is disposed on the outer cylindrical surface of the rotating component to be tested, which contains horizontal and vertical reinforcing ribs; the rotating lifting platform is located directly in front of the shooting device, the rotating lifting platform is at a certain distance from the shooting device, and the rotating lifting platform is fixed on the ground.

[0012] This invention provides a process for inspecting the component: The component to be inspected is placed on the rotating lifting platform and within the field of view of the imaging device. A segmented scanning method is used to divide the component into upper and lower parts, with the horizontal reinforcing rib as the boundary. The height of the rotating lifting platform and the top-view angle of the angle adjustment platform are adjusted so that the imaging device scans the upper half of the component from a certain angle. During scanning, the rotating lifting platform first rotates horizontally by a certain angle, then descends a certain distance, then rotates in the opposite direction by a certain angle, and then descends a certain distance, repeating this zigzag motion to complete the scanning and imaging of the upper half of the component. The height of the rotating lifting platform and the bottom-view angle of the angle adjustment platform are adjusted so that the imaging device scans the upper half of the component from a certain angle. During scanning, the rotating lifting platform first rotates horizontally by a certain angle, then rises a certain distance, then rotates in the opposite direction by a certain angle, and then rises a certain distance, repeating this zigzag motion to complete the scanning and imaging of the lower half of the component. Through these two scans, the entire surface of the component to be inspected is scanned and imaged.

[0013] This invention provides an algorithm for intelligent surface defect detection on reinforced curved surfaces. This intelligent detection algorithm mainly consists of a lightweight convolutional neural network, with three alternating convolutional layers and pooling layers, two fully connected layers, and input and output layers. The first convolutional layer uses 32 kernels of size 3×3 and depth 3 for convolution calculation; the second convolutional layer uses 64 kernels of size 3×3 and depth 32 for convolution calculation; and the third convolutional layer uses 128 kernels of size 3×3 and depth 64 for convolution calculation. The results from each convolutional layer are processed using the ReLU function. The pooling layer uses max pooling. The result of the third pooling layer is flattened into a one-dimensional vector. The first fully connected layer uses 512 neurons; the second fully connected layer uses 256 neurons; and the output layer contains one neuron, and the output of the output layer is processed using the Sigmoid activation function. For images that are determined to contain defects, an edge detection algorithm is used to identify the specific location of the defects, and they are marked with a bright color to highlight the defect features.

[0014] Based on the need for surface defect detection of curved components with reinforcing ribs, this invention designs a novel mechanical device, builds a multi-degree-of-freedom cooperative motion platform, designs a detection algorithm based on a lightweight convolutional network, and constructs an intelligent detection system, providing a system and algorithm for intelligent surface defect detection of curved surfaces with reinforcing ribs. Attached Figure Description

[0015] One or more embodiments are illustrated by way of example with reference to the accompanying drawings. These illustrations do not constitute a limitation on the embodiments, and unless otherwise stated, the figures in the drawings are not to be limited in scale. In Figures 3 and 4, the numbers 1-9 represent: 1. Fixed platform; 2. Moving platform; 3. Quick-release clamp; 4. Servo motor; 5. Servo motor base; 6. Drive gear; 7. Driven gear; 8. Rotating platform; 9. Lifting platform;

[0016] Figure 1 : Structure diagram of the detection system

[0017] Figure 2 Schematic diagram of the component to be tested

[0018] Figure 3 Angle Adjustment Platform Structure Diagram

[0019] Figure 4 Schematic diagram of rotating platform and lifting platform

[0020] Figure 5 Schematic diagram of the zigzag scanning path

[0021] Figure 6 Schematic diagram of convolutional neural network structure

[0022] Figure 7 Schematic diagram of the marked crack defect Detailed Implementation

[0023] The inventors discovered that most existing automated defect detection systems have only one degree of freedom in their motion platform (rotational or translational along an axis), thus limiting detection to planar or cylindrical surfaces. For curved surfaces with stiffeners, at least three degrees of freedom are needed to detect curved surfaces, horizontal stiffener surfaces, and vertical stiffener surfaces. However, integrating all three degrees of freedom onto the motion platform of the detection device or the component being inspected significantly increases the complexity of the mechanism. The inventors simplified the motion mechanism by distributing the degrees of freedom among the motion platforms of the detection device and the component being inspected (one degree of freedom for the detection device and two degrees of freedom for the component), achieving comprehensive detection of complex surfaces. Furthermore, convolutional neural networks (CNNs) possess excellent image discrimination capabilities; lightweight CNNs for defect recognition tasks can improve efficiency and accuracy. This innovative method enables automated detection of complex components, improving detection efficiency and accuracy, and has significant practical value and scientific significance.

[0024] This invention provides a system for intelligent detection of surface defects on ribbed curved surfaces. The system mainly includes an imaging device, an angle adjustment platform, a rotary lifting platform, a computer, etc., as shown in Figure 1.

[0025] The imaging device provided by this invention can be an infrared thermal imager or a visible light imaging device such as a common camera.

[0026] The component with a reinforced curved surface provided by the present invention has an outer cylindrical surface with a central angle of 120° as the main body of the surface to be tested. A horizontal reinforcing rib with a thickness of 5mm and a vertical reinforcing rib with a thickness of 5mm are distributed on the outer cylindrical surface, as shown in Figure 2. The surface of the component to be tested is heat-treated, fluorescently treated, or coated with magnetic powder before being photographed and recorded by the imaging device.

[0027] This invention provides an angle adjustment platform, including a fixed platform 1, a movable platform 2, a quick-release clamp 3, a servo motor 4, a servo motor base 5, a drive gear 6, and a driven gear 7, as shown in Figure 3. The fixed platform 1 has a "U"-shaped structure. The base plate of the fixed platform 1 has bolt holes for fixing to the ground. The upper left and right sides of the fixed platform 1 are respectively provided with shaft holes, and the shaft holes on both sides are concentric. The middle of the left side of the fixed platform 1 is provided with a boss and bolt holes. The movable platform 2 has a "U"-shaped structure. The base plate of the movable platform 2 is provided with bolt holes. The upper left and right sides of the movable platform 2 are respectively provided with shaft holes, and the shaft holes on both sides are concentric. The fixed platform 1 and the moving platform 2 are connected by two shafts on their left and right sides, respectively. The fixed platform 1 is fixed to the ground, and the moving platform 2 can rotate around the shaft. The quick-release clamp 3 is used to connect and fix the shooting device. The bottom of the quick-release clamp 3 is provided with bolt holes. Bolts can be used to fix the quick-release clamp 3 to the base plate of the moving platform 2. Adjusting the length of the bolts can adjust the distance between the quick-release clamp 3 and the base plate of the moving platform 2. The driven gear 7 is fixed to the boss in the middle of the left side of the fixed platform 1 by bolts. The servo motor 4 is fixed to the lower surface of the base plate of the moving platform 2 by the servo motor base 5. The servo motor 4 is connected to the driving gear 6, and the driving gear 6 and the driven gear 7 mesh with each other.

[0028] This invention provides a rotating lifting platform, comprising: a lifting platform 9 and a rotating platform 8, as shown below. Figure 4 As shown. The lifting platform 9 is an electric lifting platform capable of lifting, and the rotating platform 8 is an electric rotating platform capable of horizontal rotation; wherein, the lifting platform 9 is fixed to the rotating platform 8 by bolts to form the rotating lifting platform, and the rotating lifting platform can adjust both the horizontal rotation angle and the height;

[0029] This invention provides a relative positional relationship between the various parts: the shooting device is disposed on the angle adjustment platform, the angle adjustment platform is fixed on the ground, and the angle adjustment platform can change the pitch angle relative to the ground; the rotating lifting platform can rotate horizontally relative to the ground and change its height; the component to be tested is disposed on the outer cylindrical surface of the rotating component to be tested, which contains horizontal and vertical reinforcing ribs; the rotating lifting platform is located directly in front of the shooting device, the rotating lifting platform is at a certain distance from the shooting device, and the rotating lifting platform is fixed on the ground.

[0030] This invention provides a process for inspecting the component: the component to be inspected is placed on the rotating lifting platform and within the field of view of the imaging device. A segmented scanning method is used to divide the component into upper and lower parts, with the horizontal reinforcing rib as the boundary. The height of the rotating lifting platform and the top-view angle of the angle adjustment platform are adjusted so that the imaging device scans the upper half of the component from a certain angle. During scanning, the rotating lifting platform first rotates horizontally by a certain angle, then descends a certain distance, then rotates in the opposite direction by a certain angle, and then descends a certain distance, repeating this zigzag motion to complete the scanning and imaging of the upper half of the component. The height of the rotating lifting platform and the bottom-view angle of the angle adjustment platform are adjusted so that the imaging device scans the upper half of the component from a certain angle. During scanning, the rotating lifting platform first rotates horizontally by a certain angle, then rises a certain distance, then rotates in the opposite direction by a certain angle, and then rises a certain distance, repeating this zigzag motion to complete the scanning and imaging of the lower half of the component. Through these two scans, the entire surface of the component to be inspected is scanned and imaged. Figure 4 As shown.

[0031] This invention provides an algorithm for intelligent surface defect detection on reinforced curved surfaces. This intelligent detection algorithm mainly consists of a lightweight convolutional neural network, with alternating layers of three convolutional and pooling layers, two fully connected layers, and input and output layers, as shown below. Figure 6 As shown. The first convolutional layer uses 32 kernels of size 3×3 and depth 3 for convolution calculation; the second convolutional layer uses 64 kernels of size 3×3 and depth 32 for convolution calculation; and the third convolutional layer uses 128 kernels of size 3×3 and depth 64 for convolution calculation. By setting three convolutional layers and increasing the number of kernels from 32 to 64 and then to 128, the model can extract not only low-level features such as image edges but also high-level features such as cracks, greatly improving the model's sensitivity to defect features; the result of each convolutional layer uses the ReLU function. The feature map is processed by retaining values ​​greater than 0 and setting values ​​less than or equal to 0 to 0. This operation suppresses unimportant features and highlights significant features, resulting in faster computation and convergence speeds, and effectively mitigating the gradient vanishing problem. The pooling layer uses max pooling to halve the feature map size, extracting key features, reducing computation, and preventing overpooling. The result of the third pooling layer is flattened into a one-dimensional vector for inputting feature data into the fully connected layer. The first fully connected layer uses 512 neurons to effectively integrate the large number of features output from the previous layer. The second fully connected layer uses 256 neurons to further compress the features, retaining the most important information for a more compact and efficient feature representation. The output layer contains one neuron and uses the sigmoid activation function. The output of the output layer is processed and mapped to a range of 0 to 1 to represent the probability of defects in the image. For images determined to contain defects, an edge detection algorithm is used to identify the specific location of the defects, which are then marked with a bright color to highlight the defect features. Figure 7 As shown.

Claims

1. A system for intelligent detection of surface defects on ribbed curved surfaces, characterized in that: The device comprises a camera, an angle adjusting platform, a rotating and lifting platform and a computer. The camera is an infrared thermal imager or a visible light camera. The camera is arranged on the angle adjusting platform, which is fixed on the ground and can change the pitch angle relative to the ground. The rotating and lifting platform can rotate horizontally and change the height relative to the ground, and the component to be detected is arranged on the rotating and lifting platform. The outer cylindrical surface of the component to be detected contains horizontal and vertical reinforcing rib plates. The rotating and lifting platform is located in front of the camera at a certain distance, and is fixed on the ground. The surface of the component to be detected is heated, treated by fluorescence or treated by magnetic powder, and then photographed by the camera. The detection process comprises: Placing the component to be detected on the rotating and lifting platform and in the view angle of the camera, and dividing the component to be detected into two parts by the horizontal reinforcing rib plates by using a regional scanning method. Adjusting the height of the rotating and lifting platform and the pitch angle of the angle adjusting platform, so that the camera scans the upper half of the component to be detected at a certain angle, and the rotating and lifting platform rotates horizontally at a certain angle, then drops a certain distance, rotates in the opposite direction at a certain angle, then drops a certain distance, and so on, to complete the scanning and photographing of the upper half of the component to be detected along the zigzag path. Adjusting the height of the rotating and lifting platform and the pitch angle of the angle adjusting platform, so that the camera scans the upper half of the component to be detected at a certain angle, and the rotating and lifting platform rotates horizontally at a certain angle, then drops a certain distance, rotates in the opposite direction at a certain angle, then drops a certain distance, and so on, to complete the scanning and photographing of the upper half of the component to be detected along the zigzag path.

2. The system capable of intelligently detecting surface defects of a belt rib curved surface according to claim 1, characterized in that, Through twice scanning, the scanning and photographing of the entire surface of the component to be detected are completed. The angle adjusting platform comprises a fixed platform, a movable platform, a quick mounting clamp, a rudder, a rudder base, a driving gear and a driven gear. The fixed platform is in "U" shape structure, the bottom plate of the fixed platform is provided with bolt holes for fixing on the ground, the upper parts of the left and right sides of the fixed platform are respectively provided with shaft holes with the same center, and the left side of the fixed platform is provided with a boss and a bolt hole in the middle part. The movable platform is in "U" shape structure, the bottom plate of the movable platform is provided with bolt holes, and the upper parts of the left and right sides of the movable platform are respectively provided with shaft holes with the same center. The shaft holes on the left and right sides of the fixed platform and the movable platform are respectively connected by two shafts, the fixed platform is fixed on the ground, and the movable platform can rotate around the shafts. The quick mounting clamp is used for connecting and fixing the camera, the bottom of the quick mounting clamp is provided with bolt holes, and the quick mounting clamp can be fixed on the bottom plate of the movable platform by using bolts, and the distance between the quick mounting clamp and the bottom plate of the movable platform can be adjusted by adjusting the length of the bolts. The driven gear is fixed to the boss in the middle of the left side surface of the fixed platform through a bolt; The steering engine is fixed to the lower surface of the bottom plate of the movable platform through the steering engine base, the steering engine is connected with the driving gear, and the driving gear is engaged with the driven gear.

3. The system capable of intelligently detecting surface defects of a belt rib curved surface according to claim 1, characterized in that, The rotating lifting platform comprises a lifting platform and a rotating platform: The lifting platform is an electric lifting platform capable of lifting, and the rotating platform is an electric rotating platform capable of horizontal rotation; The lifting platform is fixed to the rotating platform through a bolt to jointly form the rotating lifting platform, which can adjust the horizontal rotation angle and the height.

4. An algorithm for intelligently detecting surface defects on a belt with a ribbed surface, using the system of claim 1, wherein, The detection algorithm comprises: processing the to-be-detected picture photographed by the photographing device into a 3-channel picture with a size of 227*227; a convolutional neural network model comprising three convolutional layers and pooling layers alternately distributed, two fully connected layers and an input-output layer is constructed; the first convolutional layer uses 32 convolutional kernels with a size of 3*3 and a depth of 3 for convolution calculation; the second convolutional layer uses 64 convolutional kernels with a size of 3*3 and a depth of 32 for convolution calculation; the third convolutional layer uses 128 convolutional kernels with a size of 3*3 and a depth of 64 for convolution calculation; the results obtained by each convolutional layer are processed using a ReLU function; the pooling layer adopts a maximum pooling manner; the result after the third pooling is flattened into a one-dimensional vector; the first fully connected layer uses 512 neurons; the second fully connected layer uses 256 neurons; the output layer comprises one neuron, and the output result of the output layer is processed using a Sigmoid activation function to obtain a final result.

Images