Automobile windscreen wiper surface defect detection device and defect detection method

Abstract

The invention discloses an automobile windscreen wiper surface defect detection device which comprises a detection table, a mechanical arm is arranged on the detection table, a detection camera and a clamping jaw are arranged at the end of the mechanical arm, and a plurality of material tables are arranged on the detection table and located on one side of the mechanical arm. A qualified product material frame, an unqualified product material frame and a sponge separation material frame are arranged on the side, away from the material table, of the detection table, and the mechanical arm, the camera and the clamping jaw are all connected with a control unit. The invention further discloses a defect detection method. The windscreen wiper image can be collected, the windscreen wipers are classified and placed according to the detection result, the detection efficiency is high, the result is reliable, the application range is wide, various types of windscreen wipers can be measured, and the good application prospect is achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of defect detection, and in particular to a device and method for detecting surface defects of automobile wipers. Background Art

[0002] A windshield wiper, also known as a water paddle, wiper or windshield wiper, is a device used to remove dust and rain from the windshield of various vehicles and is widely used in vehicles. However, due to various factors such as the process and equipment in the production process, the appearance of the wiper often has defects such as scratches, abrasions, and spots (such as pits and sand holes). These appearance defects have a significant impact on the aesthetics of the product. Therefore, wiper manufacturers attach great importance to the detection of appearance defects. Through this process, defective products can be screened out, product quality can be guaranteed, and the company's reputation can be enhanced. In addition, by analyzing the test results, companies can also promptly discover potential problems in the production process or equipment, so as to take measures to improve the product qualification rate.

[0003] As early as the mid-1980s, foreign countries began to study machine vision. In the past 30 years, machine vision technology has developed rapidly and has been widely used in industrial production, especially in the field of surface defect detection. As a kind of metal industrial product, the surface defect detection of wipers also belongs to the research scope of metal surface defect detection. At present, most domestic wiper manufacturers still rely on manual detection methods, and workers perform defect detection through equipment such as light sources and polarizers. However, this method has the problems of low efficiency and high labor cost, and the detection results are affected by the subjective judgment and fatigue level of workers, and the accuracy is low. Therefore, with the development of industrial technology, surface defect detection has gradually shifted towards intelligence and machine vision technology. Therefore, it is extremely necessary to design a set of automobile wiper surface defect detection devices that can collect wiper images and classify and place wipers according to the detection results. Summary of the invention

[0004] The purpose of the present invention is to provide a vehicle windshield wiper surface defect detection device, which can collect windshield wiper images and classify and place the wipers according to the detection results.

[0005] In order to achieve the above-mentioned purpose, the present invention adopts the following technical scheme: a vehicle wiper surface defect detection device, comprising a detection platform, a robotic arm is arranged on the detection platform, a detection camera and a clamp are respectively arranged at the ends of the robotic arm, a plurality of material tables are arranged on the detection platform and on one side of the robotic arm, a material frame for qualified products, a material frame for unqualified products and a material frame for separating sponges are arranged on the side of the detection platform away from the material table, wherein the robotic arm, the camera and the clamp are all connected to the control unit.

[0006] Further: a camera mounting plate is arranged at the end of the mechanical arm near the clamping claw, an inner folding plate is arranged at the end of the camera mounting plate, and a camera is installed at the inner folding plate.

[0007] Further: the material table has a base, a clamp is arranged at one end of the top of the base, a convex column is arranged at the other end, and inclined surfaces are arranged on both sides of the base.

[0008] Further: the detection table has a table top, and a base frame is arranged at the lower side of the table top.

[0009] Further: an auxiliary light source is set at the camera.

[0010] Further: the partition sponge is in strip shape, and a plurality of notches are arranged at equal intervals on the upper side.

[0011] Further: a display is arranged on the testing platform, and the display is connected to the control unit.

[0012] Furthermore, the boss (53) comprises a lower boss (53a) and an upper boss (53b); the lower boss is provided with a connecting shaft (53a-1) and a square stop block (53a-2) provided on the top of the connecting shaft; the upper boss is provided with a stroke adjustment cavity (53b-1) cooperating with the square stop block; and a side adjustment channel (53c) is provided between the lower boss and the upper boss;

[0013] The base (51) is provided with a horizontal adjustment hole communicated with the boss mounting hole, the hole wall of the horizontal adjustment hole is provided with a spring compression channel with an inner diameter larger than the inner diameter of the horizontal adjustment hole, the horizontal adjustment hole is installed with an adjustment rod (55), the adjustment rod is provided with an annular stopper (57), the adjustment rod is provided with a compression spring (56) located in the spring compression channel, and the end of the adjustment rod is provided with an adjustment protrusion (58) cooperating with the side adjustment channel.

[0014] On the other hand, a defect detection method is characterized in that the method uses the above-mentioned automobile wiper surface defect detection device, comprising the following steps:

[0015] (1) Collecting images of car wipers;

[0016] (2) Use the YOLOv8 convolutional neural network model to determine whether the car wiper is qualified;

[0017] (3) According to the judgment results, qualified products are placed in the qualified material box, and unqualified products are placed in the unqualified material box;

[0018] Further: Among them, the YOLOv8 convolutional neural network model training method is as follows:

[0019] (T1) Obtain a dataset of wiper surface defects and divide the dataset into training set, validation set and test set. Set training parameters based on the dataset and train the model. In the training parameters, the batch size is 16, the optimizer is AdamW, the initial learning rate is 0.001, the weight decay is 0.0005, the data enhancement ratio is 0.8, the image size is 512×512, and the training rounds are 150.

[0020] (T2) The model includes a backbone unit, a neck unit and a head unit connected in sequence; the backbone unit includes five Conv modules, four C2f_CA feature extraction modules and one SPPF module; the neck unit includes two Upsamp le upsampling modules, four Concat modules, four C2f modules and two Conv modules, three of which are connected to the head unit, and another C2f module is directly connected to a Concat module. The head unit includes three detection head modules;

[0021] (T3) The CA attention mechanism module in the C2f_CA feature extraction module uses two one-dimensional global pooling to aggregate the input features in the x and y directions into two independent direction-aware feature maps, thereby strengthening the direction and position information around the target and improving the detection accuracy of the target; the specific steps of the CA module are as follows:

[0022] (S1)CA performs coordinated information embedding, that is, input a C*H*W feature map, and two poolings (H, 1) and (1, W) are pooled along the H direction and W direction of the feature map to obtain the processed information features; the horizontal and vertical feature processing formulas are as follows:

[0023]

[0024] Where: and Represents the embedded feature maps in the H and W directions respectively; W, H and x c They are the width, height and current feature value of the feature map respectively;

[0025] (S2) Coordinate Attention Generation is performed, the obtained feature map is concatenated along the spatial dimension and activated after 1*1 convolution transformation:

[0026] f=δ(F1([Z h ,Z w ])) (3)

[0027] Where: [Z h ,Z w] means concatenating two feature maps; F1() and δ() represent 1*1 convolution kernel and activation function;

[0028] (S3) Perform a Split operation along the spatial dimension to obtain two separated feature maps and transform and activate them to obtain the attention vector g h and g w , the formula is as follows:

[0029]

[0030] g w =σ(F w (f w )) g h =σ(F h (f h )) (6)

[0031] Where: f, f w and f h They represent the feature map before separation and the feature map in the W and H directions after separation respectively; F w 、F h and σ are the convolution transformation and activation function in two directions respectively;

[0032] (S4) Obtain the corrected feature map y through the broadcast mechanism c :

[0033]

[0034] Compared with the prior art, the present invention has the following beneficial effects: the present invention can collect wiper images, classify and place the wipers according to the detection results, has high detection efficiency, reliable results, a wide range of applications, can measure various types of wipers, and has good application prospects. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] Figure 1 This is a stereoscopic view of the automobile wiper surface defect detection device from the first direction.

[0036] Figure 2 This is a stereoscopic view of the automobile wiper surface defect detection device from the second direction.

[0037] Figure 3 This is a three-dimensional picture of the material table.

[0038] Figure 4 This is a three-dimensional diagram of the robotic arm.

[0039] Figure 5 This is a three-dimensional image of the camera and the gripper.

[0040] Figure 6 This is a three-dimensional image of a separator sponge.

[0041] Figure 7 This is a schematic diagram of the wiper placement status.

[0042] Figure 8 Place a wiper status diagram for the material table.

[0043] Fig. 9 It is a side view of the base and the boss in another embodiment.

[0044] Fig.10 for Fig. 9 Partial cross-sectional view of the middle base and boss.

[0045] Fig.11 for Fig. 9 The middle boss slightly lifts the upper boss at the beginning of adjustment. DETAILED DESCRIPTION

[0046] Example 1

[0047] See also Figures 1 to 8 A vehicle wiper surface defect detection device includes a detection platform 1, a mechanical arm 2 is arranged on the detection platform, and a detection camera 3 (industrial camera) and a clamp 4 are respectively arranged at the ends of the mechanical arm 2. A plurality of material tables 5 are arranged on the detection platform 1 and on one side of the mechanical arm. A material frame 61 for qualified products, a material frame 62 for unqualified products and a material frame 63 for separating sponges are arranged on the detection platform 1 away from the material table. The mechanical arm 2, the camera 3 and the clamp 4 are all connected to a control unit 7.

[0048] In the embodiment, a camera mounting plate 31 is provided at the end of the robot arm 2 near the clamping claw, and an inner folding plate 32 is provided at the end of the camera mounting plate 31 , and the camera 3 is installed at the inner folding plate 32 .

[0049] In the embodiment, the material platform 5 has a base 51, a clamp 52 is provided at one end of the top of the base 51 (the clamp is used to place one end of the wiper), a convex column 53 is provided at the other end (the convex column is installed in the opening at the other end of the wiper), and inclined surfaces 53 are provided on both sides of the base 51. This design can stably place the wiper and facilitate detection.

[0050] In the embodiment, the testing platform 1 comprises a platform 11, and a base frame 12 is arranged under the platform 11. The frame supports the testing platform.

[0051] In the embodiment, an auxiliary light source is provided at the camera 3. This design is conducive to camera shooting.

[0052] In the embodiment, the partition sponge 8 is in strip shape, and a plurality of notches 81 are arranged at equal intervals on the upper side. This design is conducive to placing the wiper.

[0053] In the embodiment, a display 9 is provided on the detection platform 1 , and the display 9 is connected to the control unit 7 .

[0054] The wiper to be inspected is placed on the material table, and the robotic arm moves to the inspection position (wiper), sending a signal to the industrial camera to start continuous shooting and transmit the image to the control unit for defect detection. After the wiper is photographed, it moves to the wiper grabbing position. This process waits for the inspection result. If it is qualified, the robotic arm uses the gripper to put the qualified wiper into the material box for qualified products. If it is unqualified, the robotic arm uses the gripper to put the unqualified wiper into the material box for unqualified products.

[0055] After placing the corresponding material frame, if the corresponding material frame is full in a single layer, the gripper grabs the separation sponge from the separation sponge material frame and places it in the material frame with a full single layer. Taking the qualified product material frame as an example, after a single layer is full, the gripper grabs the separation sponge from the separation sponge material frame and places it in the qualified product material frame (on top of the full layer).

[0056] Then enter the next detection (detect the next wiper), and repeat this process.

[0057] It should be noted that the detection method can refer to the existing technology.

[0058] Embodiment 2: This embodiment discloses a vehicle wiper surface defect detection device, which is different from Embodiment 1 in that, see Figure 9-11 As shown, the boss 53 includes a lower boss 53a and an upper boss 53b. The height of the upper boss is greater than that of the lower boss. A connecting shaft 53a-1 and a square limit block 53a-2 are vertically arranged at the center position of the lower boss. The direction limit block can prevent unnecessary deflection of the upper and lower bosses after installation. The upper boss is provided with a stroke adjustment cavity 53b-1 cooperating with the square limit block, and a side adjustment channel 53c is arranged between the lower and upper bosses.

[0059] The side of the base 51 is provided with a horizontal adjustment hole connected with the boss mounting hole, and the hole wall of the horizontal adjustment hole is provided with a spring compression channel with an inner diameter larger than the inner diameter of the horizontal adjustment hole, and the horizontal adjustment hole is installed with an adjustment rod 55, and the adjustment rod is provided with an annular stop 57, and the adjustment rod is provided with a compression spring 56 located in the spring compression channel, and the end of the adjustment rod is provided with an adjustment protrusion 58 cooperating with the side adjustment channel. Due to the appearance of the square limit block, the stroke adjustment cavity with a square cross-section can cooperate with it, and the upper boss and the lower boss will not deflect relative to each other, so as to avoid the problem of deviation between the position of the side adjustment channel and the adjustment rod.

[0060] Fig.10 is the maximum position state after the upper boss is adjusted upward in height, and Fig.11It is the starting position of the upper boss relative to the lower boss when the upper boss starts to adjust its height (at this time, the upper boss needs to be slightly lifted so that the adjustment protrusion of the adjustment rod can be easily positioned in the side adjustment channel). When the height of different types of wiper sleeves cannot match the height of the boss, the staff will lift the upper boss to the expected height by hand (for example, Fig.10 As shown in the state, the horizontally distributed adjustment rod 55 extends the adjustment protrusion 58 completely into the side adjustment channel 53c matched therewith, and the shape of the adjustment protrusion 58 cooperates with the side adjustment channel 53c. Taking the side adjustment channel as a circular frustum as an example, the adjustment protrusion 58 adopts a circular frustum structure at this time.

[0061] This embodiment is explained by assuming that the boss has two working positions. In the original state, the upper boss and the lower boss can be completely in contact, so that the boss is in the first working state after being installed on the base. At this time, the height of the boss extending out of the upper surface of the base meets the first wiper sleeve size requirement; when it is necessary to match the second wiper sleeve size requirement, the staff slightly lifts the upper boss to allow the adjustment protrusion of the adjustment rod to enter the side adjustment channel until it is in the first working state. Fig.10 In the state shown, the spring keeps the adjusting rod stable. Fig.10 In the state shown, the height of the upper boss extending out of the upper surface of the base meets the size requirement of the second wiper sleeve.

[0062] Fig.10 In the state shown, the height adjustment requirement of the boss can be met, and the boss can be prevented from being pulled out due to the position of the adjustment protrusion.

[0063] When the upper boss and the lower boss are in contact with each other (i.e. the lower surface of the upper boss contacts the upper surface of the lower boss), the adjusting protrusion can be used to press against the sides of the two. Due to the force of the spring, the boss can also be fixed on the side, and the boss will not be pulled out due to the force of the wiper sleeve when the wiper is removed.

[0064] Example 3

[0065] This embodiment relates to a method for detecting surface defects of automobile wipers (defect detection method, which uses the automobile wiper surface defect detection device in Example 1, and includes the following steps:

[0066] (1) Collecting images of car wipers;

[0067] (2) Use the YOLOv8 convolutional neural network model to determine whether the car wiper is qualified;

[0068] (3) According to the judgment results, qualified products are placed in the qualified material box, and unqualified products are placed in the unqualified material box;

[0069] Among them, the YOLOv8 convolutional neural network model training method is as follows:

[0070] (T1) Obtain a dataset of wiper surface defects and divide the dataset into training set, validation set and test set. Set training parameters based on the dataset and train the model. In the training parameters, the batch size is 16, the optimizer is AdamW, the initial learning rate is 0.001, the weight decay is 0.0005, the data enhancement ratio is 0.8, the image size is 512×512, and the training rounds are 150.

[0071] (T2) The model includes a backbone unit, a neck unit and a head unit connected in sequence; the backbone unit includes five Conv modules, four C2f_CA feature extraction modules and one SPPF module; the neck unit includes two Upsamp le upsampling modules, four Concat modules, four C2f modules and two Conv modules, three of which are connected to the head unit, and another C2f module is directly connected to a Concat module. The head unit includes three detection head modules;

[0072] (T3) The CA attention mechanism module in the C2f_CA feature extraction module uses two one-dimensional global pooling to aggregate the input features in the x and y directions into two independent direction-aware feature maps, thereby strengthening the direction and position information around the target and improving the detection accuracy of the target; the specific steps of the CA module are as follows:

[0073] (S1)CA performs coordinated information embedding, that is, input a C*H*W feature map, and two poolings (H, 1) and (1, W) are pooled along the H direction and W direction of the feature map to obtain the processed information features; the horizontal and vertical feature processing formulas are as follows:

[0074]

[0075] Where: and Represents the embedded feature maps in the H and W directions respectively; W, H and x c They are the width, height and current feature value of the feature map respectively;

[0076] (S2) Coordinate Attention Generation is performed, the obtained feature maps are concatenated along the spatial dimension and activated after 1*1 convolution transformation:

[0077] f=δ(F1([Z h ,Z w ])) (3)

[0078] Where: [Z h ,Z w ] means concatenating two feature maps; F1() and δ() represent 1*1 convolution kernel and activation function;

[0079] (S3) Perform a Split operation along the spatial dimension to obtain two separated feature maps and transform and activate them to obtain the attention vector g h and g w , the formula is as follows:

[0080]

[0081] g w =σ(F w (f w )) g h =σ(F h (f h )) (6)

[0082] Where: f, f w and f h They represent the feature map before separation and the feature map in the W and H directions after separation respectively; F w 、F h and σ are the convolution transformation and activation function in two directions respectively;

[0083] (S4) Obtain the corrected feature map y through the broadcast mechanism c :

[0084]

[0085] The hardware and software parameters used in the present invention are as follows: The hardware environment includes Intel(R) Core i5-9400F CPU and NVIDIA GeForce RTX 2080Ti GPU. The deep learning framework uses PyTorch 1.12.0. The evaluation indicators used in the present invention include precision (P), recall (R), average precision (mAP@0.5-0.95) and frames per second (FPS). Table 1 is a network parameter detection result evaluation table.

[0086] Table 1

[0087]

[0088] It can be seen from the table that compared with other object detection models, the model of the present invention shows higher detection accuracy and faster detection speed. Compared with the original YOLOv8, mAP@0.5 is improved by 5.4%. This improvement not only improves the accuracy of detecting small defects such as pits on the wiper, but also maintains a high frame rate per second (FPS).

[0089] The above is only a preferred embodiment of the present invention, and the protection scope of the present invention is not limited to the above embodiment. All technical solutions belonging to the principle of the present invention belong to the protection scope of the present invention. For those skilled in the art, several improvements made without departing from the principle of the present invention should also be regarded as the protection scope of the present invention.

Claims

1. A vehicle wiper surface defect detection device, characterized in that: The invention comprises a testing platform (1), a mechanical arm (2) is arranged on the testing platform, a testing camera (3) and a clamp (4) are arranged at the ends of the mechanical arm (2), a plurality of material tables (5) are arranged on the testing platform (1) and on one side of the mechanical arm, a material frame (61) for qualified products, a material frame (62) for unqualified products and a material frame (63) for separating sponges are arranged on the side of the testing platform (1) away from the material table, wherein the mechanical arm (2), the camera (3) and the clamp (4) are all connected to a control unit (7).

2. The automobile wiper surface defect detection device according to claim 1, characterized in that: A camera mounting plate (31) is arranged at the end of the mechanical arm (2) near the clamping claw, an inner folding plate (32) is arranged at the end of the camera mounting plate (31), and a camera (3) is installed at the inner folding plate (32).

3. The automobile wiper surface defect detection device according to claim 1, characterized in that: The material platform (5) comprises a base (51), a clamping opening (52) is arranged at one end of the top of the base (51), a convex column (53) is arranged at the other end, and inclined surfaces (54) are arranged on both sides of the base (51).

4. The automobile wiper surface defect detection device according to claim 1, characterized in that: The detection platform (1) comprises a platform (11), and a base frame (12) is arranged at the lower side of the platform (11).

5. The automobile wiper surface defect detection device according to claim 1, characterized in that: An auxiliary light source is provided at the camera (3).

6. The automobile wiper surface defect detection device according to claim 1, characterized in that: The partition sponge (8) is in a strip shape, and a plurality of notches (81) are arranged at equal intervals on the upper side.

7. The automobile wiper surface defect detection device according to claim 1, characterized in that: A display (9) is arranged on the detection platform (1), and the display (9) is connected to the control unit (7).

8. The automobile wiper surface defect detection device according to claim 3, characterized in that: The boss (53) comprises a lower boss (53a) and an upper boss (53b); the lower boss is provided with a connecting shaft (53a-1) and a square stop block (53a-2) arranged on the top of the connecting shaft; the upper boss is provided with a stroke adjustment cavity (53b-1) cooperating with the square stop block; and a side adjustment channel (53c) is provided between the lower boss and the upper boss; The base (51) is provided with a horizontal adjustment hole communicated with the boss mounting hole, the hole wall of the horizontal adjustment hole is provided with a spring compression channel with an inner diameter larger than the inner diameter of the horizontal adjustment hole, the horizontal adjustment hole is installed with an adjustment rod (55), the adjustment rod is provided with an annular stopper (57), the adjustment rod is provided with a compression spring (56) located in the spring compression channel, and the end of the adjustment rod is provided with an adjustment protrusion (58) cooperating with the side adjustment channel.

9. A defect detection method, characterized in that: The method uses the automobile wiper surface defect detection device as described in any one of claims 1 to 7, and comprises the following steps: (1) Collecting images of car wipers; (2) Use the YOLOv8 convolutional neural network model to determine whether the car wiper is qualified; (3) According to the judgment results, qualified products are placed in the qualified material box, and unqualified products are placed in the unqualified material box.

10. The defect detection method according to claim 9, characterized in that: in, The YOLOv8 convolutional neural network model training method is as follows: (T1) obtaining a dataset of wiper surface defects, dividing the dataset into a training set, a validation set, and a test set, setting training parameters based on the dataset, and training the model; In the training parameters, the batch size is 16, the optimizer is AdamW, the initial learning rate is 0.001, the weight decay is 0.0005, the data enhancement ratio is 0.8, the image size is 512×512, and the training rounds are 150; (T2) The model includes a backbone unit, a neck unit and a head unit connected in sequence; the backbone unit includes five Conv modules, four C2f_CA feature extraction modules and one SPPF module; the neck unit includes two Upsample upsampling modules, four Concat modules, four C2f modules and two Conv modules, of which three C2f modules are connected to the head unit, and another C2f module is directly connected to a Concat module, and the head unit includes three detection head modules; (T3) The CA attention mechanism module in the C2f_CA feature extraction module uses two one-dimensional global pooling to aggregate the input features in the x and y directions into two independent direction-aware feature maps, thereby strengthening the direction and position information around the target and improving the detection accuracy of the target; the specific steps of the CA module are as follows: (S1) CA performs coordinated information embedding, that is, inputs a C*H*W feature map, and two poolings (H, 1) and (1, W) are performed along the H direction and W direction of the feature map to obtain the processed information features; The horizontal and vertical feature processing formulas are as follows: Where: and Represents the embedded feature maps in the H and W directions respectively; W, H and x c They are the width, height and current feature value of the feature map respectively; (S2) Coordinate Attention Generation is performed, the obtained feature maps are concatenated along the spatial dimension and activated after 1*1 convolution transformation: f=δ(F1([Z h ,Z w ])) (3) Where: [Z h ,Z w ] means concatenating two feature maps; F1() and δ() represent 1*1 convolution kernel and activation function; (S3) Perform a Split operation along the spatial dimension to obtain two separated feature maps and transform and activate them to obtain the attention vector g h and g w , the formula is as follows: g w =σ(F w (f w )) g h =σ(F h (f h )) (6) Where: f, f w and f h They represent the feature map before separation and the feature map in the W and H directions after separation respectively; F w 、F h and σ are the convolution transformation and activation function in two directions respectively; (S4) Obtain the corrected feature map y through the broadcast mechanism c :

Images