Automobile tire defect automatic detection system and method

Abstract

The present invention discloses an automatic detection system and method for automobile tire defects, aiming to solve the problems of automatic detection and classification of automobile tire defects. The automatic detection system for automobile tire defects mainly consists of a cylindrical guide rail (1), a slider (2), a slide base (3), an RGB-D camera (4), a lead screw (5) and a motor (6). The RGB-D camera (4) is a three-dimensional camera with depth and color measurement functions, and the RGB-D camera (4) is placed on the top of the second slider (2) and fixed with bolts. The automatic detection method for automobile tire defects consists of the steps of automatic acquisition of tire tread point cloud of the automatic detection method for automobile tire defects, automatic classification of three-dimensional point cloud of automobile tire features, etc., providing an automatic detection system and method for automobile tire defects with simple structure, high detection accuracy and small classification error.

Description

Technical Field

[0001] The invention relates to an automobile detection device and a detection method, and more specifically, to an automobile tire defect automatic detection system and method. Background Art

[0002] With the continuous development of the automobile industry, intelligent detection and diagnosis of automobiles has also received much attention. As one of the most important parts of a car, the normal working performance of the tire is an important guarantee for traffic safety. Studies have shown that tire defects are one of the main causes of traffic accidents. At present, most of the detection of tire treads is still in the stage of manual observation. This method not only causes certain errors, but also is time-consuming and laborious. How to use machine vision methods to detect the tread of automobile tires and automatically obtain the defective parts and types of tires is very important. In order to realize the three-dimensional point cloud reconstruction of automobile tire morphological characteristics, an automatic detection system and method for automobile tire defects is designed. Summary of the invention

[0003] The present invention aims to solve the problems of low automation and large detection error of the existing technical devices in the process of reconstructing the three-dimensional point cloud features of automobile tires, and proposes an automatic detection system and method for automobile tire defects to realize automatic detection and classification of automobile tire shape defects. The method mainly consists of a cross slide consisting of an RGB-D camera and two small slides. The tire rolls on the ground and uses the RGB-D camera to automatically scan the three-dimensional point cloud of the tire tread to realize automatic detection and classification of the three-dimensional point cloud of the tire.

[0004] In conjunction with the accompanying drawings of the specification, the present invention is implemented using the following technical solutions:

[0005] The automobile tire defect automatic detection system includes a cylindrical guide rail, a slider, a slide base, an RGB-D camera, a lead screw and a motor;

[0006] The slide base is placed in the foundation pit and fixed to the foundation with anchor bolts. The threaded hole of the slider is connected with the thread of the lead screw. The lead screw passes through the holes on both sides of the slide base and matches the clearance of the slide base. The bolt passes through the small circular through hole on one side of the slide base and is fixed with the motor thread. The output shaft of the motor is fixedly connected to one end of the lead screw through a coupling. The through holes of the slider are respectively slidably matched with the two cylindrical guide rails. The two cylindrical guide rails are welded and fixed to the inner side of the slide base. The top surface of the slider is coplanar with the ground. The second slide base is placed on the top surface of the slider and is fixedly connected with the slider thread. The threaded hole of the first slider is threadably matched with the second lead screw, the second lead screw passes through the holes on both sides of the second slide base and matches the clearance of the second slide base, the bolt passes through the small circular through hole on one side of the second slide base and is fixedly connected with the second motor thread, the output shaft of the second motor is fixedly connected to one end of the second lead screw through a coupling, the through holes of the second slider are respectively slidably matched with the two second cylindrical guide rails, the two second cylindrical guide rails are welded and fixed to the inner side of the second slide base, and the RGB-D camera is placed on the top of the second slider and fixed with bolts.

[0007] The cylindrical guide rail described in the technical solution is a smooth cylindrical part.

[0008] The slider described in the technical solution is a rectangular parallelepiped part made of steel material, the top surface of the slider is processed with a threaded hole, and the side surface of the slider is processed with a threaded through hole and a circular through hole.

[0009] The slide base described in the technical solution is a U-shaped part welded from steel plates, with circular through holes processed on both sides of the slide base, and small circular through holes processed around the circular through hole on one side of the slide base.

[0010] The RGB-D camera described in the technical solution is a three-dimensional camera with depth and color measurement functions.

[0011] The lead screw described in the technical solution is a slender cylindrical part made of steel material with cylindrical ends and threads in the middle.

[0012] The motor described in the technical solution is a standard stepper motor or servo motor with a threaded hole processed on one side of the output shaft.

[0013] The specific steps of the automatic detection method for automobile tire defects are as follows:

[0014] Step 1: Automatic collection of tire tread point cloud for automobile tire defect automatic detection method:

[0015] The RGB-D camera is fixedly connected to the slider of the cross slide with bolts, and the RGB-D camera can observe the tire to be inspected; when the car is moving forward, the tire rolls toward the RGB-D camera. According to the tire point cloud signal collected by the RGB-D camera, the motor and the cross slide drive the RGB-D camera to move left and right to the position of the tread of the tire to be inspected, and then keep a constant distance from the tire to be inspected and move forward and backward, so as to extract the three-dimensional point cloud of the tire characteristics to be inspected during the forward and backward movement of the car.

[0016] Step 2: Automatic classification of 3D point cloud of car tire features:

[0017] First, a linear function of the tire point cloud input is established, namely

[0018] z=f(w T x+b)

[0019] Where x is the input vector of the tire point cloud, w is the weight vector parameter of tire point cloud recognition, and b is the bias;

[0020] The common defects on the surface of automobile tires, such as aging cracks, eccentric wear, sharp object punctures, bulges, pattern pockmarks, scratches, cuts, and pits, are divided into K categories. The posterior probability of each defect category is

[0021]

[0022] Among them G k and G j is one of the K tire defect categories;

[0023] For the generative model of the tire defect multi-classification problem, the posterior probability of the tire point cloud classification is given by the activation function transformation, that is,

[0024]

[0025] where i k =w k T θ, θ is a tire defect category G k The eigenvector of

[0026] Then the maximum likelihood method is used to determine the tire point cloud recognition parameter w in the classification model. The likelihood function is:

[0027]

[0028] where y nk =y k (θ n ), t n For a tire defect category G k The eigenvector θ kThe target vector of , T is an N×K matrix of target variables, with elements t nk ;

[0029] The error function is defined by taking the negative logarithm of the likelihood function, and the cross entropy error function of the multi-classification problem of the tire point cloud is obtained, that is,

[0030]

[0031] Take the gradient of the error function with respect to the tire point cloud recognition parameter w, that is,

[0032]

[0033] The error function of the tire point cloud is updated by iteration through the sequential gradient descent method, that is,

[0034]

[0035] Where H is the Hessian matrix, whose elements consist of the second-order derivatives of the tire point cloud error function J(w) with respect to w, that is,

[0036]

[0037] Finally, the trained tire point cloud recognition parameter w is obtained, and the linear function z=f(w T x+b) classifies and predicts different tire defect conditions.

[0038] The beneficial effects of the present invention are:

[0039] 1. The present invention uses an RGB-D camera to reconstruct a three-dimensional point cloud of automobile tire features. First, the RGB-D camera is fixedly connected to the slider of the cross slide, and the tire tread and the RGB-D camera are ensured to be in the same straight line. By using the rolling of the tire during driving, the cross slide drives the RGB-D camera to move forward, backward, left and right, and the three-dimensional point cloud of the automobile tire features can be automatically and comprehensively extracted, and the automobile tire defects can be detected by classifying the point cloud.

[0040] 2. The present invention utilizes the method of machine learning logistic regression to automatically classify the defects of automobile tire features, and adopts the gradient descent method to realize iterative training of tire point cloud classification, which has the advantages of easy implementation, efficient training, and high accuracy.

[0041] 3. The tire defect detection system designed by the present invention solves the problems of low manual detection accuracy and low automation of existing tire detection methods. The RGB-D camera is used to reconstruct and classify the three-dimensional point cloud of automobile tire features, which has a beneficial effect on achieving safe driving of automobiles. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 It is the overall axonometric diagram of the automatic detection system for automobile tire defects;

[0043] Figure 2 It is an axonometric view of a cylindrical guide rail 1, a slider 2, and a slide base 3 in an automatic detection system for automobile tire defects;

[0044] Figure 3 It is an axonometric view of a slider 2 in an automatic automobile tire defect detection system;

[0045] Figure 4 It is an axonometric view of a cylindrical guide rail 1 and a slide base 3 in an automatic detection system for automobile tire defects;

[0046] Figure 5 is an axonometric view of the RGB-D camera 4 in the automatic detection system for automobile tire defects;

[0047] Figure 6 It is an axonometric view of the lead screw 5 in the automatic detection system for automobile tire defects;

[0048] Figure 7 is an axonometric view of motor 6 in an automatic automobile tire defect detection system;

[0049] Figure 8 is a flow chart of an automatic detection method for automobile tire defects;

[0050] In the figure: 1. Cylindrical guide rail, 2. Slider, 3. Slide base, 4. RGB-D camera, 5. Lead screw, 6. Motor. DETAILED DESCRIPTION

[0051] The present invention is further described in detail below in conjunction with the accompanying drawings:

[0052] See also Figures 1 to 7 The automobile tire defect automatic detection system includes a cylindrical guide rail 1, a slider 2, a slide base 3, an RGB-D camera 4, a lead screw 5 and a motor 6.

[0053] The slide base 3 is a U-shaped part welded from steel plates, and circular through holes are processed on both sides of the slide base 3, and small circular through holes are processed around the circular through holes of one example of the slide base 3. The slide base 3 is placed in a foundation pit and fixed to the foundation with anchor bolts, and the lead screw 5 is a slender cylindrical part made of steel material with cylindrical ends and threads processed in the middle, and the slider 2 is a rectangular part made of steel material, and a threaded hole is processed on the top surface of the slider 2, and a threaded through hole and a circular through hole are processed on the side of the slider 2, and the threaded hole of the slider 2 is threadedly connected with the lead screw 5, and the lead screw 5 passes through the holes on both sides of the slide base 3 and is clearance-matched with the slide base 3, and the motor 6 is a standard stepper motor or servo motor with threaded holes processed on one side of the output shaft, and the bolt passes through the small circular through hole on one side of the slide base 3 and is threadedly fixedly connected to the motor 6, and the output shaft of the motor 6 is fixedly connected to one end of the lead screw 5 through a coupling, and the cylindrical guide rail 1 is a smooth cylindrical part, and the through holes of the slider 2 are respectively The two cylindrical guide rails 1 are slidably matched, and the two cylindrical guide rails 1 are welded and fixed to the inner side of the slide base 3. The top surface of the slider 2 is coplanar with the ground. The second slide base 3 is placed on the top surface of the slider 2 and is threadedly fixed to the slider 2. The threaded hole of the second slider 2 is threadedly matched with the second lead screw 5. The second lead screw 5 passes through the holes on both sides of the second slide base 3 and is clearance-matched with the second slide base 3. The bolt passes through the small circular through hole on one side of the second slide base 3 and is threadedly fixed to the second motor 6. The output shaft of the second motor 6 is fixedly connected to one end of the second lead screw 5 through a coupling. The through holes of the second slider 2 are slidably matched with the two second cylindrical guide rails 1 respectively. The two second cylindrical guide rails 1 are welded and fixed to the inner side of the second slide base 3. The RGB-D camera 4 is a three-dimensional camera with depth and color measurement functions. The RGB-D camera 4 is placed on the top of the second slider 2 and is fixed with bolts.

[0054] See also Figure 8 ,The automatic detection method of automobile tire defects can be divided into the following two steps:

[0055] Step 1: Automatic collection of tire tread point cloud for automobile tire defect automatic detection method:

[0056] The RGB-D camera 4 is fixedly connected to the slider 2 of the cross slide with bolts, and the RGB-D camera 4 can observe the tire to be inspected; when the car is moving forward, the tire rolls toward the RGB-D camera 4, and according to the tire point cloud signal collected by the RGB-D camera 4, the motor 6 and the cross slide drive the RGB-D camera 4 to move left and right to the position of the tread of the tire to be inspected, and then keep a constant distance from the tire to be inspected and move forward and backward, so as to extract the three-dimensional point cloud of the characteristics of the tire to be inspected during the forward and backward movement of the car.

[0057] Step 2: Automatic classification of 3D point cloud of car tire features:

[0058] First, a linear function of the tire point cloud input is established, namely

[0059] z=f(w T x+b)

[0060] Where x is the input vector of the tire point cloud, w is the weight vector parameter of tire point cloud recognition, and b is the bias;

[0061] The common defects on the surface of automobile tires, such as aging cracks, eccentric wear, sharp object punctures, bulges, pattern pockmarks, scratches, cuts, and pits, are divided into K categories. The posterior probability of each defect category is

[0062]

[0063] Among them G k and G j is one of the K tire defect categories;

[0064] For the generative model of the tire defect multi-classification problem, the posterior probability of the tire point cloud classification is given by the activation function transformation, that is,

[0065]

[0066] where i k =w k T θ, θ is a tire defect category G k The eigenvector of

[0067] Then the maximum likelihood method is used to determine the tire point cloud recognition parameter w in the classification model. The likelihood function is:

[0068]

[0069] where y nk =y k (θ n ), t n For a tire defect category G k The eigenvector θ k The target vector of , T is an N×K matrix of target variables, with elements t nk ;

[0070] The error function is defined by taking the negative logarithm of the likelihood function, and the cross entropy error function of the multi-classification problem of the tire point cloud is obtained, that is,

[0071]

[0072] Take the gradient of the error function with respect to the tire point cloud recognition parameter w, that is,

[0073]

[0074] The error function of the tire point cloud is updated by iteration through the sequential gradient descent method, that is,

[0075]

[0076] Where H is the Hessian matrix, whose elements consist of the second-order derivatives of the tire point cloud error function J(w) with respect to w, that is,

[0077]

[0078] Finally, the trained tire point cloud recognition parameter w is obtained, and the linear function z=f(w T x+b) classifies and predicts different tire defect conditions.

Claims

1. A detection method of an automatic detection system for automobile tire defects, the automatic detection system for automobile tire defects comprising a cylindrical guide rail (1), a slider (2), a slide base (3), an RGB-D camera (4), a lead screw (5) and a motor (6); The slide base (3) is placed in the foundation pit and fixed to the foundation with anchor bolts. The threaded hole of the slider (2) is threadedly connected with the lead screw (5). The lead screw (5) passes through the holes on both sides of the slide base (3) and is gap-matched with the slide base (3). The bolt passes through the small circular through hole on one side of the slide base (3) and is threadedly fixedly connected with the motor (6). The output shaft of the motor (6) is fixedly connected to one end of the lead screw (5) through a coupling. The through hole of the slider (2) is respectively slidably matched with the two cylindrical guide rails (1). The two cylindrical guide rails (1) are welded and fixed to the inner side of the slide base (3). The top surface of the slider (2) is coplanar with the ground. The second slide base (3) is placed on the top surface of the slider (2) and is threadedly fixedly connected with the slider (2). The threaded holes of the two sliders (2) are threadedly matched with the second lead screw (5), the second lead screw (5) passes through the holes on both sides of the second slide base (3) and is loosely matched with the second slide base (3), the bolt passes through the small circular through hole on one side of the second slide base (3) and is threadedly fixedly connected with the second motor (6), the output shaft of the second motor (6) is fixedly connected to one end of the second lead screw (5) through a coupling, the through holes of the second slider (2) are respectively slidably matched with the two second cylindrical guide rails (1), the two second cylindrical guide rails (1) are welded and fixed to the inner side of the second slide base (3), the RGB-D camera (4) is placed on the top of the second slider (2) and fixed with bolts, characterized in that: The specific steps are as follows: Step 1: Automatic collection of tire tread point cloud for automatic tire defect detection method: The RGB-D camera (4) is fixedly connected to the slider (2) of the cross slide by bolts, and the RGB-D camera (4) can observe the tire to be inspected; when the car is moving forward, the tire rolls toward the RGB-D camera (4), and according to the tire point cloud signal collected by the RGB-D camera (4), the motor (6) and the cross slide drive the RGB-D camera (4) to move left and right to the position of the tread of the tire to be inspected, and then keep a constant distance from the tire to be inspected and move forward and backward, thereby extracting the three-dimensional point cloud of the characteristics of the tire to be inspected during the forward and backward movement of the car; Step 2: Automatic classification of 3D point cloud of car tire features: First, a linear function of the tire point cloud input is established, namely z=f(w T x+b) Where x is the input vector of the tire point cloud, w is the weight vector parameter of tire point cloud recognition, and b is the bias; The defects of aging cracks, eccentric wear, sharp object punctures, bulges, pattern pitting, scratches, cuts, and pits on the surface of automobile tires are divided into K categories, and the posterior probability of each defect category is Among them G k and G j is one of the K tire defect categories; For the generative model of the tire defect multi-classification problem, the posterior probability of the tire point cloud classification is given by the activation function transformation, that is, where i k =w k T θ, θ is a tire defect category G k The eigenvector of Then the maximum likelihood method is used to determine the weight vector parameter w of tire point cloud recognition in the classification model. The likelihood function is: where y nk =y k (θ n ), t n For a tire defect category G k The eigenvector θ k The target vector of , T is an N×K matrix of target variables, with elements t nk ; The error function is defined by taking the negative logarithm of the likelihood function, and the cross entropy error function of the multi-classification problem of the tire point cloud is obtained, that is, Take the gradient of the error function with respect to the weight vector parameter w of tire point cloud recognition, that is, The error function of the tire point cloud is updated by iteratively updating the weight vector parameter w of tire point cloud recognition through the sequential gradient descent method, that is, Where H is the Hessian matrix, whose elements are composed of the second-order derivatives of the tire point cloud error function J(w) with respect to the weight vector parameter w of tire point cloud recognition, that is, Finally, the weight vector parameter w of the trained tire point cloud recognition is obtained, and the linear function z=f(w T x+b) classifies and predicts different tire defect conditions.

2. The detection method of the automobile tire defect automatic detection system according to claim 1, characterized in that The cylindrical guide rail (1) is a smooth cylindrical part.

3. The detection method of the automobile tire defect automatic detection system according to claim 1, characterized in that The slider (2) is a rectangular parallelepiped part made of steel material. The top surface of the slider (2) is processed with a threaded hole, and the side surface of the slider (2) is processed with a threaded through hole and a circular through hole.

4. The detection method of the automobile tire defect automatic detection system according to claim 1, characterized in that The slide base (3) is a U-shaped part welded from steel plates. Circular through holes are processed on both sides of the slide base (3), and small circular through holes are processed around the circular through hole on one side of the slide base (3).

5. The detection method of the automobile tire defect automatic detection system according to claim 1, characterized in that The RGB-D camera (4) is a three-dimensional camera with depth and color measurement functions.

6. The detection method of the automobile tire defect automatic detection system according to claim 1, characterized in that The lead screw (5) is a slender cylindrical part made of steel material with cylindrical ends and a thread processed in the middle.

7. The detection method of the automobile tire defect automatic detection system according to claim 1, characterized in that The motor (6) is a standard stepping motor or a servo motor with a threaded hole processed on one side of the output shaft.

Images