A method and system for detecting internal bubble defects of tires based on tactile perception

Abstract

The present invention relates to the technical field of tire detection, and particularly to a method and system for detecting internal bubble defects of tires based on tactile perception. The method includes fixing and calibrating the spatial coordinates of the tire to be detected, driving a three-axis force sensor to scan row by row along the tire surface by using a multi-axis motion mechanism, collecting the force data in the X, Y, and Z directions under different applied normal forces, and performing noise filtering, zero-point drift correction, and interpolation processing on the data to construct a position-mechanical feature data set. By constructing a multi-force-level mechanical fingerprint map and comparing it with a standard tire mechanical fingerprint library, abnormal low-stiffness or deformation regions are automatically identified, and the bubble size and defect severity are estimated. This method overcomes the disadvantages of high cost, complex operation, and low detection efficiency of traditional X-ray and ultrasonic detection equipment, and has the advantages of high detection accuracy, good real-time performance, and strong adaptability, and is suitable for on-line detection and quality control of internal defects in tire production, retreading, and maintenance.

Description

Technical Field

[0001] The present invention relates to the technical field of tire detection, and particularly to a method and system for detecting internal bubble defects of a tire based on tactile perception. Background Art

[0002] Tires are one of the crucial components in modern transportation vehicles, and their safety directly affects driving safety. Quality control of tires, especially the detection of internal defects, is an important link to ensure tire performance and safety. During the manufacturing, use, and maintenance of tires, internal defects such as air bubbles may occur due to uneven materials, process problems, or external factors. These defects may not be easily detected in the initial stage, but during tire use, especially under high-speed driving or high-load conditions, they may lead to structural damage to the tire and even cause traffic accidents. Therefore, accurately and effectively detecting internal defects of tires is the key to improving tire quality management and use safety.

[0003] Traditional methods for detecting internal defects of tires mainly rely on technologies such as X-ray imaging, ultrasonic detection, and optical imaging. Although these methods can provide relatively accurate detection results, they usually have limitations such as high cost, complex operation, low detection efficiency, and the need for professional equipment. In addition, methods such as X-ray and ultrasonic can only provide relatively fuzzy images or data for the internal structure and defects of the tire, and are greatly affected by the complex surface morphology of the tire, resulting in low detection accuracy and sensitivity.

[0004] As an emerging non-destructive testing technology, tactile perception technology has the advantages of high sensitivity, real-time performance, and convenience. Chinese Patent Application for Invention (Publication No.: CN116818172A, Publication Date: September 29, 2023) discloses a method and device for detecting a flexible interface based on tactile perception. The method includes the following steps: controlling the three-dimensional displacement stage to move downward along the Z-axis direction, and obtaining the Z-axis force value of the three-axis force sensor in real time. When the force value reaches a preset threshold, stop moving and establish a reference plane for the sensor to scan the flexible interface; on the reference plane, synchronously control the data acquisition of the three-axis force sensor and the displacement stage based on a hard trigger method; perform a row-by-row scan of the flexible interface based on the data acquisition method, and collect the force change data in the X, Y, and Z directions of the contact between the sensor and the flexible interface as three-dimensional tactile information; perform fusion processing and calculation on the three-dimensional force data to achieve high-precision, high-resolution, multi-dimensional tactile scanning imaging of the flexible interface, and finally realize the detection of the flexible interface based on tactile perception.

[0005] However, the current tactile sensing technology used in tire internal defect detection is still immature. In particular, there are still some urgent problems to be solved in terms of how to combine scans with different force levels, how to accurately identify internal bubble defects, and how to efficiently process data. Summary of the invention

[0006] In order to solve the above-mentioned technical problems, the purpose of the present invention is to provide a method for detecting bubble defects inside tires based on tactile perception. The method combines multi-force-level scanning with data processing, uses a three-axis force sensor to obtain the force information on the tire surface, and then constructs a multi-force-level mechanical fingerprint map and compares it with a standard fingerprint library to accurately determine whether there are defects such as bubbles inside the tire, thereby overcoming the shortcomings of traditional methods and providing a high-precision and efficient solution for the field of tire detection.

[0007] In order to achieve the above-mentioned purpose, the present invention adopts the following technical solutions:

[0008] A method for detecting air bubble defects inside a tire based on tactile perception, the method comprising the following steps:

[0009] 1) Tire fixing: install the tire to be tested on a rotatable or clampable positioning device, and calibrate the tire's spatial coordinate system;

[0010] 2) Initial loading: Control the multi-axis motion mechanism to drive the three-axis force sensor to move in the vertical direction and contact the outer surface of the tire. When the normal force of the sensor reaches the first preset threshold, a detection reference surface is established;

[0011] 3) Multi-level scanning:

[0012] 3.1) With the detection reference plane as a reference, a scanning path is set on the outer surface of the tire, and normal forces of different magnitudes are applied to the same scanning point in sequence, and the force data in the three directions of X, Y, and Z output by the three-axis force sensor under each level of force loading is recorded;

[0013] 3.2) Repeat the above steps to scan the outer surface of the tire line by line or partition by partition to collect three-dimensional force data of multiple force levels;

[0014] 4) Data preprocessing: noise filtering, zero drift correction and interpolation processing are performed on the collected three-dimensional force data, and the force changes of the same scanning point under different force loading are integrated to generate the corresponding position-mechanical characteristic data set;

[0015] 5) Defect identification:

[0016] 5.1) constructing a multi-force-level mechanical fingerprint based on the position-mechanical feature data set;

[0017] 5.2) Compare the multi-force-level mechanical fingerprint map with a pre-established standard mechanical fingerprint library. When it is found that a certain local area exhibits abnormally low stiffness or deformation characteristics, it is determined that there are internal bubble defects in this area;

[0018] 5.3) Output the location and defect degree information of the bubble defect area.

[0019] Preferably, the multi-force-level scanning in step 3) includes first performing a preliminary scan with a smaller force value and then performing a secondary fine scan on the suspicious area with a higher force value to distinguish the deformation differences caused by surface micro-protrusions and deep bubbles.

[0020] Preferably, the data preprocessing in step 4) also includes filtering or compensating for the periodic fluctuations generated by the tire surface pattern to reduce the interference of the tread pattern on the identification of internal defects.

[0021] Preferably, the signals output by the three-axis force sensor are synchronously collected by hardware triggering on the FPGA or SoC platform, specifically including counting the pulse signals of the motion mechanism and triggering sampling when the count reaches a preset value, so as to ensure the accurate correspondence between the spatial coordinates and the force data in different force loading stages.

[0022] Preferably, in the defect identification step 5), the multi-task deep fingerprint network realizes intelligent defect identification based on multi-force-level scanning data through the joint tasks of defect area segmentation and bubble size regression; it specifically includes an encoder-decoder structure, an attention module, and a dual-branch output design, and uses a multi-task loss function for joint training; in the inference stage, the network can automatically extract defect features from the preprocessed mechanical fingerprint map and output the defect area, bubble size, and defect severity.

[0023] Furthermore, the present invention also provides a tire internal bubble defect detection system based on tactile perception for the above method. The system includes:

[0024] A tire positioning device for fixing the tire to be detected and providing rotation or clamping functions;

[0025] A multi-axis motion mechanism, which is disposed opposite to the tire positioning device and has at least one axis moving in the horizontal direction and one axis moving in the vertical direction;

[0026] A three-axis force sensor, which is installed at the end of the multi-axis motion mechanism and is used to detect the force values in the X, Y, and Z directions when contacting the tire surface;

[0027] A force loading unit, which cooperates with the three-axis force sensor and is used to apply multi-level normal forces of different magnitudes to the tire surface during the scanning process;

[0028] A data acquisition control unit, connected to the three-axis force sensor, is used to collect three-axis force data based on a hard trigger method and synchronize it with the spatial position of the multi-axis motion mechanism;

[0029] A data processing and defect identification module, communicating with the data acquisition control unit, is used to preprocess the collected multi-force-level three-dimensional force data and execute a defect identification algorithm, and output a judgment result on the internal defects of the tire.

[0030] Preferably, the force loading unit includes a lead screw mechanism or a cylinder mechanism, and is configured with a force closed-loop controller, which can dynamically adjust the applied force of the three-axis force sensor within a specified force value range.

[0031] Preferably, the three-axis force sensor has a replaceable wear-resistant probe or an anti-slip cushion layer to adapt to tire surfaces of different specifications and materials and reduce probe wear.

[0032] Preferably, the data acquisition control unit includes an FPGA or an SoC platform, configured with a pulse counter and a high-speed AD conversion circuit. By counting the pulse signals of the stepping or servo motor of the multi-axis motion mechanism, sampling of the three-axis force sensor is triggered when a predetermined pulse value is reached, so as to realize real-time synchronization of the scanning position and the force data.

[0033] Preferably, the data processing and defect identification module establishes a standard tire mechanical fingerprint library, compares it based on the multi-force-level scanning data of different models and specifications of tires in a normal state, and when there is a difference within a preset threshold range between the detected data and the standard data, it is determined that there are defects such as air bubbles or delamination inside the tire, and a visual report of the defects is output.

[0034] Furthermore, the present invention also provides a computer-readable storage medium, on which a computer program or instruction is stored, and when the computer program or instruction is executed by a processor, the steps 4)-5) of the method are implemented.

[0035] Furthermore, the present invention also provides a computer program product, including a computer program or instruction, and when the computer program or instruction is executed by a processor, the steps 4)-5) of the method are implemented.

[0036] Due to the adoption of the above technical solution, the present invention realizes efficient and accurate detection of internal air bubble defects of tires by innovatively combining multi-force-level scanning, data preprocessing and defect identification algorithms, overcomes the deficiencies of the prior art, and has the following remarkable technical effects:

[0037] 1. Improve the detection accuracy of internal air bubble defects in tires: The present invention applies normal forces of different magnitudes to the outer surface of the tire, gradually scans and collects the three-dimensional force data of the triaxial force sensor under different force loadings, thereby obtaining the mechanical responses at multiple force levels. Through multi-force-level scanning, it is possible to more meticulously analyze the minute deformations on the tire surface and the stiffness changes caused by internal air bubbles, achieving precise identification of internal air bubble defects in the tire. Compared with traditional detection methods such as X-rays and ultrasonic waves, the present invention can better distinguish between surface micro-protrusions and deep-layer air bubbles, significantly improving the detection sensitivity and accuracy of air bubble defects.

[0038] 2. Reduce the interference of surface patterns on detection: In the case where the surface patterns of the tire are relatively complex, traditional detection methods may be interfered by the patterns, resulting in inaccurate detection results. The present invention filters or compensates for the periodic fluctuations generated by the patterns during the data preprocessing stage, effectively reducing the interference of the patterns on the identification of internal defects and improving the reliability of the data. This technology effectively solves the problem that traditional tactile detection cannot eliminate pattern interference, enabling accurate internal defect detection even when the surface patterns of the tire are relatively complex.

[0039] 3. Enhance the synchronization and real-time performance of data acquisition and processing: The present invention uses an FPGA or SoC platform to perform hardware-triggered synchronous acquisition of the output signals of the triaxial force sensor, ensuring accurate correspondence between spatial coordinates and force data at different force loading stages. This hard-triggered synchronous acquisition method not only improves the real-time performance of data acquisition but also greatly reduces the possibility of data mismatch, guaranteeing high precision and high efficiency during the detection process. Compared with traditional software-controlled sampling methods, the present invention has significant advantages in terms of speed and accuracy.

[0040] 4. Achieve automatic identification and intelligent judgment of internal tire defects: During the defect identification process, the present invention uses machine learning or deep learning models to extract features from the multi-force-level mechanical fingerprint maps, which can automatically judge the size and distribution position of internal air bubbles in the tire and output a severity score of the defects. This intelligent defect judgment mechanism greatly improves the automation level and accuracy of tire detection, reducing the errors and subjective interference of manual judgment. This technology not only improves the detection efficiency but also realizes the intelligence and automation of the tire detection process to a certain extent.

[0041] 5. Reduce the detection cost and improve the detection efficiency: The present invention performs tire defect detection based on tactile sensing technology, eliminating the need for expensive X-ray equipment or ultrasonic detection equipment, significantly reducing the equipment cost of detection. The tactile sensing system has a simple structure, is easy to integrate and deploy, and can achieve rapid detection, making it particularly suitable for large-scale tire production lines or tire retreading factories. While improving the detection efficiency, it can also provide high-precision defect detection results to meet the requirements of industrial applications.

[0042] 6. Strong adaptability and can handle different types of tire detection requirements: The detection method and system of the present invention have strong adaptability and can handle the detection requirements of tires of different types, specifications and patterns. Whether it is a car tire, a truck tire or a construction machinery tire, the present invention can achieve precise detection by adjusting parameters such as the scanning path, loading force, and scanning accuracy. At the same time, the system can also be adaptively adjusted according to different characteristics and defect types on the tire surface, further improving the flexibility and accuracy of the detection process.

[0043] In summary, through the innovative application of tactile perception technology, the present invention realizes the efficient and precise detection of internal bubble defects in tires, with high detection accuracy, real-time performance, adaptability and intelligent level. Compared with the prior art, the present invention has significant technical advantages in improving tire quality control ability, reducing detection costs, and improving detection efficiency, and is an important technological breakthrough in the fields of tire manufacturing, retreading and repair. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 It is a flowchart of the method of the present invention.

[0045] Figure 2 It is a flowchart of the multi-force level scanning method of the present invention.

[0046] Figure 3 It is a flowchart of the data preprocessing of the present invention.

[0047] Figure 4 It is a flowchart of the defect identification of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0048] In order to enable those skilled in the art to understand and implement the technical solutions of the present invention, the following details a specific implementation manner of a method and system for detecting internal bubble defects in tires based on tactile perception. Through the introduction of this implementation manner, technicians can achieve the technical effects of the present invention and perform efficient and accurate detection of internal bubble defects in tires. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of them. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative work shall fall within the protection scope of the present invention.

[0049] Embodiment 1

[0050] As Figure 1 shown, a method for detecting internal bubble defects in tires based on tactile perception, the method comprising the following steps:

[0051] 1. Tire fixation and spatial coordinate calibration

[0052] First, install the tire to be detected onto the detection platform. The detection platform includes a tire positioning device for fixing the tire and ensuring its stability during the scanning process. The positioning device may include a clamping system or a turntable system. The turntable system allows the tire to rotate during the scanning process, enabling the sensor to scan the entire surface of the tire. To ensure the accurate spatial position of the tire, a three-dimensional positioning sensor is used for calibrating the spatial coordinates.

[0053] 2. Initial Loading and Datum Plane Establishment

[0054] After the tire is fixed, the system activates the multi-axis motion mechanism, which consists of multiple motion axes. At least one axis moves horizontally and another axis moves vertically. At this time, the three-axis force sensor is driven to move vertically and come into contact with the tire surface. The normal force of the sensor gradually increases. When the normal force reaches the preset first threshold, the system stops the vertical movement of the sensor and uses this position as the datum plane to start the subsequent scanning of the tire.

[0055] 3. Multi-Force-Level Scanning

[0056] After establishing the datum plane, the system starts multi-force-level scanning. As Figure 2 shown, the specific steps are as follows:

[0057] Step 3.1: Based on the set scanning path, the system applies different magnitudes of normal forces and records the three-axis force data (force values in the X, Y, and Z directions) of the sensor at each force level. This can fully capture the mechanical responses on and inside the tire surface.

[0058] Step 3.2: The sensor scans the tire surface row by row along the set scanning path. Each time it scans, the sensor collects multiple sets of three-dimensional force data according to the loading at different force levels. During the scanning process, the sensor can carefully capture the minute deformations on the tire surface, especially the changes in mechanical characteristics caused by internal air bubbles in the tire.

[0059] 4. Data Preprocessing and Fusion

[0060] During the multi-force-level scanning process, the collected three-dimensional force data may contain noise, so the data needs to be preprocessed. As Figure 3 shown, the steps of preprocessing include:

[0061] Noise filtering: Remove the sensor noise through digital filtering algorithms.

[0062] Zero drift correction: Correct the initial zero drift of the sensor to ensure the accuracy of the force values.

[0063] Interpolation processing: Interpolate the interval data in the scanning path to ensure the smoothness and continuity of the data.

[0064] The processed data is fused into the corresponding position-mechanical feature dataset, which is the basis for subsequent defect identification.

[0065] 5. Defect Identification and Bubble Detection

[0066] After completing the data preprocessing, the system enters the defect identification stage. As Figure 4 shown, the specific steps are as follows:

[0067] Step 5.1: Based on the position-mechanical feature dataset, construct multi-force-level mechanical fingerprint maps. These mechanical fingerprint maps are the graphical representations of the force responses at different positions on the tire surface, and can reflect the mechanical characteristics and potential defect areas of the tire.

[0068] Furthermore, according to the preprocessed position-mechanical feature dataset, the three-dimensional force data (in the X, Y, and Z directions) at each scanning position is regarded as a data point. By arranging these data in a two-dimensional space, a multi-force-level force distribution map is generated. Each row represents a scanning path of the tire (scanning along the X-axis), and each column represents the mechanical response at a certain position on the tire surface. The corresponding each point is the force values in the X, Y, and Z directions, and different force intensities can be displayed through color gradient (blue represents low force, and red represents high force).

[0069] Generation of mechanical fingerprint maps: Map the force data (the force in the X direction) of each scanning path to a heat map. The force data in the X, Y, and Z directions can be respectively plotted into separate mechanical images (X force map, Y force map, Z force map). These images can be generated from the scanning results of different force levels, and the data of multiple force levels can be spliced together to form a three-dimensional mechanical fingerprint map, where each dimension corresponds to different force conditions.

[0070] Step 5.2: Compare the constructed mechanical fingerprint maps with the standard tire mechanical fingerprint library. The standard tire mechanical fingerprint library includes the multi-force-level scanning data of different types of tires in the normal state. Through comparison, the system can judge which areas have significant differences in force characteristics from normal tires.

[0071] Standard mechanical fingerprint library: The standard fingerprint library contains the multi-force-level scanning data of different types of tires in the normal state. Each standard tire provides a mechanical fingerprint map, and these images represent the force responses of this type of tire at different scanning force levels. The standard fingerprint library is usually collected by tire manufacturers or laboratories based on a large amount of normal tire data. The dimension of each fingerprint map is the same as that of the mechanical fingerprint map of the actual detected tire.

[0072] Furthermore, the comparison method adopted by the present invention is as follows: Through an image matching algorithm, the mechanical fingerprint map of the current tire is compared with the fingerprint maps in the standard tire fingerprint library. Common image matching algorithms include:

[0073] Mean Square Error (MSE): Calculate the mean square error between two mechanical fingerprint maps. Regions with larger errors usually indicate abnormalities.

[0074] Correlation coefficient: By calculating the correlation between two images, the similarity is judged. Regions with low correlation may indicate defects.

[0075] Local feature matching: When comparing images, the local regions in the images can be selectively compared to find the parts with differences on the tire surface.

[0076] Output the comparison result: If the comparison result shows that the force characteristics in some regions are significantly different from those of normal tires (such as large normal force, frictional force or deformation amplitude), the system will mark these regions as potential defect regions, especially problems such as possible bubbles or delamination.

[0077] Step 5.3: When the comparison result shows that a certain local region exhibits abnormally low stiffness or deformation characteristics, the system determines that there are internal bubbles or other defects in this region. The system further outputs information such as the location of the defect, the bubble size, and the degree of the defect for subsequent quality assessment and processing.

[0078] Furthermore, the defect feature recognition of the present invention is as follows: When the comparison result shows that the force characteristics (low stiffness, deformation characteristics) of a certain local region are significantly different from those of normal tires, the system will automatically identify this region as a potential defect through threshold determination.

[0079] Low stiffness: The bubble region usually exhibits lower stiffness, that is, under the same applied force, the deformation amount in this region will be larger.

[0080] Abnormal deformation: By comparing the changes in normal force and lateral force, if the deformation in a certain region is more significant than other regions, it can be determined that there are internal bubbles.

[0081] Furthermore, the output defect information of the present invention is as follows: The system outputs information such as the defect location, bubble size, and degree of the defect, and generates a visualization report. This report will show the detected defect regions (through heat maps or marked regions), and at the same time provide detailed information about the defects (such as the radius, depth or deformation amplitude of the bubbles).

[0082] Furthermore, the present invention also establishes the relationship between the bubble size and the force change. The following is a simplified model based on the stiffness of the tire material and the bubble size:

[0083] Effect of Bubbles: Assume that the bubbles inside the tire have a linear effect on the mechanical response of the tire surface, i.e., the size of the bubbles is proportional to the change in force.

[0084] Estimation of Bubble Size:

[0085] 1) Select a batch of tire samples, where the known internal bubble sizes (such as diameters) are obtained manually or by means of X-ray detection, etc. At the same time, measure the normal force values in the normal areas and bubble areas of these samples under the same loading conditions.

[0086] 2) Under the same standard applied loads (10N, 50N, 100N), record for each sample:

[0087] The normal force F in the normal area normal ;

[0088] The normal force F in the bubble area defect ;

[0089] Calculate the difference in normal force in this area:

[0090] ΔF = F defect −F normal .

[0091] 3) Establish a calibration model

[0092] Perform statistical analysis on multiple sets of data (ΔF and the corresponding bubble diameter D), and use methods such as linear regression to fit and obtain an empirical formula:

[0093]

[0094] where a and b are constants obtained by experimental fitting.

[0095] Furthermore, in order to improve the accuracy of bubble detection, the present invention first performs a preliminary scan with a smaller force value during implementation, and then performs a secondary fine scan on the suspicious area with a higher force value. This method can effectively distinguish the difference between the deep deformation caused by bubbles and the surface micro-protrusions, and improve the recognition accuracy of bubble defects.

[0096] To further improve the intelligent level of detection, the present invention adopts a machine learning or deep learning model in defect recognition. Specifically, a deep neural network can be used to automatically extract the features of the multi-force-level mechanical fingerprint map, and based on these features, judge the size and distribution position of the bubbles inside the tire. This intelligent defect judgment mechanism reduces manual intervention and improves the automation degree and accuracy of detection.

[0097] 6. System Hardware Implementation

[0098] The system hardware of the present invention adopts the flexible interface detection method of a Chinese patent application for invention (Publication No.: CN116818172A, Publication Date: September 29, 2023), and specifically includes the following key components:

[0099] Tire positioning device: It is used to stably fix the tire to be detected and support rotational or clamping operations.

[0100] Multi-axis motion mechanism: It includes multiple adjustable axes and is used to drive the three-axis force sensor to scan the tire surface.

[0101] Three-axis force sensor: It is used to collect three-axis force data in real time when the tire surface contacts the sensor.

[0102] Force loading unit: It is used to apply normal forces of different magnitudes to the tire surface during the scanning process to ensure that all areas to be detected are covered.

[0103] Data acquisition and control unit: It is responsible for synchronizing the output signals of the three-axis force sensor and precisely corresponding the data with the spatial position of the tire scanning.

[0104] Data processing and defect identification module: It is used to preprocess the collected three-dimensional force data and execute the defect identification algorithm, and finally output the judgment result of the internal defects of the tire.

[0105] 7. Adaptability and scalability

[0106] The detection method and system of the present invention have strong adaptability and can be adaptively adjusted according to the specifications, patterns, and surface characteristics of different tires. The system can flexibly configure parameters such as scanning paths, loading forces, and scanning accuracies to meet the detection requirements of different types of tires. For example, car tires, truck tires, or construction machinery tires can all achieve high-precision detection by appropriately adjusting the parameters.

[0107] To better understand the application of the present invention, the following provides a specific application example of a method for detecting internal bubble defects in tires based on tactile perception. In this example, a car tire known to have internal bubble defects during the production process is selected. It is difficult to detect internal bubbles in this tire through traditional detection (manual knocking), but there are potential safety hazards in actual use. To verify the effectiveness of the method of the present invention, a multi-force-level scanning detection system based on tactile perception is used to detect the tire, and the detection results are compared with the images collected by an X-ray detection device.

[0108] 1. Detection system and experimental conditions

[0109] 1) System configuration:

[0110] Tire positioning device: A fixed clamping device is used to firmly fix the tire, and a rotating table is configured to enable the tire to rotate horizontally for comprehensive scanning.

[0111] Multi-axis motion mechanism: This mechanism controls the three-axis force sensor to move along the X and Y (or row and column) directions on the tire surface to ensure a uniform scanning path.

[0112] Three-axis force sensor: A high-precision sensor that can simultaneously collect force data in the X, Y, and Z directions; the sensor has been pre-filtered for noise and zero-corrected.

[0113] Force loading unit: A cylinder or a precision lead screw mechanism is used to apply different normal forces (10N, 50N, 100N) to the tire in different scanning areas to capture multi-force level response data.

[0114] Data acquisition and processing unit: Based on the FPGA / SoC platform, it synchronously collects and processes data in real time, constructs a position-mechanical feature dataset through steps such as preprocessing, interpolation, and normalization, and generates a multi-dimensional mechanical response map (such as an RGB fusion map or a heat map).

[0115] Defect identification module: Using a pre-established standard tire mechanical fingerprint library and a simplified model, it compares and automatically identifies the scanned data, outputs the defect area, estimates the bubble size and the severity of the defect.

[0116] 2) Experimental conditions:

[0117] Ambient temperature: 25°C, relative humidity: 60%

[0118] Scanning resolution: Sampling approximately once every 1 mm, a total of 100 scanning position data are collected.

[0119] Normal force loading scheme: First, a preliminary scan is performed with 10N, and then 50N and 100N are applied for a secondary scan for suspected defect areas.

[0120] 2. Detection steps and data acquisition

[0121] Step 1: Tire fixation and coordinate calibration

[0122] Fix the tire on the positioning device and use a laser alignment system for spatial coordinate calibration to ensure that the tire does not shift during the scanning process.

[0123] Step 2: Initial loading and reference plane establishment

[0124] Control the multi-axis motion mechanism to drive the three-axis force sensor to move vertically so that the sensor first contacts the outer surface of the tire. When the sensor detects that the normal force reaches 10N, a reference plane is established.

[0125] Step 3: Multi-force-level Scanning

[0126] Initial Scanning: Apply a load of 10 N and sequentially collect data at 100 scanning positions along the tire surface, recording the force values in the X, Y, and Z directions at each position.

[0127] Fine Scanning: For areas where abnormal forces are detected during the initial scanning (e.g., the normal force at some positions is significantly higher than the surrounding normal data), perform a secondary scan at 50 N and 100 N to collect more detailed multi-force-level data. Part of the scanning data of the tire is shown in Table 1:

[0128] Table 1 Scanning Data after Tire Pretreatment

[0129]

[0130] For scanning position 52, abnormal initial scanning data was found. Subsequently, a fine scan was performed on this position under 50 N and 100 N loads to obtain more detailed data.

[0131] Step 4: Data Pretreatment and Construction of Multi-dimensional Mechanical Response Diagram

[0132] Perform noise filtering, zero-drift correction, and interpolation on all collected data to form a complete position-mechanical feature dataset. Smooth and normalize the data matrix through Python, and then plot the data in the X, Y, and Z directions as heatmaps or generate a multi-dimensional mechanical response diagram through RGB fusion.

[0133] Step 5: Defect Identification

[0134] Construct a Mechanical Fingerprint Diagram: Generate multi-force-level heatmaps using the processed data, especially the response diagram in the Z direction (normal force). It is observed that at scanning position 52, the force in the Z direction is significantly higher than the surrounding area at all loading levels, and the image color shows obvious abnormalities.

[0135] Compare with the Standard Fingerprint Library: Compare the currently collected mechanical fingerprint diagram with the pre-established standard tire fingerprint library, and use indicators such as mean square error and correlation coefficient to determine the abnormal areas. The comparison results show that there is a deviation exceeding the set threshold between scanning position 52 and the standard data.

[0136] Estimate the Bubble Size: Based on the empirical model between the aforementioned normal force change and the bubble size, calculate the bubble size of scanning position 52.

[0137] 1) Data Extraction

[0138] According to Table 1 (data after pretreatment), the data for scanning position 52 is as follows:

[0139] Applied Load: 10 N

[0140] Normal area (scanning position 45 - 51) force in the Z direction:

[0141] 45: 2.2 N

[0142] 46: 2.3 N

[0143] 47: 2.4 N

[0144] 48: 2.3 N

[0145] 49: 2.2 N

[0146] 50: 2.3 N

[0147] 51: 2.4 N;

[0148] Take the average value:

[0149] ;

[0150] Defective area (scanning position 52) force in the Z direction:

[0151] F defect = 3.7 N

[0152] 2) Calculate the change in normal force ΔF

[0153] Substitute the values:

[0154] ΔF = 3.7 N - 2.3 N = 1.4 N

[0155] 3) Establish a calibration model and estimate the bubble size

[0156] Based on multiple experimental fittings, establish a linear calibration model:

[0157] D (mm) = a × ΔF (N) + b

[0158] Where:

[0159] a = 10 mm / N

[0160] b = 0 mm

[0161] Therefore, the estimated bubble diameter at scanning position 52 is:

[0162] D = 10 × 1.4 + 0 = 14 mm.

[0163] Step 6: Output the detection results

[0164] The detection system finally outputs the defective area and detailed information,

[0165] 1) Defective position: Scanning position 52 (and the surrounding possibly affected areas);

[0166] 2) Bubble size: The estimated diameter is about 14 mm.

[0167] 3) Defect severity: Since the abnormal force is obvious, it is judged as a serious defect.

[0168] At the same time, multi-dimensional mechanical response maps and thermal maps are generated to visually display the abnormal area in the form of images.

[0169] 4. X-ray detection verification

[0170] To verify the detection results based on tactile perception, an X-ray detection device is selected to scan the same tire. The image provided by the X-ray detection device shows the internal structure of the tire, and can directly detect internal bubbles or delamination areas.

[0171] Steps of the comparative experiment:

[0172] X-ray scanning: Conduct X-ray detection on the tire and collect internal structure images.

[0173] Result comparison: Compare the defect area shown in the X-ray image with the scanned position 52 and its surrounding areas identified by tactile detection.

[0174] If there is an obvious low-density area (representing bubbles or voids) in the scanned position 52 area in the X-ray image, it proves that the detection results of the method of the present invention are correct.

[0175] Experimental data and verification:

[0176] Tactile detection results: At the scanned position 52, the force in the Z direction increases significantly, and the system estimates that the bubble diameter in this area is about 14 mm.

[0177] X-ray detection results: The X-ray image shows that there is an obvious low-density area in the corresponding area, and the bubble size is estimated to be 1.0 - 2.0 cm, which is consistent with the tactile detection results.

[0178] Example 2

[0179] To further improve the intelligence level of detection, in Embodiment 1 of the present invention, defect recognition adopts a machine learning or deep learning model. Specifically, a deep neural network can be used to automatically extract the features of the multi-force-level mechanical fingerprint map and judge the size and distribution position of the internal bubbles of the tire based on these features. This intelligent defect judgment mechanism reduces manual intervention and improves the automation degree and accuracy of detection. The following presents a deep learning method for automatically extracting the defect features in the multi-force-level mechanical fingerprint map and judging the size and distribution position of the internal bubbles of the tire based on these features. This method adopts a multi-task learning strategy, integrating two tasks of defect region segmentation and bubble size regression, collectively referred to as the "Multi-Task Deep Fingerprint Network (MT-DFN)".

[0180] 1) Input data construction

[0181] 1.1) Data source

[0182] Use the tire detection system to collect the preprocessed multi-force-level scan data, and combine the force data in the X, Y, and Z directions of each scan point into image data. To ensure data diversity, the input image can be a grayscale image (the heat map of a single force direction) or a multi-channel image (the RGB channels respectively represent the results of the force in the X, Y, and Z directions after normalization processing).

[0183] 1.2) Data annotation

[0184] Segmentation label: Based on X-ray detection or manual annotation, generate a binary segmentation map of the defect region for each input image. The defect region is represented by "1", and the normal region is represented by "0".

[0185] Regression label: For the defect region, annotate its true bubble size (diameter, unit: mm).

[0186] 2) Network structure design

[0187] 2.1 ) Encoder

[0188] Basic architecture: Adopt a structure similar to U-Net. The input image size is set to 256×256 pixels.

[0189] Residual convolution block: In the encoder, use several residual modules (ResBlock), each module includes two layers of 3×3 convolution, batch normalization, and ReLU activation, plus skip connections.

[0190] Downsampling: Convolution or max - pooling layers with a stride of 2 are used to gradually reduce the size of the feature map (256×256 → 128×128 → 64×64 → 32×32), while increasing the number of channels (from 64 to 128, 256, 512).

[0191] 2.2) Attention Module

[0192] After the middle layer of the encoder or after each downsampling stage, an attention gate is added. Through an adaptive weighting mechanism, local features related to defects are highlighted. This module uses operations such as 1×1 convolution and Sigmoid activation to generate an attention map, and then performs a pixel - by - pixel multiplication with the feature map to obtain a weighted feature map.

[0193] 2.3) Decoder

[0194] Upsampling layer: The feature map is upsampled using transposed convolution or bilinear interpolation, and the size is gradually restored.

[0195] Skip connection: The feature map of the corresponding layer in the encoder is concatenated with the upsampled feature map in the decoder to retain detailed information.

[0196] Convolutional fusion: After upsampling and skip connection, a series of convolutional layers are used to further process the fused features to generate a high - resolution feature map.

[0197] 2.4) Multi - task output branch

[0198] Segmentation branch: At the end of the decoder, it is mapped to 2 channels (defect and normal classes) using 1×1 convolution, and then Softmax is applied to generate the classification probability for each pixel. The output size of the segmentation branch is the same as the input image (256×256), which is used to locate the bubble defect area.

[0199] Regression branch: Global average pooling is performed on the feature map at the end of the decoder to obtain a fixed - length feature vector. After passing through several fully - connected layers (FC - 256→FC - 64), a continuous real - valued output is finally obtained, representing the bubble diameter (or other size metrics) of the detection area. The regression branch can predict individually for each defect area or perform global regression across the entire image, and then map the prediction results to specific areas through connected - component analysis.

[0200] 3.) Loss function and training

[0201] 3.1) Multi - task loss function

[0202] For joint training of the segmentation and regression tasks, the total loss function is adopted:

[0203]

[0204] Where: $L_{seg}$ seg is the loss of the segmentation task, and Dice Loss, cross-entropy loss or combined loss is adopted; $L_{reg}$ reg is the loss of the regression task, and usually mean squared error (MSE) loss is adopted; $\lambda_1$ and $\lambda_2$ are weight coefficients, which are determined through experiments according to the importance of the tasks.

[0205] 3.2) Training details

[0206] Optimizer: The Adam optimizer is adopted, and the initial learning rate is set to 1e-4 and gradually decays;

[0207] Data augmentation: Operations such as rotation, translation, scaling, and noise injection are performed on the training data to expand the dataset;

[0208] Batch size: Select 32 or 16, and choose an appropriate batch size according to the GPU memory;

[0209] Training epochs: Usually train for 50 - 100 epochs until the loss on the validation set tends to be stable.

[0210] 4) Inference and post-processing

[0211] 4.1) Inference stage

[0212] Input the preprocessed multi-force-level mechanical fingerprint map, and the network outputs the segmentation map and regression results simultaneously. Using the segmentation map, connected component analysis is adopted to extract each defect region, and the regression prediction values of each region are averaged or weighted and fused to obtain the final bubble size estimate for this region.

[0213] 4.2) Post-processing

[0214] Morphological operations (such as opening and closing) are performed on the segmentation map to further remove noise points and ensure clear boundaries of the defect regions. According to the regression results and combined with the preset threshold judgment, the final defect report is output, including the coordinates of the defect regions, the bubble diameter, and the defect severity score.

[0215] The proposed Multi-Task Deep Fingerprint Network (MT-DFN) realizes intelligent defect identification based on multi-force-level scanning data by jointly performing defect area segmentation and bubble size regression tasks. This method specifically includes an encoder-decoder structure, an attention module, and a dual-branch output design, and uses a multi-task loss function for joint training. In the inference stage, the network can automatically extract defect features from the preprocessed mechanical fingerprint images and output the defect area, bubble size, and defect severity, providing a specific and innovative solution for tire internal defect detection.

[0216] The following presents an experimental data to demonstrate the performance of the Multi-Task Deep Fingerprint Network (MT-DFN) in detecting internal bubble defects of tires. The experimental data is based on an actual collected tire dataset, which contains 500 multi-force-level mechanical fingerprint images, and each image is attached with manually annotated defect areas (bubbles) and true size information. The dataset is randomly divided into a training set (400 images) and a validation set (100 images). After 80 epochs of training, the network achieved the following performance metrics:

[0217] 1. Performance of the segmentation task

[0218] Dice coefficient: 0.96

[0219] IoU (Intersection over Union): 0.92

[0220] Pixel-level accuracy: 98%;

[0221] This indicates that the network can accurately segment the internal bubble defect areas of tires, and the segmentation results are highly consistent with the defect images obtained by X-ray detection.

[0222] 2. Performance of the regression task

[0223] Mean Absolute Error (MAE): 0.25 mm

[0224] Root Mean Square Error (RMSE): 0.35 mm

[0225] R² coefficient: 0.98;

[0226] These metrics indicate that the network's prediction of the bubble diameter is very accurate, with an average prediction error of only 0.5 mm and good regression performance.

[0227] 3. Prediction results of some test samples

[0228] Table 2 shows the comparison between the true bubble diameters of 10 test samples and the network prediction results. The data are all diameter values in mm and the corresponding absolute errors:

[0229] Table 2 Comparison between the true bubble diameters of test samples and the network prediction results

[0230] Sample number True bubble diameter (mm) Predicted bubble diameter (mm) Absolute error (mm) 1 2.5 2.6 0.1 2 3.0 3.1 0.1 3 2.8 2.7 0.1 4 3.2 3.0 0.2 5 2.9 3.0 0.1 6 3.5 3.4 0.1 7 2.7 2.8 0.1 8 3.3 3.5 0.2 9 2.6 2.5 0.1 10 3.1 3.0 0.1

[0231] As can be seen from the table, the absolute error between the predicted bubble diameter and the true value is within 0.2 mm, showing extremely high prediction accuracy.

[0232] 4. Statistical analysis and comparison verification of experimental data

[0233] In addition, by training and testing a dataset composed of 500 multi-force-level mechanical fingerprint images, the improved network achieved the following overall performance on the validation set:

[0234] Segmentation task: Dice coefficient reaches 0.96, IoU reaches 0.92, and pixel accuracy is 98%;

[0235] Regression task: MAE is 0.25 mm, RMSE is 0.35 mm, and R² reaches 0.98;

[0236] Online detection speed: The inference time for each image on average is less than 100 ms, which meets the requirements of real-time detection.

[0237] Comparing with the X-ray detection results, the coincidence rate between the defect areas detected by the present invention and the low-density areas shown in the X-ray images exceeds 97%, which proves the high accuracy and reliability of the method in actual tire quality control.

[0238] Furthermore, the present invention also provides a computer-readable storage medium, including one or more programs for execution by one or more processors of an electronic device, and the one or more programs include instructions for executing the method described in this specific embodiment.

[0239] It should be noted that computer-readable media include both permanent and non-permanent, removable and non-removable media that can store information by any method or technology. The information can be computer-readable instructions, data structures, program modules, or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, compact disc read-only memory (CD-ROM), digital versatile disc (DVD) or other optical storage, magnetic cassette tapes, magnetic disk storage or other magnetic storage devices, or any other non-transmission media that can store information accessible by a computing device. As defined herein, computer-readable media does not include transitory computer-readable media such as modulated data signals and carrier waves.

[0240] The present invention may be described in the general context of computer-executable instructions, such as program modules, being executed by a computer. Generally, program modules include routines, programs, objects, components, data structures, etc. that perform particular tasks or implement particular abstract data types. The present invention may also be practiced in distributed computing environments where tasks are performed by remote processing devices that are linked through a communications network. In a distributed computing environment, program modules may be located in both local and remote computer storage media including storage devices.

[0241] The above is a description of embodiments of the present invention. Through the above description of the disclosed embodiments, those skilled in the art can implement or use the present invention. Various modifications to these embodiments will be readily apparent to those skilled in the art. The general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for detecting tire internal bubble defects based on tactile perception, characterized in that The method includes the following steps: 1) Tire fixation: Install the tire to be detected on a rotatable or clampable positioning device, and calibrate the spatial coordinate system of the tire. 2) Initial loading: Control the multi-axis motion mechanism to drive the three-axis force sensor to move in the vertical direction and contact the outer surface of the tire. When the normal force detected by the sensor reaches the first preset threshold, establish a detection reference plane. 3) Multi-force level scanning: 3.1) Taking the detection reference plane as a reference, set a scanning path on the outer surface of the tire, apply different magnitudes of normal forces to the same scanning point in sequence, and record the X, Y, and Z direction force data output by the three-axis force sensor under each level of force loading. 3.2) Repeat the above steps to perform row-by-row or partition scanning on the outer surface of the tire, and collect three-dimensional force data at multiple force levels. 4) Data preprocessing: Perform noise filtering, zero drift correction, and interpolation processing on the collected three-dimensional force data, fuse the force changes of the same scanning point under different force loadings, and generate a corresponding position-mechanical feature dataset. 5) Defect identification: 5.1) Based on the position-mechanical feature dataset, construct a multi-force level mechanical fingerprint map. 5.2) Compare the multi-force level mechanical fingerprint map with a pre-established standard mechanical fingerprint library. When it is found that a certain local area exhibits abnormally low stiffness or deformation characteristics, it is determined that there is an internal bubble defect in this area. 5.3) Output the position and defect degree information of the bubble defect area. In the defect identification step 5), the multi-task deep fingerprint network realizes intelligent defect identification based on multi-force level scanning data through the joint defect area segmentation and bubble size regression tasks; specifically, it includes an encoder-decoder structure, an attention module, and a dual-branch output design, and uses a multi-task loss function for joint training; in the inference stage, the network can automatically extract defect features from the preprocessed mechanical fingerprint map and output the defect area, bubble size, and defect severity. The dual-branch output design includes a segmentation branch and a regression branch; the segmentation branch is at the end of the decoder, uses a 1×1 convolution to map to 2 channels, and connects to Softmax to generate the classification probability of each pixel; the output size of the segmentation branch is the same as the input image, and is used to locate the bubble defect area; the regression branch performs global average pooling on the feature map at the end of the decoder to obtain a fixed-length feature vector, and after several fully connected layers, finally outputs a continuous real value representing the bubble diameter of the detection area; in the inference stage, input the preprocessed multi-force level mechanical fingerprint map, and the network outputs the segmentation map and the regression result at the same time; and using the segmentation map, extract each defect area by connected component analysis, and perform average or weighted fusion on the regression prediction values of each area to obtain the final bubble size estimate of this area.

2. The method according to claim 1, wherein: The multi-force level scanning in step 3) includes first performing a preliminary scan with a smaller force value, and then performing a secondary fine scan on the suspicious area with a higher force value to distinguish the deformation differences caused by surface micro-protrusions and deep bubbles.

3. The method according to claim 1, wherein: The data preprocessing in step 4) also includes filtering or compensating for the periodic fluctuations generated by the tire surface pattern to reduce the interference of the tread pattern on the internal defect identification.

4. The method according to claim 1, wherein: The signals output by the triaxial force sensor are synchronously collected by hardware triggering on the FPGA or SoC platform, specifically including counting the pulse signals of the motion mechanism, and triggering sampling when the count reaches a preset value, so as to ensure the accurate correspondence between the spatial coordinates and the force data at different force loading stages.

5. A tire internal bubble defect detection system based on tactile perception, for implementing the method according to any one of claims 1-4, characterized in that, The system includes: A tire positioning device for fixing the tire to be detected and providing rotation or clamping functions; A multi-axis motion mechanism arranged opposite to the tire positioning device, having at least one axis moving in the horizontal direction and one axis moving in the vertical direction; A triaxial force sensor installed at the end of the multi-axis motion mechanism for detecting the force values in the X, Y, and Z directions when contacting the tire surface; A force loading unit cooperating with the triaxial force sensor for applying normal forces of multiple different magnitudes to the tire surface during the scanning process; A data acquisition control unit connected to the triaxial force sensor for collecting triaxial force data based on the hard trigger method and synchronizing with the spatial position of the multi-axis motion mechanism; A data processing and defect identification module communicating with the data acquisition control unit for preprocessing the collected multi-force-level three-dimensional force data and executing a defect identification algorithm to output the judgment result of the internal defects of the tire.

6. The system according to claim 5, wherein: The force loading unit includes a lead screw mechanism or a cylinder mechanism and is configured with a force closed-loop controller, capable of dynamically adjusting the force applied by the triaxial force sensor within a specified force value range.

7. The system according to claim 5, characterized in that: The triaxial force sensor has a replaceable wear-resistant probe or an anti-slip cushion layer to adapt to tire surfaces of different specifications and materials and reduce probe wear.

8. The system according to claim 5, wherein: The data acquisition control unit includes an FPGA or an SoC platform, configured with a pulse counter and a high-speed AD conversion circuit, and triggers sampling of the triaxial force sensor when the count of the step or servo motor pulse signals of the multi-axis motion mechanism reaches a predetermined pulse value, so as to achieve real-time synchronization of the scanning position and the force data.

9. The system according to claim 5, wherein: The data processing and defect identification module establishes a standard tire mechanical fingerprint library, compares it based on the multi-force-level scanning data of different models and specifications of tires in the normal state, and when there is a difference within a preset threshold range between the detected data and the standard data, determines that there are bubble or delamination defects inside the tire and outputs a visual report of the defects.

10. A computer-readable storage medium having a computer program or instructions stored thereon, characterized in that, When the computer program or instruction is executed by a processor, it implements steps 4) - 5) of the method according to any one of claims 1 - 4.

11. A computer program product, comprising a computer program or instructions, characterized in that, When the computer program or instruction is executed by a processor, it implements steps 4) - 5) of the method according to any one of claims 1 - 4.

Images