3D glue coating anomaly detection method and device based on contour synchronous time sequence

Abstract

The application provides a 3D glue coating anomaly detection method and device based on contour synchronous time sequence, and the method comprises the following steps: obtaining a two-dimensional contour data set of a three-dimensional contour of glue coating, and fitting a reference straight line of the glue coating contour according to the contour data set; the contour data set is corrected by rotation and translation through the reference straight line to obtain a corrected point cloud; the glue coating contour is measured based on the corrected point cloud to obtain a measurement sequence; a teaching measurement sequence in a teaching stage is obtained, a dislocation mean square error sequence of the measurement sequence and the corresponding teaching measurement sequence is calculated, a least square problem is constructed according to the dislocation mean square error sequence to obtain a lag order; the measurement sequence is aligned and synchronized according to the lag order to obtain a target measurement sequence; a difference sequence is obtained according to the target measurement sequence and the teaching measurement sequence, the data distribution in the difference sequence is detected, and a glue coating anomaly detection result is obtained. The application fuses the contour time sequence probability distribution, and improves the accuracy and robustness of the detection algorithm.

Description

Technical Field

[0001] This invention relates to the field of adhesive coating anomaly detection technology, and in particular to a 3D adhesive coating anomaly detection method and apparatus based on contour synchronous time series. Background Technology

[0002] Adhesive application technology is widely used by major automobile manufacturers and auto parts suppliers. It is commonly seen in the body-in-white and hood application processes in automotive welding workshops, and in windshield applications in final assembly workshops. Its quality directly affects the vehicle's shock absorption, noise reduction, and windproof performance. In the automotive manufacturing industry, adhesive application quality inspection requirements mainly include adhesive width, adhesive height, and adhesive breaks. Some manufacturers also focus on detecting abnormal adhesive application trajectories and positional deviations (such as abnormalities in the reverse-engineered robotic arm), and have certain requirements for the inspection duration and accuracy.

[0003] Currently, the mainstream detection method is a non-contact glue application inspection solution based on machine vision. This solution has high accuracy, but due to the limitations of the space around the glue gun and the arbitrary spraying direction, there are blind spots in some areas that cannot be detected. Therefore, multiple laser profilometers are usually used to form a multi-view glue application inspection system to completely cover all glue application areas. Existing online glue application inspection based on 3D contours usually requires pre-modeling of standard data through teaching (establishing standards for the width, height, and continuity parameters of normal glue application trajectories using the inspection system). Only by adopting this comparative inspection method can the system achieve high detection accuracy.

[0004] However, multi-array laser profilometry systems suffer from time-variety issues due to factors such as camera exposure, algorithm processing time, and network transmission latency. This leads to increased false negatives as the contour point cloud data during the teaching and inspection phases of 3D adhesive coating inspection are not synchronized. Furthermore, contour-based anomaly detection algorithms have poor robustness, often relying on comparative analysis of contour data features at different times during teaching and inspection to arrive at detection results. This results in inconsistent accuracy and a high risk of false positives.

[0005] Therefore, there is an urgent need for a glue coating anomaly detection method that can synchronize point cloud data in time, avoid missed detections, and improve the robustness of the detection algorithm. Summary of the Invention

[0006] Therefore, it is necessary to provide a 3D coating anomaly detection method and device based on contour synchronization time series to address the above-mentioned technical problems.

[0007] A 3D adhesive coating anomaly detection method based on contour synchronization time series includes the following steps: acquiring a two-dimensional contour dataset of the adhesive coating contour; fitting a reference straight line of the adhesive coating contour based on the contour dataset; performing rotation and translation correction on the contour dataset using the reference straight line to obtain a corrected point cloud; measuring the adhesive coating contour based on the corrected point cloud to obtain a measurement sequence; acquiring a teaching measurement sequence during the teaching phase; calculating the misalignment mean square error sequence between the measurement sequence and the corresponding teaching measurement sequence; constructing a least squares problem based on the misalignment mean square error sequence and solving it to obtain the lag order; aligning and synchronizing the measurement sequence based on the lag order to obtain a target measurement sequence; obtaining the difference sequence between the target measurement sequence and the teaching measurement sequence; detecting the data distribution in the difference sequence to obtain the adhesive coating anomaly detection result.

[0008] In one embodiment, obtaining a two-dimensional contour dataset of the adhesive coating contour and fitting a reference straight line of the adhesive coating contour based on the contour dataset includes: obtaining three-dimensional point cloud data of the adhesive coating contour; projecting the three-dimensional point cloud data onto a plane to obtain a two-dimensional contour dataset of the adhesive coating contour; selecting two points in the two-dimensional contour dataset and calculating the unit direction vector passing through the two points; calculating the standardized distance from all points in the contour dataset to the unit direction vector; dividing points whose standardized distance is less than a preset distance into an interior point set; repeating the division multiple times; and obtaining the reference straight line based on the direction vector fitted by the maximum interior point set.

[0009] In one embodiment, the step of rotating and translating the contour dataset using the reference line to obtain a corrected point cloud includes: setting the direction vector of the reference line as . By solving for the tangent of the slope of the reference line, the angle between the reference line and the coordinate axis is obtained as follows:

[0010]

[0011] Based on the included angle, construct a two-dimensional rotation matrix for clockwise rotation as follows:

[0012]

[0013] Rotate the contour dataset P in space around the coordinate axes by the same angle α to obtain the rotated dataset:

[0014] P′=RP (3)

[0015] The translation coefficient is calculated using the following formula:

[0016]

[0017] The reference line is translated to coincide with the coordinate axes according to the translation coefficient to obtain the corrected point cloud, which is:

[0018] P target =RP+T (5)

[0019] In one embodiment, the step of measuring the adhesive coating contour based on the correction point cloud to obtain a measurement sequence includes: measuring the height and width of the adhesive coating contour according to the correction point cloud, and calculating the area of ​​the adhesive coating contour using numerical integration; and obtaining the corresponding measurement sequence based on the number of contours existing in the adhesive coating trajectory and the measurement data of the height, width and area of ​​the adhesive coating contour.

[0020] In one embodiment, the step of obtaining the teaching measurement sequence during the teaching phase, calculating the misalignment mean square error sequence between the measurement sequence and the corresponding teaching measurement sequence, and constructing a least squares problem based on the misalignment mean square error sequence to obtain the lag order includes: assuming the teaching measurement sequence during the teaching phase is:

[0021]

[0022] The measurement sequence is as follows:

[0023]

[0024] The teaching measurement sequence and the sequence length within the measurement sequence are selected as the target sequence length after alignment and synchronization, as follows:

[0025] l=min{m,n} (8)

[0026] Calculate the k-th order misalignment mean square error e of the teaching measurement sequence and the measurement sequence. k ,for:

[0027]

[0028] The mean square error of all misalignments in the teaching measurement sequence and the measurement sequence is calculated according to equation (9), resulting in a mean square error sequence e for different lag orders. n ,for:

[0029]

[0030] Based on the misalignment mean square error sequence, a least squares problem is constructed to solve for the lag order t of the teaching measurement sequence and the measurement sequence, as follows:

[0031]

[0032] In one embodiment, the step of obtaining the difference sequence between the target measurement sequence and the teaching measurement sequence, detecting the data distribution in the difference sequence, and obtaining the glue application anomaly detection result includes: calculating the difference sequence between the teaching measurement sequence and the measurement sequence, using the following formula:

[0033]

[0034] Based on the law of large numbers and the central limit theorem, assuming that the measurement sequence, target measurement sequence, and difference sequence all satisfy a Gaussian distribution, and considering the distribution characteristics of the Gaussian distribution, the probability value of the data in the difference sequence within the interval (u-3δ, u+3δ) is calculated using the following formula:

[0035]

[0036] Transform Summarized as follows:

[0037]

[0038] Solving for:

[0039]

[0040] The detection result of adhesive coating anomaly is obtained based on the magnitude of the probability value.

[0041] A 3D adhesive coating anomaly detection device based on contour synchronization time series is used to implement the 3D adhesive coating anomaly detection method based on contour synchronization time series as described above. The device includes: a reference line fitting module for acquiring a two-dimensional contour dataset of the adhesive coating contour and fitting a reference line of the adhesive coating contour based on the contour dataset; a contour dataset correction module for performing rotation and translation correction on the contour dataset using the reference line to obtain a correction point cloud; a measurement sequence acquisition module for measuring the adhesive coating contour based on the correction point cloud to obtain a measurement sequence; a hysteresis order acquisition module for acquiring the teaching measurement sequence during the teaching phase, calculating the misalignment mean square error sequence between the measurement sequence and the corresponding teaching measurement sequence, constructing a least squares problem based on the misalignment mean square error sequence, and solving to obtain the hysteresis order; a sequence alignment and synchronization module for aligning and synchronizing the measurement sequence according to the hysteresis order to obtain a target measurement sequence; and a detection result acquisition module for obtaining the difference sequence between the target measurement sequence and the teaching measurement sequence, detecting the data distribution in the difference sequence, and obtaining the adhesive coating anomaly detection result.

[0042] Compared with existing technologies, the advantages and beneficial effects of this invention are as follows: By acquiring a two-dimensional contour dataset of the adhesive coating contour, a reference straight line of the adhesive coating contour is obtained by fitting the contour dataset. The contour dataset is then corrected by rotation and translation using the reference straight line to obtain a corrected point cloud, thereby correcting the point cloud data and improving the accuracy of the detection results. The adhesive coating contour is measured based on the corrected point cloud to obtain a measurement sequence. A teaching measurement sequence is obtained during the teaching phase. The mean square error sequence of the misalignment between the measurement sequence and the corresponding teaching test sequence is calculated. A least squares problem is constructed based on the mean square error sequence of the misalignment to obtain the lag order. The measurement sequence is aligned and synchronized using the lag order, thereby achieving alignment and synchronization of the point cloud data in the time dimension and avoiding missed detections. A target measurement sequence is obtained. The difference sequence between the target measurement sequence and the teaching measurement sequence is obtained. The data distribution in the difference sequence is detected to obtain the adhesive coating anomaly detection result. The probability distribution of the contour time series is fused, achieving synchronization of the point cloud data in the time dimension and improving the accuracy and robustness of the detection algorithm. Attached Figure Description

[0043] Figure 1 This is a flowchart illustrating a 3D adhesive coating anomaly detection method based on contour synchronization time series in one embodiment.

[0044] Figure 2 This is a diagram illustrating the effect of asynchronous detection and teaching phases in one embodiment.

[0045] Figure 3 This is a diagram illustrating the effect of aligning and synchronizing the detection and teaching phases in one embodiment.

[0046] Figure 4 This is a diagram illustrating the performance of a traditional contour data anomaly detection algorithm in one embodiment.

[0047] Figure 5 This is a diagram illustrating the performance of the detection algorithm of the present invention in one embodiment;

[0048] Figure 6 This is a schematic diagram of a 3D adhesive coating anomaly detection device based on contour synchronization time series in one embodiment. Detailed Implementation

[0049] Before describing the specific embodiments of the present invention, the overall concept of the present invention will be explained as follows:

[0050] This invention is mainly based on the research and development of the 3D coating process. Currently, the coating process suffers from problems such as asynchronous point cloud data, leading to missed detections and low algorithm robustness.

[0051] Therefore, this invention proposes a 3D adhesive coating anomaly detection method based on contour synchronization time series. The method acquires a two-dimensional contour dataset of the adhesive coating contour, fits a reference straight line to the dataset, and corrects the contour dataset by rotation and translation using the reference straight line to obtain a corrected point cloud. This corrects the point cloud data and improves the accuracy of the detection results. The adhesive coating contour is then measured based on the corrected point cloud to obtain a measurement sequence. A teaching measurement sequence is also obtained during the teaching phase. The mean square error sequence of the misalignment between the measurement sequence and the corresponding teaching test sequence is calculated. A least squares problem is constructed based on the misalignment mean square error sequence to obtain the lag order. The measurement sequence is then aligned and synchronized using the lag order, achieving alignment and synchronization of the point cloud data in the time dimension, avoiding missed detections. A target measurement sequence is obtained. The difference sequence between the target measurement sequence and the teaching measurement sequence is obtained, and the data distribution in the difference sequence is detected to obtain the adhesive coating anomaly detection result. This method achieves synchronization of the point cloud data in the time dimension, ensuring the accuracy and completeness of the point cloud data detection results and improving the robustness of the detection algorithm.

[0052] Having introduced the overall concept of the present invention, to make the objectives, technical solutions, and advantages of the present invention clearer, the present invention will be further described in detail below through specific embodiments in conjunction with the accompanying drawings. It should be understood that the specific embodiments described herein are merely illustrative of the present invention and are not intended to limit the present invention.

[0053] In one embodiment, such as Figure 1 As shown, a 3D adhesive coating anomaly detection method based on contour synchronization time series is provided, including the following steps:

[0054] Step S101: Obtain a two-dimensional contour dataset of the adhesive coating profile, and fit the baseline line of the adhesive coating profile based on the contour dataset.

[0055] Specifically, since width and height data are mainly used in the detection of 3D adhesive coating contours, the 3D point cloud data of the adhesive coating contour can be projected onto the xoz plane to obtain a contour dataset, which is convenient for subsequent processing. Two points are randomly selected in the contour dataset, and the unit direction vector passing through the two points is calculated. Thus, the standardized distance of the remaining points in the contour dataset can be calculated based on the unit direction vector.

[0056] Step S101 includes: acquiring three-dimensional point cloud data of the adhesive coating contour, projecting the three-dimensional point cloud data onto a plane to obtain a two-dimensional contour dataset of the adhesive coating contour; selecting two points in the two-dimensional contour dataset and calculating the unit direction vector passing through the two points; calculating the standardized distance from all points in the contour dataset to the unit direction vector, dividing points whose standardized distance is less than a preset distance into an interior point set, repeating the division multiple times, and obtaining the maximum interior point set after the division is completed; obtaining the reference line based on the direction vector fitted by the maximum interior point set.

[0057] Specifically, in the contour dataset P of the glue application contour, two distinct points p1 and p2 are randomly selected, with corresponding coordinates (x1, z1) and (x2, z2) respectively. Then, the direction vector of the line passing through p1 and p2 is... for:

[0058]

[0059] Calculate the direction vector unit direction vector for:

[0060]

[0061] The standardized distance from all points in the contour dataset to the unit direction vector is calculated using the following formula:

[0062]

[0063] The statistical contour dataset is analyzed by counting the number of points whose standardized distances to the unit direction vector satisfy a preset condition. Specifically, if the standardized distance is less than the preset distance, the corresponding points are classified as inliers; if the standardized distance is greater than or equal to the preset distance, the corresponding points are classified as outliers. The size of the current inlier set is denoted as n. i :

[0064] n i =|P inner |=|{l i <τ|i∈N +}|

[0065] Repeatedly count the numbers to obtain n1, n2, n3, ..., n k The largest interior set n is obtained by selecting it. max ,for:

[0066]

[0067] Calculate the maximum interior set n max Corresponding direction vector The direction vector of the optimal fitted line is obtained as follows:

[0068]

[0069] The reference line is obtained based on the direction vector of the best-fit line.

[0070] Specifically, although the uneven surface in the adhesive application scenario results in a large number of curved features in the scanned contour data, all adhesive application contour curves contain a reference straight line. Even if the position of the adhesive-applied part is uncertain and the surface is uneven, causing this straight line to present different postures in three-dimensional space, it can basically reflect the spatial directional characteristics of all curves in the contour. Therefore, the reference straight line can be obtained from the various curves of the contour, so as to correct the adhesive application contour based on the reference straight line.

[0071] After obtaining the unit direction vector, the standardized distance from other points in the contour dataset to the unit direction vector is calculated sequentially based on the unit direction vector. The points in the contour dataset can be classified based on the standardized distance, dividing them into an interior point set and an exterior point set. After dividing all points, the maximum interior point set is obtained. The target direction vector is obtained by fitting the maximum interior point set, and the line corresponding to the direction vector is used as the reference line. The reference line can facilitate the correction of the adhesive coating contour, thereby facilitating the detection of the height and width of the adhesive coating.

[0072] When obtaining the baseline line, the standardized distance from all points in the contour dataset to the unit direction vector is calculated. The number of points whose standardized distance meets a preset requirement is that the standardized distance is less than a preset distance, which can be set according to actual conditions. When the standardized distance of a point is less than the preset distance, the point is assigned to the inlier set; when the standardized distance is greater than or equal to the preset distance, the point is assigned to the outer set, and the size of the current inlier set is recorded. This statistical process is repeated to obtain several inlier sets. The largest inlier set is selected, and a direction vector is fitted based on the largest inlier set. This is the direction vector of the optimal fitted line. The baseline line is obtained based on this direction vector, which facilitates subsequent correction of the contour dataset based on the baseline line, ensuring the accuracy of the contour data.

[0073] Step S102: Rotate and translate the contour dataset using the reference straight line to obtain the corrected point cloud.

[0074] Specifically, after obtaining the reference line from the contour dataset, the contour dataset is corrected using the reference line. The correction includes rotation and translation. The main purpose is to transform the contour dataset so that the reference line is parallel and coincident with the coordinate axis, ensuring that the corrected projection data conforms to the glued contour and improving the accuracy of anomaly detection results.

[0075] Step S102 includes: setting the direction vector of the reference line as . By inversely solving for the tangent of the slope of the reference line, the angle between the reference line and the coordinate axes is obtained as follows:

[0076]

[0077] Construct a two-dimensional rotation matrix for clockwise rotation based on the included angle, as follows:

[0078]

[0079] Rotate the contour dataset in space around the coordinate axes by the same angle to obtain the rotated dataset:

[0080] P′=RP (3)

[0081] The translation coefficient is calculated using the following formula:

[0082]

[0083] The reference line is translated to coincide with the coordinate axes based on the translation coefficient, resulting in the corrected point cloud:

[0084] P target =RP+T (5)

[0085] Specifically, when correcting the contour dataset, the angle between the reference line and the coordinate axis is obtained by inversely solving for the tangent of the slope of the reference line, for example, the angle with the x-axis is α. A two-dimensional rotation matrix is ​​constructed based on this angle. Note that this rotation is clockwise. The contour dataset P is rotated by α around the x-axis in space to obtain P′, completing the translation operation. At this point, the reference line is rotated to be parallel to the x-axis. By calculating the translation coefficient, the reference line is moved to coincide with the x-axis, resulting in the corrected point cloud. The accuracy of subsequent anomaly detection results can be improved through the corrected point cloud.

[0086] The unevenness and tilt of the workpiece surface can lead to significant errors when detecting features of the adhesive strip, such as width and height. After performing the aforementioned rotation and translation contour correction, the tilted contour of the adhesive strip is corrected to a relatively horizontal state. The gray point used to detect the highest point of height and the two reference points for calculating the width are both correctly searched, greatly improving the accuracy of detecting the width and height of the adhesive strip.

[0087] Step S103: Measure the adhesive coating profile based on the corrected point cloud to obtain a measurement sequence.

[0088] Specifically, after obtaining the transformed correction point cloud, the height and width of the adhesive coating contour are measured based on the correction point cloud, and the area of ​​the contour is calculated using numerical integration. Based on the measurement data of the height, width, and area of ​​the adhesive coating contour, the corresponding measurement sequence is obtained, thereby enabling multi-dimensional comprehensive detection of adhesive coating trajectories of multiple contours, improving the accuracy of the detection results, and expanding the applicability of the detection method.

[0089] Step S103 includes: measuring the height and width of the adhesive coating contour based on the calibration point cloud, and calculating the area of ​​the adhesive coating contour using numerical integration; obtaining the corresponding measurement sequence based on the number of contours existing in the adhesive coating trajectory and the height, width and area of ​​the adhesive coating contour.

[0090] Specifically, after obtaining the corrected point cloud through rotation and translation transformation, the height and width of the adhesive coating contour are measured based on the corrected point cloud, and the area of ​​the contour is calculated based on the height and width of the contour. Combining the number of contours of the adhesive coating trajectory with the measurement data of the height, width and area of ​​the adhesive coating contour, the measurement sequences corresponding to the height, width and area are obtained respectively. Through the measurement sequences, the detection of height anomalies, width anomalies and area anomalies of adhesive coating trajectories with multi-curve features can be realized, ensuring the accuracy of the detection results and the applicability of the detection method.

[0091] For example, assuming the complete 3D adhesive application trajectory includes M contours, then three measurement sequences of size M are obtained respectively:

[0092]

[0093] Therefore, when the 3D adhesive application trajectory includes M contours, there are M corresponding heights, widths, and areas, forming three measurement sequences of size M, which facilitates the detection of anomalies in the adhesive application height, width, and area.

[0094] Step S104: Obtain the teaching measurement sequence of the teaching phase, calculate the misalignment mean square error sequence between the measurement sequence and the corresponding teaching measurement sequence, construct a least squares problem based on the misalignment mean square error sequence, and solve for the lag order.

[0095] Specifically, since the data from the teaching and testing phases are not synchronized in time, the misalignment mean square error sequence is calculated by acquiring the teaching measurement sequence and combining it with the measurement sequence. A least squares problem is then constructed based on the misalignment mean square error sequence to calculate the lag order between the two sequences. This allows for time alignment and synchronization of the measurement sequences based on the lag order, avoiding missed detections and improving the reliability of the testing results.

[0096] Step S104 includes: Assuming the teaching measurement sequence for the teaching phase is:

[0097]

[0098] The measurement sequence is as follows:

[0099]

[0100] The teaching measurement sequence and the sequence length within the measurement sequence are selected as the target sequence length after alignment and synchronization, as follows:

[0101] l=min{m,n} (8)

[0102] Calculate the k-th order misalignment mean square error e of the teaching measurement sequence and the measurement sequence. k ,for:

[0103]

[0104] The mean square error of all misalignments in the teaching measurement sequence and the measurement sequence is calculated according to equation (9), resulting in a mean square error sequence e for different lag orders. n ,for:

[0105]

[0106] Construct a least-squares problem based on the misalignment mean square error sequence, and solve for the teaching measurement sequence and the lag order t of the measurement sequence, as follows:

[0107]

[0108] Specifically, taking the height data of the adhesive application as an example, the alignment and synchronization issues in the teaching and testing phases manifest in the data as two sequences h. m ,h′ n There is a certain lag, such as Figure 2 As shown, and since the lengths of the two sequences are different, the problem is transformed into finding the lag order t of the two time series.

[0109] Based on this lag order t, the alignment and synchronization of the two height sequences in the teaching and testing phases can be completed, and the alignment effect is as follows: Figure 3 As shown, the two sequences have achieved a high degree of alignment and synchronization. Similarly, the alignment and synchronization of width measurement data and area measurement data can be achieved.

[0110] Step S105: Align and synchronize the measurement sequence according to the hysteresis order to obtain the target measurement sequence.

[0111] Specifically, after obtaining the hysteresis order of the measurement sequence based on the misalignment mean square error, the measurement sequence is aligned and synchronized using this hysteresis order to obtain the target measurement sequence. This ensures that the measurement sequence is time-synchronized and aligned with the teaching measurement sequence, avoids missed detection of contour data, and ensures the accuracy and reliability of the detection results.

[0112] Step S106: Obtain the difference sequence between the target measurement sequence and the teaching measurement sequence, detect the data distribution in the difference sequence, and obtain the glue coating anomaly detection result.

[0113] Specifically, since the data from the detection and teaching phases are essentially identical after contour alignment and synchronization, significant differences only occur in the data from the glue breakage area. Based on this, by correcting and aligning the target measurement sequence and the teaching measurement sequence, a difference sequence is calculated. Based on the probability distribution of the data in the difference sequence, the glue application anomaly detection result can be obtained. By fusing the contour time series probability distribution, time synchronization of contour data can be achieved, effectively avoiding missed detections and improving the accuracy and robustness of the submitted anomaly detection results.

[0114] The steps for obtaining the glue coating anomaly detection results are as follows: Calculate the difference sequence between the teaching measurement sequence and the measurement sequence, using the following formula:

[0115]

[0116] Based on the law of large numbers and the central limit theorem, assuming that the measurement sequence, target measurement sequence, and interpolation sequence all satisfy a Gaussian distribution, and considering the distribution characteristics of the Gaussian distribution, the probability value of data in the difference sequence within the interval (u-3δ, u+3δ) is calculated using the following formula:

[0117]

[0118] Transform Summarized as follows:

[0119]

[0120] Solving for:

[0121]

[0122] The results of adhesive coating anomaly detection are obtained based on the probability values.

[0123] Specifically, taking the corrected and aligned height data as an example, the height difference sequence is calculated based on the two height sequences before and after correction. Considering the distribution pattern of the data in the difference sequence, normal data is distributed around the mean, while a small number of outlier abnormal data are distributed outside a certain range of the mean. Based on this consideration, it is only necessary to solve for this range threshold to detect the abnormal data that occurs when the glue breaks. Since the infinite integral Φ(k) is convergent in the real number domain, its value can be solved using the Gaussian quadrature formula or the Romberg quadrature formula, or the integral value can be obtained by looking up a table. The corresponding probability values ​​when k = 1, 2, 3 are shown in Table 1.

[0124] Table 1

[0125] 1 0.8413 68.26％ 2 0.9772 95.44％ 3 0.99865 99.74％

[0126] Clearly, when k=3, the probability that the data distribution of the difference sequence falls within (u-3δ, u+3δ) is approximately 99.74%. Therefore, we set the threshold to k=3. According to the principle of low probability, the normal data of the difference sequence after glue application is unlikely to fall outside this range. The abnormal data are outside the range of (u-3δ, u+3δ). Considering the influence of noise, it can also be determined that the adhesive application is highly abnormal when multiple abnormal data are detected consecutively.

[0127] The above method enables the detection of adhesive coating anomalies based on width and area. By combining the anomaly detection of three sets of data sequences, a comprehensive judgment is made on whether anomalies have occurred in the adhesive coating detection. Through multi-dimensional detection methods, the robustness of the detection results and adaptability to multiple scenarios are ensured.

[0128] In this embodiment, a two-dimensional contour dataset of the adhesive coating contour is acquired. A reference straight line for the adhesive coating contour is fitted based on the contour dataset. The contour dataset is then corrected by rotation and translation using the reference straight line to obtain a corrected point cloud. This corrects the point cloud data and improves the accuracy of the detection results. The adhesive coating contour is measured based on the corrected point cloud to obtain a measurement sequence. A teaching measurement sequence for the teaching phase is also obtained. The mean square error sequence of the misalignment between the measurement sequence and the corresponding teaching test sequence is calculated. A least squares problem is constructed based on the mean square error sequence of the misalignment to obtain the lag order. The measurement sequence is then aligned and synchronized using the lag order to achieve alignment and synchronization of the point cloud data in the time dimension, avoiding missed detections. A target measurement sequence is obtained. The difference sequence between the target measurement sequence and the teaching measurement sequence is obtained. The data distribution in the difference sequence is detected to obtain the adhesive coating anomaly detection result. By fusing the probability distribution of the contour time series, the synchronization of the point cloud data in the time dimension is achieved, improving the accuracy and robustness of the detection algorithm.

[0129] In one embodiment, taking a high-resolution anomaly detection algorithm as an example, the superiority of the detection algorithm of the present invention compared with conventional detection algorithms is illustrated. Conventional anomaly detection algorithms based on contour data, such as... Figure 4 As shown, directly comparing the feature data of the adhesive strip in the detection stage and the teaching stage, such as the height data in the figure, can clearly distinguish the abnormality of adhesive breakage from the image. However, due to the unevenness and tilt of the workpiece surface, it is difficult to directly distinguish the part of adhesive breakage and the length of adhesive breakage from the data dimension. Furthermore, it is easily interfered with by abnormal noise data, resulting in false detection.

[0130] Unlike traditional glue application anomaly detection algorithms, the detection algorithm of this invention is as follows: Figure 5As shown, after calculating the upper and lower boundaries of the height sequence, the abnormal data of the adhesive strip height and the number of abnormal contours can be obtained directly, and the influence of noise is avoided. The algorithm has strong robustness and anti-interference ability, and the detection rate can reach more than 99%.

[0131] like Figure 6 As shown, a 3D adhesive coating anomaly detection device 60 based on contour synchronization time series is provided to implement the 3D adhesive coating anomaly detection method based on contour synchronization time series as described above. It includes: a baseline line fitting module 61, a contour dataset correction module 62, a measurement sequence acquisition module 63, a hysteresis order acquisition module 64, a sequence alignment and synchronization module 65, and a detection result acquisition module 66, wherein:

[0132] The reference line fitting module 61 is used to obtain a two-dimensional contour dataset of the adhesive coating contour and to fit the reference line of the adhesive coating contour based on the contour dataset.

[0133] The contour dataset correction module 62 is used to perform rotation and translation correction on the contour dataset using a reference straight line to obtain a corrected point cloud.

[0134] The measurement sequence acquisition module 63 is used to measure the adhesive coating contour based on the calibration point cloud and acquire a measurement sequence;

[0135] The lag order acquisition module 64 is used to acquire the teaching measurement sequence during the teaching phase, calculate the misalignment mean square error sequence between the measurement sequence and the corresponding teaching measurement sequence, construct a least squares problem based on the misalignment mean square error sequence, and solve it to obtain the lag order.

[0136] The sequence alignment and synchronization module 65 is used to align and synchronize the measurement sequence according to the hysteresis order to obtain the target measurement sequence.

[0137] The detection result acquisition module 66 is used to obtain the difference sequence between the target measurement sequence and the teaching measurement sequence, detect the data distribution in the difference sequence, and obtain the glue coating anomaly detection result.

[0138] In one embodiment, the device further includes: a point cloud data projection module, used to acquire three-dimensional point cloud data of the adhesive coating contour, project the three-dimensional point cloud data onto a plane to obtain a two-dimensional contour dataset of the adhesive coating contour; a unit direction vector calculation module, used to select two points in the two-dimensional contour dataset and calculate the unit direction vector passing through the two points; an interior point set partitioning module, used to calculate the standardized distance from all points in the contour dataset to the unit direction vector, partition points whose standardized distance is less than a preset distance into an interior point set, and obtain the maximum interior point set after partitioning; and a reference line acquisition module, used to obtain a reference line based on the direction vector fitted by the maximum interior point set.

[0139] In one embodiment, the device further includes: a measurement calculation module, configured to measure the height and width of the adhesive coating contour based on the correction point cloud, and calculate the area of ​​the adhesive coating contour using numerical integration; and a measurement sequence acquisition module, configured to acquire a corresponding measurement sequence based on the number of contours existing in the adhesive coating trajectory and the height, width, and area of ​​the adhesive coating contour.

[0140] It is obvious to those skilled in the art that the modules or steps of the present invention described above can be implemented using general-purpose computing devices. They can be centralized on a single computing device or distributed across a network of multiple computing devices. Optionally, they can be implemented using computer-executable program code, thereby storing them in a computer storage medium (ROM / RAM, magnetic disk, optical disk) for execution by the computing device. In some cases, the steps shown or described can be performed in a different order than those described herein, or they can be fabricated as separate integrated circuit modules, or multiple modules or steps can be fabricated as a single integrated circuit module. Therefore, the present invention is not limited to any particular hardware and software combination.

[0141] The above description, in conjunction with specific embodiments, provides a further detailed explanation of the present invention. It should not be construed that the specific implementation of the present invention is limited to these descriptions. For those skilled in the art, various simple deductions or substitutions can be made without departing from the concept of the present invention, and all such deductions or substitutions should be considered within the scope of protection of the present invention.

Claims

1. A method for detecting 3D adhesive coating anomalies based on contour-synchronized time series, characterized in that, Includes the following steps: Obtain a two-dimensional contour dataset of the adhesive application profile, and fit a reference straight line of the adhesive application profile based on the contour dataset. The contour dataset is rotated and translated using the reference straight line to obtain the corrected point cloud. The measurement sequence is obtained by measuring the adhesive coating contour based on the correction point cloud, including: measuring the height and width of the adhesive coating contour according to the correction point cloud, and calculating the area of ​​the adhesive coating contour using numerical integration; and obtaining the corresponding measurement sequence based on the number of contours existing in the adhesive coating trajectory and the measurement data of the height, width and area of ​​the adhesive coating contour. Obtain the teaching measurement sequence during the teaching phase, calculate the misalignment mean square error sequence between the measurement sequence and the corresponding teaching measurement sequence, construct a least squares problem based on the misalignment mean square error sequence, and solve it to obtain the lag order. The measurement sequence is aligned and synchronized according to the hysteresis order to obtain the target measurement sequence; Based on the target measurement sequence and the teaching measurement sequence, a difference sequence between the two is obtained. The data distribution in the difference sequence is detected to obtain the glue coating anomaly detection result, including: The difference sequence between the taught measurement sequence and the target measurement sequence is calculated using the following formula: ， Based on the law of large numbers and the central limit theorem, assuming that the measurement sequence, target measurement sequence, and difference sequence all satisfy a Gaussian distribution, then, combining the distribution characteristics of the Gaussian distribution, the data in the difference sequence within the interval [missing information]. The probability value within is given by the formula: ， Transform After sorting, we get: ， Solving for: ， The detection result of adhesive coating anomaly is obtained based on the magnitude of the probability value.

2. The 3D coating anomaly detection method based on contour synchronization time series according to claim 1, characterized in that, The process of obtaining a two-dimensional contour dataset of the adhesive application contour, and fitting a reference straight line of the adhesive application contour based on the contour dataset, includes: Obtain the three-dimensional point cloud data of the adhesive coating contour, and project the three-dimensional point cloud data onto a plane to obtain a two-dimensional contour dataset of the adhesive coating contour. Select two points from the two-dimensional contour dataset and calculate the unit direction vector passing through the two points; Calculate the standardized distance from all points in the contour dataset to the unit direction vector, divide the points whose standardized distance is less than a preset distance into an interior point set, repeat the division multiple times, and obtain the maximum interior point set after the division is completed; The reference line is obtained by fitting the direction vector to the maximum set of interior points.

3. The 3D coating anomaly detection method based on contour synchronization time series according to claim 1, characterized in that, The step of rotating and translating the contour dataset using the reference straight line to obtain the corrected point cloud includes: Let the direction vector of the reference line be . By solving for the tangent of the slope of the reference line, the angle between the reference line and the coordinate axis is obtained as follows: ， Based on the included angle, construct a two-dimensional rotation matrix for clockwise rotation as follows: ， The contour dataset Rotation around the coordinate axes in space and the included angle With the same degree, the resulting rotation dataset is: ， The translation coefficient is calculated using the following formula: ， The reference line is translated to coincide with the coordinate axes according to the translation coefficient to obtain the corrected point cloud, which is: 。 4. The 3D coating anomaly detection method based on contour synchronization time series according to claim 1, characterized in that, The process of acquiring the teaching measurement sequence during the teaching phase, calculating the misalignment mean square error sequence between the measurement sequence and the corresponding teaching measurement sequence, constructing a least squares problem based on the misalignment mean square error sequence, and solving for the lag order includes: Let the teaching measurement sequence in the teaching phase be: ， The measurement sequence is as follows: ， The teaching measurement sequence and the sequence length within the measurement sequence are selected as the target sequence length after alignment and synchronization, as follows: ， Calculate the k-th order misalignment mean square error of the teaching measurement sequence and the measurement sequence. ,for: ， The mean square error of all misalignments in the teaching measurement sequence and the measurement sequence is calculated according to equation (9), resulting in a mean square error sequence of misalignments with different lag orders. ,for: ， Construct a least-squares problem based on the misalignment mean square error sequence, and solve for the lag order of the teaching measurement sequence and the measurement sequence. ,for: 。 5. A 3D adhesive coating anomaly detection device based on contour-synchronized time series, characterized in that, A method for implementing a 3D adhesive coating anomaly detection based on contour synchronization time series as described in any one of claims 1-4, comprising: The baseline line fitting module is used to obtain a two-dimensional contour dataset of the adhesive coating contour, and to fit a baseline line of the adhesive coating contour based on the contour dataset. The contour dataset correction module is used to perform rotation and translation correction on the contour dataset using the reference straight line to obtain a corrected point cloud. The measurement sequence acquisition module is used to measure the adhesive coating contour based on the calibration point cloud and acquire a measurement sequence, including: measuring the height and width of the adhesive coating contour according to the calibration point cloud and calculating the area of ​​the adhesive coating contour using numerical integration; and acquiring the corresponding measurement sequence based on the number of contours existing in the adhesive coating trajectory and the measurement data of the height, width and area of ​​the adhesive coating contour. The lag order acquisition module is used to acquire the teaching measurement sequence during the teaching phase, calculate the misalignment mean square error sequence between the measurement sequence and the corresponding teaching measurement sequence, construct a least squares problem based on the misalignment mean square error sequence, and solve it to obtain the lag order. A sequence alignment and synchronization module is used to align and synchronize the measurement sequence according to the hysteresis order to obtain a target measurement sequence; The detection result acquisition module is used to obtain the difference sequence between the target measurement sequence and the teaching measurement sequence, detect the data distribution in the difference sequence, and obtain the glue application anomaly detection result, including: calculating the difference sequence between the teaching measurement sequence and the target measurement sequence, using the following formula: ， Based on the law of large numbers and the central limit theorem, assuming that the measurement sequence, target measurement sequence, and difference sequence all satisfy a Gaussian distribution, then, combining the distribution characteristics of the Gaussian distribution, the data in the difference sequence within the interval [missing information]. The probability value within is given by the formula: ， Transform After sorting, we get: ， Solving for: ， The detection result of adhesive coating anomaly is obtained based on the magnitude of the probability value.

Images