A system and method for synchronously completing part mistake proofing detection and break clearance measurement

Abstract

The application provides a system and method for synchronously completing part mistake-proofing detection and gap difference measurement, wherein the method comprises the following steps: acquiring current vehicle detection information; acquiring vehicle outer surface image information; operating a cooperative mechanical arm to a preset detection point; acquiring an RGB color image and a point cloud image of the outer surface of the vehicle at the detection point by using a 3D camera; combining the color image, the point cloud image and vehicle digital model information to calculate the working position of a gap difference detector; operating the cooperative mechanical arm to the working position according to a preset path and motion parameters; controlling a PLC controller to control a slide rail to start moving at a speed synchronized with a conveying belt; after the speed of the slide rail is synchronized with the conveying belt, the gap difference detector starts to measure the gap and the surface difference of the vehicle; after the measurement is completed, the cooperative mechanical arm operates to a next detection point or returns to an initial position after the current vehicle detection is completed, and waits for the next vehicle to enter.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent manufacturing and machine vision, in particular, to a system and method for simultaneously completing part mistake-proof detection and gap measurement. BACKGROUND

[0002] In the process of automobile production, there are many types of parts with high similarity, low efficiency of manual visual inspection, and easy missed detection and error detection. Gap is an important indicator affecting the appearance and sealing performance. The existing laser measurement equipment needs to measure the vehicle when it is stationary, which is low in efficiency and cannot be integrated with appearance detection.

[0003] In the prior art, automobile part mistake-proof detection and gap measurement mainly have the following main types:

[0004] Manual detection and measurement method: At present, the appearance characteristics of parts rely on memory to identify the differences between different parts. The online quality detection site is deployed by manual detection. The measurement is mainly performed by manual measurement using a gap ruler, a surface difference table, and other tools by measurement personnel. This method is low in efficiency, the measurement result is greatly affected by human factors, and it is difficult to meet the needs of large-scale production and quality big data.

[0005] Multi-view visual detection system based on vision: industrial cameras are arranged online to collect images, and algorithms are integrated to output detection results in real time. A plurality of 2D cameras at different points reconstruct the workpiece to be measured, but the precision is low and the measurement deviation is large, sometimes it is difficult to meet the requirements. For example, the accuracy of the measurement result of some multi-view vision-based systems is greatly affected under complex curved surfaces or different lighting conditions.

[0006] Traditional automobile production line layout separates the scenes of visual detection and gap measurement, installs hardware devices at different stations to realize their respective functions, occupies a large space, has high investment cost, and cannot adapt to mixed-line multi-model production. At the same time, the traditional visual algorithm has poor generalization and cannot quickly adapt to new models.

[0007] Therefore, a multifunctional integrated system that simultaneously completes appearance detection and gap measurement under the dynamic running state of the vehicle is needed to improve detection efficiency, reduce cost, and improve versatility. SUMMARY

[0008] In view of the defects in the prior art, the purpose of the present application is to provide a system and method for simultaneously completing part mistake-proof detection and gap measurement. The 3D camera used for visual detection and the gap surface difference detector are ingeniously installed and integrated on a slide rail and a set of collaborative robot arm, so that the two sets of equipment can work simultaneously during the movement of the production line. The vehicle appearance part photo can be taken, and real-time online measurement can be completed, saving equipment investment, optimizing the production line layout, and ensuring the flexibility of future new model adaptation function expansion.

[0009] A system for simultaneously completing part error-proof detection and gap measurement is provided, which simultaneously completes part error-proof detection and gap measurement in a moving state of a vehicle, and the system comprises:

[0010] 3D camera: for taking photos of vehicle appearance features and obtaining point position information of the vehicle surface;

[0011] Gap measurement module, comprising a gap and surface difference detector, for accurately measuring the gap and surface difference of the vehicle;

[0012] Part error-proof detection module: based on a pre-constructed initial sample set, a feature extraction network and a detection judgment network are constructed, product feature information of a search hit is sent to the feature extraction network according to image data obtained by the 3D camera, a multi-dimensional feature map is obtained, the multi-dimensional feature map is converted into a sequence vector through a visual embedding layer and sent to the detection judgment network for judgment, and finally a detection result is obtained;

[0013] Collaborative robot: according to the point position information obtained by the 3D camera, it moves to the corresponding gap detection position for detection; the 3D camera and the gap and surface difference detector are integrated and installed at the end of the collaborative robot;

[0014] Photoelectric sensor: providing information about the detection range of the vehicle entering and leaving the system;

[0015] Encoder: obtaining the speed of the conveyor belt, for calculating the distance of the vehicle running on the conveyor belt during the detection process of the system;

[0016] Slide rail: after the collaborative robot sends the gap and surface difference detector to the designated position, the slide rail is synchronized with the speed of the conveyor belt to ensure that the gap and surface difference detector remains relatively stationary relative to the gap measurement position, and accurate measurement values are obtained;

[0017] Training server: for deploying AI visual detection algorithms, providing corresponding point detection computing power for the system, analyzing image features and calculating point gap and clearance information in real time and feeding back;

[0018] Edge server: storing program code, vehicle data and detection results, and obtaining the VIN code of the vehicle currently triggering the photoelectric sensor;

[0019] PLC controller: integrating vehicle information obtained by the photoelectric sensor and the edge server, for motion control of the collaborative robot.

[0020] As Figure 1As shown, preferably, the part mistake-proof detection module further comprises a sample feature library, the part mistake-proof detection module can accumulate sample information based on continuous image input, by judging the accumulation of the current sample feature library, the sample feature library is updated iteratively, and the algorithm model is further updated incrementally in the form of an additional parameter patch.

[0021] Preferably, the part mistake-proof detection module introduces an LLM model as a basic abnormality recognition network, which combines prior information of the feature library as external Few-Shot Prompt information for abnormal information judgment; at this time, the top-k result is obtained by comparing the fused image features and the sample feature library, and the LLM model judgment result is cross-validated.

[0022] Preferably, the part mistake-proof detection module realizes incremental update of the algorithm model through T-Patcher patch technology.

[0023] As Figure 3 shown, the application also provides a method for synchronously completing part mistake-proof detection and gap measurement, the method uses the system for synchronously completing part mistake-proof detection and gap measurement as described above, and the method comprises the following steps:

[0024] Step S1: acquiring current detection vehicle information, when the front end of the vehicle head parked on the conveying belt triggers the photoelectric sensor, the detection process of the current vehicle is started; recording the triggering time of the current photoelectric sensor, the encoder real-time acquires the conveying belt speed, the edge server acquires the VIN code of the current vehicle through the workshop MES system, loads the configuration information of the current vehicle through the VIN code, and associates the digital-analog gap information, and the preset detection position of the vehicle is called coarse positioning trajectory;

[0025] Step S2: acquiring vehicle outer surface image information, the cooperative mechanical arm runs to the preset detection point, at this time, the 3D camera is used to scan the surface of the vehicle at the detection point, and the RGB color image and the point cloud image of the vehicle at the detection point are obtained;

[0026] Step S3: combining the color image and the point cloud image and the vehicle digital-analog information, the working position of the gap surface difference detector is calculated;

[0027] Step S4: the cooperative mechanical arm transports the 3D camera and the gap surface difference detector to the working position according to the preset path and motion parameters; the PLC controller controls the slide rail to start moving, and the speed is synchronized with the conveying belt;

[0028] Step S5: When the slide rail speed is synchronized with the conveyor belt, the gap and face difference detector starts to measure the gap and face difference of the vehicle; after the measurement is completed, the collaborative robot moves to the next detection point or returns to the initial position after the current vehicle detection is completed, and waits for the next vehicle to enter.

[0029] Preferably, the step S1 further comprises:

[0030] When the vehicle on the conveyor belt blocks and triggers the photoelectric sensor, the triggering time t0 is recorded, and the PLC controller obtains the conveyor belt speed calculated by the encoder in real time, wherein the calculation formula of the encoder is:

[0031] V = ( N / t ) × ( P / L );

[0032] Wherein, V is the speed of the conveyor belt; t is the time used for measuring the pulse number; N is the number of pulses received in t time; P is the number of pulses generated by the encoder per revolution; L is the circumference of the conveyor belt.

[0033] Preferably, the speed V is filtered by Kalman filtering algorithm to remove noise and interference.

[0034] Preferably, the step S2 further comprises point cloud preprocessing of the RGB color image and the point cloud image, and the point cloud preprocessing comprises denoising and down-sampling.

[0035] Preferably, the denoising step comprises:

[0036] Neighborhood search: for each point P in the point cloud, calculate its k nearest neighbors;

[0037] Calculate the average distance μi: calculate the distance between each point Pi and its k nearest neighbors, and calculate the average value μi of these distances:

[0038] ;

[0039] Calculate the standard deviation σi:

[0040] ;

[0041] Outlier determination: determine whether the point Pi is an outlier according to the average distance and the standard deviation, and set a threshold τ, if d(Pi)>τ, then the point Pi is an outlier, and the formula is:

[0042] ;

[0043] Remove outliers: according to the determination result, remove all points marked as outliers.

[0044] Preferably, the down-sampling step comprises:

[0045] setting a proportion r of points to be reserved;

[0046] randomly selecting P' = r x P from the denoised point cloud, wherein P is the denoised point cloud, and P' is the down-sampled point cloud.

[0047] Compared with the prior art, the application has the following beneficial effects:

[0048] 1. The system and method for synchronously completing part mistake-proof detection and gap measurement provided by the application can simultaneously complete online intelligent detection and measurement in a single station, which can save about 300,000 yuan of installation cost of mechanical device layout of a station. The appearance detection efficiency is improved by 70%, the accuracy is 99.95%, and the detection is completely completed by artificial replacement. The measurement efficiency and sample coverage rate are improved: 3 hours of inspection time is saved for a single vehicle, the whole vehicle off-line quality qualification rate is improved by 17%, more than 100,000 vehicles are detected annually, more than 40 times of misloading problems are found, 2 inspection workers are reduced, and 3,000,000 yuan of after-sales quality cost is saved.

[0049] 2. The system and method for synchronously completing part mistake-proof detection and gap measurement provided by the application introduce a self-learning method, which can automatically identify future new vehicle models or part drawings, greatly reducing the time of manual data collection and labeling. Through incremental updating, the model is continuously optimized and the feature library is continuously updated, which can efficiently perceive new data and realize self-adaptive iteration of the feature model.

[0050] 3. The system and method for synchronously completing part mistake-proof detection and gap measurement provided by the application can automatically adapt to feature points of different vehicle models by combining key point information obtained by a 3D camera and vehicle digital model information, realize detection of various vehicle body gap, and do not need to adjust mechanical arm trajectories and separately set measurement parameters and paths for each vehicle model, greatly improving the universality and flexibility of the system. BRIEF DESCRIPTION OF DRAWINGS

[0051] Other features, objects and advantages of the application will become more apparent from the following detailed description of non-limiting embodiments, made with reference to the accompanying drawings:

[0052] Figure 1 The structural schematic diagram of the part mistake-proof detection module in embodiment 1 is shown in the figure.

[0053] Figure 2 The schematic diagram of the T-Patcher patch technology in embodiment 1 is shown in the figure.

[0054] Figure 3 The flowchart of the method for synchronously completing part mistake-proof detection and gap measurement in embodiment 2 is shown in the figure. DETAILED DESCRIPTION

[0055] The technical solutions and advantages of the embodiments of the present application will be more apparent from the following description of the embodiments of the present application in conjunction with the accompanying drawings. Obviously, the described embodiments are only some of the embodiments of the present application, rather than all the embodiments. The components of the embodiments of the present application described and shown in the accompanying drawings can be arranged and designed in various different configurations.

[0056] Therefore, the following detailed description of the embodiments of the present application provided in the accompanying drawings is not intended to limit the scope of the claimed present application, but only represents selected embodiments of the present application. All other embodiments obtained by those of ordinary skill in the art based on the embodiments in the present application without creative labor are within the scope of protection of the present application.

[0057] It should be noted that: similar reference numbers and letters represent similar items in the following drawings, therefore, once an item is defined in one drawing, it does not need to be further defined and explained in subsequent drawings. In addition, all directional indications (such as up, down, left, right, front, back, bottom, etc.) in the present application are only used to explain the relative positional relationship, movement condition, etc. between components in a certain posture (as shown in the drawings), if the certain posture changes, the directional indications also change accordingly. Further, the description involving "first", "second", etc. in the application is only for the purpose of description, and cannot be understood as indicating or implying the relative importance of the indicated technical features or implicitly indicating the number of the indicated technical features.

[0058] Embodiment 1

[0059] The present embodiment provides a system for simultaneously completing part mistake-proof detection and gap clearance measurement, which simultaneously completes vehicle part mistake-proof detection and gap clearance measurement in a moving state of the vehicle, and the system comprises:

[0060] 3D camera: for shooting photos of vehicle appearance features, and acquiring point position information of the vehicle outer surface;

[0061] Gap clearance measurement module, comprising a gap surface difference detector, for accurately measuring the gap and surface difference of the vehicle;

[0062] Part mistake-proof detection module: based on a pre-constructed initial sample set, a feature extraction network and a detection judgment network are constructed, product feature information of a search hit is sent to the feature extraction network according to image data acquired by the 3D camera, a multi-dimensional feature map is obtained, the multi-dimensional feature map is converted into a sequence vector through a visual embedding layer and sent to the detection judgment network for judgment, and finally a detection result is obtained;

[0063] Collaborative robot: according to the point information obtained by the 3D camera, it moves to the corresponding gap detection position for detection; the 3D camera and the gap face difference detector are integrated and installed at the end of the collaborative robot;

[0064] Optical sensor: provides information on the detection range of vehicles entering and leaving the system;

[0065] Encoder: obtains the speed of the conveyor belt, which is used to calculate the distance traveled by the vehicle on the conveyor belt during the detection process;

[0066] Slide rail: when the collaborative robot sends the gap face difference detector to the specified position, the slide rail is synchronized with the speed of the conveyor belt to ensure that the gap face difference detector remains relatively stationary relative to the gap measurement point, obtaining accurate measurement values;

[0067] Training server: used to deploy AI visual detection algorithms, providing corresponding point detection power for the system, real-time analyzing image features and calculating point gap information, and feeding back;

[0068] Edge server: stores program code, vehicle data and detection results, and obtains the VIN code of the vehicle that triggers the optical sensor;

[0069] PLC controller: integrates vehicle information obtained by the optical sensor and the edge server, and is used for motion control of the collaborative robot.

[0070] Further, the part mistake-proofing detection module further comprises a sample feature library, and the part mistake-proofing detection module can accumulate sample information based on continuous image input, autonomously select to update and iterate the sample feature library by judging the accumulation of the current sample feature library, and further perform incremental update of the algorithm model in the form of an additional parameter patch.

[0071] Further, the part mistake-proofing detection module introduces an LLM model as a basic abnormality recognition network, and the LLM model combines prior information of the feature library as external Few-Shot Prompt information to determine abnormal information; at this time, the top-k result is obtained based on comparison of the fused image features and the sample feature library, and cross-validation is performed with the LLM model determination result.

[0072] Specifically, the top-k result is obtained by using feature similarity to find the k samples most similar to the given sample. In the calculation, it is assumed that the fused feature vector has been obtained.

[0073] First, the similarity measurement method between feature vectors needs to be defined. Common similarity measurements include cosine similarity, Euclidean distance, etc. In this embodiment, cosine similarity is taken as an example.

[0074] Let and is the cosine similarity between two samples is defined as:

[0075] ,

[0076] where denotes the dot product of two vectors, denotes the norm of a vector.

[0077] Secondly, the similarity matrix is computed. Suppose there are samples, whose fused feature vectors are respectively. A similarity matrix can be computed, where the element in the matrix represents the similarity between the -th sample and the -th sample:

[0078] ,

[0079] where , .

[0080] Finally, the top-k results are obtained. For each sample , the top-k samples with the highest similarity are needed. Let denote the -th row vector of the similarity matrix , which contains the similarity between the sample and all other samples. Define a sorting function

[0081] , which sorts the elements in the vector from large to small and returns the sorted element value vector and the corresponding index vector . That is:

[0082] .

[0083] Then, the index set of the top-k similar samples of the sample can be represented as:

[0084] .

[0085] where denotes the index of the -th element in the original vector after sorting.

[0086] ​​Further, the part mistake-proofing detection module realizes incremental updating of the algorithm model through T-Patcher patch technology. Specifically, an additional parameter is introduced in the model by using the T-Patcher additional parameter method, which adds an additional parameter in the full connection feedforward layer of the last Transformer layer of the model, i.e., a patch, and then the patch is trained to complete the editing of specific knowledge. When to update depends on the monitored model performance, and iterative updating is performed during idle time. The form of the patch is as shown in Figure 2 .

[0087] After adding the patch, the output of the full connection feedforward layer is adjusted as follows:

[0088] ,

[0089] .

[0090] By designing a loss function considering accuracy and locality, the patch can correctly modify the output of the model and also reduce the impact on other problems, thereby realizing accurate and reliable model editing. The relevant expression is as follows:

[0091] .

[0092] Embodiment 2

[0093] The embodiment provides a method for synchronously completing part mistake-proofing detection and gap measurement, and the method uses the system for synchronously completing part mistake-proofing detection and gap measurement in the embodiment 1, and the method comprises the following steps:

[0094] Step S1: acquiring current vehicle information, when the front end of the vehicle head parked on the conveying belt triggers the photoelectric sensor, the detection process of the current vehicle is started; the triggering time of the current photoelectric sensor is recorded, the encoder acquires the conveying belt speed in real time, the edge server acquires the VIN code of the current vehicle through the workshop MES system, loads the configuration information of the current vehicle through the VIN code, and associates the digital-analog gap information, and the coarse positioning trajectory of the preset detection position of the vehicle model is called;

[0095] Step S2: acquiring vehicle outer surface image information, the cooperative mechanical arm runs to the preset detection point, at this time, the 3D camera is used to scan the surface of the detection point of the vehicle, and the RGB color image and the point cloud image of the surface of the detection point of the vehicle are acquired;

[0096] Step S3: combining the color image and the point cloud image and the vehicle digital-analog information, the working position of the gap surface difference detector is calculated;

[0097] Step S4: The collaborative robot carries the 3D camera and the gap and flatness detector to the working position according to the preset path and motion parameters; the PLC controller controls the slide rail to start moving at a speed synchronized with the conveyor belt;

[0098] Step S5: After the speed of the slide rail is synchronized with the conveyor belt, the gap and flatness detector starts measuring the gap and flatness of the vehicle; after the measurement is completed, the collaborative robot moves to the next detection point or returns to the initial position after the current vehicle detection is completed, and waits for the next vehicle to enter.

[0099] Further, step S1 further includes:

[0100] When the vehicle on the conveyor belt blocks and triggers the photoelectric sensor, the triggering time t0 is recorded, and the PLC controller obtains the speed of the conveyor belt calculated by the encoder in real time, wherein the calculation formula of the encoder is:

[0101] V = (N / t) x (P / L);

[0102] Wherein, V is the speed of the conveyor belt; t is the time for measuring the pulse number; N is the number of pulses received in t time; P is the number of pulses generated by the encoder per revolution; L is the circumference of the conveyor belt.

[0103] Further, the speed V is filtered by Kalman filtering algorithm to remove noise and interference and improve the stability and reliability of the data. Through integration, the distance moved by the vehicle within any time after triggering the photoelectric sensor can be known, which is recorded as CarDis. At the same time, the mechanical arm on the slide rail is also controlled by the PLC controller to slide forward and backward, and the distance of the mechanical arm relative to the initial mechanical arm can be obtained in real time, which is recorded as ArmDis, and the coordinates of the mechanical arm need to be adjusted according to the displacement of the slide rail. Assuming that the origin of the original coordinate system is O, the origin O is the original coordinate system of the mechanical arm at the triggering time of the photoelectric sensor, the initial position of the mechanical arm is P, and the gap measurement points DetP1(x1, y1, z1), DetP2(x2, y2, z2), DetP3(x3, y3, z3) and the like of the vehicle are set, according to the on-site environment, the movement direction of the slide rail is the positive movement of the X axis, after the slide rail moves X meters in the positive direction, the new position P' can be mapped by the following way, the initial coordinates on the slide rail are (Px, Py, Pz), then at a certain time, after the slide rail moves X meters, the new coordinates P' of the mechanical arm can be expressed as:

[0104] P'=(Px+X,Py,Pz);

[0105] Similarly, the new coordinates of the preset gap measurement points are:

[0106] DetP1'=(x1+CarDis,y1,z1);

[0107] DetP2' = (x2 + CarDis, y2, z2);

[0108] DetP3' = (x3 + CarDis, y3, z3).

[0109] In addition, the formula for mapping each point Pcar = (Xcar, Ycar, Zcar) in the vehicle digital model to a point Parm = (Xarm, Yarm, Zarm) in the mechanical arm base coordinate system is:

[0110] Parm = R · Pcar + T.

[0111] Where: R is the rotation matrix, T is the translation vector.

[0112] Further, step S2 further includes point cloud preprocessing of the RGB color image and the point cloud image, and the point cloud preprocessing includes denoising and downsampling. Due to environmental influences, the vehicles parked on the conveyor belt are not centrally placed, with left and right deviations, but the gap face difference detector must be vertically above the working range of 5 cm ± 1 cm of the gap, so it is necessary to first roughly set a coarse sampling point within a safe range at the preset gap measurement point, and after the collaborative mechanical arm operates the 3D camera to the coarse sampling point, the point cloud and the RGB image are collected, the actual position of the gap is obtained through the point cloud preprocessing and fusion gap positioning operation, and the working position P lasers (x, y, z) of the gap face difference detector is calculated, and then the collaborative mechanical arm delivers the gap face difference detector to the calculated working position, and then the gap face difference detector starts working to measure the distance of the gap.

[0113] Specifically, the step of denoising includes:

[0114] Neighborhood search: for each point P in the point cloud, calculate its k nearest neighbors;

[0115] Calculate the average distance μi: calculate the distance of each point Pi from its k nearest neighbors, and find the average value μi of these distances:

[0116] ;

[0117] Calculate the standard deviation σi:

[0118] ;

[0119] Outlier determination: determine whether point Pi is an outlier based on the average distance and the standard deviation, and set a threshold τ, if d(Pi) > τ, then point Pi is considered an outlier, and the formula is:

[0120] ;

[0121] Remove outliers: according to the determination result, remove all points marked as outliers.

[0122] Specifically, the down-sampling step includes:

[0123] Set the proportion r of points to be retained;

[0124] Randomly select P' = r x P from the denoised point cloud, where P is the denoised point cloud, and P' is the down-sampled point cloud.

[0125] After the point cloud preprocessing, the color image obtained is segmented by a deep learning image segmentation algorithm such as Mask R-CNN, SegNet, DeepLab series, etc. The gap is segmented in the image, the mask edge after segmentation is matched with the point cloud graph, and the matched point cloud is extracted. In these selected points, the average values of x, y and z are calculated, denoted as Point(x', y', z'), and the normal vector of the point cloud plane. At a distance of 5 cm above the average value along the normal vector, it is the working point of the gap face difference detector. Finally, the collaborative robot transports the gap face difference detector to the working point, and the gap face difference detector starts measuring, calculates the gap size, and returns to the computer storage system.

[0126] The system and method for synchronously completing part mistake-proof detection and gap measurement provided by the application have the following advantages: first, the self-learning method is introduced, which can automatically identify future new vehicle models or part drawings, greatly reducing the time for manual data collection and labeling. Through incremental updating, the model is continuously optimized and the feature library is continuously updated, which can efficiently perceive new data and realize self-adaptive iteration of the feature model; second, by combining the key point information obtained by the 3D camera and the vehicle numerical model information, the feature points of different vehicle models can be automatically adapted, the detection of various vehicle gap gaps can be realized, and the mechanical arm trajectory and measurement parameters and path for each vehicle model do not need to be adjusted separately, greatly improving the universality and flexibility of the system; third, online intelligent detection and measurement can be completed at a single station, which can save about 300,000 yuan of installation cost of mechanical device layout of a station, improve the appearance detection efficiency by 70%, and the accuracy rate is 99.95%, which completely replaces manual detection, and the measurement efficiency and sample coverage rate are improved: a single vehicle saves 3 hours of inspection time, the whole vehicle off-line quality qualification rate is improved by 17%, more than 100,000 vehicles are detected annually, more than 40 times of misloading problems are found, 2 inspection workers are reduced, and 3 million yuan of after-sales quality cost is saved.

[0127] The specific embodiments of the application are described above, and through the above description, relevant personnel can make various changes and modifications without deviating from the technical idea of the application.

Claims

1. A system for synchronously completing a parts-poka-yoke inspection and a break-gap measurement, the system comprising: The system simultaneously completes error-proof detection of vehicle parts and gap measurement in a vehicle motion state, and the system comprises: a 3D camera: used for taking photos of vehicle appearance features and acquiring point position information of a vehicle outer surface; a gap measurement module: comprising a gap and surface difference detector, used for accurately measuring the gap and surface difference of the vehicle; a part error-proof detection module: based on a pre-constructed initial sample set, a feature extraction network and a detection judgment network are constructed, product feature information of a search hit is sent to the feature extraction network according to image data acquired by the 3D camera, a multi-dimensional feature map is obtained, the multi-dimensional feature map is converted into a sequence vector through a visual embedding layer and is sent to the detection judgment network for judgment, and finally a detection result is obtained; a collaborative robot arm: according to the point position information acquired by the 3D camera, the collaborative robot arm moves to a corresponding gap detection position for detection; the 3D camera and the gap and surface difference detector are integrally installed at the end of the collaborative robot arm; a photoelectric sensor: providing information about the detection range of the system for vehicle entry and exit; an encoder: acquiring the speed of a conveyor belt, used for calculating the distance of a vehicle running on the conveyor belt during the detection process of the system; a slide rail: when the collaborative robot arm sends the gap and surface difference detector to a specified position, the slide rail is synchronized with the speed of the conveyor belt, ensuring that the gap and surface difference detector remains relatively stationary relative to the gap measurement position, and accurate measurement values are obtained; a training server: used for deploying an AI visual detection algorithm, providing corresponding point detection computing power for the system, and analyzing image features and calculating point gap information in real time and feeding back; an edge server: storing program codes, vehicle data and detection results, and acquiring the VIN code of a vehicle currently triggering the photoelectric sensor; a PLC controller: integrating vehicle information acquired by the photoelectric sensor and the edge server, used for motion control of the collaborative robot arm; wherein the part error-proof detection module further comprises a sample feature library, the part error-proof detection module can accumulate sample information based on continuous image input, autonomously select to update and iterate the sample feature library by judging the accumulation of the current sample feature library, and further update the algorithm model in the form of an additional parameter patch.

2. The system of claim 1, wherein, The part error-proof detection module introduces an LLM model as a basic abnormality recognition network, the LLM model combines prior information of the sample feature library as external Few-Shot Prompt information for abnormal information judgment; at this time, the image features and the sample feature library are compared based on fusion to obtain top-k results, and the LLM model judgment results are cross-validated.

3. The system of claim 1, wherein, The part error-proof detection module realizes incremental update of the algorithm model through T-Patcher patch technology.

4. A method of synchronously completing mistake proofing inspection and gap measurement of a part, the method employing the system for synchronously completing mistake proofing inspection and gap measurement of a part according to any one of claims 1 to 3, characterized in that, The method comprises: Step S1: Obtain current vehicle information, start the detection process of the current vehicle when the vehicle head of the vehicle parked on the conveyor belt triggers the photoelectric sensor; record the triggering time of the current photoelectric sensor, the encoder real-time obtains the conveyor belt speed, the edge server obtains the VIN code of the current vehicle through the workshop MES system, loads the configuration information of the current vehicle through the VIN code, and associates the digital-analog gap information, and the preset detection position of the vehicle model is called; Step S2: Obtain vehicle surface image information, and the cooperative robot arm runs to the preset detection point, at which time the 3D camera scans the surface of the vehicle at the detection point to obtain the RGB color image and the point cloud image of the surface of the vehicle at the detection point; Step S3: Combine the color image and the point cloud image and the vehicle digital-analog information to calculate the working position of the gap and surface difference detector; Step S4: The cooperative robot arm carries the 3D camera and the gap and surface difference detector to the working position according to the preset path and motion parameters; the PLC controller controls the slide rail to start moving at a speed synchronized with the conveyor belt; Step S5: After the speed of the slide rail is synchronized with the conveyor belt, the gap and surface difference detector starts to measure the gap and surface difference of the vehicle; after the measurement is completed, the cooperative robot arm runs to the next detection point or returns to the initial position after the current vehicle detection is completed, and waits for the next vehicle to enter.

5. The method of claim 4, wherein, The step S1 further comprises: When the vehicle on the conveyor belt blocks and triggers the photoelectric sensor, record the triggering time t0, and the PLC controller real-time obtains the conveyor belt speed calculated by the encoder, wherein the calculation formula of the encoder is: V = ( N / t ) × ( P / L ); Wherein, V is the speed of the conveyor belt; t is the time for measuring the pulse number; N is the number of pulses received in t time; P is the number of pulses generated by the encoder per revolution; L is the circumference of the conveyor belt.

6. The method of claim 5, wherein, The speed V is filtered by Kalman filtering algorithm to remove noise and interference.

7. The method of claim 4, wherein, The step S2 further comprises point cloud preprocessing of the RGB color image and the point cloud image, and the point cloud preprocessing comprises denoising and downsampling.

8. The method of claim 7, wherein, The denoising step comprises: neighborhood search: for each point P in the point cloud, calculate its k nearest neighbors; calculate the average distance μi: calculate the distance between each point Pi and the k nearest neighbors, and find the average value μi of the distances: ； calculate the standard deviation σi: ； outlier point determination: determine whether the point Pi is an outlier point according to the average distance and the standard deviation, set a threshold τ, if d(Pi)>τ, then the point Pi is an outlier point, the formula is: ； remove outliers: remove all points marked as outliers according to the determination result.

9. The method of claim 7, wherein, The downsampling step comprises: set the proportion r of points to be retained; randomly select P' = r × P from the denoised point cloud, wherein P is the denoised point cloud and P' is the downsampled point cloud.

Images