High-precision intelligent detection system based on AGV and three-coordinate measuring machine

Abstract

The invention discloses a high-precision intelligent detection system based on an AGV and a three-coordinate measuring machine, and relates to the technical field of high-precision intelligent detection of workpieces, the high-precision intelligent detection system comprises a path planning module and a type detection module, the path planning module combines a production line three-dimensional layout map, a workpiece preset placement area and real-time station occupation information, and the type detection module detects the type of the workpieces; an initial target station for loading the workpiece AGV is planned; when the AGV runs to the monitoring range of the vision sensor of the first station along the optimal path, the vision sensor begins to collect images of workpieces carried by the AGV, an image sequence of the workpieces changing along with time in the monitoring range is obtained, a type candidate set of the workpieces is preliminarily screened based on the image sequence, and after the AGV sends the workpieces to a detection station, the type candidate set of the workpieces is detected; the workpiece is further detected through a three-coordinate measuring machine, and the specific type of the workpiece is determined; the system improves the detection cooperation efficiency and the preliminary screening reliability, and guarantees the detection precision.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of high-precision intelligent detection of workpieces, in particular to a high-precision intelligent detection system based on AGV and three-coordinate measuring machines. BACKGROUND

[0002] In the field of industrial manufacturing, accurate detection of workpiece types is a key link to ensure production quality and process efficiency. However, existing workpiece type detection technologies still have the following defects: Firstly, existing solutions only consider distance factors when planning paths, without considering real-time workstation occupancy status, workstation release potential, and other dynamic information, resulting in invalid waiting of AGV carrying workpieces due to target workstation occupation, and low overall process efficiency. Secondly, existing technologies lack effective processing logic for dynamic image sequences, making it difficult to ensure the accuracy of preliminary screening and unable to provide reliable candidate sets for subsequent accurate detection. Thirdly, three-coordinate measuring machines in existing solutions are often used to directly detect the full size of workpieces without being linked with visual preliminary screening results, resulting in a large amount of redundant measurement operations, which not only takes a long time but also wastes equipment resources, making it difficult to adapt to the rapid detection needs of multiple types of workpieces. Therefore, there is an urgent need for a high-precision intelligent detection system based on AGV and three-coordinate measuring machines. SUMMARY

[0003] To address the deficiencies of existing technologies, the present application provides a high-precision intelligent detection system based on AGV and three-coordinate measuring machines, which solves the problems of insufficient workstation path coordination, unreliable visual preliminary screening, and weak device linkage in existing solutions.

[0004] To achieve the above purpose, the present application realizes the following technical solutions: a high-precision intelligent detection system based on AGV and three-coordinate measuring machines, comprising: A path planning module that combines a production line three-dimensional layout map, a workpiece preset placement area, and real-time workstation occupancy information to plan an initial target workstation for a workpiece-carrying AGV. A type detection module that, when the AGV travels along the optimal path to the monitoring range of the first workstation visual sensor, the visual sensor starts collecting workpiece images carried by the AGV, obtaining an image sequence of the workpiece within the monitoring range over time, and based on the image sequence, preliminarily screening a type candidate set of the workpiece, and when the AGV delivers the workpiece to the detection workstation, further detecting the workpiece through the three-coordinate measuring machine to determine its specific type.

[0005] As a further solution of the present application, planning an initial target workstation for a workpiece-carrying AGV specifically includes: The use state of all stations is collected in real time through a production line loT terminal, and a station-path association map is constructed in combination with a production line topological structure map, an optimal path from an AGV starting point to a current idle station is found, and the position coordinates, number and current occupation time length of all occupied stations on the optimal path are automatically marked to construct an optimal path list of the current idle station, and the use state includes two kinds of idle and occupied; For the optimal path bound to each idle station, the residual occupation estimated time of all occupied stations in the path is counted, and the minimum value of the residual occupation estimated time of all occupied stations is taken as the release potential of the path, and the residual occupation estimated time represents the difference between the historical detection time length average and the current occupied time length; The idle station with the minimum release potential is selected as the initial target station, and if the minimum release potential corresponds to multiple idle stations, the idle station with the shortest path physical length is selected as the initial target station.

[0006] As a further scheme of the application, the specific steps for finding the optimal path from the AGV starting point to the current idle station are as follows: Based on the association map, the node types contained in each candidate path are checked one by one, and the node types include travel nodes, turning nodes and obstacle adjacent nodes; The candidate paths containing the following nodes are directly rejected: turning nodes with an included angle < 90° and obstacle adjacent nodes, forming a preliminary screening path set; For each candidate path in the preliminary screening path set, the effective association density of all nodes is counted: taking each node on the candidate path as the center, all nodes directly connected to it are found, the number of travel nodes and turning nodes with an included angle ≥ 90° is counted, and the number is taken as the number of associated edges, and the average number of associated edges of all nodes in the candidate path is taken, i.e. the effective association density of the candidate path is obtained; The effective association densities of all candidate paths in the preliminary screening path set are counted, only the candidate paths with the top 3 effective association densities are retained, forming a refined screening path set, and the AGV driving process along the candidate paths in the refined screening path set is simulated according to the connection order of the nodes in the association map: starting from the starting node, the continuous traffic state of each node on the candidate path is checked in turn, the continuous traffic state represents the node state updated in real time through the map, and it is confirmed whether the connection between the current node and the next node is uninterrupted and without temporary occupation; For each candidate path in the refined screening path set, the number of continuous uninterrupted nodes is recorded, and the candidate path with the highest proportion of continuous uninterrupted nodes to the total number of nodes and without more than two consecutive turning nodes is selected as the optimal path, and the number of continuous uninterrupted nodes represents that if the connection between a node and the next node is temporarily occupied during driving, the count is interrupted, and the maximum number of continuous uninterrupted nodes is taken; The candidate path with the highest proportion of continuous uninterrupted node number to the total node number of the path and without more than two consecutive turning nodes is selected as the optimal path.

[0007] As a further scheme of the present application, after the initial target station is determined, the AGV starts and travels along the optimal path of the initial target station, when the use state of a certain occupied station in front of the travel direction of the AGV changes to idle, the idle station is directly updated as a new target station, and the travel trajectory is automatically corrected through the path topology map, so that the AGV drives to the new target station.

[0008] As a further scheme of the present application, the specific steps of preliminarily screening the candidate set of the workpiece type based on the image sequence are as follows: For the image sequence, the global pixel overlap degree of adjacent two frames is calculated in frame order to form an inter-frame overlap degree sequence, and the deviation value of each frame overlap degree from the sequence mean value is calculated to obtain a deviation value sequence, when the deviation value changes from positive to negative or from negative to positive, it is marked as a change inflection point, and the frame number between all adjacent inflection points is counted to obtain an inter-frame interval sequence; For the values in the deviation value sequence, if the value > 0, it is replaced with 1, if the value < 0, it is replaced with -1, and if the value is 0, it remains unchanged; Based on the similarity S1 of the deviation value sequence and the similarity S2 of the inter-frame interval sequence, the cooperative fitting coefficient C is calculated ; Based on the cooperative fitting coefficient C of all standard workpieces, the dynamic screening threshold T is calculated as the mean value of all C values multiplied by 1.1, the type corresponding to the standard workpiece with C≥T is screened out, if the types meeting the condition exceed three, the TOP3 is taken according to the C value from high to low, if there are less than three, the first three with the highest C value are taken, and the types with C < T but ranking in the front, finally forming the candidate set of the workpiece type.

[0009] As a further scheme of the present application, the specific operation of calculating the global pixel overlap degree of adjacent two frames in frame order is as follows: The adjacent two frames of images are respectively preprocessed: ① converted to 8-bit grayscale image; ② fixed image resolution to 1920x1080; ③ removed noise in industrial environment by using 3x3 mean filter to obtain two frames of standardized grayscale images, denoted as G1 and G2; The standardized G1 and G2 are completely aligned with the image center as the reference, and the normalized cross-correlation algorithm is directly applied to the two frames of complete images: the grayscale value covariance of G1 and G2 is calculated pixel by pixel, and divided by the product of the standard deviations of the grayscale values of the two frames to obtain the normalized cross-correlation coefficient R; The cross-correlation coefficient R is converted into the global pixel overlap degree P, and the specific formula is: P=(R+1)×50%.

[0010] As a further scheme of the present application, the deviation value sequence and the standard deviation value sequence are aligned in the same length, and the last symbol is supplemented when it is insufficient, the number of the same position symbol alternation consistent is counted, and the deviation value sequence similarity S1=consistent number / total alternation number is obtained.

[0011] As a further scheme of the present application, the high frequency interval of the real-time frame interval sequence and the standard frame interval sequence is counted respectively, the overlap degree=intersection length / union length of the two intervals is calculated, and the proportion of the values falling in the high frequency interval of the other party in the two sequences is counted, and the frame interval sequence similarity S2=(overlap degree+two-way proportion average) / 2 is obtained, wherein the high frequency interval represents the value range with the appearance frequency≥30%.

[0012] As a further scheme of the present application, the specific steps for further detecting the specific type of the workpiece by the three-coordinate measuring machine are as follows: For the workpiece type candidate set, the unique three-dimensional reference coordinate system of each candidate workpiece and two sets of size association rules are called from the system database, the three-dimensional reference coordinate system is composed of three non-collinear reference points, and the size association rule is a fixed logical relationship of two independent linear dimensions, the logical relationship and the numerical proportion of each rule only correspond to the candidate workpiece, and the rules of the three candidate types have no repetition; After the three-coordinate measuring machine is started, the three reference points of the first candidate workpiece are measured in contact: the actual three-dimensional coordinates of each reference point are collected by the measuring head, the deviation values of the actual coordinates and the reference coordinates in the database are calculated, the single-axis deviation of X, Y and Z axes is calculated, if the single-axis deviation of the three reference points is less than or equal to the specified threshold, it is determined that the reference adaptation of the candidate workpiece is passed, if the single-axis deviation of any reference point is greater than the specified threshold, the candidate workpiece is directly removed from the workpiece type candidate set, the above operation is repeated for the next candidate workpiece according to the sequence, and the candidate workpiece with reference adaptation passed is filtered out. For the candidate workpiece with reference adaptation passed, whether it meets the two sets of size association rules is verified: based on the actual coordinates of the reference points measured above, the actual values of the two linear dimensions in each rule are calculated, and whether the actual values meet the fixed logical relationship is calculated, if the two sets of rules of a candidate workpiece meet, the candidate workpiece is retained and marked as “rule full satisfaction”; if only one set of rules meets, it is marked as “rule partial satisfaction”; if neither meets, the candidate workpiece is directly removed. After the rule verification of all the adaptation candidates is completed, only the candidate workpieces with “rule full satisfaction” and “rule partial satisfaction” are retained, and the fine screening workpiece type is formed. If the precision screening workpiece type is only one, it is directly determined as the final type, if the precision screening workpiece type is more than one, the unique difference size of the precision screening workpiece type is called, the three-coordinate measuring machine only measures the actual value of the difference size, the deviation value of the actual value and the preset difference size of each precision screening workpiece is obtained, the precision screening workpiece type corresponding to the minimum deviation value is selected as the final type, and the unique difference size is a linear physical size that is preset for each workpiece in the precision screening workpiece type and can be absolutely distinguished from other workpieces.

[0013] The application provides a high-precision intelligent detection system based on an AGV and a three-coordinate measuring machine, and has the following beneficial effects compared with the prior art: (1) The application collects real-time station states through a production line IoT terminal, constructs a station-path association graph, and selects an initial target station in combination with path release potential, supports dynamic update of idle stations during AGV travel, avoids transportation stagnation caused by station occupation in traditional schemes, and plans an optimal path through node type screening, effective correlation density calculation and continuous trafficability verification, thereby guaranteeing AGV travel stability. (2) The application constructs a sequence through image preprocessing and interframe global pixel overlap calculation, and selects a type candidate set in combination with deviation values and interframe interval sequence similarity, thereby effectively avoiding interference of industrial environment noise and illumination fluctuation on preliminary screening. (3) The application links three-coordinate measurement and visual preliminary screening candidates, removes inconsistent candidates through reference point adaptation, verifies size correlation rules, and finally measures only a unique difference size to determine a final type, thereby avoiding redundant operations of full-size measurement and saving device resources. BRIEF DESCRIPTION OF DRAWINGS

[0014] Figure 1 The application provides a system principle frame. DETAILED DESCRIPTION

[0015] The technical solutions in the embodiments of the application will be clearly and completely described below with reference to the drawings in the embodiments of the application. Obviously, the described embodiments are only part of the embodiments of the application, rather than all the embodiments of the application. Based on the embodiments in the application, all other embodiments obtained by a person of ordinary skill in the art without creative labor fall within the protection scope of the application.

[0016] As Figure 1 The application provides a high-precision intelligent detection system based on an AGV and a three-coordinate measuring machine, comprising: A path planning module plans an initial target station of a workpiece-loaded AGV in combination with a production line three-dimensional layout map, a workpiece preset placement area and real-time station occupation information, and the specific operation is as follows: The use state of all stations is collected in real time through the production line loT terminal, a station-path correlation map is constructed in combination with a production line topological structure map, an optimal path from an AGV starting point to a current idle station is found according to the correlation map, and the position coordinates, number and current occupation time length of all occupied stations on the optimal path are automatically marked to construct an optimal path list of the current idle station, the use state including two states of idle and occupied; The optimal path from the AGV starting point to the current idle station according to the correlation map specifically includes: Based on the correlation map, the types of nodes contained in each candidate path are checked one by one, and the types of nodes include travel nodes, turning nodes and obstacle adjacent nodes: The travel node refers to a node in the path without turning, without obstacles and extending in a straight line, such as a straight section of a production line trunk road and a straight channel between stations, and the AGV does not need to change speed and turn on the node and can travel at a stable speed; The turning node refers to a node in the path that needs to change the travel direction, such as a 90° turning point and a 120° turning point, and the node determines the turning speed of the AGV, and the smaller the angle, the more urgent the turning, and the more likely the workpiece carried by the AGV to shake; The obstacle adjacent node refers to a node with a distance from a fixed obstacle (such as a column or a device base) or a dynamic obstacle (such as temporarily stacked materials) ≤ a specified threshold, and the AGV needs to reserve enough space to avoid collision when carrying a workpiece, and also needs to be prepared for unexpected situations; The candidate paths containing the following nodes are directly removed: turning nodes with an angle < 90° and obstacle adjacent nodes, forming a preliminary screening path set; For each candidate path in the preliminary screening path set, the effective correlation density of all nodes is counted: taking each node on the candidate path as the center, all nodes directly connected to it are found, the number of travel nodes and turning nodes with an angle ≥ 90° is counted, and the number is taken as the number of associated edges, and the average number of associated edges of all nodes in the candidate path is taken, that is, the effective correlation density of the candidate path is obtained; The effective correlation densities of all candidate paths in the preliminary screening path set are counted, only the candidate paths with the top three effective correlation densities are retained, forming a fine screening path set, and the higher the correlation density, the smoother the path, and at the same time, if an unexpected situation occurs subsequently, it can be quickly adjusted through the surrounding associated nodes; According to the connection order of the nodes in the correlation map, the AGV travel process along the candidate paths in the fine screening path set is simulated: starting from the starting node, the continuous traffic state of each node on the candidate path is checked in turn, and the continuous traffic state represents the node state updated in real time through the map, and whether the connection between the current node and the next node is uninterrupted and temporarily occupied is confirmed; For each candidate path in the refined path set, record the number of consecutive uninterrupted nodes: if the connection between a node and the next node is temporarily occupied during travel, the count is interrupted, and the maximum number of consecutive uninterrupted nodes is taken; The more consecutive uninterrupted nodes, the stronger the dynamic stability of the candidate path, that is, the lower the probability of temporary occupation of node connection, the fewer the sudden conditions during AGV travel, even if interrupted, the maximum continuous segment length can ensure most of the route to be smooth, avoiding the risk of workpiece collision, equipment wear and tear and other risks caused by frequent interruptions; Select the candidate path with the highest proportion of consecutive uninterrupted nodes to the total number of nodes and no more than two consecutive turning nodes as the optimal path; The total number of nodes of different candidate paths may be different, and if only the number of consecutive uninterrupted nodes is considered, it will be distorted. By calculating the proportion of consecutive uninterrupted nodes to the total number of nodes, the candidate path with frequent interruptions but fewer total nodes can be avoided. Consecutive turns will cause the superposition of turning actions, which will destroy the stability of AGV travel and increase the risk of workpiece shaking; For each optimal path bound to each idle station, calculate the remaining occupation estimation time of all occupied stations in the path, and take the minimum value of the remaining occupation estimation time of all occupied stations as the release potential of the path. The smaller the value, the faster the occupied station on the candidate path will become idle. The remaining occupation estimation time represents the difference between the historical detection time length average and the current occupied time length; Preferentially select the idle station with the smallest release potential as the initial target station. If the smallest release potential corresponds to multiple idle stations, preferentially select the idle station with the shortest physical length of the path; The release potential of the optimal path of the idle station is used as the core screening dimension, rather than simply comparing the distance of the idle station, so that the screening result is more in line with the demand for efficiency maximization in the dynamic scenario; After determining the initial target station, the AGV starts and travels according to the optimal path of the initial target station. When the use state of a certain occupied station in front of the AGV travel direction becomes idle, there is no need to recalculate, and the idle station is directly updated as a new target station, and the travel trajectory is automatically corrected through the path topological map, so that the AGV drives to the new target station; For example, the AGV drives to the idle station A along the optimal path P1. During the travel, it is found that the use state of a certain occupied station B in front of the current position becomes idle. At this time, the AGV immediately switches the target station to station B, corrects the path and directly drives to station B, saving detection waiting time compared with the original plan.

[0017] The type detection module, when the AGV travels along the optimal path to the monitoring range of the vision sensor of the first station, the vision sensor starts collecting the image of the workpiece carried by the AGV, and obtains the image sequence of the workpiece changing over time; When the AGV travels along the optimal path, the first station is the first stable travel interval after the path starts. At this time, the AGV has just completed the start-up acceleration and has entered the uniform speed travel state. The slight shaking of the workpiece caused by the start-up has been completely eliminated, and the workpiece has not yet experienced the turning, path switching and other operations that may cause shaking. The position of the workpiece on the worktable is the most stable, and compared with the subsequent stations, the vision sensor of the first station can collect the clearest and most stable initial image frame. According to the image sequence, a type candidate set of the workpiece is obtained, and the specific steps are as follows: For the image sequence, the global pixel overlap degree of the adjacent two frames is calculated in sequence to form an inter-frame overlap degree sequence. Different workpieces have different volume and outline in the dynamic shooting of the AGV, and the overlap degree of adjacent frames will also be completely different, for example, when a large-size flat workpiece moves, the workpiece area range of adjacent frames has large overlap, the overlap degree fluctuation amplitude is small, and the sequence curve is flat. When a small-size columnar workpiece moves, the workpiece area range of adjacent frames has small overlap, the overlap degree fluctuation amplitude is large, and the sequence curve fluctuates obviously. The specific operation of calculating the global pixel overlap degree of the adjacent two frames in sequence is as follows: The adjacent two frames of images are respectively preprocessed: ① converted into 8-bit grayscale images; ② fixed image resolution to 1920x1080; ③ removed noise in industrial environment by using 3x3 mean filter to obtain two frames of standardized grayscale images, denoted as G1 and G2. The color of the workpiece in the industrial scene is not the key to type differentiation, and the color image contains redundant data of three channels of RGB, which increases the calculation complexity. The core of overlap degree calculation is the consistency of pixel grayscale value, and the use of grayscale image can directly strip the color interference. In the dynamic shooting of the AGV, the resolution of some frames may be abnormal due to sensor parameter fluctuation and surrounding environment influence. If directly calculated, the number of pixels and physical size of the two frames of images may not match, and due to noise influence, the local grayscale values of adjacent frames are inconsistent, which interferes with the authenticity of the overlap degree. The standardized G1 and G2 are completely aligned with the image center as the reference, and the normalized cross-correlation algorithm is directly applied to the two frames of complete images: the grayscale value covariance of G1 and G2 is calculated pixel by pixel, and divided by the product of the standard deviations of the two frames of grayscale values to obtain the normalized cross-correlation coefficient R. The closer R is to 1, the stronger the consistency of the grayscale value distribution of the two frames of pixels. The cross-correlation coefficient R is converted into the global pixel overlap degree P, and the specific formula is: P=(R+1)×50%; For the inter-frame overlap degree sequence, the deviation value of each frame overlap degree from the sequence mean value is calculated to obtain a deviation value sequence, when the deviation value changes from positive to negative or from negative to positive, it is marked as a change inflection point (representing a key change node of the workpiece shape or position), the frame number between all adjacent inflection points is counted to obtain an inter-frame interval sequence; For the values in the deviation value sequence, if the value > 0, replace it with 1, if the value < 0, replace it with -1, if the value is 0, do not change; Calculate the similarity S1 of the deviation value sequence: the deviation value sequence consists of -1, 0, and 1, and the core value is the sign alternation pattern, therefore, align the deviation value sequence with the standard deviation value sequence according to the same length, and if insufficient, complete it with the last sign, count the number of consistent sign alternation directions at the same position, such as the 2→3 frame of the real-time sequence is 1→-1, and the same position of the standard sequence is also 1→-1, that is, consistent, and S1=consistent number / total alternation number is obtained; Calculate the similarity S2 of the inter-frame interval sequence: the inter-frame interval sequence is a numerical sequence, and the core value is the interval distribution rule, the high-frequency interval of the frame interval value is counted for the real-time inter-frame interval sequence and the standard inter-frame interval sequence respectively, the overlap degree of the two intervals = intersection length / union length is calculated, and then the proportion of the values falling in the high-frequency interval of the other sequence in both sequences is counted, and S2=(overlap degree+two-way proportion mean) / 2 is obtained, wherein the high-frequency interval represents the value range with an occurrence frequency ≥30%; Calculate the synergistic fit coefficient The numerator part of the formula adopts the multiplication of the two similarities, only when both similarities are in the high value interval, the product will reach a high level, thereby limiting the misjudgment caused by the excellent single dimension from the source; and the core role of the denominator is to amplify the difference between the two indexes, when the two similarities are highly close, the value of the denominator is small, and the final obtained C is a relatively large value, and the constant 0.5 is added mainly to ensure that the denominator is not 0; Based on the synergistic fit coefficients C of all standard workpieces, a dynamic screening threshold T is calculated, wherein T=mean value of all C values×1.1; The types corresponding to the standard workpieces with C≥T are screened out, if the types meeting the conditions exceed three, the top 3 are taken from high to low according to the C value, if less than three, the first three with the highest C value are taken, and the types with C<T but ranking in the front, finally form a workpiece type candidate set; After the AGV delivers the workpiece to the detection station, it is further detected by the three-coordinate measuring machine to determine the specific type of the workpiece, and the specific operation is as follows: For each candidate workpiece type, a unique three-dimensional reference coordinate system and two sets of dimension association rules are retrieved from the system database. The three-dimensional reference coordinate system consists of three non-collinear reference points, such as the lower left corner point Q1, the lower right corner point Q2, and the center hole point Q3 on the top surface. The coordinate values ​​are fixed based on the workpiece design drawings. The dimension association rules are fixed logical relationships between two independent linear dimensions, such as "the distance between Q1 and Q2 = 1.5 times the distance between Q2 and Q3" and "the distance between Q1 and Q3 = 1.5 times the distance between Q2 and Q3". "Times", the logical relationship and numerical ratio of each set of rules correspond only to the candidate workpiece, and there is no repetition of rules among the three candidate types; After the coordinate measuring machine is started, it first performs contact measurement on the three reference points of the first candidate workpiece according to the workpiece type candidate set: the actual three-dimensional coordinates of each reference point are collected by the probe, and the deviation value between the actual coordinates and the reference coordinates in the database is calculated. The deviations of the X, Y, and Z axes are calculated separately. If the single-axis deviation of the three reference points is less than or equal to the specified threshold, the candidate workpiece is determined to have passed the reference adaptation. If the single-axis deviation of any reference point is greater than the specified threshold, the candidate workpiece is directly removed from the workpiece type candidate set without further measurement. The above operation is repeated for the next candidate workpiece according to the sorting until a candidate workpiece that has passed the reference adaptation is selected. For candidate workpieces that pass the benchmark adaptation, further verification is conducted to determine whether they meet the two sets of dimensional association rules: based on the actual coordinates of the benchmark points measured above, the actual values ​​of the two linear dimensions in each set of rules are calculated, and then it is checked whether the actual values ​​meet the fixed logical relationship. If both sets of rules for a candidate workpiece are met, the candidate workpiece is retained and marked as "fully satisfied"; if only one set of rules is satisfied, it is marked as "partially satisfied"; if neither is satisfied, the candidate workpiece is directly eliminated. After completing the rule verification of all matching candidates, only the candidate workpieces that "fully satisfy the rules" and "partially satisfy the rules" are retained to form the fine screening workpiece type; If there is only one type of workpiece for fine screening, it is directly determined as the final type. If there is more than one type of workpiece for fine screening, the unique difference dimension of the workpiece type is retrieved. The coordinate measuring machine only measures the actual value of the difference dimension and obtains the deviation value between the actual value and the preset difference dimension of each workpiece. The workpiece type corresponding to the smallest deviation value is selected as the final type. The unique difference dimension is a linear physical dimension that is preset individually for each workpiece in the workpiece type and can be absolutely distinguished from other workpieces.

[0018] Some of the data in the above formulas are numerical calculations with dimensions removed, and the contents not described in detail in this specification are all prior art known to those skilled in the art.

[0019] The above embodiments are only used to illustrate the technical methods of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical methods of the present invention without departing from the spirit and scope of the technical methods of the present invention.

Claims

1. A high-precision intelligent detection system based on an AGV and a three-coordinate measuring machine, characterized in that, Comprise: The path planning module, combined with the three-dimensional layout map of the production line, the preset placement area of the workpiece and the real-time station occupation information, plans the initial target station of the workpiece-loaded AGV; The type detection module, when the AGV travels to the monitoring range of the first station visual sensor along the optimal path, the visual sensor starts collecting the workpiece image carried by the AGV, obtains the image sequence of the workpiece in the monitoring range changing with time, and preliminarily screens the type candidate set of the workpiece based on the image sequence, and further detects the workpiece through the three-coordinate measuring machine after the AGV delivers the workpiece to the detection station to determine the specific type to which the workpiece belongs.

2. The high-precision intelligent detection system based on AGV and three-coordinate measuring machine according to claim 1, characterized in that, Planning the initial target station of the workpiece-loaded AGV specifically comprises: Collecting the use state of all stations in real time through the production line loT terminal, and combining the production line topological structure map, constructing a station-path correlation graph, finding the optimal path from the AGV starting point to the current idle station, and automatically marking the position coordinates, number and current occupation time length of all occupied stations on the optimal path, constructing the optimal path list of the current idle station, the use state including two kinds of idle and occupied; For each optimal path bound to an idle station, the remaining occupation estimated time of all occupied stations in the path is counted, and the minimum value of the remaining occupation estimated time of all occupied stations is taken as the release potential of the path, and the remaining occupation estimated time represents the difference between the historical detection time length average and the current occupied time length; Select the idle station with the smallest release potential as the initial target station, if the smallest release potential corresponds to multiple idle stations, select the idle station with the shortest path physical length as the initial target station.

3. The high-precision intelligent detection system based on AGV and three-coordinate measuring machine according to claim 2, characterized in that, The specific steps for finding the optimal path from the AGV starting point to the current idle station are: Based on the correlation graph, check the node types contained in each candidate path one by one, the node types include travel nodes, turning nodes and obstacle adjacent nodes; Directly eliminate candidate paths containing the following nodes: turning node angle < 90°, obstacle adjacent node, form the initial screening path set; For each candidate path in the initial screening path set, count the effective correlation density of all nodes in the path: take each node on the candidate path as the center, find all nodes directly connected to it, count the number of travel nodes and turning nodes with an angle ≥ 90° among them, and take the number as the number of associated edges, and take the average number of associated edges of all nodes in the candidate path as the effective correlation density of the candidate path; Count the effective correlation density of all candidate paths in the initial screening path set, only keep the candidate paths with the top 3 effective correlation densities, form the fine screening path set, and simulate the AGV travel process along the candidate paths in the fine screening path set according to the connection order of the nodes in the correlation graph: from the starting node, check the continuous traffic state of each node on the candidate path in turn, the continuous traffic state represents the node state updated in real time through the graph, and confirm whether the connection between the current node and the next node is uninterrupted and without temporary occupation; For each candidate path in the refined path set, record the number of consecutive uninterrupted nodes, select the candidate path with the highest proportion of consecutive uninterrupted nodes in the total number of nodes and without more than two consecutive turning nodes as the optimal path, the number of consecutive uninterrupted nodes represents that the connection between a node and the next node is temporarily occupied during driving, then the count is interrupted, and the maximum number of consecutive uninterrupted nodes is taken; Select the candidate path with the highest proportion of consecutive uninterrupted nodes in the total number of nodes and without more than two consecutive turning nodes as the optimal path.

4. The high-precision intelligent detection system based on AGV and three-coordinate measuring machine according to claim 1, characterized in that, After determining the initial target station, the AGV starts and travels according to the optimal path of the initial target station, when the use state of a certain occupied station in front of the AGV travel direction changes to idle, the idle station is directly updated as a new target station, and the travel trajectory is automatically corrected through the path topological map, so that the AGV drives to the new target station.

5. The high-precision intelligent detection system based on AGV and three-coordinate measuring machine according to claim 1, characterized in that, The specific steps of preliminarily screening the candidate set of the type to which the workpiece belongs based on the image sequence are: For the image sequence, the global pixel overlap degree of adjacent two frames is calculated in frame order to form an inter-frame overlap degree sequence, and the deviation value of each frame overlap degree from the sequence average is calculated to obtain a deviation value sequence, when the deviation value changes from positive to negative or from negative to positive, it is marked as a change inflection point, and the frame number between all adjacent inflection points is counted to obtain an interval sequence; For the values in the deviation value sequence, if the value is greater than 0, replace it with 1, if the value is less than 0, replace it with -1, and if the value is 0, do not change it; Based on the bias value sequence similarity S1 and the calculated inter-frame interval sequence similarity S2, a synergistic fit coefficient is calculated ; Based on the cooperative fitting coefficient C of all standard workpieces, a dynamic screening threshold T is calculated as the average of all C values multiplied by 1.1, and the type corresponding to the standard workpiece with C greater than or equal to T is screened out, if the number of types meeting the condition exceeds three, the top three types are selected according to the C value from high to low, if there are less than three, the first three types with the highest C value are selected, and the types with C less than T but ranking high are finally formed, and the candidate set of the type of the workpiece is finally formed.

6. The high-precision intelligent detection system based on AGV and three-coordinate measuring machine according to claim 5, characterized in that, The specific operation of calculating the global pixel overlap degree of adjacent two frames in frame order is: The adjacent two frames of images are preprocessed respectively: ① converted to 8-bit grayscale images; ② fixed image resolution to 1920x1080; ③ removed noise in industrial environment by using 3x3 mean filter to obtain two standardized grayscale images G1 and G2; The standardized G1 and G2 are completely aligned with the image center as the reference, and the normalized cross-correlation algorithm is directly applied to the two complete images: the gray value covariance of G1 and G2 is calculated pixel by pixel, and then divided by the product of the standard deviations of the gray values of the two frames to obtain the normalized cross-correlation coefficient R; The cross-correlation coefficient R is converted into the global pixel overlap degree P, and the specific formula is: P=(R+1)×50%.

7. The high-precision intelligent detection system based on AGV and three-coordinate measuring machine according to claim 5, characterized in that, The deviation value sequence and the standard deviation value sequence are aligned in the same length, and the tail symbol is supplemented when it is insufficient, the number of times of the same position symbol alternation direction consistency is counted to obtain the similarity S1 of the deviation value sequence.

8. The high-precision intelligent detection system based on AGV and three-coordinate measuring machine according to claim 5, characterized in that, The high-frequency interval of the frame interval values in the real-time frame interval sequence and the standard frame interval sequence is counted respectively, the overlap degree = intersection length / union length of the two intervals is calculated, and then the proportion of the values falling in the high-frequency interval of the other party in the two sequences is counted to obtain the frame interval sequence similarity S2 = (overlap degree + average of bidirectional proportion) / 2, wherein the high-frequency interval represents a value range with an occurrence frequency of more than 30%.

9. The high-precision intelligent detection system based on AGV and three-coordinate measuring machine according to claim 1, characterized in that, The specific steps for further detecting the specific type of the workpiece by the three-coordinate measuring machine are as follows: For the candidate set of workpiece types, the unique three-dimensional reference coordinate system of each candidate workpiece and two sets of size association rules are retrieved from the system database, the three-dimensional reference coordinate system is composed of three non-collinear reference points, and the size association rules are the fixed logical relationship of two independent linear dimensions, the logical relationship and numerical proportion of each rule only correspond to the candidate workpiece, and the rules of the three candidate types have no repetition; After the three-coordinate measuring machine is started, the three reference points of the first candidate workpiece are measured in contact: the actual three-dimensional coordinates of each reference point are collected by the measuring head, the deviation values of the actual coordinates and the reference coordinates in the database are calculated, the X, Y and Z axis deviations are calculated respectively, if the single-axis deviation of the three reference points is less than or equal to the specified threshold, it is determined that the reference adaptation of the candidate workpiece is passed, if the single-axis deviation of any reference point is greater than the specified threshold, the candidate workpiece is directly excluded from the candidate set of workpiece types, the above operation is repeated for the next candidate workpiece according to the order, and the candidate workpiece with reference adaptation passed is filtered out. For the candidate workpiece with reference adaptation passed, whether it meets the two sets of size association rules is verified: based on the actual coordinates of the reference points measured above, the actual values of the two linear dimensions in each rule are calculated, and then whether the actual values meet the fixed logical relationship is calculated, if the two sets of rules of a candidate workpiece meet, the candidate workpiece is retained and marked as "rule full satisfaction"; if only one set of rules meets, it is marked as "rule partial satisfaction"; if neither set of rules meets, the candidate workpiece is directly excluded; After the rule verification of all the adaptation candidates is completed, only the candidate workpieces with "rule full satisfaction" and "rule partial satisfaction" are retained to form the fine-screened workpiece types. If the fine-screened workpiece types are only one, it is directly determined as the final type, if the fine-screened workpiece types are more than one, the unique difference size of the fine-screened workpiece types is retrieved, the three-coordinate measuring machine only measures the actual value of the difference size, the deviation value of the actual value and the preset difference size of each fine-screened workpiece is calculated, and the fine-screened workpiece type corresponding to the minimum deviation value is selected as the final type, wherein the unique difference size is a linear physical size preset for each workpiece in the fine-screened workpiece types, which can absolutely distinguish other workpieces.

Images