Reducer box defect identification method and system based on industrial vision

Abstract

The invention discloses a speed reducer box defect identification method and system based on industrial vision, and relates to the technical field of vision inspection.The method comprises the steps that three-dimensional design information of a speed reducer box is collected, hole-shaped structure identification is conducted, and hole-shaped structure marking information is obtained; the conveying characteristics of the reducer box on a conveying line are obtained, the inner wall 360-degree imaging motion demand analysis of the multi-axis mechanical arm is executed in combination with the hole-shaped structure mark information, and a multi-axis motion control decision is constructed; and executing control of the multi-axis mechanical arm according to a multi-axis motion control decision, collecting an inner wall image set of the hole-shaped structure, and performing defect identification on the inner wall structure. The technical problem that in the prior art, comprehensive, accurate and efficient detection of the inner wall of the hole-shaped structure of the speed reducer box body is difficult to achieve in a dynamic conveying environment is solved, and the technical effect that through cooperative control of industrial vision and a multi-axis mechanical arm, the defect detection precision and efficiency of the inner wall of the speed reducer box body and the dynamic environment adaptability are improved is achieved.

Description

Technical Field

[0001] The present invention relates to the field of visual inspection technology, and in particular to a method and system for identifying defects in a reducer housing based on industrial vision. Background Art

[0002] In the manufacturing process of reducers, the housing serves as a key load-bearing and connecting structure, and its interior usually contains multiple hole-like structures for assembly, lubrication, or positioning. These holes are prone to defects such as cracks, sand holes, and burrs during casting, machining, or post-processing due to their large processing depth, complex structure, and irregular angles. Traditional detection methods mostly rely on manual visual inspection or single-angle photography inspection, which not only has many blind spots and poor repeatability, but also is difficult to adapt to the dynamic detection needs on the assembly line. Some automatic detection solutions that use fixed industrial cameras or single-axis robotic arms often have incomplete imaging or synchronization failures due to the inability to accurately match the hole distribution and transmission speed. Especially when faced with high-beat, multi-model mixed-line production scenarios, existing technologies have significant deficiencies in imaging angle control, detection rhythm coordination, and inner wall defect recognition accuracy, making it difficult to achieve stable and reliable full-coverage intelligent detection. Summary of the Invention

[0003] The present application provides a method and system for identifying defects in a reducer housing based on industrial vision, which is used to solve the technical problem that the existing technology is difficult to achieve comprehensive, accurate and efficient detection of the inner wall of the porous structure of the reducer housing in a dynamic transmission environment.

[0004] The first aspect of the present application provides a method for identifying defects in a reducer housing based on industrial vision, the method comprising: acquiring three-dimensional design information of the reducer housing to identify the porous structure and obtain porous structure marking information; obtaining the transmission characteristics of the reducer housing on the transmission line; performing a 360° imaging motion demand analysis of the inner wall of a multi-axis robot arm in combination with the porous structure marking information and the transmission characteristics, and constructing a multi-axis motion control decision; executing control of the multi-axis robot arm with the multi-axis motion control decision, and acquiring a set of images of the inner wall of the porous structure; and identifying inner wall structure defects based on the set of images of the inner wall of the porous structure.

[0005] The second aspect of the present application provides a reducer case defect recognition system based on industrial vision, the system comprising: a hole structure recognition module, the hole structure recognition module is used to perform hole structure recognition by collecting three-dimensional design information of the reducer case, and obtain hole structure marking information; a transmission feature acquisition module, the transmission feature acquisition module is used to obtain the transmission characteristics of the reducer case on the transmission line; an imaging motion demand analysis module, the imaging motion demand analysis module is used to perform 360° imaging motion demand analysis of the inner wall of the multi-axis robot arm in combination with the hole structure marking information and the transmission characteristics, and construct a multi-axis motion control decision; an image acquisition module, the image acquisition module is used to execute the control of the multi-axis robot arm with the multi-axis motion control decision, and collect a set of images of the inner wall of the hole structure; an inner wall structure defect recognition module, the inner wall structure defect recognition module is used to perform inner wall structure defect recognition based on the inner wall image set of the hole structure.

[0006] One or more technical solutions provided in this application have at least the following technical effects or advantages: The industrial vision-based reducer case defect recognition method and system provided in the present application relate to the field of visual inspection technology. By collecting the three-dimensional design information and transmission characteristics of the reducer case, identifying the porous structure and generating marking information, combining with a multi-axis robotic arm to perform 360° imaging path analysis, controlling the robotic arm to perform image acquisition, and identifying the inner wall structure defects based on the acquired image set, high-precision automated defect detection of the reducer case's porous structure is achieved, solving the technical problem that the existing technology is difficult to achieve comprehensive, accurate and efficient detection of the inner wall of the reducer case's porous structure in a dynamic transmission environment, and achieving the technical effect of improving the accuracy, efficiency and dynamic environment adaptability of the reducer case inner wall defect detection through the coordinated control of industrial vision and a multi-axis robotic arm. BRIEF DESCRIPTION OF THE DRAWINGS

[0007] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0008] Figure 1 A schematic flow chart of a method for identifying defects in a reducer housing based on industrial vision provided in an embodiment of the present application; Figure 2 Schematic diagram of the structure of the reducer housing defect recognition system based on industrial vision provided in an embodiment of the present application.

[0009] Explanation of the reference numerals: hole structure recognition module 11 , transmission feature acquisition module 12 , imaging motion demand analysis module 13 , image acquisition module 14 , inner wall structure defect recognition module 15 . DETAILED DESCRIPTION

[0010] The present application provides a method and system for identifying defects in a reducer housing based on industrial vision, which is used to solve the technical problem that the existing technology is difficult to achieve comprehensive, accurate and efficient detection of the inner wall of the porous structure of the reducer housing in a dynamic transmission environment.

[0011] The following will be combined with the accompanying drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only some of the embodiments of this application, not all of them. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0012] It should be noted that the terms "first", "second", etc. in the specification of the present application and the above-mentioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or server that includes a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or modules that are not clearly listed or inherent to these processes, methods, products or devices.

[0013] Example 1, as Figure 1 As shown, the present application provides a method for identifying defects in a reducer housing based on industrial vision, the method comprising: P10: By collecting the three-dimensional design information of the reducer housing, the hole structure is identified and the hole structure marking information is obtained.

[0014] Specifically, the first step is to comprehensively collect the 3D design information of the reducer housing. This 3D design information refers to the detailed 3D model data of the reducer housing generated using computer-aided design (CAD) software. This data includes not only the housing's external contours and dimensions, but also detailed information about its internal structure, particularly the geometric characteristics of pores, such as their diameter, depth, location, shape, and connections to other components. This information forms the basis for subsequent pore structure identification and defect detection.

[0015] In order to collect three-dimensional design information, the following technical means can be used: If the design of the reducer housing has been completed using CAD software, a three-dimensional model file containing all the necessary information can be directly exported from the CAD software. These files can provide high-precision geometric descriptions and are suitable for subsequent processing and analysis. For reducer housings that have been manufactured but do not retain CAD models, they can be scanned using three-dimensional scanning equipment, such as laser scanners or structured light scanners. The scanning equipment generates point cloud data on the surface of the housing by emitting laser or structured light and capturing the pattern of reflected light or deformed light. Subsequently, the point cloud data is processed and fitted to reconstruct the three-dimensional model of the reducer housing. Scanning technology can obtain high-precision three-dimensional data in a non-contact manner, and is particularly suitable for modeling complex shapes and internal structures.

[0016] After collecting the 3D design information, the pore structure recognition phase begins. This phase involves automatically detecting and labeling pore features within the 3D model using image processing and computer vision algorithms. Specifically, the collected 3D design information undergoes preprocessing, including noise removal, smoothing, and data format conversion, to ensure data quality and consistency. For example, point cloud data obtained through scanning requires filtering and downsampling to remove noise generated during the scanning process and reduce the data volume for subsequent processing. Subsequently, computer vision algorithms are used to analyze the preprocessed 3D model and extract pore structure features. Commonly used algorithms include geometry-based detection algorithms, such as cylinder detection, and deep learning-based feature extraction algorithms. For example, for regular circular pores, the Hough Transform algorithm can be used to detect circular features. For complex pore shapes, a convolutional neural network (CNN) can be used to learn and classify features from sliced ​​images of the 3D model to identify the location and shape of the pore structure. After recognition, the pore structure features are labeled, and the relevant information is stored as pore structure labeling information. Hole structure marking information should include key parameters such as hole number, location coordinates, diameter, and depth, allowing for rapid location and recall in subsequent steps. For example, marking information can be stored as a structured data file, such as XML or JSON format, with each hole's information as a data node for easy subsequent reading and processing. This information serves as a key input for subsequent imaging path planning, robotic arm scheduling, and defect identification, ensuring that each target inspection area has complete, clear, and executable structural data support.

[0017] P20: Acquire the transmission characteristics of the reducer housing on the transmission line.

[0018] Optionally, in order to achieve effective detection of the reducer housing, it is necessary to obtain the transmission characteristics of the reducer housing on the conveyor line, that is, the dynamic and static characteristics exhibited by the reducer housing during the transmission process. Specifically, the transmission characteristics include but are not limited to the instantaneous position of the housing on the conveyor belt, movement speed, acceleration, attitude angle (including pitch angle, yaw angle and roll angle) and the operating stability parameters of the conveyor belt itself. In order to ensure that the subsequent imaging path of the multi-axis robotic arm can accurately match the actual position of the housing, it is necessary to first establish a standard coordinate reference for the conveying system, and track the position of the housing by arranging laser beam sensors and encoders on both sides of the conveyor line or using high-frame-rate industrial cameras, and complete the calibration of its reference starting point position before the housing enters the imaging area for the first time.

[0019] Next, the aforementioned sensors are used to continuously record the motion trajectory of the box along the conveyor line, and its linear velocity is calculated in real time in combination with the pulse count output by the encoder. The calculation of the linear velocity is usually based on the ratio between the unit length of the conveyor belt and the rotation angle of the encoder, and a high-speed sampling cycle is used to achieve millisecond-level accuracy in velocity curve analysis. In addition, if the conveyor line has a slope or an intermittent propulsion mechanism, a gyroscope or inertial measurement unit (IMU) can be used to obtain the posture change parameters of the box in different time periods to form a multi-dimensional dynamic posture feature vector to assist in path matching. This posture change information is extremely critical for determining the imaging blind spots of the box at certain special angles, especially when the axis of the hole structure is not perpendicular to the conveying direction.

[0020] All of this transmission feature information is combined with the housing number and current timestamp to form a structured transmission feature record unit, which is then fed into the path planning module as input to the motion control logic. This step enables precise quantification of the dynamic characteristics of the reducer housing under actual operating conditions, providing real-time and spatial consistency support for subsequent imaging path generation and synchronized robotic arm control, thereby improving the environmental adaptability and imaging coverage efficiency of the industrial vision system in assembly line environments.

[0021] P30: Combine the hole structure marking information and the transmission characteristics to perform 360° imaging motion demand analysis of the inner wall of the multi-axis robot arm and build a multi-axis motion control decision.

[0022] Furthermore, step P30 in the embodiment of the present application further includes: P31: Extract the transmission speed characteristics and box transmission posture characteristics from the transmission characteristics; P32: Based on the hole structure marking information, perform hole positioning based on the box transmission posture characteristics and the transmission speed characteristics, and determine the hole coordinate timing; P33: Perform 360° imaging simulation of the inner wall of the probe according to the hole structure marking information, and determine the inner wall imaging time; P34: Combine the inner wall imaging time, the transmission speed characteristics and the hole coordinate timing to perform motion control analysis of the multi-axis robotic arm and generate the multi-axis motion control decision.

[0023] It should be understood that the next step is to enter the imaging path construction stage based on the collaborative analysis of the detected target structure and the actual transmission status, that is, combining the hole structure marking information and transmission characteristics, to conduct a detailed analysis of the 360° imaging motion requirements of the inner wall of the multi-axis robot arm, and to build a multi-axis motion control decision.

[0024] Specifically, two key variables are first extracted from the acquired transmission characteristics: the transmission speed characteristic, which is the distance the reducer housing moves per unit time on the conveyor belt, used to characterize the overall linear displacement behavior; and the transmission posture characteristic of the housing, which is the sequence of spatial rotation angles (pitch, yaw, and roll) during the transmission process, used to reflect the spatial orientation dynamics of the hole structure during transmission. These characteristics directly influence the position determination of the imaging window and the configuration of the path timing.

[0025] After extracting the above two features, the established hole structure marking information is used as a reference, and the coordinates of the hole structure in three-dimensional space are dynamically mapped to the motion trajectory of the conveyor line through geometric mapping. Combined with the changes in the posture of the box, the effective accessibility of the hole structure during the transmission process is corrected in real time. Specifically, based on the initial position and transmission speed of the box, the moving distance and time relationship of the box on the conveyor line can be calculated. At the same time, combined with the posture changes of the box, the coordinates of the hole structure can be corrected to ensure that the robot arm can accurately locate the hole structure during inspection. At this point, the system not only completes the spatial positioning of the hole structure, but also maps it into a time-related hole coordinate sequence, that is, the position and direction state sequence corresponding to each hole structure at a specific time point. This information will be used as a time control reference for imaging scheduling to ensure that image acquisition is triggered at the optimal time.

[0026] The probe's 360° imaging of the inner wall is then simulated using the pore structure marker information. This simulation determines the time required to complete a complete 360° image of the inner wall. This time depends on the size and shape of the pore structure and the imaging speed of the probe. For example, larger pore structures may require longer imaging times to ensure sufficient image data. The probe's imaging resolution and imaging range also need to be considered to ensure full coverage of the inner wall of the pore structure. Simulation allows the probe's imaging path and imaging parameters to be optimized, thereby determining the shortest inner wall imaging time and improving inspection efficiency.

[0027] Finally, after determining the inner wall imaging time and the hole coordinate timing, the multi-axis robotic arm's motion control is analyzed in detail, combined with the conveyor speed characteristics. For example, the inner wall imaging time, conveyor speed characteristics, and hole coordinate timing are integrated into a motion control analysis model. This model, based on inverse kinematics and a time-position joint optimization algorithm, generates a motion sequence for the multi-axis robotic arm, ensuring that the robotic arm can reach the designated hole location, complete positioning, perform imaging, and retreat to a safe position within a defined time window throughout the conveyor's movement. Specifically, the robotic arm's motion path is planned based on the hole coordinate timing, ensuring that it reaches the hole structure at the appropriate time. The robotic arm's motion speed and posture are adjusted based on the inner wall imaging time. Furthermore, variations in conveyor speed are considered to ensure synchronization between the robotic arm's motion and the conveyor's operation. Finally, multi-axis motion control decisions are generated, including the robotic arm's motion path, velocity profile, and posture adjustment instructions. These decisions are transmitted to the robotic arm's control system, enabling efficient and accurate inspection of the reducer housing's inner wall.

[0028] This step enables precise control of the movement of the multi-axis robotic arm, ensuring accurate 360° imaging and detection of the inner wall of the porous structure during the movement of the reducer housing on the conveyor line, reflecting the intelligent scheduling capability of the highly complex space-time coupling system in industrial vision scenarios.

[0029] Furthermore, the motion control analysis of the multi-axis robotic arm is performed in combination with the inner wall imaging time, the transmission speed characteristics, and the hole coordinate timing to generate the multi-axis motion control decision. In this embodiment of the application, step P34 further includes: Step 1: Determine the predetermined imaging distance relative to the conveyor belt length based on the inner wall imaging time and the transmission speed characteristics; Step 2: Obtain the fixed position of the multi-axis robot arm and the degree of freedom range based on the fixed position; Step 3: With the degree of freedom range as a constraint and the predetermined imaging distance as a target, perform the imaging position analysis closest to the fixed position to generate a preset imaging position interval; Step 4: With the head end position in the preset imaging position interval as a target, perform the first stage motion control analysis of the multi-axis robot arm to generate a first stage control decision; Step 5: Within the preset imaging position interval, perform synchronous motion matching of the multi-axis robot arm with the transmission speed characteristics to generate a second stage control decision; Step 6: With the head end position as the predetermined acquisition activation position, combine the first stage control decision and the second stage control decision to generate the multi-axis motion control decision.

[0030] Optionally, the generation process of the multi-axis robot arm motion control decision can be further refined to ensure that the robot arm can efficiently and accurately complete the 360° imaging task of the inner wall during the movement of the reducer box on the conveyor belt.

[0031] First, based on the inner wall imaging time determined for each hole structure during the simulated imaging process and combined with the corresponding conveyor speed characteristics at that moment, dynamic spatial displacement conversion is performed to calculate the coverage length of the imaging window along the conveyor belt, recorded as the predetermined imaging distance. This predetermined imaging distance is used to quantify the spatial range of the hole position on the conveyor belt that can be stably imaged, providing a basic length parameter for the robot arm to determine the starting position of the window when entering and the ending position of the window when leaving.

[0032] Next, the device's position information and configuration parameters are retrieved to obtain the current multi-axis robotic arm's fixed installation reference point. The maximum reach and working envelope of each degree of freedom (such as the X, Y, and Z translation axes and the revolute joints) are then extracted. This range of freedom serves as the boundary constraint for the entire path planning process, ensuring that the generated motion path is realistically reachable based on the physical structure and kinematic model.

[0033] Then, using the aforementioned degrees of freedom as constraints and the predetermined imaging distance as the target projection length, an analysis of the imaging interval closest to the manipulator's fixed point along the conveyor belt is performed. Combined with a multi-objective dynamic programming approach, the boundaries of the areas preferentially covered in the achievable path are searched, ultimately generating a preset imaging position interval. This interval represents the spatial window within which the manipulator can achieve stable imaging without violating the degree of freedom constraints.

[0034] After entering the imaging window, the first phase of motion control analysis begins, targeting the starting position within the window. This phase primarily addresses how the robotic arm can accurately reach the predetermined imaging starting point from its current resting position or its retreat position after the previous task. A path planning algorithm, combined with the robotic arm's kinematic model, calculates the optimal motion path and velocity profile, generating the first-phase control decision. This decision includes the motion angles, velocities, and acceleration commands for each joint, ensuring the robotic arm reaches the starting imaging position smoothly and efficiently.

[0035] Once the robotic arm has precisely entered the preset imaging position range, the second phase of synchronous motion control analysis can be performed, combined with the real-time transmission speed of the box within this range. The key to this stage is to ensure a one-to-one mapping between the robotic arm's movement speed and the box's transmission speed. That is, while tracking the target hole position, the robotic arm remains relatively stationary or in a state of fine-tuning compensation. By monitoring the transmission speed changes in real time, the robotic arm's movement speed and posture are dynamically adjusted to enable it to accurately track the box's movement trajectory. During this process, the system performs differential prediction and adjustment based on the transmission speed change trend to avoid imaging errors caused by speed fluctuations, and ultimately generates the second-phase control decision.

[0036] Finally, the beginning of the preset imaging position interval is used as the predetermined acquisition activation position. At this position, the camera assembly and the ring light source are triggered to initiate synchronous acquisition. The first-stage control decision and the second-stage control decision are logically merged to uniformly generate the multi-axis motion control decision corresponding to the current hole position task. This decision set includes not only the spatial path, time trigger point, joint angle and speed parameters, but also the task completion flag, anomaly avoidance mechanism, and connection logic with subsequent hole position tasks. This forms a continuous, stable, and real-time motion control solution, ensuring that the system can still achieve high-precision and high-consistency industrial visual imaging and defect recognition operations under high-speed transmission conditions.

[0037] Furthermore, to generate the multi-axis motion control decision, the embodiment of the present application further includes step P34a, which further includes: P34-1a: Determine whether the conveyor line contains multiple robotic arms. If so, perform a one-to-one matching check between the robotic arm and the hole structure in combination with the hole structure marking information; P34-2a: If the matching check passes, perform motion control analysis of the hole structure and the corresponding robotic arm according to the one-to-one matching relationship, and generate multi-axis motion control decisions corresponding to each robotic arm.

[0038] Specifically, a task allocation and matching mechanism can be introduced in a multi-robot environment to improve overall efficiency and resource utilization during multi-target imaging. This step is primarily targeted at conveyor line scenarios with the ability for multiple robotic arms to work together. It focuses on resolving spatial binding, task scheduling, and path concurrency conflicts between the porous structure and multiple robotic arms, ensuring that the system maintains high-precision imaging control even in parallel operation.

[0039] First, it's necessary to determine whether the conveyor line contains multiple robotic arms. This determination can be made by reading pre-configured information about the conveyor line layout or by monitoring the distribution of equipment along the conveyor line in real time. If there is only one robotic arm on the conveyor line, the subsequent motion control analysis steps proceed directly. If there are multiple robotic arms on the conveyor line, a one-to-one matching verification between the robotic arms and the hole structure is performed, combining the hole structure marking information.

[0040] Specifically, the key features of each hole-like structure are extracted from the hole-like structure marking information obtained in step P10, including parameters such as the hole number, position coordinates, diameter, and depth. At the same time, the detection capability range of each robotic arm is determined, and the range of hole-like structures that it can detect is calculated based on its fixed position and degree of freedom range. By simulating the motion path and imaging range of the robotic arm, it is ensured that each robotic arm can reach and complete the detection task within its working range. Subsequently, the hole-like structure marking information is matched and verified with the detection capability range of each robotic arm. Based on the position coordinates of the hole-like structure and the fixed position and degree of freedom range of the robotic arm, it is determined whether each hole-like structure can be detected by a certain robotic arm. If each hole-like structure can be detected by a unique robotic arm, the matching verification passes; otherwise, the layout of the robotic arm or the distribution of detection tasks needs to be readjusted.

[0041] If the matching check passes, a motion control analysis is performed for each robot arm, comparing the hole structure to the corresponding robot arm, based on a one-to-one matching relationship. Based on the inner wall imaging time, conveyor speed characteristics, and hole coordinate timing determined in steps P31 to P33, the motion requirements of each robot arm are analyzed separately. Each robot arm's analysis process is performed independently to ensure that its motion path, speed profile, and posture adjustment instructions meet the inspection requirements of its corresponding hole structure. Finally, based on the motion control analysis results for each robot arm, its corresponding multi-axis motion control decision is generated. These decisions include the motion instructions for the robot arm to reach the imaging position from its initial position, as well as the motion adjustment instructions for synchronization with the conveyor belt during the imaging process.

[0042] All generated motion control decisions will eventually be uploaded to the local control module of each robotic arm through the central scheduling unit, and jointly scheduled with the global timestamp synchronization mechanism of the visual recognition system to ensure that each robotic arm can efficiently perform its own dedicated inner wall imaging task without cross-intervention.

[0043] Furthermore, the embodiment of the present application further includes step P34-3a, and step P34-3a further includes: P34-31a: If the matching check fails, the hole-like structures with the closest distance are combined in pairs according to the number of mismatches to generate a structural combination result; P34-32a: Based on the structural combination result, motion control analysis of the two hole-like structures is performed in batches by a robotic arm to generate multi-axis motion control decisions corresponding to each robotic arm.

[0044] Optionally, when the one-to-one matching verification between the robotic arm and the hole structure in steps P34-1a and P34-2a fails, that is, there are cases where some hole structures cannot be detected by a single robotic arm, a task reallocation mechanism based on structural combination can be designed to solve the task conflict problem where the number of hole positions is greater than the number of schedulable robotic arms, ensuring that all target hole structures can complete the image acquisition task within a reasonable error range, thereby improving the redundant adaptability of system resources and the flexibility of imaging scheduling.

[0045] Specifically, if the matching check fails, the system first performs a pairwise combination of the closest hole structures based on the number of mismatches. This process calculates the distance between each hole structure based on the position coordinates in the hole structure marking information, and then combines the two closest hole structures to generate a structure combination result. This combination method is designed to optimize the robot's motion path, reduce the time it takes to move between different hole structures, and improve inspection efficiency.

[0046] Then, for each resulting structural combination, a robotic arm performs motion control analysis on the two hole structures in each combination. For example, the optimal path and timing for the robotic arm to move to the second hole structure after completing 360° imaging of the inner wall of the first hole structure is determined. This analysis requires comprehensive consideration of factors such as the robotic arm's range of freedom, transfer speed characteristics, and inner wall imaging time. By simulating the robotic arm's motion, it is ensured that the robotic arm can efficiently switch between the two imaging sessions while maintaining imaging stability and accuracy.

[0047] Ultimately, based on these analysis results, multi-axis motion control decisions are generated for each robotic arm. These decisions include not only the robotic arm's motion control instructions for a single hole structure, but also the path planning and timing for switching between different hole structures. This way, even when a one-to-one match isn't possible, appropriate combination and optimization can ensure that each robotic arm can efficiently complete its assigned inspection task.

[0048] Furthermore, step P34-32a of the embodiment of the present application also includes: P34-321a: Calculate the distance between the two hole-like structures and determine the relay movement time of the robotic arm; P34-322a: Under the constraint of the relay movement time of the robotic arm, perform imaging position analysis on the two hole-like structures that are closest to the fixed position and whose two imaging position intervals do not overlap, and generate two imaging position intervals; P34-323a: Based on the two imaging position intervals, execute steps four and five in batches, and determine the relay route according to the relative position between the two hole-like structures, and generate multi-axis motion control decisions corresponding to each robotic arm.

[0049] In a possible embodiment of the present application, the specific operating procedures when the robotic arm needs to perform motion control analysis of two hole-like structures in batches can be further refined to ensure that the imaging task generates neither path conflicts nor acquisition blind spots during continuous execution, thereby improving the task coverage and execution stability of the robotic arm in multi-target tasks.

[0050] First, when the matching check fails and the pairwise combination of the closest hole structures is performed based on the number of mismatches, the structural combination result is generated. Next, for the two hole structures in each combination, the distance between the two hole structures is first calculated. This distance is calculated based on the position coordinates in the hole structure marking information and is obtained by the Euclidean distance formula or other applicable distance calculation methods. Based on the calculated distance, the time required for the robot arm to move from relaying one hole structure to another hole structure is further determined, that is, the robot arm relay movement time. The determination of this time needs to take into account the maximum moving speed, acceleration and possible path planning constraints of the robot arm to ensure that the robot arm can complete the movement from the relay point to the next imaging position within a reasonable time.

[0051] Under the constraint of the relay movement time of the robot arm, imaging position analysis is performed on the two hole-like structures respectively. The goal of this analysis is to find the imaging position closest to the fixed position of the robot arm, while ensuring that the two imaging position intervals do not overlap. Specifically, based on the fixed position of the robot arm, combined with the position coordinates of the hole-like structure and the range of freedom of the robot arm, the optimal imaging position interval of each hole-like structure is determined through kinematic analysis and path planning algorithm. The determination of these two imaging position intervals needs to take into account the range of motion and transmission speed characteristics of the robot arm to ensure that the robot arm can complete the imaging task within the predetermined time while avoiding mutual interference between the two imaging position intervals.

[0052] Taking these two imaging position intervals as a reference, step four and step five are performed in sequence. For each hole-like structure, step four is performed first, that is, taking the head end position in the preset imaging position interval as the target, a first-stage control decision is generated to ensure that the robot arm can quickly and accurately move from the initial position to the head end position of the imaging position interval. Subsequently, within the imaging position interval, step five is performed according to the conveying speed characteristics to generate a second-stage control decision to ensure that the robot arm can maintain a stable relative motion relationship with the reducer housing on the conveyor belt during the imaging process. In the process of executing these two steps, it is also necessary to determine the relay route of the robot arm based on the relative position between the two hole-like structures. The planning of this relay route needs to take into account the movement efficiency of the robot arm and the continuity of the imaging task to ensure that the robot arm can quickly and smoothly move to the imaging position of another hole structure after completing the imaging of one hole-like structure.

[0053] Finally, based on the above analysis and planning results, multi-axis motion control decisions are generated for each robotic arm. These decisions integrate the entire process of the robotic arm moving from its initial position to the first imaging position, performing the imaging task at the first imaging position, moving from the intermediate position to the second imaging position, and finally performing the imaging task at the second imaging position. This ensures that each robotic arm can efficiently and accurately complete its assigned inspection task, even when the robotic arm needs to perform imaging tasks for two hole-shaped structures in batches, further improving the adaptability and reliability of the industrial vision inspection system.

[0054] Furthermore, to determine whether the conveyor line includes multiple robotic arms, the embodiment of the present application further includes step P34b, which further includes: P34-1b: If not, identify multiple porous structure distribution information based on the porous structure marking information; P34-2b: Perform imaging position interval fitting of multiple porous structures using the multiple porous structure distribution information and the transmission characteristics to determine whether there are multiple imaging position intervals that do not overlap at all; P34-3b: If so, execute steps four and five in batches according to the multiple imaging position intervals.

[0055] Optionally, for application scenarios where only a single multi-axis robotic arm is equipped on the conveyor line, a time-sharing task scheduling mechanism based on structural distribution and imaging interval fitting can be further introduced to solve task execution problems under typical single-arm multi-target working conditions such as dense distribution of multiple porous structures to be detected and limited acquisition windows, ensuring the integrity of imaging tasks and non-interference of control paths under limited resource conditions.

[0056] First, based on the hole structure marking information obtained in step P10, the distribution of the multiple hole structures on the reducer housing is identified. This distribution information includes the location coordinates, size, shape, and relative positional relationships of each hole structure. By analyzing this marking information, the layout of the multiple hole structures across the housing can be determined, providing basic data for subsequent imaging position interval fitting.

[0057] Next, the imaging position intervals for the multiple hole structures are fitted, combining the distribution information of the multiple hole structures with conveying characteristics (such as conveying speed and box posture). Specifically, the optimal imaging position interval for each hole structure is calculated based on the position of each hole structure and the range of the robot's degrees of freedom. Fitting these imaging position intervals requires considering the following factors: the robot's range of motion and degrees of freedom to ensure that each imaging position interval is within the robot's reach; conveying speed characteristics to determine the relative position of each imaging position interval on the conveyor belt based on the conveying speed and imaging time; and the relative position between the hole structures to avoid overlap between imaging position intervals and ensure the robot's motion efficiency when switching between different hole structures. Through this fitting process, it is determined whether there are multiple completely non-overlapping imaging position intervals. If there are completely non-overlapping imaging position intervals, it means that the robot can complete the imaging task of each hole structure in sequence without path conflicts.

[0058] If there are multiple imaging position intervals that do not overlap at all, then steps four and five are performed in sequence according to these imaging position intervals. The specific operations are as follows: for each imaging position interval, with the starting position of the interval as the target, a first-stage control decision is generated, and the robot arm moves from the initial position to the starting position of each imaging position interval according to the decision, preparing to perform the imaging task; within each imaging position interval, the second-stage control decision is executed according to the transmission speed characteristics, and the robot arm maintains a stable relative motion relationship with the reducer box on the conveyor belt during the imaging process according to the decision, completing the 360° imaging task of the inner wall of the hole-shaped structure; after completing the task of an imaging position interval, the robot arm quickly moves to the starting position of the next imaging position interval according to the relay movement time constraint, and continues to perform the imaging task. This process requires optimizing the relay route of the robot arm to ensure motion efficiency and the continuity of the imaging task.

[0059] Through this mechanism, this embodiment breaks through the limitations of resource allocation under the single-arm operation model. While ensuring image quality and execution accuracy, it significantly improves the flexibility, adaptability and task throughput efficiency of the robotic arm scheduling system in high-rate industrial visual inspection scenarios.

[0060] Furthermore, the embodiment of the present application further includes step P34-4b: If there are no multiple imaging position intervals that do not completely overlap, perform deletion fitting iterations of the pore structure to determine the number of detectable pore structures; and perform sampling detection of different pore structures based on the number of detectable pore structures.

[0061] Specifically, when there is only one robotic arm on the conveyor line and after analysis it is found that there are no multiple imaging position intervals that are completely non-overlapping, a deletion fitting iteration and sampling detection mechanism can be further introduced to achieve task priority reconstruction and target selection strategy optimization under the conditions where the resources of a single multi-axis robotic arm are limited and the imaging window cannot be completely decoupled, thereby ensuring the detection coverage of key structures while maintaining the overall operation efficiency of the system.

[0062] Specifically, after completing the identification of the distribution information of multiple pore structures and the fitting of the imaging position intervals, if the judgment result shows that there are no multiple imaging position intervals that do not overlap completely, this indicates that the robot arm is unable to complete the detection task of all pore structures at the same time within the current layout and motion capability. At this point, it is necessary to perform an iterative process of deleting and fitting the pore structures. Specifically, some pore structures are gradually deleted from the multiple pore structures, and the imaging position intervals are refitted after each deletion to determine the maximum number of pore structures that can be detected under the current conditions, that is, to determine the number of detectable pore structures. This iterative process needs to comprehensively consider factors such as the importance of the pore structure, the position distribution, and the motion efficiency of the robot arm to ensure that the final number of detectable pore structures can not only meet the detection requirements, but also achieve efficient detection within the motion range of the robot arm.

[0063] Based on the determined number of detectable pore structures, sampling inspection of different pore structures is further performed. The purpose of sampling inspection is to evaluate the overall quality status by selecting representative pore structures for inspection when it is impossible to conduct comprehensive inspections on all pore structures. The sampling inspection strategy can be formulated based on factors such as the location distribution, size, and importance level of the pore structures. For example, pore structures distributed in key positions can be selected for inspection, or sampling can be performed from pore structures in different areas according to a certain proportion. Through sampling inspection, the quality status of the reducer housing can be understood as comprehensively as possible while ensuring the inspection efficiency.

[0064] This sampling inspection mechanism supports dynamic configuration. This means the system can increase sampling frequency based on historical defect statistics and the risk level of a specific area, or implement periodic sampling rotations based on changes in inspection strategies. This prevents the accumulation of potential defects from long periods of uninspected fixed holes. Furthermore, the sampling targets are timestamped and archived with structural marking information, ensuring traceability and visual auditability of their identification results.

[0065] Through the introduction of this step, when faced with resource bottlenecks and task conflicts, the system can use structural priorities as the core and combine dynamic adjustment strategies to make intelligent cutting and sampling detection decisions, thereby maximizing the detection effectiveness and stable operation capabilities of the system under limited resource conditions, and reflecting the robust scheduling and hierarchical detection capabilities for highly complex scenarios in actual industrial production environments.

[0066] P40: Controlling the multi-axis robotic arm using the multi-axis motion control decision to capture an image set of the inner wall of the hole-like structure. The multi-axis robotic arm is provided with a probe carrying a camera and a ring-shaped LED assembly at its distal end. When the multi-axis robotic arm reaches a predetermined acquisition activation position according to the multi-axis motion control decision, the ring-shaped LED assembly is activated, and image acquisition is performed using the camera by controlling the probe's rotation.

[0067] Optionally, based on the multi-axis motion control decision generated above, the multi-axis robotic arm is actually driven to perform the imaging task of the target hole structure, thereby completing the closed-loop control of the image acquisition stage.

[0068] In this embodiment, an integrated imaging probe is mounted at the end of a multi-axis robotic arm. This probe comprises a high-resolution industrial-grade camera and an integrated ring-shaped LED light source. The camera captures images of the inner wall of the hole-like structure, capable of high-speed frame rates and low-light imaging. The ring-shaped LED assembly provides uniform illumination within the hole, preventing image degradation caused by structural obstructions or localized shadows. Once the robotic arm reaches the acquisition activation position, the system immediately activates the LED assembly, emitting a stable, controllable ring-shaped light beam that illuminates the inner wall of the hole from all angles. A brightness feedback loop adjusts the light source's brightness in real time to accommodate varying hole depths, material reflectivity, or ambient light conditions.

[0069] Simultaneously, the control system activates the camera control module, controlling the probe's axial rotation, allowing the probe to rotate along the hole axis, enabling 360° continuous scanning image acquisition. During probe rotation, the camera continuously captures images of the hole's inner wall at a preset frame rate (e.g., above 60 fps), recording the spatial angle information of each frame by angle segment, thus binding the image to the pose. Immediately after acquisition, the image data is cached in the local image processing module and annotated with metadata such as the hole position number, timestamp, imaging sequence number, and imaging angle range, providing a foundation for subsequent defect identification and structural alignment.

[0070] To ensure image acquisition quality and task stability, the system continuously monitors the probe rotation speed, light source working status, image frame rate and acquisition focus position during image acquisition. If abnormalities such as image blur, underexposure or excessive mechanical disturbance occur, the pause mechanism is triggered and the current frame index and robotic arm status information are recorded for subsequent task compensation or reshooting.

[0071] Through the complete execution of this step, multi-angle, high-quality image acquisition of the inner wall of the target porous structure is efficiently completed, and highly synchronized control between the industrial vision system and the multi-axis control system is achieved, further laying the image foundation and data guarantee for the subsequent defect recognition algorithm.

[0072] P50: Identify inner wall structural defects based on the inner wall image set of the porous structure.

[0073] It should be understood that based on the image set of the inner wall of the hole structure collected by the multi-axis robotic arm in the previous step, the defect recognition module is started to complete the image analysis and defect judgment process of the internal surface state of each hole position, and finally output the recognition result of the structural defect.

[0074] First, the collected images of the inner wall of the porous structure are preprocessed. The goal of preprocessing is to remove noise, enhance contrast, and convert the images into a format suitable for subsequent analysis. This includes using image filtering algorithms (such as Gaussian filtering and median filtering) to remove random noise and improve image clarity. Histogram equalization or adaptive contrast enhancement algorithms are used to enhance image contrast and make defect features more visible. Finally, image segmentation techniques are used to separate the inner wall of the porous structure from the background, allowing subsequent defect detection algorithms to focus on the area of ​​interest.

[0075] After preprocessing is completed, the feature extraction stage begins. The purpose of feature extraction is to extract parameters that can characterize the inner wall structure from the preprocessed image. These parameters will be used for subsequent defect identification. Common feature extraction methods include using edge detection algorithms (such as Canny edge detection, Sobel operator, etc.) to extract edge information of the inner wall structure. Edge information can reflect the contour and shape changes of the inner wall and is an important basis for detecting defects such as cracks and scratches. By calculating the texture features of the image (such as gray-level co-occurrence matrix, local binary pattern, etc.), the texture information of the inner wall surface is extracted. Changes in texture features may indicate the presence of defects such as sand holes and holes in the inner wall. For some regularly shaped porous structures, their shape features (such as roundness, ellipticity, etc.) are extracted to detect whether there is shape deviation or deformation.

[0076] After extracting the features of the inner wall structure, defects are identified using a pre-trained defect recognition model. This model can be based on traditional image processing algorithms or deep learning algorithms (such as convolutional neural networks (CNNs)). If a deep learning algorithm is used, the model must be trained using a large number of labeled images of the inner walls of porous structures (including both defective and non-defective samples). During the training process, the model learns the differences between defect features and those of normal inner walls. The pre-processed images are input into the defect recognition model, which automatically identifies the presence of defects and classifies the defect type (e.g., cracks, pinholes, holes, etc.). The model also outputs the defect's location, typically expressed as coordinates. For identified defects, their severity is further assessed. This can be achieved by calculating parameters such as the defect's size, area, and depth. Defects can be categorized as minor, moderate, or severe based on their severity.

[0077] Finally, the recognition results are organized into a structured output data package containing fields such as the number of each inspection hole, imaging time, defect presence, defect type label, defect location coordinates, and defect severity level (which can be comprehensively evaluated by area, depth, or type). This data is then used by the back-end equipment management system for quality grading, alarm processing, or automatic rejection. Furthermore, the system supports real-time display of defect images and their identified boundary information on the industrial human-machine interface, facilitating manual review or supporting decision-making. All recognition results are linked and stored in a quality traceability database, supporting subsequent data analysis and production optimization.

[0078] Through the implementation of this step, the intelligent conversion from high-resolution image data to semantic information of structural defects is completed, and non-contact automatic defect detection of the inner wall structure is realized. This method not only improves the detection efficiency, but also reduces the interference of human factors and improves the accuracy and consistency of the detection results.

[0079] In summary, the embodiments of the present application have at least the following technical effects: This application collects 3D design information of the reducer housing, identifies the porous structure, generates marking information, and simultaneously captures the transmission characteristics of the conveyor line. Combining this information, it performs a 360° imaging requirement analysis of the inner wall of a multi-axis robotic arm, builds a multi-axis motion control decision, and controls the robotic arm to perform the imaging task. Defect identification is performed using the collected image set of the inner wall of the porous structure, enabling high-precision, full-coverage defect detection of the inner wall of the porous structure in a dynamic transmission environment.

[0080] The technical effect of improving the accuracy, efficiency and dynamic environment adaptability of reducer case inner wall defect detection through the coordinated control of industrial vision and multi-axis robotic arms has been achieved.

[0081] The second embodiment is based on the same inventive concept as the reducer housing defect recognition method based on industrial vision in the above embodiment. Figure 2 As shown, the present application provides a reducer housing defect recognition system based on industrial vision. The system and method embodiments in the present application are based on the same inventive concept. The system includes: The hole structure recognition module 11 is used to collect the three-dimensional design information of the reducer housing, perform hole structure recognition, and obtain hole structure marking information.

[0082] The transmission characteristic acquisition module 12 is used to obtain the transmission characteristics of the reducer housing on the transmission line.

[0083] The imaging motion demand analysis module 13 is used to perform 360° imaging motion demand analysis of the inner wall of the multi-axis robot arm in combination with the hole structure marking information and the transmission characteristics, and to construct a multi-axis motion control decision.

[0084] The image acquisition module 14 is used to execute the control of the multi-axis robotic arm based on the multi-axis motion control decision and acquire an image set of the inner wall of the hole-shaped structure.

[0085] The inner wall structure defect recognition module 15 is used to recognize inner wall structure defects based on the inner wall image set of the hole-like structure.

[0086] Furthermore, the imaging motion demand analysis module 13 is further configured to perform the following steps: Extract the transmission speed characteristics and the box transmission posture characteristics from the transmission characteristics; take the hole structure marking information as a reference, perform hole positioning based on the box transmission posture characteristics and the transmission speed characteristics, and determine the hole coordinate timing; perform 360° imaging simulation of the inner wall of the probe according to the hole structure marking information, and determine the inner wall imaging time; combine the inner wall imaging time, the transmission speed characteristics and the hole coordinate timing to perform motion control analysis of the multi-axis robot arm and generate the multi-axis motion control decision.

[0087] Furthermore, the imaging motion demand analysis module 13 is further configured to perform the following steps: Step 1: Determine the predetermined imaging distance relative to the conveyor belt length based on the inner wall imaging time and the transmission speed characteristics; Step 2: Obtain the fixed position of the multi-axis robot arm and the degree of freedom range based on the fixed position; Step 3: With the degree of freedom range as a constraint and the predetermined imaging distance as a target, perform the imaging position analysis closest to the fixed position to generate a preset imaging position interval; Step 4: With the head end position in the preset imaging position interval as a target, perform the first stage motion control analysis of the multi-axis robot arm to generate a first stage control decision; Step 5: Within the preset imaging position interval, perform synchronous motion matching of the multi-axis robot arm with the transmission speed characteristics to generate a second stage control decision; Step 6: With the head end position as the predetermined acquisition activation position, combine the first stage control decision and the second stage control decision to generate the multi-axis motion control decision.

[0088] Furthermore, the imaging motion demand analysis module 13 is further configured to perform the following steps: Determine whether the conveyor line includes multiple robotic arms. If so, perform a one-to-one matching check between the robotic arm and the hole structure in combination with the hole structure marking information; if the matching check passes, perform motion control analysis of the hole structure and the corresponding robotic arm according to the one-to-one matching relationship, and generate multi-axis motion control decisions corresponding to each robotic arm.

[0089] Furthermore, the imaging motion demand analysis module 13 is further configured to perform the following steps: If the matching check fails, the hole structures with the closest distance are combined in pairs according to the number of mismatches to generate a structural combination result; based on the structural combination result, a robotic arm is used to perform motion control analysis of the two hole structures in batches to generate multi-axis motion control decisions corresponding to each robotic arm.

[0090] Furthermore, the imaging motion demand analysis module 13 is further configured to perform the following steps: Calculate the distance between the two hole-like structures and determine the relay movement time of the robotic arm; under the constraint of the relay movement time of the robotic arm, perform imaging position analysis on the two hole-like structures that are closest to the fixed position and whose two imaging position intervals do not overlap, and generate two imaging position intervals; based on the two imaging position intervals, execute steps four and five in sequence, and determine the relay route according to the relative position between the two hole-like structures, and generate multi-axis motion control decisions corresponding to each robotic arm.

[0091] Furthermore, the imaging motion demand analysis module 13 is further configured to perform the following steps: If not, identify multiple hole structure distribution information based on the hole structure marking information; perform imaging position interval fitting of multiple hole structures using the multiple hole structure distribution information and the transmission characteristics to determine whether there are multiple imaging position intervals that do not overlap at all; if so, execute steps four and five in batches according to the multiple imaging position intervals.

[0092] Furthermore, the imaging motion demand analysis module 13 is further configured to perform the following steps: If there are no multiple imaging position intervals that do not completely overlap, perform deletion fitting iterations of the pore structure to determine the number of detectable pore structures; and perform sampling detection of different pore structures based on the number of detectable pore structures.

[0093] Furthermore, the image acquisition module 14 is further configured to perform the following steps: The end of the multi-axis robotic arm is equipped with a probe carrying a camera and a ring-shaped LED assembly. When the multi-axis robotic arm reaches the predetermined acquisition activation position according to the multi-axis motion control decision, the ring-shaped LED assembly is activated, and the camera is used to acquire images by controlling the rotation of the probe.

[0094] It should be noted that the order in which the embodiments of the present application are presented is for illustrative purposes only and does not necessarily represent the superiority or inferiority of the embodiments. Furthermore, the foregoing descriptions of specific embodiments of this specification are provided. Furthermore, the processes depicted in the accompanying drawings do not necessarily require the specific order or sequential sequence shown to achieve the desired results. In certain embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0095] The above description is only a preferred embodiment of the present application and is not intended to limit the present application. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present application should be included in the scope of protection of the present application.

[0096] This specification and drawings are merely illustrative of the present application and are intended to cover any and all modifications, variations, combinations, or equivalents within the scope of this application. Obviously, those skilled in the art may make various modifications and variations to this application without departing from the scope of this application. Thus, this application is intended to include such modifications and variations as fall within the scope of this application and its equivalents.

Claims

1. A method for identifying defects in a reducer housing based on industrial vision, characterized in that: include: By collecting the three-dimensional design information of the reducer housing, the hole structure is identified and the hole structure marking information is obtained; Acquiring the transmission characteristics of the reducer housing on the transmission line; Combining the hole structure marking information and the transmission characteristics, performing a 360° imaging motion demand analysis of the inner wall of the multi-axis robot arm, and constructing a multi-axis motion control decision; executing control of the multi-axis robotic arm using the multi-axis motion control decision to collect an image set of the inner wall of the hole-like structure; Inner wall structural defects are identified based on the inner wall image set of the porous structure.

2. The method for identifying defects in a reducer housing based on industrial vision according to claim 1, characterized in that: The end of the multi-axis robotic arm is equipped with a probe carrying a camera and a ring-shaped LED assembly. When the multi-axis robotic arm reaches the predetermined acquisition activation position according to the multi-axis motion control decision, the ring-shaped LED assembly is activated, and the camera is used to acquire images by controlling the rotation of the probe.

3. The method for identifying defects in a reducer housing based on industrial vision according to claim 2, characterized in that: Combining the hole structure marking information and the transmission characteristics, a 360° imaging motion demand analysis of the inner wall of the multi-axis robot is performed to construct a multi-axis motion control decision, including: Extracting a conveying speed feature and a box conveying posture feature from the conveying features; Taking the hole structure marking information as a reference, hole positioning is performed based on the box conveying posture characteristics and the conveying speed characteristics to determine the hole coordinate time sequence; Performing a 360° imaging simulation of the inner wall of the probe according to the hole structure marking information to determine the inner wall imaging time; The motion control analysis of the multi-axis robotic arm is performed in combination with the inner wall imaging time, the transmission speed characteristics, and the hole-shaped coordinate timing to generate the multi-axis motion control decision.

4. The method for identifying defects in a reducer housing based on industrial vision according to claim 3, characterized in that: The motion control analysis of the multi-axis robotic arm is performed in combination with the inner wall imaging time, the transmission speed characteristics, and the hole coordinate timing to generate the multi-axis motion control decision, including: Step 1: determining a predetermined imaging distance relative to the conveyor belt length according to the inner wall imaging time and the conveying speed characteristics; Step 2: Obtaining a fixed position of the multi-axis robotic arm and a range of degrees of freedom based on the fixed position; Step 3: using the range of degrees of freedom as a constraint and the predetermined imaging distance as a target, performing an imaging position analysis closest to the fixed position to generate a preset imaging position interval; Step 4: Taking the head end position in the preset imaging position interval as the target, performing the first stage motion control analysis of the multi-axis robotic arm and generating a first stage control decision; Step 5: performing synchronous motion matching of the multi-axis robotic arm within the preset imaging position interval using the transmission speed characteristics to generate a second-stage control decision; Step 6: Using the head end position as the predetermined acquisition activation position, combining the first stage control decision and the second stage control decision to generate the multi-axis motion control decision.

5. The method for identifying defects in a reducer housing based on industrial vision according to claim 4, characterized in that: Generating the multi-axis motion control decision further includes: Determine whether the conveyor line includes multiple robotic arms, and if so, perform a one-to-one matching check between the robotic arms and the hole-shaped structures in combination with the hole-shaped structure marking information; If the matching verification passes, the motion control analysis of the hole structure and the corresponding robotic arm is performed according to the one-to-one matching relationship, and the multi-axis motion control decisions corresponding to each robotic arm are generated.

6. The method for identifying defects in a reducer housing based on industrial vision according to claim 5, characterized in that: Performing a one-to-one matching check between the robot arm and the hole structure in combination with the hole structure marking information further includes: If the matching check fails, the nearest hole structures are combined in pairs according to the number of mismatches to generate a structural combination result; According to the structural combination result, motion control analysis of the two hole-shaped structures is performed in batches by a robotic arm to generate multi-axis motion control decisions corresponding to each robotic arm.

7. The method for identifying defects in a reducer housing based on industrial vision according to claim 6, characterized in that: Based on the structural combination results, a robotic arm performs motion control analysis of the two hole-shaped structures in sequence to generate multi-axis motion control decisions corresponding to each robotic arm, including: Calculate the distance between the two hole-like structures and determine the relay movement time of the robotic arm; Under the relay movement time constraint of the robot arm, performing imaging position analysis on the two hole-shaped structures that are closest to the fixed position and whose two imaging position intervals do not overlap, respectively, to generate two imaging position intervals; Based on the two imaging position intervals, step four and step five are performed in sequence, and a relay route is determined according to the relative position between the two hole-shaped structures, and multi-axis motion control decisions corresponding to each robotic arm are generated.

8. The method for identifying defects in a reducer housing based on industrial vision according to claim 1, characterized in that: Determining whether the conveyor line includes multiple robotic arms further includes: If not, identifying a plurality of pore structure distribution information based on the pore structure label information; performing imaging position interval fitting of the plurality of hole structures using the plurality of hole structure distribution information and the transmission characteristics, and determining whether there are a plurality of imaging position intervals that are completely non-overlapping; If so, step 4 and step 5 are performed in batches according to multiple imaging position intervals.

9. The method for identifying defects in a reducer housing based on industrial vision according to claim 8, characterized in that: If there are no multiple imaging position intervals that do not completely overlap, perform deletion fitting iterations of the pore structure to determine the number of detectable pore structures; and perform sampling detection of different pore structures based on the number of detectable pore structures.

10. The reducer housing defect recognition system based on industrial vision is characterized by: The system comprises: A hole structure recognition module is used to identify the hole structure by collecting three-dimensional design information of the reducer housing to obtain hole structure marking information; A transmission characteristic acquisition module, the transmission characteristic acquisition module is used to obtain the transmission characteristics of the reducer housing on the transmission line; An imaging motion demand analysis module, the imaging motion demand analysis module being used to perform a 360° imaging motion demand analysis of the inner wall of the multi-axis manipulator in combination with the hole structure marking information and the transmission characteristics, and to construct a multi-axis motion control decision; An image acquisition module, configured to execute control of the multi-axis robotic arm using the multi-axis motion control decision and acquire an image set of an inner wall of the hole-shaped structure; An inner wall structure defect recognition module is used to identify inner wall structure defects based on the inner wall image set of the hole-like structure.

Images