Waste automobile engine disassembling system and method based on artificial intelligence
Through an artificial intelligence-based system, visual recognition, data processing and robotic arm execution modules are used to solve the problem of low manual operation efficiency in disassembly of used car engines, and an efficient, safe and automated disassembly process is realized, which improves disassembly Quality and safety.
Patent Information
- Application Number
- CN202410952253.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-16
- Publication Date
- 2024-11-12
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
In the prior art, the disassembly of used automobile engines relies on manual operation, with high labor intensity, low efficiency and unstable quality, and lack of efficient automation solutions.
Adopting a system based on artificial intelligence, including visual recognition module, data processing module, artificial intelligence module, robotic arm execution module and sensor module, obtain engine image information through the camera, image processing and analysis recognition components, generate disassembly paths, and robotic arm execution Disassembly operation, sensors monitor and adjust the process in real time.
It significantly improves the efficiency and quality of engine disassembly, reduces the uncertainty and safety risks of manual operation, enhances the flexibility and scalability of the system, and realizes an intelligent, efficient and safe disassembly process.
Smart Images

Figure CN118941758A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field related to the dismantling of scrap automobile engines, and in particular to a system and method for dismantling scrap automobile engines based on artificial intelligence. Background Art
[0002] With the development of the automobile industry, the number of scrapped cars has increased year by year. Among them, the engine is the most important part of the car, and its dismantling and reuse are of great significance. At present, most engine dismantling work still relies on manual operation, which is not only labor-intensive, but also inefficient and unstable dismantling quality. Therefore, the development of a scrap automobile engine dismantling system and method based on artificial intelligence technology will help improve the dismantling efficiency and quality, and has broad application prospects. Summary of the invention
[0003] The purpose of the present invention is to provide a system and method for dismantling scrap automobile engines based on artificial intelligence to solve the problems raised in the above background technology.
[0004] To achieve the above-mentioned purpose, the present invention provides the following technical solutions: one, including a visual recognition module, a data processing module, an artificial intelligence module, a mechanical wall execution module and a sensor module, wherein the visual recognition module includes several cameras for obtaining image information of the appearance and internal structure of the engine;
[0005] The data processing module comprises an image processing and analysis unit for processing and analyzing images from the visual recognition module;
[0006] The artificial intelligence module: includes a pre-trained model for identification, classification and generation of disassembly solutions;
[0007] The robot execution module comprises a plurality of robot arms, and is used to execute the disassembly operation according to the disassembly plan generated by the data processing module;
[0008] The sensor module is used to monitor the force and position changes during the disassembly process in real time through various sensors and provide feedback information.
[0009] Preferably, the visual recognition module includes a camera unit, an image acquisition unit, a light source system unit, and an image preprocessing unit. The camera unit adopts multiple different types of cameras according to needs, specifically RGB cameras, infrared cameras and depth cameras. The number of cameras depends on specific application requirements and needs to meet multi-angle and multi-position coverage to ensure comprehensive acquisition of engine image information; the image acquisition unit is equipped with image acquisition software for controlling the working state of the camera, including exposure, focus, and viewing angle adjustment, to achieve fast and stable acquisition and transmission of image data; the light source system unit adopts LED lights or laser light sources to enhance the contrast of engine components and avoid shadows and reflections affecting image acquisition; the image preprocessing unit adopts mean filtering to remove noise in the image of the image acquisition unit and improve image quality, and then uses the Sobel edge detection algorithm to highlight the contours of various engine components to provide a clear image basis for subsequent processing, and then eliminates the lens distortion and color deviation of the camera through geometric correction and color correction methods to ensure that the image truly reflects the actual situation of the engine.
[0010] Preferably, the data processing module includes an image processing unit, which stitches the images from the visual recognition module into a complete engine 3D model, uses a SIFT image registration algorithm to achieve accurate stitching of multiple images, and improves the image quality by adjusting contrast, brightness, and sharpening image enhancement technology.
[0011] Preferably, the image processing unit uses the SIFT image registration algorithm to achieve the precise stitching of multiple images in the following specific steps:
[0012] Step A: performing Gaussian blurring on the image from the visual recognition module multiple times to generate images of multiple scales, and performing differentiation on the images of different scales to obtain a Gaussian difference image sequence;
[0013] Step B: in the Gaussian difference image in step A, local extreme points, i.e., image key points, are found, where the key points are invariant at different scales and different viewing angles; each extreme point found is accurately located, including removing low-contrast points and unstable edge response points, and using Taylor series expansion to perform detailed processing to ensure the stability and accuracy of the key points;
[0014] Step C: Calculate the gradient direction histogram for each key point, and select the peak direction as the main direction of the key point;
[0015] Step D: In the local neighborhood of the key point, a feature descriptor is generated by calculating the gradient histogram. The descriptor captures the intensity gradient information of the local area of the image, and the descriptor is normalized to enhance the illumination invariance.
[0016] Step E: Find the matching feature point pairs between different images by calculating the Euclidean distance of the feature descriptors, i.e., feature point matching;
[0017] Step F: Use the RANSAC algorithm to further screen the matching point pairs, calculate the homography matrix between the images, use the homography matrix to perform perspective transformation, align the images, and finally complete the image stitching through Laplacian pyramid fusion.
[0018] Preferably, the artificial intelligence module includes an image analysis unit, a model training and reasoning unit and a disassembly solution generation unit. The image analysis unit uses a convolutional neural network algorithm to extract features from the image spliced by the image processing unit to identify key components and connection points of the engine. The U-Net image segmentation algorithm is used to separate the various components of the engine from the image, and the components are labeled to clarify the position, size and shape of each component, and then the relative position and connection relationship of each component are analyzed to construct the structure diagram of the engine. The model training and reasoning unit selects the ResNet deep learning model to accurately identify and locate the engine components. In the initial stage, a large amount of waste engine image data is used for model pre-training to improve the recognition accuracy and generalization ability of the ResNet deep learning model. The disassembly solution generation unit combines the structure recognition results and uses the Dijkstra algorithm to generate the optimal disassembly path to ensure that the disassembly process is efficient and safe, and then generates specific robot arm operation instructions based on the disassembly path and specific information of the components to guide the robot arm to perform the disassembly task.
[0019] Preferably, the disassembly solution generation unit combines the structure recognition result and uses the Dijkstra algorithm to generate the optimal disassembly path. The specific contents are as follows:
[0020] Suppose a simple engine model, which includes 5 components. The connection relationship between the components is represented by a weighted undirected graph, and the weight represents the time or distance required to disassemble a component. Suppose the nodes of the weighted undirected graph represent engine components, the edges represent the connection relationship between the components, and the weight represents the difficulty or resource consumption of disassembling each pair of connected components. The node is V = {A, B, C, D, E}, the edge is E, and the weight is as follows: (A, B, 2), (A, C, 1), (B, D, 3), (C, D, 1), (C, E, 4), (D, E, 1), where the number represents the weight. Then the optimal disassembly path is generated from node A to find the optimal disassembly path to node E. The specific steps are as follows:
[0021] a. For each node in the graph, initialize the distance vector d and set up a priority queue to store the nodes;
[0022] The distance d(A) = 0, that is, the distance from the starting point to itself is 0; the initial distances of all other nodes are infinite (d(B) = ∞, d(C) = ∞, d(D) = ∞, d(E) = ∞); initialize the predecessor node P of the current node, and P(v) = \text{None} for all nodes;
[0023] b. Continuously select the node with the shortest current distance for processing and update the distance of its neighboring nodes, as follows:
[0024] Iteration 1, current node: A; neighbor nodes and weights: B(2), C(1); update distance: predecessor node update: (d(B)=\min(∞,0+2)=2), d(C)=\min(∞,0+1)=1; predecessor node update: (P(B)=A), (P(C)=A); current shortest path tree: A→C=1, A→B=2;
[0025] Iteration 2, current node: C, because 1 is the shortest; neighbor nodes and weights: D(1), E(4); update distance: d(D) = \min(∞,1+1=2) = 2, d(E) = \min(∞,1+4=5); predecessor node update: (P(D) = C), P(E) = C; current shortest path tree: A→C = 1, A→B→D = 2, A→C→E = 5;
[0026] Iteration 3, current node: B, because 2 is less than 5; neighbor nodes and weights: D(3); update distance: d(D) = \min(2,2+3) = 2;
[0027] Iteration 4, current node: D; neighbor nodes and weights: E(1); update distance: d(E) = min(5,2+1) = 3; predecessor node update: P(E) = D; current shortest tree path: A→C = 1, A→C→D = 2, A→C→D→E = 3;
[0028] Iteration 5, current node: E; neighbor nodes and weights: empty;
[0029] c. Starting from the end point E, reconstruct the path through the predecessor node set P: D→E, D→C, C→A; the final path is A→C→D→E, with a total weight of 3, that is, the shortest disassembly path from A to E is calculated.
[0030] Preferably, the sensor module includes a force sensor, a position sensor, an acceleration sensor, a torque sensor, a proximity sensor, a vision sensor and a laser rangefinder, wherein the force sensor is used to monitor the force acting on the engine components, including the torque and pressure required for disassembly; the position sensor is used to monitor the spatial position of the robotic arm and the engine components to ensure accurate movement and positioning during the disassembly process; the acceleration sensor is used to monitor the acceleration of the robotic arm and the engine components to help detect any abnormal vibration or impact to avoid damage; the torque sensor is used to monitor the torque applied during the disassembly process to ensure that the bolts and fasteners are disassembled according to the specified torque value; the proximity sensor is used to detect whether an object is approaching, prevent collision and ensure the correct movement of the components; the vision sensor is used to obtain visual information of the disassembly scene, including component identification, positioning and posture estimation, and the laser rangefinder is used to measure the distance to the object, provide high-precision position information, and help the robotic arm perform precise operations in three-dimensional space.
[0031] Preferably, a method for dismantling a scrap automobile engine based on artificial intelligence comprises the following steps:
[0032] Step S1, information acquisition: the visual recognition module acquires image information of the waste engine through a camera and transmits it to the data processing module;
[0033] Step S2, image processing: the data processing module processes the image information to identify the various components of the engine and their connection relationships;
[0034] Step S3, solution generation: the artificial intelligence module combines the pre-trained model to analyze the identified engine structure and generate the optimal disassembly path and operation plan;
[0035] Step S4, executing disassembly: the robot arm execution module uses the robot arm to perform specific disassembly operations according to the generated disassembly plan, and the sensor module monitors the operation process in real time and provides feedback;
[0036] Step S5, self-adjustment: Through the information fed back by the sensor module, the artificial intelligence module adjusts the disassembly plan in real time to ensure the efficiency and accuracy of the disassembly process.
[0037] Compared with the prior art, the beneficial effects of the present invention are as follows: through the pre-training model and real-time image processing, the system of the present invention can quickly identify the structure and components of the engine, thereby generating an efficient disassembly path and operation steps, and the robotic arm executes according to the optimal disassembly plan, avoiding delays and uncertainties in manual operation, and greatly improving the disassembly speed; the real-time data provided by the sensor module enables the artificial intelligence module to dynamically adjust the disassembly process, correct potential deviations and errors, and ensure the consistency and stability of the operation; force sensors and position sensors realize real-time monitoring of the disassembly process, can timely detect and avoid dangerous operations, and improve the safety of the overall operation; the system is composed of different functional modules, and each module can be upgraded or replaced as needed, thereby enhancing the flexibility and scalability of the system. In summary, the present invention integrates advanced image recognition, data processing, artificial intelligence and robotic arm control technologies to establish an intelligent, efficient, safe and sustainable disassembly platform, thereby better serving the modern scrap automobile engine disassembly industry. BRIEF DESCRIPTION OF THE DRAWINGS
[0038] Figure 1 It is a schematic diagram of the system structure of the present invention;
[0039] Figure 2 It is a schematic diagram of the method flow of the present invention. DETAILED DESCRIPTION
[0040] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0041] See also Figure 1-2 ,The present invention provides a technical solution: a ,technical solution comprising a visual recognition module, a data ,processing module, an artificial intelligence module, a mechanical wall ,execution module and a sensor module, wherein the visual recognition ,module comprises a number of cameras for acquiring image information of the appearance and ,internal structure of the engine;
[0042] Data processing module: including image processing and analysis unit, used to process and analyze images from visual recognition module;
[0043] Artificial Intelligence Module: Contains pre-trained models for identification, classification, and generation of disassembly solutions;
[0044] Robotic arm execution module: includes multiple robotic arms, used to perform disassembly operations according to the disassembly plan generated by the data processing module;
[0045] Sensor module: used to monitor the force and position changes during the disassembly process in real time through various sensors and provide feedback information.
[0046] Furthermore, the visual recognition module includes a camera unit, an image acquisition unit, a light source system unit, and an image preprocessing unit. The camera unit adopts multiple different types of cameras according to needs, specifically RGB cameras, infrared cameras and depth cameras. The number of cameras depends on the specific application needs and needs to meet multi-angle and multi-position coverage to ensure comprehensive acquisition of engine image information; the image acquisition unit is equipped with image acquisition software for controlling the working status of the camera, including exposure, focus, and viewing angle adjustment, to achieve fast and stable acquisition and transmission of image data; the light source system unit adopts LED lights or laser light sources to enhance the contrast of engine components and avoid shadows and reflections affecting image acquisition; the image preprocessing unit adopts mean filtering to remove noise in the image of the image acquisition unit and improve image quality, and then uses the Sobel edge detection algorithm to highlight the contours of various engine components to provide a clear image basis for subsequent processing, and then eliminates the lens distortion and color deviation of the camera through geometric correction and color correction methods to ensure that the image truly reflects the actual situation of the engine.
[0047] Furthermore, the data processing module includes an image processing unit, which stitches the images from the visual recognition module into a complete engine 3D model, uses the SIFT image registration algorithm to achieve accurate stitching of multiple images, and improves the image quality by adjusting the contrast, brightness, and sharpening image enhancement technology.
[0048] Furthermore, the image processing unit uses the SIFT image registration algorithm to achieve accurate stitching of multiple images. The specific steps are as follows:
[0049] Step A: performing Gaussian blurring on the image from the visual recognition module multiple times to generate images of multiple scales, and performing differentiation on the images of different scales to obtain a Gaussian difference image sequence;
[0050] Step B: in the Gaussian difference image in step A, local extreme points, i.e., image key points, are found, where the key points are invariant at different scales and different viewing angles; each extreme point found is accurately located, including removing low-contrast points and unstable edge response points, and using Taylor series expansion to perform detailed processing to ensure the stability and accuracy of the key points;
[0051] Step C: Calculate the gradient direction histogram for each key point, and select the peak direction as the main direction of the key point;
[0052] Step D: In the local neighborhood of the key point, a feature descriptor is generated by calculating the gradient histogram. The descriptor captures the intensity gradient information of the local area of the image, and the descriptor is normalized to enhance the illumination invariance.
[0053] Step E: Find the matching feature point pairs between different images by calculating the Euclidean distance of the feature descriptors, i.e., feature point matching;
[0054] Step F: Use the RANSAC algorithm to further screen the matching point pairs, calculate the homography matrix between the images, use the homography matrix to perform perspective transformation, align the images, and finally complete the image stitching through Laplacian pyramid fusion.
[0055] Furthermore, the artificial intelligence module includes an image analysis unit, a model training and reasoning unit, and a disassembly plan generation unit. The image analysis unit uses a convolutional neural network algorithm to extract features from the image spliced by the image processing unit and identify the key components and connection points of the engine. The U-Net image segmentation algorithm is used to separate the various components of the engine from the image and mark the components to clarify the position, size and shape of each component, and then the relative position and connection relationship of each component are analyzed to construct the structure diagram of the engine. The model training and reasoning unit selects the ResNet deep learning model to accurately identify and locate the engine components. In the initial stage, a large amount of waste engine image data is used for model pre-training to improve the recognition accuracy and generalization ability of the ResNet deep learning model. The disassembly plan generation unit combines the structure recognition results and uses the Dijkstra algorithm to generate the optimal disassembly path to ensure the efficiency and safety of the disassembly process. Then, according to the specific information of the disassembly path and components, specific robot arm operation instructions are generated to guide the robot arm to perform the disassembly task.
[0056] Furthermore, the disassembly solution generation unit combines the structural recognition results and uses the Dijkstra algorithm to generate the optimal disassembly path. The specific contents are as follows:
[0057] Suppose a simple engine model, which includes 5 components. The connection relationship between the components is represented by a weighted undirected graph, and the weight represents the time or distance required to disassemble a component. Suppose the nodes of the weighted undirected graph represent engine components, the edges represent the connection relationship between the components, and the weight represents the difficulty or resource consumption of disassembling each pair of connected components. The node is V = {A, B, C, D, E}, the edge is E, and the weight is as follows: (A, B, 2), (A, C, 1), (B, D, 3), (C, D, 1), (C, E, 4), (D, E, 1), where the number represents the weight. Then the optimal disassembly path is generated from node A to find the optimal disassembly path to node E. The specific steps are as follows:
[0058] a. For each node in the graph, initialize the distance vector d and set up a priority queue to store the nodes;
[0059] The distance d(A) = 0, that is, the distance from the starting point to itself is 0; the initial distances of all other nodes are infinite (d(B) = ∞, d(C) = ∞, d(D) = ∞, d(E) = ∞); initialize the predecessor node P of the current node, and P(v) = \text{None} for all nodes;
[0060] b. Continuously select the node with the shortest current distance for processing and update the distance of its neighboring nodes, as follows:
[0061] Iteration 1, current node: A; neighbor nodes and weights: B(2), C(1); update distance: predecessor node update: (d(B)=\min(∞,0+2)=2), d(C)=\min(∞,0+1)=1; predecessor node update: (P(B)=A), (P(C)=A); current shortest path tree: A→C=1, A→B=2;
[0062] Iteration 2, current node: C, because 1 is the shortest; neighbor nodes and weights: D(1), E(4); update distance: d(D) = \min(∞,1+1=2) = 2, d(E) = \min(∞,1+4=5); predecessor node update: (P(D) = C), P(E) = C; current shortest path tree: A→C = 1, A→B→D = 2, A→C→E = 5;
[0063] Iteration 3, current node: B, because 2 is less than 5; neighbor nodes and weights: D(3); update distance: d(D) = \min(2,2+3) = 2;
[0064] Iteration 4, current node: D; neighbor nodes and weights: E(1); update distance: d(E) = min(5,2+1) = 3; predecessor node update: P(E) = D; current shortest tree path: A→C = 1, A→C→D = 2, A→C→D→E = 3;
[0065] Iteration 5, current node: E; neighbor nodes and weights: empty;
[0066] c. Starting from the end point E, reconstruct the path through the predecessor node set P: D→E, D→C, C→A; the final path is A→C→D→E, with a total weight of 3, that is, the shortest disassembly path from A to E is calculated
[0067] Furthermore, the sensor module includes a force sensor, a position sensor, an acceleration sensor, a torque sensor, a proximity sensor, a vision sensor and a laser rangefinder. The force sensor is used to monitor the force acting on the engine components, including the torque and pressure required for disassembly; the position sensor is used to monitor the spatial position of the robotic arm and the engine components to ensure accurate movement and positioning during the disassembly process; the acceleration sensor is used to monitor the acceleration of the robotic arm and the engine components to help detect any abnormal vibration or impact to avoid damage; the torque sensor is used to monitor the torque applied during the disassembly process to ensure that the bolts and fasteners are disassembled according to the specified torque value; the proximity sensor is used to detect whether an object is approaching, prevent collision and ensure the correct movement of the components; the vision sensor is used to obtain visual information of the disassembly scene, including component identification, positioning and posture estimation, and the laser rangefinder is used to measure the distance to the object, provide high-precision position information, and help the robotic arm perform precise operations in three-dimensional space.
[0068] Furthermore, a method for dismantling a scrap automobile engine based on artificial intelligence comprises the following steps:
[0069] Step S1, information acquisition: the visual recognition module acquires image information of the waste engine through a camera and transmits it to the data processing module;
[0070] Step S2, image processing: the data processing module processes the image information to identify the various components of the engine and their connection relationships;
[0071] Step S3, solution generation: the artificial intelligence module combines the pre-trained model to analyze the identified engine structure and generate the optimal disassembly path and operation plan;
[0072] Step S4, executing disassembly: the robot arm execution module uses the robot arm to perform specific disassembly operations according to the generated disassembly plan, and the sensor module monitors the operation process in real time and provides feedback;
[0073] Step S5, self-adjustment: Through the information fed back by the sensor module, the artificial intelligence module adjusts the disassembly plan in real time to ensure the efficiency and accuracy of the disassembly process.
[0074] The present invention realizes an efficient, accurate and automated engine disassembly process by integrating a visual recognition module, a data processing module, an artificial intelligence module, a robotic arm execution module and a sensor module. The specific workflow includes: information acquisition, image processing, solution generation, disassembly execution and self-adjustment.
[0075] The visual recognition module uses several cameras to obtain image information of the engine, which is processed and analyzed by the data processing module to identify the connection relationship between the various components. The artificial intelligence module generates the optimal disassembly path and operation plan based on the pre-trained model, and controls the robotic arm execution module to perform the disassembly operation according to the plan. At the same time, the sensor module monitors the force and position changes in real time and provides feedback information, enabling the artificial intelligence module to make real-time adjustments during the disassembly process.
[0076] By adopting this system and method, the dismantling efficiency and quality are significantly improved, the operational safety is enhanced, the operation and maintenance costs are reduced, and a positive impact is produced on environmental protection. This invention is not only innovative in technology, but also shows extremely high promotion value in practical applications.
[0077] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.
Claims
1. A waste automobile engine disassembly system based on artificial intelligence, comprising a visual recognition module, a data processing module, an artificial intelligence module, a mechanical wall execution module and a sensor module, characterized in that: The visual recognition module includes several cameras for acquiring image information of the appearance and internal structure of the engine; The data processing module comprises an image processing and analysis unit for processing and analyzing images from the visual recognition module; The artificial intelligence module: includes a pre-trained model for identification, classification and generation of disassembly solutions; The robot execution module comprises a plurality of robot arms, and is used to execute the disassembly operation according to the disassembly plan generated by the data processing module; The sensor module is used to monitor the force and position changes during the disassembly process in real time through various sensors and provide feedback information.
2. The artificial intelligence-based waste automobile engine disassembly system according to claim 1 is characterized in that: The visual recognition module includes a camera unit, an image acquisition unit, a light source system unit, and an image preprocessing unit. The camera unit adopts multiple different types of cameras according to needs, specifically RGB cameras, infrared cameras and depth cameras. The number of cameras depends on specific application needs and needs to meet multi-angle and multi-position coverage to ensure comprehensive acquisition of engine image information; the image acquisition unit is equipped with image acquisition software for controlling the working state of the camera, including exposure, focus, and viewing angle adjustment, to achieve fast and stable acquisition and transmission of image data; the light source system unit adopts LED lights or laser light sources to enhance the contrast of engine components and avoid shadows and reflections affecting image acquisition; the image preprocessing unit adopts mean filtering to remove noise in the image of the image acquisition unit and improve image quality, and then uses the Sobel edge detection algorithm to highlight the contours of various engine components to provide a clear image basis for subsequent processing, and then eliminates the lens distortion and color deviation of the camera through geometric correction and color correction methods to ensure that the image truly reflects the actual situation of the engine.
3. The artificial intelligence-based waste automobile engine disassembly system according to claim 1 is characterized in that: The data processing module includes an image processing unit, which stitches the images from the visual recognition module into a complete engine 3D model, uses the SIFT image registration algorithm to achieve accurate stitching of multiple images, and improves the image quality by adjusting the contrast, brightness, and sharpening image enhancement technology.
4. The artificial intelligence-based waste automobile engine disassembly system according to claim 1 is characterized in that: The image processing unit uses the SIFT image registration algorithm to achieve accurate stitching of multiple images in the following specific steps: Step A: performing Gaussian blurring on the image from the visual recognition module multiple times to generate images of multiple scales, and performing differentiation on the images of different scales to obtain a Gaussian difference image sequence; Step B: in the Gaussian difference image in step A, local extreme points, i.e., image key points, are found, where the key points are invariant at different scales and different viewing angles; each extreme point found is accurately located, including removing low-contrast points and unstable edge response points, and using Taylor series expansion to perform detailed processing to ensure the stability and accuracy of the key points; Step C: Calculate the gradient direction histogram for each key point, and select the peak direction as the main direction of the key point; Step D: In the local neighborhood of the key point, a feature descriptor is generated by calculating the gradient histogram. The descriptor captures the intensity gradient information of the local area of the image, and the descriptor is normalized to enhance the illumination invariance. Step E: Find the matching feature point pairs between different images by calculating the Euclidean distance of the feature descriptors, i.e., feature point matching; Step F: Use the RANSAC algorithm to further screen the matching point pairs, calculate the homography matrix between the images, use the homography matrix to perform perspective transformation, align the images, and finally complete the image stitching through Laplacian pyramid fusion.
5. A waste automobile engine disassembly system based on artificial intelligence according to any one of claims 1 or 3, characterized in that: The artificial intelligence module includes an image analysis unit, a model training and reasoning unit, and a disassembly solution generation unit. The image analysis unit uses a convolutional neural network algorithm to extract features from the image spliced by the image processing unit and identify key components and connection points of the engine. The U-Net image segmentation algorithm is used to separate the various components of the engine from the image and mark the components to clarify the position, size and shape of each component, and then the relative position and connection relationship of each component are analyzed to construct the structure diagram of the engine. The model training and reasoning unit selects the ResNet deep learning model to accurately identify and locate the engine components. In the initial stage, a large amount of waste engine image data is used for model pre-training to improve the recognition accuracy and generalization ability of the ResNet deep learning model. The disassembly solution generation unit combines the structure recognition results and uses the Dijkstra algorithm to generate the optimal disassembly path to ensure the efficiency and safety of the disassembly process. Then, according to the disassembly path and the specific information of the components, specific robot arm operation instructions are generated to guide the robot arm to perform the disassembly task.
6. The artificial intelligence-based waste automobile engine disassembly system according to claim 5 is characterized by: The disassembly solution generation unit combines the structure recognition results and uses the Dijkstra algorithm to generate the optimal disassembly path. The specific contents are as follows: Suppose a simple engine model, which includes 5 components. The connection relationship between the components is represented by a weighted undirected graph, and the weight represents the time or distance required to disassemble a component. Suppose the nodes of the weighted undirected graph represent engine components, the edges represent the connection relationship between the components, and the weight represents the difficulty or resource consumption of disassembling each pair of connected components. The node is V = {A, B, C, D, E}, the edge is E, and the weight is as follows: (A, B, 2), (A, C, 1), (B, D, 3), (C, D, 1), (C, E, 4), (D, E, 1), where the number represents the weight. Then the optimal disassembly path is generated from node A to find the optimal disassembly path to node E. The specific steps are as follows: a. For each node in the graph, initialize the distance vector d and set up a priority queue to store the nodes; The distance d(A) = 0, that is, the distance from the starting point to itself is 0; the initial distances of all other nodes are infinite (d(B) = ∞, d(C) = ∞, d(D) = ∞, d(E) = ∞); initialize the predecessor node P of the current node, and P(v) = \text{None} for all nodes; b. Continuously select the node with the shortest current distance for processing and update the distance of its neighboring nodes, as follows: Iteration 1, current node: A; neighbor nodes and weights: B(2), C(1); update distance: predecessor node update: (d(B)=\min(∞,0+2)=2), d(C)=\min(∞,0+1)=1; Predecessor node update: (P(B)=A), (P(C)=A); Current shortest path tree: A→C=1, A→B=2; Iteration 2, current node: C, because 1 is the shortest; neighbor nodes and weights: D(1), E(4); update distance: d(D) = \min(∞,1+1=2) = 2, d(E) = \min(∞,1+4=5); predecessor node update: (P(D) = C), P(E) = C; current shortest path tree: A→C = 1, A→B→D = 2, A→C→E = 5; Iteration 3, current node: B, because 2 is less than 5; neighbor nodes and weights: D(3); update distance: d(D) = \min(2,2+3) = 2; Iteration 4, current node: D; neighbor nodes and weights: E(1); update distance: d(E) = min(5,2+1) = 3; predecessor node update: P(E) = D; current shortest tree path: A→C = 1, A→C→D = 2, A→C→D→E = 3; Iteration 5, current node: E; neighbor nodes and weights: empty; c. Starting from the end point E, reconstruct the path through the predecessor node set P: D→E, D→C, C→A; the final path is A→C→D→E, with a total weight of 3, that is, the shortest disassembly path from A to E is calculated.
7. The artificial intelligence-based waste automobile engine disassembly system according to claim 1 is characterized by: The sensor module includes a force sensor, a position sensor, an acceleration sensor, a torque sensor, a proximity sensor, a vision sensor and a laser rangefinder. The force sensor is used to monitor the force acting on the engine components, including the torque and pressure required for disassembly; the position sensor is used to monitor the spatial position of the robotic arm and the engine components to ensure accurate movement and positioning during the disassembly process; the acceleration sensor is used to monitor the acceleration of the robotic arm and the engine components to help detect any abnormal vibration or impact to avoid damage; the torque sensor is used to monitor the torque applied during the disassembly process to ensure that the bolts and fasteners are disassembled according to the specified torque value; the proximity sensor is used to detect whether an object is approaching, prevent collision and ensure the correct movement of the components; the vision sensor is used to obtain visual information of the disassembly scene, including component identification, positioning and posture estimation, and the laser rangefinder is used to measure the distance to the object, provide high-precision position information, and help the robotic arm perform precise operations in three-dimensional space.
8. A method for dismantling scrap automobile engines based on artificial intelligence according to any one of claims 1 to 7, characterized in that: The following steps are involved: Step S1, information acquisition: the visual recognition module acquires image information of the waste engine through a camera and transmits it to the data processing module; Step S2, image processing: the data processing module processes the image information to identify the various components of the engine and their connection relationships; Step S3, solution generation: the artificial intelligence module combines the pre-trained model to analyze the identified engine structure and generate the optimal disassembly path and operation plan; Step S4, executing disassembly: the robot arm execution module uses the robot arm to perform specific disassembly operations according to the generated disassembly plan, and the sensor module monitors the operation process in real time and provides feedback; Step S5, self-adjustment: Through the information fed back by the sensor module, the artificial intelligence module adjusts the disassembly plan in real time to ensure the efficiency and accuracy of the disassembly process.
Citation Information
Cited By
Intelligent control method and system for waste automobile disassembling robot
CN121821413A
