Pipeline warehouse dynamic space optimization method and system based on AI vision

Abstract

The application discloses a pipeline warehouse dynamic space optimization method and system based on AI vision, which firstly acquires pipe image, infrared thermal image, temperature and humidity, goods location, AGV state and ERP construction demand data, and then generates a pipe entity record through AI vision recognition; then, the pipe state, the goods location state and the environmental parameters are spatio-temporally registered to construct a dynamic digital twin model; then, candidate stacking actions are generated based on the goods location size, the load capacity, the pipe compression resistance, the environmental adaptation and the warehouse-out priority, and a target stacking scheme is determined according to the space utilization, the compression safety, the carrying efficiency and the environmental risk; then, the AGV carries out carrying and updates in a closed loop through visual verification, and the scheme can improve the pipe recognition accuracy, the warehouse space utilization rate and the warehouse-in and warehouse-out collaborative efficiency.

Description

Technical Field

[0001] This invention relates to the fields of intelligent warehouse management, machine vision, digital twins, and automated warehouse logistics, and particularly to a method and system for dynamic space optimization of pipeline storage based on AI vision. It is applicable to the identification of pipe materials entering the warehouse, spatial planning, dynamic stacking, automatic handling, inventory linkage, and outbound scheduling in the process of municipal engineering, petrochemical, building construction, energy transmission, and industrial pipeline network construction. Background Technology

[0002] Pipes, as essential materials commonly used in engineering construction and industrial production, are characterized by a wide range of specifications, long lengths, unique cross-sectional shapes, significant material differences, and high requirements for stacking stability. Pipes of different diameters, wall thicknesses, lengths, and materials have varying requirements for storage location dimensions, load-bearing capacity, support methods, stacking layers, and environmental conditions. Relying solely on manual experience for classification, registration, and stacking can easily lead to problems such as inaccurate pipe specification records, delayed updates on storage location occupancy status, mixing of pipes of similar specifications, and low efficiency in retrieval.

[0003] Current pipeline storage management methods typically rely on manual inventory checks, static location coding, and standard inventory systems, with space planning largely dependent on pre-defined areas or manual scheduling. These methods struggle to accurately monitor the real-time three-dimensional occupancy of pipes and fail to incorporate factors such as surface defects, internal thermal anomalies, temperature and humidity conditions, location load-bearing capacity, and handling routes into stacking decisions. For long metal pipes, plastic pipes, or pipes with localized defects, improper stacking can lead to pressure deformation, crack propagation, accelerated corrosion, or increased localized damage, thereby reducing storage quality and posing safety hazards.

[0004] Furthermore, traditional warehousing systems often lack real-time closed-loop communication with construction plans, procurement plans, and AGV handling systems. The quantities and locations recorded in the inventory system may not match the actual storage status. Even after the construction team submits material requests, the warehouse still requires manual searching, manual determination of the outbound order, and manual adjustment of the stacking method for remaining pipes. This process is not only inefficient but also fails to generate dynamic optimization solutions based on inbound time, pipe health status, outbound distance, and the urgency of construction needs. Therefore, it is necessary to propose a diversified dynamic spatial optimization method and system for pipeline warehousing that integrates AI visual recognition, digital twin modeling, AGV automated handling, and ERP linkage. Summary of the Invention

[0005] In view of this, the purpose of this invention is to propose a dynamic space optimization method and system for pipeline storage based on AI vision, which aims to solve the problems of inaccurate pipe identification, low space utilization, poor stacking security, insufficient linkage between inbound and outbound operations, and difficulty in real-time updating of storage status in the existing technology.

[0006] To achieve the above-mentioned technical objectives, the technical solution adopted by this invention is as follows: A dynamic space optimization method for pipeline storage based on AI vision, comprising: S1. Obtain multi-source status data of the pipeline storage area, including pipe images, pipe thermal images, temperature and humidity data, cargo location data, AGV status data, and construction requirement data; S2. Perform AI visual recognition on the pipe images and pipe thermal imaging data to obtain a pipe entity record corresponding to each pipe. The pipe entity record includes pipe identification, specification information, surface defect information, thermal anomaly information, warehousing time and location information. S3. Perform spatiotemporal registration of the pipe entity records, temperature and humidity data, cargo location data and AGV status data to construct a dynamic digital twin model corresponding to the actual storage area; S4. Based on the pipe material entity record, dynamic digital twin model and construction requirement data, generate a storage constraint set and a candidate stacking action set for each pipe material to be put into storage or rearranged. S5. For each candidate stacking action, calculate the space utilization index, pipe pressure safety index, AGV handling efficiency index and environmental risk index, and then construct a comprehensive evaluation value. S6. The current state of the dynamic digital twin model is used as the input of the strategy model. Combined with the comprehensive evaluation value, the strategy model determines the target stacking scheme from the candidate stacking actions that satisfy the storage constraint set. The target stacking scheme includes the target storage location, the target number of stacking layers, the target placement angle, the target support method, and the target AGV handling task. S7. The target AGV handling task is sent to the AGV handling equipment, and the pipe handling, placement and re-stacking process is visually verified by a multi-view camera device. When the visual verification result is inconsistent with the target stacking scheme, or when the temperature and humidity data, pipe pressure status, or AGV operation status are abnormal, the dynamic digital twin model is updated and steps S4 to S6 are re-executed.

[0007] As one possible implementation, this solution further includes: S8. After the pipes are put into storage, rearranged or put out of storage, the pipe inventory change information, storage location change information, defect change information and inbound / outbound execution information are synchronized to the ERP system, and the AI ​​visual recognition model and strategy model are iteratively updated using visual recognition error, stacking execution deviation, pipe damage record and AGV handling time.

[0008] As a preferred implementation method, in step S1 of this solution, the multi-source status data includes pipe images collected by multi-view camera equipment, pipe thermal image data collected by infrared thermal imaging equipment, temperature and humidity data collected by environmental sensing equipment, cargo location data provided by warehouse management equipment, AGV status data provided by handling equipment, and construction demand data provided by ERP system.

[0009] As a preferred implementation method, preferably, in step S2 of this solution, the pipe entity record includes pipe identification, pipe diameter, length, wall thickness, material, weight, surface defect information, thermal anomaly information, warehousing time, and current location.

[0010] As a preferred implementation method, step S2 of this solution, which involves performing AI visual recognition on the pipe image and pipe thermal image data, includes the following steps: Target detection and instance segmentation are performed on pipe images from multiple perspectives to determine the contour region of each pipe. Three-dimensional reconstruction of the pipe outline region from multiple perspectives is performed to obtain pipe point cloud data. Pipe diameter, length, and wall thickness are extracted from the pipe point cloud data. Defect identification is performed on the surface images of the pipe to obtain the defect type, location, and severity corresponding to corrosion, cracks, and deformation; Temperature field anomaly analysis was performed on the thermal imaging data of the pipe to obtain thermal anomaly information used to characterize the risk of internal damage to the pipe. Pipe diameter, length, wall thickness, material, weight, defect type, defect location, defect severity, and thermal anomaly information are bound to the same pipe identifier to form the pipe entity record.

[0011] As a preferred implementation method, in step S3 of this solution, the dynamic digital twin model includes the three-dimensional occupancy status of the pipe, the idle status of the storage location, the distribution status of environmental parameters, and the operating status of the AGV.

[0012] As a preferred implementation method, the method for constructing a dynamic digital twin model in step S3 of this solution preferably includes: Establish a unified three-dimensional coordinate system for the warehouse area; Transform the coordinate systems of the camera equipment, cargo location, AGV navigation, and environmental sensing equipment to the unified three-dimensional coordinate system. A 3D occupancy raster map is generated based on pipe point cloud data and storage location data; Generate a spatial distribution map of environmental parameters based on temperature and humidity data; Generate a map of accessible AGV paths based on AGV status data; By overlaying the three-dimensional occupancy grid map, the spatial distribution map of environmental parameters, and the AGV passable path map, a dynamic digital twin model that can be updated as pipes are put into storage, put out of storage, rearranged, and as the environment changes.

[0013] As a preferred implementation method, preferably, in step S4 of this solution, the storage constraint set includes storage location size constraints, storage location load-bearing constraints, pipe pressure resistance constraints, environmental adaptability constraints, defect avoidance constraints, and outbound priority constraints, and the candidate stacking action set includes candidate storage locations, number of stacking layers, placement angle of each layer, support method, and AGV handling path.

[0014] As a preferred implementation method, in step S4 of this solution, the compressive strength constraint of the pipe is determined in the following way: The self-weight of the pipe is determined based on its material, diameter, wall thickness, and length. The stacking pressure that the target pipe will withstand is determined based on the target number of stacking layers and the weight of the upper pipe. The allowable pressure value of the pipe is determined based on the allowable stress of the pipe material, the cross-sectional dimensions of the pipe, the severity of surface defects, and the environmental degradation coefficient. When the stacking pressure is less than or equal to the allowable pressure value, the candidate stacking action satisfies the pipe pressure resistance constraint; when the stacking pressure is greater than the allowable pressure value, the corresponding candidate stacking action is eliminated.

[0015] As a preferred implementation method, in step S5 of this solution, the comprehensive evaluation value is calculated according to the following formula:

[0016] in, For candidate stacking actions The overall evaluation value; As a space utilization indicator; This represents the pressure value of the pipe under candidate stacking operations; This refers to the allowable pressure value for the corresponding pipe material; The estimated handling time required for the AGV to perform the candidate stacking action; Based on the transport time; As an environmental risk indicator; , , , These are the weighting coefficients corresponding to space utilization, pressure safety, handling efficiency, and environmental risk, respectively.

[0017] As a preferred implementation method, in step S6 of this scheme, the strategy model is trained using a near-end strategy optimization algorithm. Its training state space includes pipe parameters, storage location parameters, environmental parameters, AGV state parameters, and construction requirement parameters. Its action space includes storage location selection action, stacking layer selection action, placement angle selection action, support method selection action, and AGV path selection action. Its reward function consists of space utilization reward, pressure safety reward, handling efficiency reward, and environmental risk penalty.

[0018] As a preferred implementation method, in step S7 of this solution, the target AGV handling task includes the pipe picking position, pipe placing position, handling path, handling sequence, and obstacle avoidance strategy; when the AGV handling equipment performs the handling task, it uses a combination of magnetic strip navigation, QR code positioning, and laser navigation for positioning. When any one of the navigation information fails, the handling task continues to be performed based on the remaining navigation information, and the current position, remaining power, load status, and task progress are fed back to the dynamic digital twin model.

[0019] As a preferred implementation method, in step S8 of this solution, when matching outbound materials based on the construction demand data provided by the ERP system, the outbound priority is determined according to the matching degree of pipe specifications, the time of entry into the warehouse, the health status of the pipe materials, the distance from the outbound port, and the construction demand time. Based on the outbound priority, an outbound pipe material list, an AGV outbound path, and a re-stacking scheme for the remaining pipe materials are generated.

[0020] Based on the above, this solution also proposes an AI vision-based dynamic space optimization system for pipeline storage, which includes: The multi-source data acquisition module is used to acquire pipe images, pipe thermal imaging data, temperature and humidity data, cargo location data, AGV status data, and construction requirement data from the ERP system. The AI ​​visual recognition module is used to identify pipe images and pipe thermal imaging data, and generate pipe entity records including pipe diameter, length, wall thickness, material, weight, surface defect information and thermal anomaly information; The digital twin construction module is used to build a dynamic digital twin model based on pipe entity records, temperature and humidity data, cargo location data, and AGV status data. The constraint generation module is used to generate and store a set of constraints and a set of candidate stacked actions based on pipe entity records, dynamic digital twin models, and construction requirement data. The dynamic space optimization module is used to calculate the comprehensive evaluation value of candidate stacking actions and determine the target stacking scheme through the strategy model; The AGV scheduling and execution module is used to generate AGV handling tasks based on the target stacking scheme and control the AGV handling equipment to complete the pipe material warehousing, rearrangement and warehousing. The ERP linkage module is used to synchronize information on changes in pipe inventory, storage location, defects, and inbound / outbound execution to the ERP system, and to receive construction demand data sent by the ERP system. The model iteration module is used to update the AI ​​visual recognition model and strategy model based on visual recognition errors, stacking execution deviations, pipe damage records, and AGV handling time.

[0021] As a preferred implementation method, the multi-source data acquisition module of this solution preferably includes multi-view camera devices set in different locations in the storage area, infrared thermal imaging devices set in association with the multi-view camera devices, temperature and humidity sensing devices arranged according to the storage function zones, and AGV communication interfaces for collecting AGV operating status. The ERP linkage module interacts with the ERP system via API interface and generates early warning information when inventory falls below the safety threshold, construction needs change, or pipe material status is abnormal.

[0022] By adopting the above technical solution, the present invention has the following beneficial effects compared with the prior art: The present invention automatically identifies pipes using multi-view camera equipment, infrared thermal imaging equipment, and AI visual recognition models. It can uniformly bind pipe diameter, length, wall thickness, material, weight, surface defects, thermal anomalies, and warehousing time to the pipe entity record, reducing specification misjudgments and missed inspections caused by manual registration. Simultaneously, by spatiotemporally registering the pipe entity record, location status, temperature and humidity data, and AGV status data, a dynamic digital twin model is constructed. This enables the warehousing system to reflect the three-dimensional occupancy status of pipes, environmental risk status, and the operating status of handling equipment in real time, providing an accurate data foundation for subsequent space optimization and inbound / outbound scheduling.

[0023] This invention's solution simultaneously considers space utilization, pipe pressure safety, environmental adaptability, defect avoidance, and AGV handling efficiency in stacking decisions. It avoids problems such as pipe deformation, defect propagation, and space waste caused by solely relying on available storage space or manual experience to determine stacking locations. By using a strategy model to determine the target stacking scheme from candidate stacking actions and employing visual verification and digital twin models for closed-loop updates during execution, it achieves dynamic space optimization during pipe warehousing, rearrangement, and warehousing processes, improving warehouse space utilization and pipe storage security.

[0024] This invention also integrates construction needs, inventory changes, storage location changes, and outbound execution information into a single closed-loop management process through an ERP linkage module. This enables the warehousing system to automatically match pipe materials, determine outbound priorities, plan AGV routes, and update the remaining inventory layout based on construction needs. When inventory is insufficient, construction needs change, or pipe material status is abnormal, the system can promptly send early warning information to the ERP system, thereby improving the matching degree between inventory and construction progress and reducing operating costs caused by pipe material backlog, shortages, and repeated handling. Attached Figure Description

[0025] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0026] Figure 1 This is a simplified implementation flowchart of the AI ​​vision-based dynamic space optimization method for pipeline storage in this solution. Figure 2 This is a schematic diagram of the unit module connections of the AI ​​vision-based pipeline storage dynamic space optimization system in this solution. Detailed Implementation

[0027] The present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be particularly noted that the following embodiments are for illustrative purposes only and do not limit the scope of the invention. Similarly, the following embodiments are only some, not all, embodiments of the present invention, and all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0028] Combination Figure 1 As shown in the figure, this embodiment proposes a dynamic space optimization method for pipeline storage based on AI vision, which includes: S1. Obtain multi-source status data of the pipeline storage area, including pipe images, pipe thermal images, temperature and humidity data, cargo location data, AGV status data, and construction requirement data; S2. Perform AI visual recognition on the pipe images and pipe thermal imaging data to obtain a pipe entity record corresponding to each pipe. The pipe entity record includes pipe identification, specification information, surface defect information, thermal anomaly information, warehousing time and location information. S3. Perform spatiotemporal registration of the pipe entity records, temperature and humidity data, cargo location data and AGV status data to construct a dynamic digital twin model corresponding to the actual storage area; S4. Based on the pipe material entity record, dynamic digital twin model and construction requirement data, generate a storage constraint set and a candidate stacking action set for each pipe material to be put into storage or rearranged. S5. For each candidate stacking action, calculate the space utilization index, pipe pressure safety index, AGV handling efficiency index and environmental risk index, and then construct a comprehensive evaluation value. S6. The current state of the dynamic digital twin model is used as the input of the strategy model. Combined with the comprehensive evaluation value, the strategy model determines the target stacking scheme from the candidate stacking actions that satisfy the storage constraint set. The target stacking scheme includes the target storage location, the target number of stacking layers, the target placement angle, the target support method, and the target AGV handling task. S7. The target AGV handling task is sent to the AGV handling equipment, and the pipe handling, placement and re-stacking process is visually verified by a multi-view camera device. When the visual verification result is inconsistent with the target stacking scheme, or when the temperature and humidity data, pipe pressure status, or AGV operation status are abnormal, the dynamic digital twin model is updated and steps S4 to S6 are re-executed. S8. After the pipes are put into storage, rearranged or put out of storage, the pipe inventory change information, storage location change information, defect change information and inbound / outbound execution information are synchronized to the ERP system, and the AI ​​visual recognition model and strategy model are iteratively updated using visual recognition error, stacking execution deviation, pipe damage record and AGV handling time.

[0029] Traditional warehousing typically only records pipe specifications and quantities, failing to simultaneously obtain information on the actual location of the pipes, their stacking status, surface damage, ambient temperature and humidity, AGV availability, and construction requirements. This results in a lack of unified data foundation for subsequent space optimization.

[0030] Based on this, as a preferred implementation method, in step S1 of this solution, the multi-source status data includes pipe images collected by multi-view camera equipment, pipe thermal image data collected by infrared thermal imaging equipment, temperature and humidity data collected by environmental sensing equipment, storage location data provided by warehouse management equipment, AGV status data provided by handling equipment, and construction demand data provided by the ERP system. This solution solves the problems of incomplete pipe storage status perception, fragmented data sources, and asynchrony between inbound information and the actual on-site status in existing pipe storage solutions.

[0031] As an example, step S1 of this solution may include the following details: At any moment The following multi-source state data were collected. :

[0032] in, For multi-view camera equipment at all times A collection of visible light images of the pipe material; For infrared thermal imaging equipment at all times The collection of thermal image data of the pipes; A collection of temperature and humidity data collected by environmental sensing devices; This is a collection of warehouse storage location data; This is a collection of AGV status data. A collection of construction requirement data provided to the ERP system.

[0033] In this plan, if the storage area is set up One camera device (multi-view camera device, infrared thermal imaging device), An environmental sensing device, A multi-source status data system for AGVs. The sets of visible light images of pipes, the sets of thermal images of pipes, and the sets of AGV status data can be defined as follows:

[0034]

[0035]

[0036]

[0037] Among them, the definition For the first Visible light images captured by a camera device, i.e. For the first Visible light images captured by a camera device; For the first Thermal images acquired by an infrared thermal imaging device; For the first Temperature collected by an environmental sensor; For the first Humidity collected by an environmental sensor; For the first Spatial coordinates of an environmental sensing device; , For the first AGVs in the warehouse layout , Position coordinates; For the first The remaining power of the AGV; For the first The current load of the AGV; For the first The task status of the AGV.

[0038] ERP construction requirements data It can be represented as:

[0039] in, For the first Construction requirements; For the required pipe diameter; For the required length; For demand-side wall thickness; For the required materials; For the required quantity; For the required time; Prioritize the requirements. This is the requirement number, and its maximum value is equal to... .

[0040] Because visual data, environmental data, AGV data, and ERP data have different sampling frequencies, they need to be unified into the same time window. Let the unified decision-making cycle of the system be... Then at time The effective data window is: For data collection time of... The data is only valid if the following conditions are met: When that time, it is included in the state data set of the current period. .

[0041] To address the need for space optimization, which requires simultaneous knowledge of pipe type, location, damage status, environmental suitability, AGV capability, and construction requirements, this solution merges vision, infrared, environmental, storage location, AGV, and ERP data into a unified status set in step S1. The next steps will be from the set Visible light image collection of pipes and pipe thermal imaging data set Identify the physical pipe structure; utilize temperature and humidity data sets. Warehouse location data set AGV status data set Build a digital twin model; then base it on ERP construction requirement data. A stacking scheme for generating digital twin models.

[0042] To address the problems of low efficiency, large errors in dimensional recording, and difficulty in timely detection of surface defects and internal thermal anomalies in manual pipe specification identification, this solution uses a camera array, deep learning algorithms, and an infrared thermal imager to identify pipe diameter, length, wall thickness, corrosion, cracks, and internal damage.

[0043] Specifically, as a preferred implementation method, in step S2 of this solution, the pipe entity record includes pipe identification, pipe diameter, length, wall thickness, material, weight, surface defect information, thermal anomaly information, warehousing time, and current location.

[0044] In step S2 of this solution, the AI ​​visual recognition of the pipe image and pipe thermal image data includes the following steps: Target detection and instance segmentation are performed on pipe images from multiple perspectives to determine the contour region of each pipe. Three-dimensional reconstruction of the pipe outline region from multiple perspectives is performed to obtain pipe point cloud data. Pipe diameter, length, and wall thickness are extracted from the pipe point cloud data. Defect identification is performed on the surface images of the pipe to obtain the defect type, location, and severity corresponding to corrosion, cracks, and deformation; Temperature field anomaly analysis was performed on the thermal imaging data of the pipe to obtain thermal anomaly information used to characterize the risk of internal damage to the pipe. Pipe diameter, length, wall thickness, material, weight, defect type, defect location, defect severity, and thermal anomaly information are bound to the same pipe identifier to form the pipe entity record.

[0045] As an example of implementation, step S2 of this solution includes the following details: The visible light image obtained in step S1 and infrared thermal imaging data The process yields a pipe entity record for each pipe, defined as follows:

[0046] in, For the first Root canal material at all times Entity records; This serves as a unique identifier for the pipe material. The outer diameter of the pipe; This refers to the length of the pipe. For pipe wall thickness; For pipe material; This refers to the weight of the pipe. Information on surface defects; Defect severity; This is information about thermal anomalies; This refers to the time of warehousing; This indicates the current location of the pipe.

[0047] 1. Identification of three-dimensional geometric parameters of pipes First, target detection and instance segmentation are performed on the multi-view images to obtain the first... Root canal material in the first The image contour region under each viewpoint is defined as follows:

[0048] in, For the first From the perspective of the first The segmented area of ​​the root canal material; This is an instance segmentation model.

[0049] Then, 3D reconstruction is performed using multi-view geometric relationships. Let 3D points... Projected to the Two-dimensional points on the image plane of a camera device are The camera intrinsic parameter matrix is The extrinsic parameter matrix is Then we have:

[0050] in, A two-dimensional pixel in the image plane; For the first The intrinsic parameter matrix of each camera device; For the first The rotation matrix of a camera device relative to the world coordinate system; For the first The translation vector of a camera device relative to the world coordinate system; Let be the three-dimensional point to be solved; This is to represent the equivalence relation in the sense of homogeneous coordinates.

[0051] To obtain 3D points The minimum reprojection error is used for estimation, which is defined as:

[0052] in, The optimal 3D points obtained from the reconstruction; It is the projection function from homogeneous coordinates to two-dimensional pixel coordinates; The number of camera devices involved in the reconstruction.

[0053] A point cloud set of pipes, composed of multiple three-dimensional points, is defined as follows:

[0054] in, For the first Point cloud set of root canal material; definition For the first The first on the root canal A three-dimensional point; This represents the number of point clouds.

[0055] Since the pipe is approximately cylindrical, its outer diameter and length can be obtained through cylindrical fitting, and are defined as follows:

[0056]

[0057]

[0058] in, For the first Reference point on the axis of the root canal; For the first The unit vector along the axial direction of the root canal material; For the first The outer radius of the root canal material; For the first Outer diameter of the root canal; For the first Root canal length; For vector cross product; This is the dot product of vectors.

[0059] If the system can identify the inner circle boundary of the pipe, the wall thickness can be expressed as:

[0060] in, For the first Inner diameter of the root canal; For the first Root canal wall thickness.

[0061] If the inner circle is not visible, the wall thickness can be estimated by combining the ERP inbound specifications, image recognition labels, and weighing results.

[0062] 2. Pipe weight estimation If the material has been identified Its density is The weight of the pipe can be derived from its volume.

[0063] The pipe is a hollow cylinder with a cross-sectional area of:

[0064] And because:

[0065] therefore:

[0066] The volume of the pipe is:

[0067] The weight of the pipe is:

[0068] Right now:

[0069] in, This refers to the cross-sectional area of ​​the pipe. This refers to the volume of the pipe. Material The corresponding density; It is the acceleration due to gravity; This refers to the weight of the pipe.

[0070] 3. Identification of surface defects and thermal anomalies Surface defect information is represented as follows:

[0071] in, For the first The defect types include rust, cracks, and deformation; Location of the defect; The defect area; The defect severity level is indicated. For the first The number of defects on the root canal material.

[0072] The severity of the defect can be normalized as follows:

[0073] in, For the first Overall severity of defects in root canal materials; These are weighting coefficients for different defect types; For the first External surface area of ​​the tube material; This represents the percentage of the defective area. The defect severity level is: To calculate the index.

[0074] The outer surface area of ​​the pipe is:

[0075] Thermal anomaly information can be determined from the temperature deviation of the thermal image. Let the first... The local temperature of the thermal imaging area of ​​the root canal is The average temperature of the thermal imaging area of ​​the pipe is The standard deviation is Then the thermal anomaly fraction is:

[0076] in, This represents the thermal anomaly score. For the first The temperature of a thermal image pixel or thermal image point; The average temperature of the thermal imaging area of ​​the pipe; The standard deviation of temperature in the thermal imaging area of ​​the pipe; To prevent extremely small positive numbers with a denominator of zero.

[0077] In this step, a two-dimensional contour is first obtained from multi-view images, and then a three-dimensional point cloud is inferred from the camera projection relationship. Since the geometric shape of the pipe is approximately cylindrical, the pipe diameter and length are extracted by cylinder fitting. The wall thickness is then determined by combining the inner diameter or specification information. The weight is derived from the material density and the volume of the hollow cylinder. Finally, the defect severity and thermal anomaly score are constructed by the defect area ratio, defect type weight and thermal image temperature difference.

[0078] Through step S2, this scheme obtains a set of records for all pipe entities. Its definition is:

[0079] in, This represents the number of pipes identified within the current storage area.

[0080] To address the issue of the inability to integrate visual recognition results, environmental monitoring results, cargo location status, and AGV status under the same spatial reference, step S3 of this solution spatiotemporally aligns the 3D point cloud of the pipe, surface defect information, and temperature and humidity data, and constructs a digital twin corresponding to the actual warehousing environment.

[0081] As a preferred implementation method, in step S3 of this solution, the dynamic digital twin model includes the three-dimensional occupancy status of the pipe, the idle status of the storage location, the distribution status of environmental parameters, and the operating status of the AGV.

[0082] In step S3 of this scheme, the method for constructing a dynamic digital twin model includes: Establish a unified three-dimensional coordinate system for the warehouse area; Transform the coordinate systems of the camera equipment, cargo location, AGV navigation, and environmental sensing equipment to the unified three-dimensional coordinate system. A 3D occupancy raster map is generated based on pipe point cloud data and storage location data; Generate a spatial distribution map of environmental parameters based on temperature and humidity data; Generate a map of accessible AGV paths based on AGV status data; By overlaying the three-dimensional occupancy grid map, the spatial distribution map of environmental parameters, and the AGV passable path map, a dynamic digital twin model that can be updated as pipes are put into storage, put out of storage, rearranged, and as the environment changes.

[0083] As an example, step S3 of this solution includes the following details: Let the unified world coordinate system of the storage area be . , No. The transformation matrix from the camera device coordinate system to the world coordinate system is:

[0084] Then the points in the camera equipment coordinate system Transformed to world coordinates:

[0085] in, For the first Homogeneous transformation matrix from the camera device coordinate system to the world coordinate system; It is a rotation matrix; It is a translation vector; A three-dimensional point in the coordinate system of the camera device; A three-dimensional point in the world coordinate system.

[0086] The pipe point cloud, pipe entity record, storage location data, environmental data, and AGV data obtained in step S2 are all converted to the world coordinate system to form a dynamic digital twin model, which is defined as follows:

[0087] in, For a moment A dynamic digital twin model; A 3D raster map showing the occupancy of the warehouse area; This is a diagram showing the status of the storage locations. Spatial distribution map of environmental parameters; This is a diagram showing the operating status of the AGV. This is a set of physical states of the pipe.

[0088] 1. 3D Occupancy Raster Modeling The storage space is divided into several voxel units. Define the occupancy function:

[0089] in, For a moment voxels The occupancy status; It is a three-dimensional voxel unit in the storage space.

[0090] No. The remaining available volume for each storage location is:

[0091] in, For the first Each storage space at any time The remaining available volume; For the first Total volume of each storage space; The volume of a single voxel; For the first The volume already occupied in each storage location.

[0092] 2. Spatial distribution modeling of environmental parameters Since the temperature and humidity sensors are discretely arranged, it is necessary to calculate the environmental conditions at the storage location based on the discrete measuring points. Let the first... The center coordinates of each storage location are: , No. The coordinates of the sensors are Then the first The estimated temperature for each storage location is:

[0093] The humidity estimate is:

[0094] in, For the first Estimated temperature at each cargo location; For the first Estimated humidity at each storage location; For the first The center coordinates of each cargo location; For the first Coordinates of each sensing device; For the first Temperature collected by a single sensor; For the first Humidity collected by a single sensor device; To prevent extremely small positive numbers with a distance of zero.

[0095] Under the definition logic of the above formula, the closer the sensor is to the target cargo location, the greater the impact of its environmental readings on the target cargo location. Therefore, the reciprocal of the distance is used as the weight.

[0096] 3. AGV operating status modeling Abstracting the AGV's drivable path into a path diagram Its definition is:

[0097] in, For the set of channel nodes; For the set of channel edges; time Let be the passage cost function for each edge.

[0098] For the channel edge The cost of its passage for:

[0099] in, For the channel edge Length; The average speed of the AGV; For the edge At any moment Indicator of congestion level; For the edge At any moment An indicator of whether there is an obstacle avoidance risk; , This represents the congestion and risk penalty coefficient.

[0100] In this step, the digital twin model essentially maps the physical warehouse to a computable virtual warehouse. Therefore, a unified coordinate system and time base are needed, along with a 3D occupancy grid to represent space, environmental interpolation to represent the storage location environment, and a path map to represent AGV reachability. Based on this, subsequent steps determine whether a particular pipe can be placed in a specific storage location, whether it is safe, and whether the AGV can reach it.

[0101] To address the issues arising from unconstrained screening of candidate storage locations and stacking actions, which can lead to problems such as pipes not fitting, being damaged, being unsuitable for the environment, inconvenient outbound operations, or AGV malfunctions, this solution's dynamic stacking strategy comprehensively considers factors such as pipe weight, diameter, length, material, defect severity, storage location dimensions and load-bearing capacity, temperature and humidity, and AGV status.

[0102] As a preferred implementation method, preferably, in step S4 of this solution, the storage constraint set includes storage location size constraints, storage location load-bearing constraints, pipe pressure resistance constraints, environmental adaptability constraints, defect avoidance constraints, and outbound priority constraints, and the candidate stacking action set includes candidate storage locations, number of stacking layers, placement angle of each layer, support method, and AGV handling path.

[0103] In step S4 of this scheme, the compressive strength constraint of the pipe is determined in the following way: The self-weight of the pipe is determined based on its material, diameter, wall thickness, and length. The stacking pressure that the target pipe will withstand is determined based on the target number of stacking layers and the weight of the upper pipe. The allowable pressure value of the pipe is determined based on the allowable stress of the pipe material, the cross-sectional dimensions of the pipe, the severity of surface defects, and the environmental degradation coefficient. When the stacking pressure is less than or equal to the allowable pressure value, the candidate stacking action satisfies the pipe pressure resistance constraint; when the stacking pressure is greater than the allowable pressure value, the corresponding candidate stacking action is eliminated.

[0104] As an example, step S4 of this solution includes the following technical details: For the For pipes awaiting warehousing, rearrangement, or outbound adjustment, refer to the pipe entity record from step S2. and the digital twin model of step S3 Generate a set of storage constraints, defined as follows:

[0105] in, Due to space constraints; For the load-bearing capacity constraints of the storage space; For the compressive strength constraint of the pipe; Environmental adaptation constraints; Defect avoidance constraints; This is a priority constraint for outbound shipments.

[0106] Candidate stacking actions are defined as:

[0107] in, For a candidate stacking action; Candidate storage location number; The number of candidate stacking layers; The angle at which the pipes are placed; As a support method; This is the AGV transport path.

[0108] The candidate set of stacked actions is:

[0109] in, The number of candidate actions to meet the preliminary feasibility requirements.

[0110] 1. Storage space size constraints Let the first The available length, width, and height for each storage location are as follows: If the pipes are to be placed in this storage location, the following conditions must be met:

[0111]

[0112]

[0113] in, For the first Root canal length; For the first Outer diameter of the root canal; The number of stacking layers; This is a safety margin in the length direction; This is a safety margin in the width direction; This is for the height allowance of interlayer supports or wooden blocks; The remaining available length, width, and height dimensions of the storage space.

[0114] 2. Load-bearing capacity constraints of the storage space Let the first The remaining load-bearing capacity of each storage space is: The total weight of the pipes stacked at this location is:

[0115] in, Candidate actions The total weight in the corresponding storage location; This refers to the weight of the pipes to be stacked. Currently at the th A collection of pipes in each storage location; For existing pipe materials The weight.

[0116] The load-bearing constraints are:

[0117] 3. Pipe compressive strength constraints For the Root tube, if there is a tube assembly above it Then the pressure it bears from the upper layer is:

[0118] in, Candidate actions Next The pressure exerted on the root canal material by the upper layer; It is the acceleration due to gravity; In candidate actions The next one is in the The assembly of pipes above the root canal; For the above number Weight of root canal material.

[0119] No. The allowable pressure value for root canal material is:

[0120] in, For the first Permissible pressure value for root canal materials; For material The corresponding allowable stress; The effective pressure-bearing area; This is the defect reduction factor; This is the environmental reduction factor; Defect severity; The target cargo location's temperature and humidity.

[0121] The more severe the defect, the lower the allowable pressure value; therefore, we can set:

[0122] The greater the environmental deviation, the lower the permissible pressure value; therefore, we can set:

[0123] in, , , To reduce the weight; For material The optimal center value for storage temperature; For material The optimal storage humidity center value; For material The allowable temperature deviation range; For material The allowable humidity deviation range.

[0124] The compressive constraint is:

[0125] 4. Environmental Adaptation Constraints Material The permissible storage temperature range is The allowable humidity range is ,but:

[0126]

[0127] 5. Defect avoidance constraints For pipes with severe defects, avoid placing them at the bottom or in high-pressure locations. Definition:

[0128] The defect avoidance constraint is then:

[0129] in, This indicates the pressure position. The allowable pressure threshold for defects.

[0130] 6. Outbound priority constraints According to ERP construction requirements If the first Pipe materials that need to be shipped out soon should be prioritized for placement in easily accessible locations. Define the relevance of outbound demand:

[0131] in, For the first Relevance of root canal material outbound demand; For pipe material physical record and construction requirements Specification matching degree; For the required time; This is the current time.

[0132] like If the cost is high, then the path cost from the target storage location to the outbound port must be calculated. Not greater than a set threshold, which is defined as:

[0133] In this step, candidate actions cannot be directly passed to the strategy model for selection; otherwise, the model might select solutions with mismatched dimensions, insufficient load-bearing capacity, unsuitable temperature and humidity, or inaccessible AGV locations. Therefore, this approach first eliminates infeasible solutions using deterministic constraints, and then passes feasible solutions to subsequent steps for evaluation and selection.

[0134] To address the issue of the inability to quantitatively compare multiple candidate stacking actions, this stacking strategy uses space utilization, pipe pressure value, and AGV handling efficiency as the core indicators of the reward function, with space utilization, pipe safety, and handling efficiency each corresponding to different weights.

[0135] As a preferred implementation method, in step S5 of this solution, the comprehensive evaluation value is calculated according to the following formula:

[0136] in, For candidate stacking actions The overall evaluation value; As a space utilization indicator; This represents the pressure value of the pipe under candidate stacking operations; This refers to the allowable pressure value for the corresponding pipe material; The estimated handling time required for the AGV to perform the candidate stacking action; Based on the transport time; As an environmental risk indicator; , , , These are the weighting coefficients corresponding to space utilization, pressure safety, handling efficiency, and environmental risk, respectively.

[0137] As a possible implementation example, step S5 of this solution includes the following technical details: For each candidate stacking action output in step S4 The space utilization index, pipe pressure safety index, AGV handling efficiency index and environmental risk index were calculated respectively.

[0138] 1. Space utilization indicators Candidate Action The first Pipes are placed in the storage location. The additional volume occupied by this action is the pipe envelope volume, which is defined as:

[0139] in, For the first The spatial envelope volume of the root canal material; This refers to the length of the pipe. This refers to the outer diameter of the pipe.

[0140] Storage location The space utilization index is:

[0141] in, Candidate actions Space utilization indicators; For storage location At any moment Volume used; For storage location Total volume.

[0142] 2. Pipe safety indicators under pressure The pressure value under the candidate action is obtained from step S4. and permissible pressure value Define security indicators:

[0143] in, For the pressure resistance safety indicators of pipe materials; The pressure value of the pipe under the candidate action; This represents the allowable pressure value for the pipe.

[0144] when The closer The smaller the safety margin, the better; when If the above condition is met, it indicates that the proposed solution is unacceptable and has been eliminated in step S4.

[0145] 3. AGV handling efficiency indicators Candidate actions The corresponding AGV path is If the path consists of several channel edges, the estimated transport time is:

[0146] in, The estimated transport time for the candidate action; For path One of the passageways in the middle; The cost of border crossing; This refers to the time for tube retrieval; Pipeline placement time; To allow time for waiting or to make way.

[0147] The handling efficiency index is defined as follows:

[0148] in, This refers to the AGV handling efficiency index. The baseline handling time can be the average time for manual handling or the historical average handling time for AGVs.

[0149] 4. Environmental risk indicators Environmental risk arises from deviations between the target storage location environment and the suitable storage environment for the pipes, and is defined as follows:

[0150] in, Environmental risk indicators for candidate actions; For storage location Temperature; For storage location Humidity; For material The suitable temperature range; For material The suitable humidity range; The normalized scale is the temperature scale; A normalized scale for humidity; Risk weights for temperature and humidity.

[0151] 5. Overall Evaluation Value The overall evaluation value of the candidate action is:

[0152] Substitution and back:

[0153] in, For candidate stacking actions The overall evaluation value; As a space utilization indicator; This refers to the pressure value of the pipe. This refers to the allowable pressure value for the pipe material. For estimated handling time; Based on the transport time; As an environmental risk indicator; Weighting for space utilization; For pressure safety weights; Weighted by handling efficiency; Environmental risk penalty weights.

[0154] In this step, pipeline storage optimization is not a single-objective problem. Simply improving space utilization might lead to excessive pressure on the pipes; prioritizing safety might result in wasted space; and focusing solely on short transport distances might disrupt the outbound order. Therefore, a comprehensive evaluation function is constructed after normalizing the various indicators. This makes different candidate actions comparable.

[0155] Through step S5, this scheme obtains the evaluation result for each candidate action, which is defined as: The evaluation result will then serve as input for subsequent steps, allowing the strategy model to further combine the current state and long-term benefits to determine the target stacking scheme.

[0156] To address the issue that a single evaluation can only reflect the current optimal state and cannot consider the impact of subsequent inbound and outbound processes, AGV scheduling, and spatial evolution, this solution's dynamic stacking strategy module employs the PPO deep reinforcement learning algorithm to output the optimal stacking scheme, including location selection, number of stacking layers, and placement angle.

[0157] As a preferred implementation method, in step S6 of this scheme, the strategy model is trained using a near-end strategy optimization algorithm. Its training state space includes pipe parameters, storage location parameters, environmental parameters, AGV state parameters, and construction requirement parameters. Its action space includes storage location selection action, stacking layer selection action, placement angle selection action, support method selection action, and AGV path selection action. Its reward function consists of space utilization reward, pressure safety reward, handling efficiency reward, and environmental risk penalty.

[0158] Specifically, step S6 of this solution includes the following sub-steps: The dynamic digital twin model obtained in step S3 The candidate action set obtained in step S4 Evaluation value of step S5 Common input strategy model.

[0159] The state definition of the policy model is:

[0160] in, For the policy model at time The state; For the physical record of the pipe material to be processed; The status of the storage location; Environmental state; AGV status; This is the state required for construction.

[0161] The actions of the policy model are the candidate stacking actions, which are defined as follows:

[0162] in, For the target storage location; The target number of stacked layers; Position it at the target angle; As a means of supporting the target; The target AGV transport path.

[0163] The policy model outputs the probability of selecting each candidate action in the current state: ;in, For parameters The strategy network; For candidate stacking actions; This is the current state.

[0164] The target action is determined as follows:

[0165] in, For the target stacking scheme; This is the comprehensive evaluation value.

[0166] PPO Training Derivation To ensure the strategy model not only selects actions with high current evaluation values ​​but also adapts to long-term warehouse operation performance, a near-end strategy optimization approach is employed for training. Let the old strategy be... The new strategy is Both affect the action The probability ratio is:

[0167] in, The ratio of the probability of actions under the new and old strategies; Selecting actions for the new strategy The probability of; Choose an action for the old strategy The probability of.

[0168] The advantage function represents the payoff advantage of the current action relative to the average strategy, and is defined as follows:

[0169] in, The dominant function; An immediate reward after an action is performed; Discount factor; The state value function; The next state value function.

[0170] The immediate reward can be composed of the comprehensive evaluation value from step S5, which is defined as:

[0171] in, For candidate stacking actions The overall evaluation value, Penalty for any pipe damage that occurs after execution; Penalties for delays in handling or outbound shipments; , For penalty weights.

[0172] The objective function for PPO shearing is:

[0173] in, The objective function for PPO shearing; To calculate the expectation over time steps of the training samples; To limit the probability ratio to Within the range; Update the magnitude limit parameter for the strategy.

[0174] Step S5 of this solution provides an immediate assessment of the merits of the current candidate actions, but pipe warehousing is a continuous process. Occupying a storage location may affect subsequent outbound shipments; placing a pipe on the outside may improve the efficiency of outbound shipments for recent construction work. Therefore, step S6 employs a strategy model to combine the current state, candidate actions, and long-term rewards to determine a target stacking scheme more suitable for dynamic warehousing scenarios.

[0175] Through step S6, this scheme obtains the target stacking scheme, which is defined as:

[0176] in, For the target storage location; The target number of stacked layers; Position it at the target angle; As a means of supporting the target; The target AGV path is defined. This result will serve as the basis for step S7, used to generate AGV handling tasks and perform on-site handling, placement, and visual verification.

[0177] To address the discrepancies between theoretical stacking schemes and actual on-site implementation, the potential impact of navigation errors or obstacles on AGV handling, and the lack of automatic verification after pipe placement, this solution utilizes a fusion navigation system for AGVs, incorporating magnetic strips, QR codes, and lasers. Furthermore, area cameras can collaboratively handle handling and restacking.

[0178] As a preferred implementation method, in step S7 of this solution, the target AGV handling task includes the pipe picking position, pipe placing position, handling path, handling sequence, and obstacle avoidance strategy; when the AGV handling equipment performs the handling task, it uses a combination of magnetic strip navigation, QR code positioning, and laser navigation for positioning. When any one of the navigation information fails, the handling task continues to be performed based on the remaining navigation information, and the current position, remaining power, load status, and task progress are fed back to the dynamic digital twin model.

[0179] As an example, step S7 of this solution includes the following technical details: Based on the target stacking scheme output in step S6 Generate AGV handling tasks, defined as follows:

[0180] in, For AGV handling tasks; This is the location for taking the tube; This indicates the location for pipe placement; Path for transporting goods to the target location; This refers to the weight of the pipes to be transported. For the target storage location; The target number of stacked layers; Position it at the target angle; The target is supported by a certain method.

[0181] 1. AGV Integrated Positioning AGVs can simultaneously obtain magnetic stripe positioning results, QR code positioning results, and laser positioning results, which are denoted as follows: .

[0182] The fusion localization result can be expressed as a weighted estimate defined as:

[0183] in, The AGV fusion positioning results; The result of magnetic stripe positioning; The location result is the QR code. The result is the laser positioning. For the first Error covariance of various positioning methods; This represents the confidence weight for the corresponding positioning method.

[0184] Under the definition of this formula, the smaller the positioning error, the smaller its covariance, the larger its inverse covariance, and the higher its contribution to the fusion result.

[0185] 2. Execution process monitoring During the AGV's task execution, the following information is continuously provided:

[0186] in, For the first The execution status of the AGV; This indicates the current position of the AGV. Remaining battery level; Current load; Task status; This represents the time the task has been executed.

[0187] If the AGV path deviation exceeds the threshold:

[0188] This is then determined to be a path execution exception, where, To plan the route at time The target location; This is the allowed path deviation threshold.

[0189] 3. Visual verification of placement results After the pipes are placed, the vision equipment re-identifies the actual position, actual placement angle, and actual number of stacked layers of the pipes:

[0190] in, This refers to the actual placement location of the pipes; This refers to the actual angle at which the pipes are placed. This represents the actual number of stacked layers. It is a visual verification model.

[0191] Verification deviation is:

[0192]

[0193]

[0194] When the following conditions are met:

[0195] When the stacking process is deemed successful, it is determined that the stacking operation is qualified.

[0196] in, Allowable positional deviation; This is the allowable deviation in angle.

[0197] If the above conditions are not met, or if abnormalities occur in temperature and humidity, pipe pressure conditions, or AGV operating status, the digital twin model definition will be updated as follows:

[0198] Then repeat steps S4 to S6 to generate an adjustment plan.

[0199] In this step, the target stacking scheme is only the result of system calculation. The actual execution is affected by factors such as AGV positioning errors, robotic arm gripping errors, obstacles in the storage location, and pipe rolling. Therefore, visual verification must be performed after execution, and the actual execution results must be fed back to the digital twin model. If the execution result deviates from the target scheme, constraints are regenerated and strategies are selected again to achieve closed-loop control.

[0200] Through step S7, this solution obtains its execution status. This includes success, deviation, and anomaly; actual stacking results (actual placement of pipes). Actual placement angle of the pipes Actual number of stacked layers AGV execution status .

[0201] If execution status If successful, proceed to step S8; if there is a deviation or anomaly, update the... Return to step S4.

[0202] To address the issues of outdated inventory records, disconnect between construction needs and warehouse status, and the inability of AI models and stacking strategies to adaptively optimize based on operational data, this solution integrates with the ERP system via API interfaces to achieve inventory data synchronization, construction demand matching, procurement alerts, and strategy updates.

[0203] As a preferred implementation method, in step S8 of this solution, when matching outbound materials based on the construction demand data provided by the ERP system, the outbound priority is determined according to the matching degree of pipe specifications, the time of entry into the warehouse, the health status of the pipe materials, the distance from the outbound port, and the construction demand time. Based on the outbound priority, an outbound pipe material list, an AGV outbound path, and a re-stacking scheme for the remaining pipe materials are generated.

[0204] As an example, step S8 of this solution includes the following details: After step S7 is completed, an inventory change record is generated, which is defined as:

[0205] in, Record inventory changes; For pipe material identification; The operation type includes inbound, rearrangement, and outbound. Original cargo location; For the new storage location; For quantity changes; Operation time; Information on surface defects; This is information about thermal anomalies; In execution status.

[0206] ERP inventory quantity updated as follows:

[0207] in, Specifications category Inventory quantities in the updated ERP system; Specifications category Inventory levels in the ERP system before the update; This represents the quantity received into the warehouse this period. This represents the quantity shipped out during this period.

[0208] If inventory falls below the safety threshold:

[0209] Then a procurement alert is sent to the ERP system, wherein, Specifications category The safety stock threshold.

[0210] 1. Outbound priority update When the ERP system provides construction requirements, the system can calculate the outbound priority based on specification matching, inbound time, health status, outbound distance, and required time, which is defined as:

[0211] in, For the first Priority for the release of root canal materials; The degree of matching between pipe materials and construction requirements; The score is normalized to the time of entry into the warehouse; the earlier the entry into the warehouse, the higher the score. Score the health status; This is the normalized distance from the pipe to the outlet. Based on the urgency of the construction needs; This refers to priority weights.

[0212] The health status score can be determined by the defect severity and thermal anomaly score, and is defined as follows:

[0213] in, Defect severity; This represents the thermal anomaly score. , Weights for defects and thermal anomalies.

[0214] 2. Iteration of AI visual recognition models If the true parameters obtained through manual verification or weighing are:

[0215] The model recognition result is as follows:

[0216] The recognition error loss is:

[0217] in, For AI visual recognition model loss; To verify the actual value; Model recognition value; The weights for each error term are denoted as .

[0218] The model parameters are updated as follows:

[0219] in, For the updated visual model parameters; These are the parameters of the visual model before the update. The learning rate for the visual model; This is the gradient of the loss function with respect to the model parameters.

[0220] 3. Strategy Model Iteration The system calculates operating costs over a period of time:

[0221] in, For overall operating costs; This represents the average space utilization rate. This represents the average pipe damage rate. This represents the average AGV transport time. To average environmental risk; This represents the average outbound delay. The weights for each cost item.

[0222] If operating costs increase, adjust the weights in step S5:

[0223] Make it satisfy:

[0224] in, The updated overall evaluation weights; To use weights Operating costs at the time.

[0225] In this solution, the warehousing system generates real-world feedback after operation, such as pipe identification errors, handling time, damage status, space utilization, and outbound delays. This data reflects the discrepancies between the model and the actual scenario. Therefore, step S8 synchronizes actual inventory changes to the ERP system and utilizes operational feedback to correct the AI ​​visual model and stacking strategy model, transforming the system from static rule execution to dynamic self-learning optimization.

[0226] Combination Figure 2 As shown above, this solution also proposes an AI vision-based dynamic space optimization system for pipeline storage, which includes: The multi-source data acquisition module is used to acquire pipe images, pipe thermal imaging data, temperature and humidity data, cargo location data, AGV status data, and construction requirement data from the ERP system. The AI ​​visual recognition module is used to identify pipe images and pipe thermal imaging data, and generate pipe entity records including pipe diameter, length, wall thickness, material, weight, surface defect information and thermal anomaly information; The digital twin construction module is used to build a dynamic digital twin model based on pipe entity records, temperature and humidity data, cargo location data, and AGV status data. The constraint generation module is used to generate and store a set of constraints and a set of candidate stacked actions based on pipe entity records, dynamic digital twin models, and construction requirement data. The dynamic space optimization module is used to calculate the comprehensive evaluation value of candidate stacking actions and determine the target stacking scheme through the strategy model; The AGV scheduling and execution module is used to generate AGV handling tasks based on the target stacking scheme and control the AGV handling equipment to complete the pipe material warehousing, rearrangement and warehousing. The ERP linkage module is used to synchronize information on changes in pipe inventory, storage location, defects, and inbound / outbound execution to the ERP system, and to receive construction demand data sent by the ERP system. The model iteration module is used to update the AI ​​visual recognition model and strategy model based on visual recognition errors, stacking execution deviations, pipe damage records, and AGV handling time.

[0227] As a preferred implementation method, the multi-source data acquisition module of this solution preferably includes multi-view camera devices set in different locations in the storage area, infrared thermal imaging devices set in association with the multi-view camera devices, temperature and humidity sensing devices arranged according to the storage function zones, and AGV communication interfaces for collecting AGV operating status. The ERP linkage module interacts with the ERP system via API interface and generates early warning information when inventory falls below the safety threshold, construction needs change, or pipe material status is abnormal.

[0228] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0229] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) or processor to execute all or part of the steps of the methods of various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0230] The above description is only a part of the embodiments of the present invention and does not limit the scope of protection of the present invention. Any equivalent device or equivalent process transformation made based on the content of the present invention specification and drawings, or direct or indirect application in other related technical fields, are similarly included within the patent protection scope of the present invention.

Claims

1. A dynamic space optimization method for pipeline storage based on AI vision, characterized in that, It includes: S1. Obtain multi-source status data of the pipeline storage area, including pipe images, pipe thermal imaging data, temperature and humidity data, cargo location data, AGV status data, and construction requirement data; S2. Perform AI visual recognition on the pipe images and pipe thermal imaging data to obtain a pipe entity record corresponding to each pipe. The pipe entity record includes pipe identification, specification information, surface defect information, thermal anomaly information, warehousing time and location information. S3. Perform spatiotemporal registration of the pipe entity records, temperature and humidity data, cargo location data and AGV status data to construct a dynamic digital twin model corresponding to the actual storage area; S4. Based on the pipe material entity record, dynamic digital twin model and construction requirement data, generate a storage constraint set and a candidate stacking action set for each pipe material to be put into storage or rearranged. S5. For each candidate stacking action, calculate the space utilization index, pipe pressure safety index, AGV handling efficiency index and environmental risk index, and then construct a comprehensive evaluation value. S6. The current state of the dynamic digital twin model is used as the input of the strategy model. Combined with the comprehensive evaluation value, the strategy model determines the target stacking scheme from the candidate stacking actions that satisfy the storage constraint set. The target stacking scheme includes the target storage location, the target number of stacking layers, the target placement angle, the target support method, and the target AGV handling task. S7. The target AGV handling task is sent to the AGV handling equipment, and the pipe handling, placement and re-stacking process is visually verified by a multi-view camera device. When the visual verification result is inconsistent with the target stacking scheme, or when the temperature and humidity data, pipe pressure status, or AGV operation status are abnormal, the dynamic digital twin model is updated and steps S4 to S6 are re-executed.

2. The AI ​​vision-based dynamic space optimization method for pipeline storage as described in claim 1, characterized in that, It also includes: S8. After the pipes are put into storage, rearranged or put out of storage, the pipe inventory change information, storage location change information, defect change information and inbound / outbound execution information are synchronized to the ERP system, and the AI ​​visual recognition model and strategy model are iteratively updated using visual recognition error, stacking execution deviation, pipe damage record and AGV handling time.

3. The AI ​​vision-based dynamic space optimization method for pipeline storage as described in claim 1 or 2, characterized in that, In step S1, the multi-source status data includes pipe images collected by multi-view camera equipment, pipe thermal image data collected by infrared thermal imaging equipment, temperature and humidity data collected by environmental sensing equipment, cargo location data provided by warehouse management equipment, AGV status data provided by handling equipment, and construction demand data provided by ERP system.

4. The AI ​​vision-based dynamic space optimization method for pipeline storage as described in claim 3, characterized in that, In step S2, the pipe entity record includes pipe identification, pipe diameter, length, wall thickness, material, weight, surface defect information, thermal anomaly information, warehousing time, and current location; Step S2, performing AI visual recognition on the pipe image and pipe thermal image data includes the following steps: Target detection and instance segmentation are performed on pipe images from multiple perspectives to determine the contour region of each pipe. Three-dimensional reconstruction of the pipe outline region from multiple perspectives is performed to obtain pipe point cloud data. Pipe diameter, length, and wall thickness are extracted from the pipe point cloud data. Defect identification is performed on the surface images of the pipe to obtain the defect type, location, and severity corresponding to corrosion, cracks, and deformation; Temperature field anomaly analysis was performed on the thermal imaging data of the pipe to obtain thermal anomaly information used to characterize the risk of internal damage to the pipe. Pipe diameter, length, wall thickness, material, weight, defect type, defect location, defect severity, and thermal anomaly information are bound to the same pipe identifier to form the pipe entity record.

5. The AI ​​vision-based dynamic space optimization method for pipeline storage as described in claim 2, characterized in that, In step S3, the dynamic digital twin model includes the three-dimensional occupancy status of the pipe, the idle status of the storage location, the distribution status of environmental parameters, and the operating status of the AGV. In step S3, the method for constructing a dynamic digital twin model includes: Establish a unified three-dimensional coordinate system for the warehouse area; Transform the coordinate systems of the camera equipment, cargo location, AGV navigation, and environmental sensing equipment to the unified three-dimensional coordinate system. A 3D occupancy raster map is generated based on pipe point cloud data and storage location data; Generate a spatial distribution map of environmental parameters based on temperature and humidity data; Generate a map of accessible AGV paths based on AGV status data; By overlaying the three-dimensional occupancy grid map, the spatial distribution map of environmental parameters, and the AGV passable path map, a dynamic digital twin model that can be updated as pipes are put into storage, put out of storage, rearranged, and as the environment changes.

6. The AI ​​vision-based dynamic space optimization method for pipeline storage as described in claim 2, characterized in that, In step S4, the storage constraint set includes storage location size constraints, storage location load-bearing constraints, pipe pressure resistance constraints, environmental adaptability constraints, defect avoidance constraints, and outbound priority constraints. The candidate stacking action set includes candidate storage locations, number of stacking layers, placement angle of each layer, support method, and AGV handling path. In step S4, the compressive strength constraint of the pipe is determined in the following way: The self-weight of the pipe is determined based on its material, diameter, wall thickness, and length. The stacking pressure that the target pipe will withstand is determined based on the target number of stacking layers and the weight of the upper pipe. The allowable pressure value of the pipe is determined based on the allowable stress of the pipe material, the cross-sectional dimensions of the pipe, the severity of surface defects, and the environmental degradation coefficient. When the stacking pressure is less than or equal to the allowable pressure value, the candidate stacking action satisfies the pipe pressure resistance constraint. When the stacking pressure exceeds the allowable pressure value, the corresponding candidate stacking action is eliminated.

7. The AI ​​vision-based dynamic space optimization method for pipeline storage as described in claim 2, characterized in that, In step S5, the comprehensive evaluation value is calculated according to the following formula: in, For candidate stacking actions The overall evaluation value; As a space utilization indicator; This represents the pressure value of the pipe under candidate stacking operations; This refers to the allowable pressure value for the corresponding pipe material; The estimated handling time required for the AGV to perform the candidate stacking action; Based on the transport time; As an environmental risk indicator; , , , These are the weighting coefficients corresponding to space utilization, pressure safety, handling efficiency, and environmental risk, respectively.

8. The AI ​​vision-based dynamic space optimization method for pipeline storage as described in claim 2, characterized in that, In step S6, the strategy model is trained using a near-end strategy optimization algorithm. Its training state space includes pipe parameters, storage location parameters, environmental parameters, AGV state parameters, and construction requirement parameters. Its action space includes storage location selection action, stacking layer selection action, placement angle selection action, support method selection action, and AGV path selection action. Its reward function consists of space utilization reward, pressure safety reward, handling efficiency reward, and environmental risk penalty.

9. The AI ​​vision-based dynamic space optimization method for pipeline storage as described in claim 2, characterized in that, In step S7, the target AGV handling task includes the pipe picking position, pipe placing position, handling path, handling sequence and obstacle avoidance strategy; when the AGV handling equipment performs the handling task, it uses a combination of magnetic strip navigation, QR code positioning and laser navigation for positioning. When any one of the navigation information fails, the handling task continues to be performed based on the remaining navigation information, and the current position, remaining power, load status and task progress are fed back to the dynamic digital twin model. In step S8, when matching outbound materials based on the construction demand data provided by the ERP system, the outbound priority is determined according to the pipe specification matching degree, warehousing time, pipe health status, distance from the outbound port and construction demand time. Based on the outbound priority, an outbound pipe list, AGV outbound path and remaining pipe re-stacking scheme are generated.

10. A dynamic space optimization system for pipeline storage based on AI vision, characterized in that, It includes: The multi-source data acquisition module is used to acquire pipe images, pipe thermal imaging data, temperature and humidity data, cargo location data, AGV status data, and construction requirement data from the ERP system. The AI ​​visual recognition module is used to identify pipe images and pipe thermal imaging data, and generate pipe entity records including pipe diameter, length, wall thickness, material, weight, surface defect information and thermal anomaly information; The digital twin construction module is used to build a dynamic digital twin model based on pipe entity records, temperature and humidity data, cargo location data, and AGV status data. The constraint generation module is used to generate and store a set of constraints and a set of candidate stacked actions based on pipe entity records, dynamic digital twin models, and construction requirement data. The dynamic space optimization module is used to calculate the comprehensive evaluation value of candidate stacking actions and determine the target stacking scheme through the strategy model; The AGV scheduling and execution module is used to generate AGV handling tasks based on the target stacking scheme and control the AGV handling equipment to complete the pipe material warehousing, rearrangement and warehousing. The ERP linkage module is used to synchronize information on changes in pipe inventory, storage location, defects, and inbound / outbound execution to the ERP system, and to receive construction requirement data sent by the ERP system. The model iteration module is used to update the AI ​​visual recognition model and strategy model based on visual recognition errors, stacking execution deviations, pipe damage records, and AGV handling time. The multi-source data acquisition module includes multi-view camera devices set in different locations in the storage area, infrared thermal imaging devices associated with the multi-view camera devices, temperature and humidity sensing devices arranged according to the storage function zones, and AGV communication interfaces for collecting AGV operating status. The ERP linkage module interacts with the ERP system via API interface and generates early warning information when inventory falls below the safety threshold, construction needs change, or pipe material status is abnormal.

Images