A Method and System for Production Information Collection Based on AI Video Recognition and RFID Positioning

Abstract

A production information collection method and system based on AI video recognition and RFID positioning is disclosed. The method includes: constructing and training a mold state recognition model based on deep learning using images captured by AI video cameras deployed on the production line; establishing a mold information database and binding RFID chips to the molds; deploying RFID readers on the production line to locate mold positions and prefabricated component positions; scheduling daily production tasks and establishing a correspondence between mold types and prefabricated components; determining the mold load status through the mold state recognition model; when the mold status is detected to be non-empty, extracting the corresponding type of prefabricated component information from the production task and initiating a production process tracking program; and recording the production process information and storage location information of the prefabricated components based on the RFID chip data read in real time by the RFID readers and the location of the RFID readers.

Description

Technical Field

[0001] This invention relates to the field of precast segment production information management technology, and in particular to a production information collection method and system based on AI video recognition and RFID positioning. Background Technology

[0002] As the core structural component of shield tunnels, the production quality and efficiency of precast tunnel segments directly affect the safety and economy of tunnel engineering. In existing precast tunnel segment production plants, each production line can produce 30 to 50 segments per day. Each segment requires more than ten processes, including steel reinforcement fabrication, mold preparation, concrete pouring, steam curing, water curing, demolding, quality inspection, and storage. The production process is complex and requires high efficiency and accuracy in information collection.

[0003] Traditional precast tunnel segment production information collection relies primarily on manual recording: on-site workers manually fill out production record sheets after each process, which are then entered into the information system by management personnel. This method has the following problems: 1) Poor real-time information collection, as manual recording often lags behind actual production progress, failing to reflect the true situation on-site in a timely manner, leading to inaccurate production scheduling decisions; 2) Low data accuracy, due to the complex environment of the production site, manual recording is prone to omissions or errors, resulting in discrepancies between statistical data and actual conditions, thus affecting the effectiveness of production analysis; 3) High workload, especially in the case of large-scale production, information recording requires a large amount of human resources, significantly increasing production costs and easily causing excessive workload for on-site personnel; 4) Difficult problem tracing, due to the lack of complete and accurate production process data, it is difficult to quickly locate the problem link and cause when quality problems occur. Summary of the Invention

[0004] To address the shortcomings of existing technologies, the present invention aims to provide a production information collection method and system based on AI video recognition and RFID positioning, which can achieve automated monitoring of the entire precast segment production process while improving data collection efficiency and accuracy.

[0005] To achieve the above objectives, the present invention provides a production information collection method based on AI video recognition and RFID positioning, comprising: constructing and training a mold state recognition model based on deep learning based on images collected by AI video cameras deployed on the production line;

[0006] Establish a mold information database and bind RFID chips to the molds;

[0007] RFID readers are deployed on the production line to locate the positions of molds and prefabricated parts;

[0008] Arrange daily production tasks and establish the correspondence between mold types and prefabricated parts;

[0009] The mold load status is determined by the mold status recognition model; when the mold status is detected to be non-empty, the corresponding type of prefabricated part information in the production task is extracted, and the production process tracking program is started.

[0010] Based on the RFID chip data read in real time by the RFID reader and the location of the RFID reader, the production process information and storage location information of the prefabricated parts are recorded.

[0011] Furthermore, the step of constructing and training the deep learning-based mold state recognition model includes:

[0012] Collect image samples of empty molds, molds loaded with steel cages, and molds loaded with concrete under different lighting conditions;

[0013] The image samples are labeled, and mold status classification labels are established;

[0014] A mold state recognition model is constructed using a convolutional neural network. The mold state recognition model includes an input layer, a convolutional layer, a pooling layer, a fully connected layer, and an output layer.

[0015] The model is trained according to the set learning rate and number of iterations until the model recognition accuracy reaches the threshold, thus obtaining a trained mold state recognition model.

[0016] Furthermore, the step of establishing a mold information database and binding the RFID chip to the mold further includes: using a 12-digit number to encode the mold, with the first 4 digits representing the production line number, the middle 4 digits representing the mold type, and the last 4 digits representing the mold serial number; and writing the mold code, production date, and applicable prefabricated part specifications into the RFID chip.

[0017] Furthermore, the step of deploying RFID readers on the production line for locating the mold position and the preform position further includes: installing RFID readers at the entrance of each production mold table, with the reading range covering the mold table feeding area; installing RFID readers at the entrance and exit of the steam curing chamber to monitor the time when the preform enters and exits the steam curing chamber; installing RFID readers at the entrance and exit of the water curing chamber to record the time nodes of the water curing process; and installing RFID readers at the entrance of each channel in the storage area to locate the final storage position of the preform.

[0018] Furthermore, the steps of arranging daily production tasks and establishing the correspondence between mold types and prefabricated parts include: formulating a daily production plan based on production needs, including the type, quantity, and quality requirements of prefabricated parts; establishing a matching table between molds and prefabricated parts to ensure that mold specifications match prefabricated part specifications; and setting task priorities, with urgent tasks having a priority weight of 0.8-1.0 and regular tasks having a priority weight of 0.3-0.7.

[0019] Furthermore, the step of determining the mold load state through the mold state recognition model further includes:

[0020] A confidence threshold of 0.85 is set. When the confidence of the identification result is greater than the confidence threshold, the corresponding mold state is confirmed. The correspondence between the mold state and the production process is established as follows: an empty mold corresponds to the preparation process, and a mold for loading the rebar cage corresponds to the rebar binding completion process.

[0021] Furthermore, it also includes: when the recognition results of three consecutive frames of images are consistent, confirm the change in mold state to avoid misjudgment.

[0022] Furthermore, the step of recording the production process information and storage location information of the prefabricated part based on the RFID chip data read in real time by the RFID reader and the location of the RFID reader further includes: when the RFID reader detects the RFID chip signal, automatically recording the timestamp, location information, and mold code; establishing a mapping relationship between the RFID reader and the production process, so as to automatically determine the current stage of the production process based on the location of the RFID reader.

[0023] Furthermore, it also includes a data fusion processing step: fusion processing is performed based on the identification results of the mold status recognition model and the data read by the RFID reader, and a prefabricated part production status model is established to update the production progress of each prefabricated part in real time; daily output, process time, mold utilization rate are statistically analyzed, and production reports are generated.

[0024] Furthermore, the data fusion processing step further includes:

[0025] Time series analysis is used to statistically analyze the time taken in production processes and identify production bottlenecks;

[0026] Establish a correlation model between precast component quality and production parameters; the production parameters include steam curing time, steam curing temperature, and concrete strength.

[0027] An abnormality warning mechanism is set up; the abnormality warning mechanism is to automatically alarm when the process time exceeds 120% of the standard time.

[0028] Furthermore, it also includes steps for visual management: establishing a 3D digital production line model to display the location and status of each mold in real time; designing a production management dashboard, including real-time output, plan completion rate, and equipment operating status; and providing a mobile application so that managers can view production progress and abnormal situations at any time.

[0029] To achieve the above objectives, the present invention also provides a production information collection system based on AI video recognition and RFID positioning, comprising:

[0030] The production task management module is used to arrange daily production tasks according to order requirements and establish the correspondence between mold types and prefabricated parts;

[0031] The RFID positioning module is used to collect mold location information and prefabricated component location information through an RFID reader, perform RFID data processing, obtain the positioning data of the mold and prefabricated component, and send it to the data platform.

[0032] The AI ​​video recognition module is used to collect and process image data of the mold, and identify the mold status based on a deep learning mold status recognition model, and send the mold status data to the data platform.

[0033] The data platform is used to process the positioning data sent by the RFID positioning module and the mold status data sent by the AI ​​video recognition module to obtain the corresponding production information and update it to the production management module in real time.

[0034] The production management module is used to collect the production information, generate a complete production history, and provide it to the visualization management platform; the visualization management platform is used to visualize the production history.

[0035] The production information collection method based on AI video recognition and RFID positioning provided by this invention collects and integrates multi-source production information by deploying intelligent sensing devices at key nodes of the production line, thereby realizing automated monitoring of the entire production process and significantly improving the efficiency and accuracy of production information collection.

[0036] Other features and advantages of the invention will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing the invention. Attached Figure Description

[0037] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used together with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0038] Figure 1This is a flowchart of a production information collection method based on AI video recognition and RFID positioning according to Embodiment 1 of the present invention;

[0039] Figure 2 This is a schematic diagram of the production line process flow and data acquisition nodes according to Embodiment 1 of the present invention;

[0040] Figure 3 The flowchart is shown below for the mold state recognition algorithm according to Embodiment 1 of the present invention.

[0041] Figure 4 This is a rendering of the production management visualization interface according to Embodiment 1 of the present invention;

[0042] Figure 5 This is a schematic diagram comparing the effects of the manual method in Embodiment 1 and the method of the present invention.

[0043] Figure 6 This is a flowchart of a production information collection method based on AI video recognition and RFID positioning according to Embodiment 2 of the present invention;

[0044] Figure 7 This is a flowchart of the data fusion processing according to Embodiment 2 of the present invention;

[0045] Figure 8 This is a schematic diagram illustrating the configuration of application scenario 1 according to embodiment 3 of the present invention;

[0046] Figure 9 This is a schematic diagram of the production information collection system based on AI video recognition and RFID positioning according to Embodiment 4 of the present invention;

[0047] Figure 10 This is a flowchart for identifying the mold state according to Embodiment 4 of the present invention. Detailed Implementation

[0048] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are for illustration and explanation only and are not intended to limit the present invention.

[0049] Embodiments of the present invention will now be described in more detail with reference to the accompanying drawings. While some embodiments of the invention are shown in the drawings, it should be understood that the invention can be implemented in various forms and should not be construed as limited to the embodiments set forth herein. Rather, these embodiments are provided to provide a more thorough and complete understanding of the invention. It should be understood that the accompanying drawings and embodiments are for illustrative purposes only and are not intended to limit the scope of protection of the invention.

[0050] The term "comprising" and its variations as used herein are open-ended inclusion, meaning "including but not limited to". The term "based on" means "at least partially based on". The term "one embodiment" means "at least one embodiment"; the term "another embodiment" means "at least one additional embodiment"; and the term "some embodiments" means "at least some embodiments".

[0051] It should be noted that the concepts of "first" and "second" may be mentioned in this invention only to distinguish different devices, components or parts, and are not used to limit the order of the functions performed by these devices, components or parts or their interdependence.

[0052] It should be noted that the terms "one" and "multiple" used in this invention are illustrative rather than restrictive. Those skilled in the art should understand that, unless explicitly stated otherwise in the context, they should be understood as "one or more". "Multiple" should be understood as two or more.

[0053] The following terms may be used in this invention:

[0054] AI video recognition: A technical method for automatically identifying the state of target objects in video images based on artificial intelligence and computer vision technology;

[0055] RFID positioning: A positioning method that uses radio frequency identification technology to track the location of objects and collect information;

[0056] Precast tunnel segments: Concrete segment components for shield tunnels that are prefabricated in a production environment (such as a factory);

[0057] Convolutional Neural Network: A deep learning network architecture particularly suitable for image recognition and processing tasks;

[0058] Data platform: A centralized data processing and management platform that enables unified processing and analysis of multi-source data;

[0059] Edge computing: A computing paradigm that performs data processing and analysis at the network edge, close to the data source;

[0060] Steam curing process: A production process that uses steam to accelerate the curing of concrete components;

[0061] Water curing process: A production process that involves curing concrete components in water at room temperature;

[0062] Confidence threshold: A numerical standard for judging the credibility of AI model recognition results;

[0063] Process nodes: key control points and state transition points in the production process;

[0064] RFID reader: A device responsible for communicating with RFID electronic tags, realizing radio frequency identification of tag identification codes and reading memory data, used to read the serial number of production equipment;

[0065] Digital twin: A virtual copy of a physical system created using digital technology;

[0066] Production history: A data archive that records the complete production process of prefabricated components.

[0067] Definitions for other terms will be provided in the following description.

[0068] Example 1

[0069] Figure 1 The flowchart below shows the production information collection method based on AI video recognition and RFID positioning according to Embodiment 1 of the present invention. Figure 1 A detailed description of one specific embodiment of the present invention will be provided.

[0070] First, in step S101: construct a mold state recognition model based on deep learning and deploy an AI video monitoring system on the precast segment production line.

[0071] In embodiments of the present invention, this step includes: constructing a mold state recognition model based on deep learning, such as using a convolutional neural network architecture, which includes an input layer, a feature extraction layer, a classification layer, and an output layer. The training dataset includes mold state images under different lighting conditions and from different angles, covering states such as empty molds, molds with steel cages, molds for pouring concrete, and molds after demolding. The dataset for each state must have a certain number of images, such as no less than 2000 images per state. Based on the process flow and spatial layout of the precast tunnel segment production line, an AI video monitoring system is deployed, mainly by placing high-definition AI video cameras at the starting point, key process nodes, and quality inspection points of the production line.

[0072] In embodiments of the present invention, to ensure image acquisition accuracy, the camera resolution is required to be no less than 1920×1080 pixels and the frame rate is required to be no less than 25fps.

[0073] In this embodiment of the invention, the model training employs the backpropagation algorithm, and the learning rate, batch size, number of training iterations, and termination condition are set. For example, the learning rate is set to 0.001, the batch size is 32, and the number of training iterations is 1000-1500. Training stops when the validation set recognition accuracy reaches above 95% and there is no significant improvement for 50 consecutive epochs. One epoch is defined as a complete traversal and learning process of the entire training dataset during neural network model training.

[0074] In this embodiment, a production line is designed to produce 45 shield tunnel segments with a diameter of 6 meters per day. It is equipped with 15 sets of steel molds and includes five main functional areas: a rebar fabrication area, a casting area, a steam curing workshop, a water curing tank, and a finished product storage area. For example, four high-definition network cameras, such as Hikvision DS-2CD2T85G1 cameras, are installed at the beginning of the production line. These cameras have a resolution of 2560×1920 pixels, support H.265 encoding, and have infrared night vision capabilities. The cameras are installed at a height of 4.5 meters and an angle of 15° to ensure clear capture of the molds and their loading status. The cameras are connected to an edge computing server via gigabit Ethernet. The server is configured with an Intel i7-10700 processor, 32GB of RAM, and an NVIDIA RTX 3060 graphics card.

[0075] A deep learning model based on ResNet-50 was constructed. The network structure includes: an input layer (224×224×3), 4 residual blocks (each block contains 3 convolutional layers), a global average pooling layer, a fully connected layer (512 nodes), and an output layer (4 state classifications: empty mold, loaded steel cage, loaded concrete, and demolded state).

[0076] The training dataset contains 6000 labeled images, including 1500 images of empty molds, 1500 images of rebar cages being loaded, 1500 images of concrete being loaded, and 1500 images of demolding. Image acquisition covered different lighting conditions at dawn, noon, and dusk, and included multiple perspectives such as front, side, and oblique angles. Data augmentation was performed: the dataset was expanded to 12000 images using methods such as random rotation (±15°), brightness adjustment (±20%), and contrast variation (±15%). Model training parameters were set as follows: batch size of 32, learning rate of 0.001, optimizer Adam (an adaptive algorithm commonly used for deep neural network training), loss function cross-entropy, and 800 training epochs. 80% of the data was used for training, and 20% for validation. After training, the model achieved an accuracy of 96.8% on the validation set, meeting the requirements for production applications.

[0077] In step S102: Establish a mold information database and bind the RFID chip to the mold.

[0078] In this invention, mold information is entered into the production management system and bound to an RFID chip, thereby establishing a complete mold information database within the system. The mold information includes: mold code (using a 12-digit encoding rule, where the first 4 digits represent the production line number, the middle 4 digits represent the mold type code, and the last 4 digits represent the mold serial number), mold name, geometric dimensions, applicable precast component type, manufacturing date, maintenance records, and other basic information. A high-frequency RFID chip with a working frequency of 13.56MHz (such as the NXP NTAG213 RFID chip) is selected, with a storage capacity of at least 8KB and an operating temperature range of -40℃ to +85℃ to meet the environmental requirements for precast segment production. The RFID chip is securely attached to or embedded in the non-working surface of the mold to ensure it does not fall off or get damaged during production. Key data such as the mold's unique identifier, production line information, and mold specifications are written into the RFID chip, establishing a one-to-one correspondence between the chip ID and the mold information.

[0079] Continuing with the example above, a mold information database was established for this production line, and detailed information for a total of 15 molds was entered:

[0080] Mold codes: 0002-A001-0001 to 0002-A001-0015;

[0081] Mold type: 6-meter diameter standard segment mold;

[0082] Geometric specifications: outer diameter 6000mm, inner diameter 5400mm, thickness 300mm;

[0083] Manufacturer: A heavy equipment company;

[0084] Production commencement date: March 2023;

[0085] Maintenance cycle: Maintenance is required once every 100 uses.

[0086] For data binding, the following chip was selected: the NXP NTAG213 RFID chip, with an operating frequency of 13.56MHz, a storage capacity of 180 bytes, a reading distance of 1-4 meters, and an operating temperature range of -25℃ to +70℃. This chip was encapsulated in a PVC card measuring 85.6×54×0.84mm and adhered to the non-working area on the side of the mold using epoxy resin adhesive.

[0087] The following data structure is written into the chip:

[0088] Bytes 0-11: Unique identifier for the mold;

[0089] Bytes 12-27: Mold type code;

[0090] Bytes 28-43: Manufacturing date and maintenance records;

[0091] Bytes 44-59: Production line number and mold specifications;

[0092] Bytes 60-75: Reserved for expansion area.

[0093] In step S103: Establish a mapping table between RFID readers and production processes and storage locations, and deploy the RFID system on the production line.

[0094] In embodiments of the present invention, RFID system deployment and integration are completed by installing RFID readers at various key nodes of the production line and establishing a mapping relationship between RFID readers and production processes and storage locations.

[0095] Figure 2 This is a schematic diagram of the production line process flow and data acquisition nodes according to Embodiment 1 of the present invention, as shown below. Figure 2 As shown, the RFID reader deployment includes: fixed readers installed at the feed inlet of the production mold, covering a 3m x 2m area; access control readers installed at the entrance and exit of the curing chamber to monitor the time points of precast components entering and leaving the curing process; waterproof readers installed at the inlet and outlet of the water curing tank, with an operating environment humidity up to 95% RH; and handheld or fixed combined readers installed at the entrances of each passage in the finished product storage area for precise location of precast components. All readers are connected to the central data processing system via Ethernet or Wi-Fi, enabling real-time data transmission. A mapping table between each reader and the production process and storage location is established in the central data processing system to ensure accurate correspondence of location information.

[0096] For example, a total of 16 Impinj R220 RFID readers were installed in the following key locations:

[0097] Production line entrance: 1 unit, model R220-001, reading range 3×2 meters;

[0098] Rebar cage installation area: 5 units, models R220-002 to R220-006, covering 5 work stations;

[0099] Steam curing chamber entrance: 1 unit, model R220-007, access control system installation;

[0100] Inside the steam curing chamber: 3 units, models R220-008 to R220-010, to monitor the steam curing process;

[0101] Steam curing chamber exit: 1 unit, model R220-011, access control type installation;

[0102] Hydroponic tank area: 3 units, models R220-012 to R220-014, waterproof rating IP67;

[0103] Finished product storage area: 2 units, models R220-015 to R220-016, covering the storage channel.

[0104] All readers are connected to the central data server via an industrial Ethernet switch, and use the LLRP protocol for data communication, with a data transmission latency of less than 100ms to ensure timely reading.

[0105] In step S104: Create a task through the production task management system and establish a standardized matching database of mold type and prefabricated part specifications.

[0106] In this embodiment of the invention, the step includes: scheduling production tasks and matching molds based on the project schedule and customer order requirements. Specifically, based on the project schedule and customer order requirements, daily and weekly production task plans are formulated in the production task management system, including detailed information such as prefabricated part specifications, quantity, quality grade, and delivery time, and task creation is completed. A standardized matching database of mold types and prefabricated part specifications is established to ensure that the mold size and shape fully match the prefabricated part design requirements. Task priorities are calculated using a priority algorithm, which calculates priority weights based on factors such as project urgency, production difficulty, and mold availability. The weight for urgent projects is set to 0.8-1.0, and the weight for regular projects is set to 0.3-0.7 to ensure optimal allocation of production resources.

[0107] For example, a production task management system is developed within an enterprise ERP system to achieve the following functions:

[0108] Task Creation: Automatically generates production tasks based on customer orders, including information such as segment specifications, quantity, quality requirements, and delivery date. Taking a subway project as an example, the order requires the production of 1200 standard segments, delivered in batches of 120 segments each.

[0109] Mold matching: Establish a matching matrix between mold type and product specifications to ensure that mold size error is within ±2mm. The production management system can automatically check mold availability and prioritize the allocation of molds in good maintenance condition.

[0110] Priority calculation: A weighted scoring method is used, with urgency weighted at 40%, technical difficulty weighted at 30%, customer importance weighted at 20%, and inventory pressure weighted at 10%. Urgent projects with a total score > 80 points will be given priority for production.

[0111] In step S105: Monitor the mold status during the production process using an AI video monitoring system.

[0112] In this embodiment of the invention, starting from the beginning of the production line, an AI video monitoring system continuously monitors changes in the mold's state, enabling intelligent mold state recognition and production tracking. When the mold enters the camera's monitoring range, the AI ​​video monitoring system automatically captures the current frame image, preprocesses it, and inputs it into a trained mold state recognition model for state determination. The output of the mold state recognition model includes: mold state category (empty, loaded with reinforcing bars, loaded with concrete, etc.) and a confidence score. A confidence threshold of 0.85 is set; when the confidence score of the recognition result is greater than this threshold, the mold state is confirmed.

[0113] In embodiments of the present invention, to avoid misjudgment, a continuous frame verification mechanism is adopted. A state change is only confirmed when the recognition results of three consecutive frames are consistent. The specific process of the mold state recognition algorithm is as follows: Figure 3 As shown. When the mold is detected to be in a non-idle state, the production management system automatically queries the production task plan for the day, matches the corresponding prefabricated part task according to the mold code and type, and initiates the full-process production tracking record for the prefabricated part.

[0114] For example, during the operation of the AI ​​video surveillance system, images are continuously acquired at a frequency of 25fps. When the mold enters the monitored area, the AI ​​video surveillance system executes the following processing flow:

[0115] Image acquisition: The camera captures 1920×1080 resolution images in real time and transmits them to the processing server via the RTSP protocol; Preprocessing: The image is resized (224×224 pixels), normalized (pixel value range 0-1), and mean centered, etc.

[0116] Model inference: The preprocessed image is input into the trained mold state recognition model (CNN model) to obtain the probability distribution of the four state categories. A confidence threshold of 0.85 is set, and the recognition result is confirmed when the maximum probability value is greater than this confidence threshold; State verification: A sliding window mechanism is used, and the mold state change is only confirmed when the recognition results are consistent for three consecutive frames to avoid misjudgment caused by changes in lighting or occlusion;

[0117] Task matching: When the mold is identified as being loaded with steel cages, the RFID reader automatically reads the mold's RFID information, queries the production tasks for the day, matches the corresponding prefabricated component orders, and initiates the production tracking process.

[0118] In step S106: The production process is tracked in real time through the RFID system, and production data is recorded.

[0119] In embodiments of this invention, an RFID reader continuously scans RFID chip signals within the working range at a frequency of 2-5Hz for real-time positioning and automatic recording of production processes. When an RFID chip on a mold is detected entering the reading range, the RFID reader automatically acquires the mold identification information stored on the chip and records the current timestamp (accurate to the second) and the reader's location code. Based on the RFID reader's location, the current production process is automatically determined: a reader at the mold table corresponds to the casting process, a reader at the steam curing chamber entrance corresponds to the start of steam curing, a reader at the steam curing chamber exit corresponds to the end of steam curing, a reader in the water curing tank corresponds to the water curing process, and a reader in the storage area corresponds to finished product warehousing. Through continuous location reading and recording, the movement trajectory and process progress of each precast component on the production line can be tracked in real time, and the start time, end time, and duration of each process can be automatically calculated, thereby generating a complete production history for easy querying and traceability.

[0120] Continuing with the example above, the RFID reader scans the working area at a frequency of 5Hz. When a chip signal is detected, the following process is executed: Signal acquisition: The reader sends a 13.56MHz radio frequency signal to activate the passive RFID chip on the mold and obtain the mold identification information stored in the chip;

[0121] Data analysis: The acquired information is analyzed to extract the mold code and type, and then matched and verified with the mold files in the database;

[0122] Location Recording: Determine the current position of the mold based on the reader number, record the timestamp (accurate to the second), and establish the mold movement trajectory. Process Judgment: The system automatically determines the current production process based on the reader position.

[0123] R220-002 to R220-006: Reinforcing cage installation procedures;

[0124] R220-007: Steam curing process begins;

[0125] R220-008 to R220-010: During steam curing;

[0126] R220-011: Steam curing process completed;

[0127] R220-012 to R220-014: Water conditioning process;

[0128] R220-015 to R220-016: Finished products are put into storage.

[0129] In embodiments of the present invention, a production management visualization interface is established, such as... Figure 4This demonstrates the visual interface effect of production management according to an embodiment of the present invention. The specific steps of visual management include: establishing a three-dimensional digital production line model to display the position and status of each mold in real time; designing a production management dashboard, including information such as real-time output, plan completion rate, and equipment operating status; and providing a mobile application so that managers can view production progress and abnormal situations at any time.

[0130] In the embodiments of the present invention, after a three-month trial run, it was confirmed that the above-mentioned production information collection method based on AI video recognition and RFID positioning has achieved significant results in actual production, such as... Figure 5 As shown:

[0131] Data accuracy: The method of the present invention achieves an identification accuracy of 94.2% and an RFID reading success rate of 99.1% in actual production environments, with overall data accuracy improved by 22 percentage points compared to manual recording.

[0132] Real-time performance: From the change in mold status to the completion of recording, the average time is 3.2 seconds, which is a significant improvement compared to manual recording (average delay of 4.5 hours).

[0133] Labor costs: The number of on-site record-keeping personnel was reduced by 3, resulting in monthly labor cost savings of 24,000 yuan and annual cost savings of 288,000 yuan. Production optimization: Through precise process time statistics, the steam curing process was identified as the production bottleneck, and optimization improved overall production efficiency by 18%.

[0134] The production information collection method based on AI video recognition and RFID positioning provided by this invention has the following advantages:

[0135] (1) Significantly improved data collection accuracy: Through the dual technologies of AI video recognition and RFID positioning, the production process is objectively and accurately recorded. Compared with manual recording, the data accuracy rate has increased from 70-80% to over 95%, effectively avoiding recording errors and omissions caused by human factors.

[0136] (2) Significantly improved the real-time performance of information collection: It can perceive changes in mold status and position movement in real time, and the data delay time is shortened from several hours in the traditional way to seconds, providing timely and accurate data support for production scheduling decisions and improving the response speed of production management.

[0137] (3) Significantly reduced labor costs: Automated information collection methods reduce the workload of on-site recorders. Each production line can save 2-3 full-time recorders, reducing labor costs by more than 30%, while avoiding the problem of reduced record quality caused by personnel fatigue and lack of concentration.

[0138] (4) Full-process traceability has been achieved: a complete data chain from raw materials to finished products has been established. Each prefabricated part has a detailed production history, including information such as molds used, production time, process parameters, and quality inspection results, which provides a reliable basis for tracing quality issues and optimizing production.

[0139] (5) Provide decision support data: Through the accumulation and analysis of a large amount of production data, production bottlenecks can be identified, process parameters can be optimized, and equipment maintenance needs can be predicted, providing a scientific basis for continuous improvement of production management, which is expected to improve overall production efficiency by 15-20%.

[0140] (6) It has innovative technology integration: For the first time, AI video recognition technology and RFID positioning technology are deeply integrated and applied to the production of precast segments, realizing the combination of visual recognition status judgment capability and radio frequency positioning precise tracking capability, solving the technical problem that a single technical means cannot fully cover the production information collection needs.

[0141] (7) A breakthrough in intelligence: Instead of relying on manual judgment or simple sensor triggering, this invention achieves fully automated status recognition and information recording. The AI ​​mold status recognition model can accurately identify the mold status in complex production environments, greatly improving the intelligence level of production information collection.

[0142] Example 2

[0143] This embodiment, based on embodiment 1, employs more advanced technical configurations and optimized algorithms to further optimize the information collection method.

[0144] In terms of visual recognition, this embodiment uses the YOLOv5s object detection model instead of the classification model to achieve accurate localization and recognition of the mold state. The YOLOv5s object detection model consists of three parts: Backbone (CSPDarknet53), Neck (PANet), and Head (detection head), which can simultaneously detect multiple molds and identify their states. The training dataset was expanded to 15,000 images, adding scenarios such as complex lighting, partial occlusion, and multiple molds in the same frame. Data annotation was performed using the YOLO format, including bounding box coordinates and class labels. The model training parameters were optimized: input size 640×640 pixels, batch size 16, learning rate 0.01, using a cosine annealing scheduler, and training for 300 epochs. The final model achieved a recognition accuracy of 97.3% and a detection speed of 45 FPS on the test set.

[0145] In terms of RFID reading, this embodiment upgrades to a UHF band RFID reader, operating at a frequency of 860-960MHz, extending the reading distance to 8-12 meters, enabling automatic identification over a wider range. Using the Alien ALR-9900+ reader and ALN-9662 tag provides stronger resistance to metal interference. Simultaneously, an RFID antenna array is added, deploying a 4-antenna array at each monitoring point, achieving precise mold positioning with a positioning accuracy of ±30cm through phase difference ranging technology.

[0146] In edge computing, this embodiment deploys an NVIDIA Jetson AGX Xavier edge computing device, integrating AI inference and RFID data processing functions to achieve localized real-time processing and reduce network latency to less than 5ms. TensorRT (a high-performance deep learning inference SDK that includes a deep learning inference optimizer and runtime environment, providing low latency and high throughput for deep learning inference applications) is used to optimize the model, improving inference speed by 60% and supporting parallel processing of multiple video streams. Performance in all aspects is improved.

[0147] AI recognition accuracy improved from 96.8% to 98.1%;

[0148] RFID read success rate: increased from 99.1% to 99.7%;

[0149] System response time: improved from 3.2 seconds to 1.8 seconds;

[0150] Simultaneously monitor the number of molds: increase from 15 to 25;

[0151] Positioning accuracy: improved from ±2 meters to ±30 cm.

[0152] Figure 6 The flowchart below shows the production information collection method based on AI video recognition and RFID positioning according to Embodiment 2 of the present invention. Figure 6 The second embodiment of the present invention will be described in detail.

[0153] The production information collection method based on AI video recognition and RFID positioning according to embodiments of the present invention includes the following steps:

[0154] S601 inputs multiple video streams to the edge computing device for data preprocessing.

[0155] S602, using the YOLOv5s object detection model to perform parallel detection of preprocessed multi-channel video data, completes multi-mold state synchronous recognition. This step also includes: using TensorRT to optimize the YOLOv5s object detection model, improving the model's inference speed.

[0156] S603 scans the work area using a UHF RFID array.

[0157] S604 uses a phase difference positioning algorithm to calculate the precise position of multiple molds based on the data obtained from UHF RFID array scanning.

[0158] S605 integrates precise position data from multiple molds to confirm mold status and update it in real time.

[0159] S606 synchronizes mold status data to the cloud to support intelligent decision-making.

[0160] In embodiments of the present invention, a fusion processing mechanism for multi-source heterogeneous data is established, such as... Figure 7 As shown, by unifying and analyzing video image data, RFID radio frequency data, and production task data, a complete production information profile is formed, which cannot be achieved by a single technical path. Therefore, the production information collection method based on AI video recognition and RFID positioning of this invention has the innovation of data fusion processing.

[0161] Example 3

[0162] This embodiment uses the adaptation of various types of precast component production lines as an example to illustrate the application flexibility of the production information collection method based on AI video recognition and RFID positioning in different scale production environments.

[0163] Application Scenario 1: Small-scale precast component production line. For example, a precast box girder production line of a precast concrete company has a daily production capacity of 15 pieces, is equipped with 8 sets of molds, and has a relatively simplified production process.

[0164] Figure 8 This is a configuration diagram for application scenario 1 according to embodiment 3 of the present invention, as shown below. Figure 8 As shown, the configuration adjustments for this production line are as follows:

[0165] 1) Simplified configuration of the AI ​​video recognition system: Considering the cost control needs of small-scale production lines, two Dahua IPC-HFW2431S cameras with a resolution of 1920×1080 pixels are used. The MobileNetV2 lightweight model is used, with a model size of only 14MB, making it suitable for running on low-configuration hardware.

[0166] The edge computing device uses a Raspberry Pi 4B to process data, costing approximately 500 yuan, which can meet the AI ​​inference needs of small-scale production. The model is quantized to 8-bit precision, and the inference speed is 20 FPS.

[0167] 2) Simplified RFID System Deployment: Using a low-cost HF RFID system, the number of readers is reduced to 6, covering key process nodes. A timed scanning mode is adopted, scanning once every 30 seconds, reducing system power consumption and cost.

[0168] 3) Data processing optimization: Production data is stored using the lightweight database SQLite, reducing server configuration requirements. A simplified web-based management interface has been developed, supporting mobile access.

[0169] After the adjustment, the cost-benefit ratio is as follows:

[0170] Total system investment: 65% lower than standard configuration;

[0171] Implementation period: Deployment and debugging to be completed in 2 weeks;

[0172] Operation and maintenance costs: Annual maintenance fee of 12,000 yuan;

[0173] Investment payback period: 8 months.

[0174] Application Scenario 2: Multi-line Parallel Production Application. Taking a large prefabrication plant with 3 parallel production lines as an example, the collection and management of production information adopts the following distributed architecture:

[0175] Each production line is independently equipped with an AI video surveillance system and an RFID system;

[0176] The central data center processes all production line data in a unified manner;

[0177] Cross-line resource scheduling optimization enables dynamic allocation of molds and tasks;

[0178] Establish a plant-wide production coordination mechanism.

[0179] This distributed architecture has the following technical features: it supports parallel monitoring of up to 8 production lines; cross-line data correlation analysis to identify systemic bottlenecks; and pooled management of mold resources to improve utilization by 15%.

[0180] Application Scenario 3: Applications adapted to special environments, such as production environments with high temperature, high humidity, and strong electromagnetic interference. Configuration requirements are as follows: The camera adopts an explosion-proof design with an IP68 protection rating;

[0181] RFID systems use anti-metal tags and high-power readers;

[0182] Add environmental monitoring sensors to record parameters such as temperature and humidity;

[0183] Data transmission uses fiber optic communication, which enhances anti-interference capabilities.

[0184] In embodiments of the present invention, the applicable environment includes: an operating temperature range of -20°C to +80°C;

[0185] The relative humidity is between 5% and 95% RH;

[0186] It has electromagnetic compatibility and complies with GB / T 17626 standard.

[0187] Application Scenario 4: Enhanced Quality Control. In this application scenario, quality monitoring of the production process is achieved by integrating quality inspection equipment. The configuration adjustments are as follows:

[0188] Add industrial CT scanning equipment to detect internal defects;

[0189] Deploy a laser measurement system to inspect geometric dimensional accuracy;

[0190] An integrated strength detection sensor is used to monitor the maintenance process;

[0191] Establish a correlation model between quality and process parameters.

[0192] In embodiments of the present invention, quality control indicators include: dimensional accuracy detection (e.g., ±1mm), internal defect identification (e.g., φ>2mm), and strength prediction accuracy (e.g., >92%).

[0193] The effectiveness of the production information collection method based on AI video recognition and RFID positioning of this invention is demonstrated through several experiments. Experiment 1: Comparison test of recognition accuracy. The test conditions were as follows: In an actual production environment, production data from 7 consecutive days was selected, totaling 315 mold state change events for statistical analysis. The statistical results are shown in Table 1:

[0194] Table 1

[0195] Method type Correctly identify numbers Number of misjudgments Number of missed judgments accuracy Recall rate F1 score Manual records 251 38 26 79.7％ 88.9％ 84.0％ Method of the present invention 304 7 4 96.5％ 98.7％ 97.6％

[0196] Test results show that the method provided by this invention improves the recognition accuracy by 16.8 percentage points compared with manual recording, and reduces the false judgment rate by 82.6%.

[0197] Experiment 2: System Response Time Test. The test scenario involved recording and comparing the time taken for a mold to move from one process to the next. The test results are shown in Table 2.

[0198] Table 2

[0199]

[0200]

[0201] Experiment 3: Impact Analysis on Production Efficiency. As shown in Table 3, this experiment compared production data for six months before and after implementation:

[0202] Table 3

[0203] efficiency indicators Before implementation After implementation Increase Daily output 38.2 pieces 45.1 pieces +18.1％ Equipment utilization rate 72.5％ 86.3％ +13.8％ Process connection time 25 minutes 12 minutes -52％ First pass rate 94.2％ 97.8％ +3.6％

[0204] Experiment 4: Cost-Benefit Analysis. As shown in Table 4, this experiment is based on a production scale of 15,000 pieces per year:

[0205] Table 4

[0206] Cost items Annual expenses (ten thousand yuan) Savings (ten thousand yuan) Labor cost savings - 28.8 Reduced quality loss - 15.2 Efficiency Improvement Benefits - 42.5 System investment cost 35.0 - Operation and maintenance costs 8.5 - Net income - 42.0

[0207] The embodiments of this application provide fault analysis and corresponding improvement measures. Several common fault types and solutions are as follows: 1) Decreased AI recognition accuracy. The fault phenomenon is that the recognition accuracy drops below 85% under specific lighting conditions. The reason is that the training dataset lacks samples under extreme lighting conditions, resulting in insufficient model generalization ability. Improvement measures include: adding 2000 training samples under extreme lighting conditions; using data augmentation technology to simulate different lighting environments; implementing an online learning mechanism to continuously optimize the model based on production feedback; and adding auxiliary lighting equipment to improve the shooting environment. The improvements will bring the following technical effects: the recognition accuracy will recover to over 96%, and the system robustness will be improved.

[0208] 2) RFID Read Failure. The fault symptom is that the RFID read success rate drops below 90% in environments with dense metal elements. This is because metal materials such as steel bars interfere with the 13.56MHz radio frequency signal, thus affecting signal propagation. The improvements include: using anti-metal RFID tags to improve anti-interference capabilities; optimizing antenna layout and using directional antennas to reduce multipath interference; increasing reader power to expand the effective reading range; and implementing a multiple-read verification mechanism to improve reliability. Through these improvements, the RFID read success rate increased to 99.5%, and system stability was significantly improved.

[0209] 3) Data transmission latency. The fault symptom was a data transmission latency exceeding 5 seconds under high concurrency. The cause was insufficient network bandwidth and a bottleneck in data processing capabilities. The following improvements were implemented: upgrading network equipment to Gigabit Ethernet; adopting an edge computing architecture to reduce data transmission volume; optimizing data compression algorithms to reduce transmission load; and implementing a data caching mechanism to improve system response speed. Ultimately, the data transmission latency was controlled to within 1 second, effectively improving the real-time performance of information acquisition.

[0210] 4) System Integration Compatibility. This type of failure manifests as data format incompatibility when interfacing with existing ERP systems. The cause is the lack of standardized data interfaces and effective data conversion mechanisms between different systems. Improvement measures include: developing standardized data interfaces to support multiple data formats; establishing data mapping rules for automatic format conversion; providing API interfaces to facilitate third-party system integration; and developing data exchange standards to ensure system compatibility. Through the combination of these improvements, the system compatibility issue is completely resolved, enabling seamless integration with over 95% of mainstream ERP systems.

[0211] Example 4

[0212] In the embodiments of this application, a production information collection system based on AI video recognition and RFID positioning is also provided. Figure 9 This is a schematic diagram of the production information collection system based on AI video recognition and RFID positioning according to Embodiment 4 of this application. Figure 9 As shown, the production information collection system based on AI video recognition and RFID positioning of this application includes:

[0213] The production task management module 802 is used to arrange daily production tasks based on the entered mold information and production line information, according to production order requirements, and to establish the correspondence between mold types and prefabricated parts. The mold information includes mold code, name, and applicable prefabricated part types. The production line information includes production line name, production mold table information, steam curing chamber information, water curing chamber information, storage area information, production process information, and production equipment information.

[0214] The RFID positioning module 803 is used to collect mold position information and prefabricated component position information to obtain RFID data; process the RFID data to obtain positioning data of the mold and prefabricated component and send it to the data platform 806. The RFID positioning module 803 includes: an RFID chip 809 and an RFID reader 810, which are respectively deployed at various key nodes on the production line; and an RFID data processing module 811, used to process the RFID data acquired by the RFID reader.

[0215] The AI ​​video recognition module 805 processes image data and identifies the mold state from the processed image data based on a deep learning-based mold state recognition model, then sends the mold state data to the data platform 806. The image data is collected by the camera device 804.

[0216] The data platform 806 integrates and processes positioning data sent by the RFID positioning module 803 and mold status data sent by the AI ​​video recognition module 805. Through data analysis, it obtains the latest production information and updates it in real time to the production management module 807, providing decision support. For example, it uses time series analysis to statistically analyze the time taken for each production process to identify bottlenecks; it establishes a correlation model between precast component quality and production parameters, including curing time, curing temperature, and concrete strength; and it sets up an anomaly warning mechanism that automatically alarms when a process takes 120% more than the standard time.

[0217] The production management module 807 is used to collect production information, generate a complete production history, and provide it to the visualization management platform 808. The visualization management platform 808 is used to visualize the production history and provide functions such as production progress query and problem tracing.

[0218] In the embodiments of this application, the process by which the AI ​​video recognition module 805 identifies the mold state based on the recognition model is as follows: Figure 10 As shown. In the embodiments of this application, the mold state recognition model includes: an input layer, a feature extraction layer, a first convolutional layer, a first pooling layer, a second convolutional layer, a third convolutional layer, a fully connected layer, and a classification output layer.

[0219] It will be understood by those skilled in the art that the above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for collecting production information based on AI video recognition and RFID positioning, characterized in that, include: Based on images captured by AI video cameras deployed on the production line, a mold state recognition model based on deep learning is constructed and trained. Establish a mold information database and bind RFID chips to the molds; deploy RFID readers on the production line to locate the mold and preform positions; schedule daily production tasks and establish the correspondence between mold types and preforms; determine the mold load status through the mold status recognition model. When the mold status is detected to be non-empty, the corresponding type of prefabricated part information in the production task is extracted, and the production process tracking program is started. Based on the RFID chip data read in real time by the RFID reader and the location of the RFID reader, the production process information and storage location information of the prefabricated parts are recorded. The steps of constructing and training a deep learning-based mold state recognition model include: collecting image samples of empty molds, molds loaded with steel cages, and molds loaded with concrete under different lighting conditions; labeling the image samples and establishing mold state classification labels; constructing a mold state recognition model using a convolutional neural network; and training the model according to a set learning rate and number of iterations until the model recognition accuracy reaches a threshold, thereby obtaining a trained mold state recognition model. The step of deploying RFID readers on the production line to locate the mold position and the preform position further includes: installing RFID readers at the entrance of each production mold table, with the reading range covering the mold table feeding area; installing RFID readers at the entrance and exit of the steam curing chamber to monitor the time when the preform enters and exits the steam curing chamber; installing RFID readers at the entrance and exit of the water curing chamber to record the time node of the water curing process; and installing RFID readers at the entrance of each channel in the storage area to locate the final storage position of the preform.

2. The production information collection method based on AI video recognition and RFID positioning according to claim 1, characterized in that, The steps of constructing and training the mold state recognition model based on deep learning further include: the number of image samples is not less than 5,000; the number of network layers of the mold state recognition model is 16, including an input layer, 3 convolutional layers, 2 pooling layers, 1 fully connected layer and an output layer.

3. The production information collection method based on AI video recognition and RFID positioning according to claim 1, characterized in that, The step of establishing a mold information database and binding an RFID chip to the mold further includes: using a 12-digit number to encode the mold, with the first 4 digits representing the production line number, the middle 4 digits representing the mold type, and the last 4 digits representing the mold serial number; and writing the mold code, production date, and applicable prefabricated part specifications into the RFID chip.

4. The production information collection method based on AI video recognition and RFID positioning according to claim 1, characterized in that, The steps of scheduling daily production tasks and establishing the correspondence between mold types and prefabricated parts include: developing a daily production plan based on production needs, including the type, quantity, and quality requirements of prefabricated parts; establishing a matching table between molds and prefabricated parts to ensure that mold specifications match prefabricated part specifications; and setting task priorities, with urgent tasks having a priority weight of 0.8-1.0 and regular tasks having a priority weight of 0.3-0.

7.

5. The production information collection method based on AI video recognition and RFID positioning according to claim 1, characterized in that, The step of determining the mold load state through the mold state recognition model further includes: setting a confidence threshold of 0.85, and confirming the corresponding mold state when the confidence of the recognition result is greater than the confidence threshold; establishing a correspondence between the mold state and the production process as follows: an empty mold corresponds to the preparation process, and a mold for loading the rebar cage corresponds to the rebar binding completion process.

6. The production information collection method based on AI video recognition and RFID positioning according to claim 5, characterized in that, Also includes: When the recognition results of three consecutive frames are consistent, the change in mold state is confirmed to avoid misjudgment.

7. The production information collection method based on AI video recognition and RFID positioning according to claim 1, characterized in that, The step of recording the production process information and storage location information of the prefabricated part based on the RFID chip data read in real time by the RFID reader and the location of the RFID reader further includes: when the RFID reader detects the RFID chip signal, automatically recording the timestamp, location information, and mold code; establishing a mapping relationship between the RFID reader and the production process, so as to automatically determine the current stage of the production process based on the location of the RFID reader.

8. The production information collection method based on AI video recognition and RFID positioning according to claim 1, characterized in that, It also includes a data fusion processing step: fusion processing is performed based on the identification results of the mold status identification model and the data read by the RFID reader, and a prefabricated component production status model is established to update the production progress of each prefabricated component in real time; Statistics on daily output, process time, and mold utilization rate are compiled to generate production reports.

9. The production information collection method based on AI video recognition and RFID positioning according to claim 8, characterized in that, The data fusion processing steps also include: using time series analysis to statistically analyze the production process time and identify production bottlenecks; establishing a correlation model between precast component quality and production parameters; the production parameters include steam curing time, steam curing temperature, and concrete strength; setting an anomaly warning mechanism; the anomaly warning mechanism is to automatically alarm when the process time exceeds 120% of the standard time.

10. The production information collection method based on AI video recognition and RFID positioning according to claim 1, characterized in that, It also includes steps for visual management: establishing a 3D digital production line model to display the location and status of each mold in real time; designing a production management dashboard, including real-time output, plan completion rate, and equipment operating status; and providing a mobile application so that managers can view production progress and abnormal situations at any time.

11. A production information collection system based on AI video recognition and RFID positioning, characterized in that, include: The production task management module is used to arrange daily production tasks according to order requirements and establish the correspondence between mold types and prefabricated parts; The RFID positioning module is used to collect mold location information and preform location information through RFID readers, perform RFID data processing, obtain positioning data of the mold and preform, and send it to the data platform. The RFID readers are deployed as follows: RFID readers are installed at the entrance of each production mold table, and the reading range covers the mold table feeding area; RFID readers are installed at the entrance and exit of the steam curing chamber to monitor the time when the preform enters and exits the steam curing chamber. RFID readers are installed at the entrance and exit of the water curing room to record the time points of the water curing process; RFID readers are installed at the entrance of each passage in the storage area to locate the final storage location of the prefabricated components. The AI ​​video recognition module is used to collect and process image data of the mold, and identify the mold status based on a deep learning-based mold status recognition model, sending the mold status data to the data platform. The mold status recognition model is constructed in the following way: collecting image samples of empty molds, molds loaded with steel cages, and molds loaded with concrete under different lighting conditions; labeling the image samples and establishing mold status classification labels; constructing the mold status recognition model using a convolutional neural network; and training the model according to the set learning rate and number of iterations until the model recognition accuracy reaches a threshold. The data platform is used to process the positioning data sent by the RFID positioning module and the mold status data sent by the AI ​​video recognition module to obtain the corresponding production information and update it to the production management module in real time. The production management module is used to collect the production information, generate a complete production history, and provide it to the visualization management platform; the visualization management platform is used to visualize the production history.

Images