Intelligent packaging system for full-sealed sample collection tank

Abstract

The invention relates to the technical field of automation, in particular to an intelligent packaging system for a fully-sealed sample collecting tank. The system comprises a sample collection tank transmission module, wherein one-way transmission of a sample collection tank is realized through a multi-station rotating table; the model training module adopts a deep learning algorithm based on Vision Transform, constructs a sample collection tank packaging and positioning model by using historical data, and identifies a cover locking state and a tank opening position of the sample collection tank; the packaging judgment module is used for judging whether the sample collection tank needs to be packaged or not, and invalid operation is avoided; the packaging execution module is used for executing lock cover packaging by utilizing a mechanical arm according to a model positioning result; the locking control module is used for locking the sample collecting tank and confirming the state; in addition, the system is further provided with a wireless power supply module to intelligently monitor and manage the electric quantity of the lock cover. On the whole, the problem that a traditional packaging mode is low in packaging efficiency is effectively solved, and the packaging accuracy and safety of the full-sealed sample collecting tank are improved.

Description

Technical Field

[0001] The invention relates to the field of automation technology, and in particular to an intelligent packaging system for a fully sealed sample collecting tank. Background Art

[0002] In the existing fully sealed sample can packaging process, there are problems of insufficient automation and low packaging efficiency. The traditional method relies on manual operation to package and lock the sample can, which not only increases labor costs, but is also easily affected by human factors and cannot quickly respond to the processing needs of large quantities of sample cans; in addition, manual packaging also faces safety risks, such as possible contamination of samples during operation. Therefore, there is an urgent need for an intelligent packaging system that can automatically identify the status of the sample can, accurately perform packaging operations, and provide real-time feedback on packaging results, so as to improve packaging efficiency, ensure packaging quality, and reduce operational risks. Summary of the invention

[0003] In view of the above-mentioned technical deficiencies, an object of the present invention is to provide an intelligent packaging system to solve the problem of low packaging efficiency in the prior art.

[0004] In order to solve the above technical problems, the present invention adopts the following technical solutions: In a first aspect, an intelligent packaging system for a fully sealed sample collection tank comprises: Sample collection tank transmission module: used for one-way transmission of sample collection tanks, including waiting area, packaging area and completion area; the waiting area is used to wait for packaging, the packaging area performs packaging operations, and the completion area stores packaged sample collection tanks; the sample collection tanks are moved between the areas through a multi-station rotary table; the first infrared sensor is configured at the edge of the packaging area near the waiting area, and the second infrared sensor is configured at the edge of the packaging area near the completion area; Model training module: Using the deep learning algorithm based on Vision Transformer, the packaging positioning model of the sample collection tank is constructed by training historical data. The model is used to identify the lock cover status and tank mouth position of the sample collection tank. After training, the model is deployed in the packaging judgment module and the packaging execution module. Packaging judgment module: The camera is used to capture the image of the sample collection tank in the packaging area, and the deployed sample collection tank packaging positioning model is used to determine whether there is a lock cover on the sample collection tank; if the judgment result is that there is no lock cover, the sample collection tank enters the packaging execution module to perform the lock cover packaging operation; if the judgment result is that there is a lock cover, the sample collection tank directly enters the locking control module to perform the locking operation; Packaging execution module: Use the sample collection tank packaging positioning model to locate the can mouth of the sample collection tank, and use the robotic arm to transport the lock cover to the can mouth to perform the packaging operation, so that the lock cover and the can mouth are tightly fitted; Locking control module: sends a locking signal through the communication interface between the robotic arm and the lock cover, controls the lock cover to complete the locking action, and confirms the completion of the locking operation by receiving a feedback signal.

[0005] Preferably, in a possible implementation manner of the first aspect, the PLC controls the rotation of the workstation according to the feedback signals of the first infrared sensor and the second infrared sensor to ensure that there is only one sample collecting tank in the packaging area. The specific control logic is as follows: The PLC has a built-in first infrared signal counter and a second infrared signal counter. When the first infrared signal count is 1 more than the second infrared signal count, the PLC sends a station rotation stop command to the multi-station rotary table controller, and the multi-station rotary table stops transmitting. When the packaging operation starts, the first infrared signal count and the second infrared signal count are set to 0; When the first sample collecting tank enters the packaging area, the sample collecting tank triggers the first infrared sensor, and the first infrared signal count is increased by 1. At this time, the first infrared signal count is 1 more than the second infrared signal count. The PLC sends a station rotation stop instruction to the multi-station rotary table controller, and the multi-station rotary table stops transmission, so that the sample collecting tank is packaged in the packaging area; For the continuously transmitted sample collection tanks, each time a sample collection tank enters and completes its task in the packaging area, it leaves the packaging area to trigger the second infrared sensor, and the second infrared signal count is increased by 1. This signal serves as confirmation that the processing of the previous sample collection tank has been completed; When the sample collecting can to be packaged enters the packaging area, the first infrared signal count is increased by 1. At this time, the first infrared signal count is 1 more than the second infrared signal count. There is only one sample collecting can in the current packaging area and the previous sample collecting can has left the packaging area. The PLC sends a station rotation stop command to the multi-station rotary table controller, and the multi-station rotary table stops transmission.

[0006] Preferably, in a possible implementation manner of the first aspect, the Vision Transformer-based deep learning algorithm network structure includes an input layer, a Transformer Encoder layer, and a task-specific layer, which are respectively used for image preprocessing, feature extraction, classification, and positioning output; Input layer: resizes the image to a preset size, splits it into blocks of preset size, flattens it and embeds it into the specified dimensions, and appends classification tags to support classification; Transformer Encoder layer: stacks multiple encoder layers, each of which contains multi-head self-attention, layer normalization, and feed-forward networks, combined with residual connections and dropout to optimize training and prediction; Task-specific layer: It is divided into classification and regression. The classification head extracts the classification label and maps it to the number of categories, and predicts it through softmax normalization. The regression head performs a pooling operation on the image block embedding vector and then maps it to the regression target.

[0007] Preferably, in a possible implementation manner of the first aspect, the Vision Transformer-based deep learning algorithm further includes a loss function, an optimizer, and a data loader; Loss function: The classification task uses cross entropy loss for image classification; the regression task uses mean square error loss to approximate continuous value prediction; Optimizer: Adopt Adam optimizer to accelerate model convergence by adaptively adjusting the learning rate of each parameter; Data Loader: Use DataLoader to achieve batch processing, data shuffling, and multi-process acceleration.

[0008] Preferably, in a possible implementation manner of the first aspect, the historical data is historical sample collecting jar mouth data, including photos of the sample collecting jar mouth with a locked cover and photos of the sample collecting jar mouth without a locked cover.

[0009] Preferably, in a possible implementation manner of the first aspect, the encapsulation operation specifically includes: Control the camera to take photos of the mouth of the sample collecting tank; Use the sample collection tank packaging positioning model and the sample collection tank mouth photo to locate the tank mouth; The robot arm takes the lock cap and places it on the can mouth for pressurized packaging.

[0010] Preferably, in a possible implementation of the first aspect, after the locking control module confirms that the locking operation is completed, it sends a station rotation start instruction to the multi-station rotary table controller to control the multi-station rotary table to transfer the sample collection tank to the completion area, and synchronously transfer the next sample collection tank to the packaging area.

[0011] Preferably, in a possible implementation manner of the first aspect, a wireless power supply module is further included, which is used to monitor the power of the lock cover and perform a lock cover charging operation.

[0012] Preferably, in a possible implementation manner of the first aspect, the wireless power supply module specifically includes: The remaining power of the current lock cover is read through the communication interface between the front gripper of the robot arm and the lock cover; When it is detected that the remaining power of the lock cover is lower than the preset power threshold, the charging process is triggered; The robotic arm moves the lock cover to the preset charging area for wireless charging; During the charging process, the lock cover maintains communication and receives charging status feedback; When the battery is fully charged, charging stops automatically and the robot arm resumes packaging operations.

[0013] The beneficial effects of the present invention are: through automated transmission, deep learning positioning, intelligent judgment of packaging status, accurate execution of packaging and locking operations, and equipped with wireless power supply function, the packaging efficiency and quality are significantly improved, and the operational risks and costs are reduced. The system realizes intelligent management of the entire process from waiting to completion, ensuring the accuracy and safety of the packaging process. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0015] Figure 1 A structural schematic diagram of an intelligent packaging system for a fully sealed sample collection tank is provided for this application.

[0016] Explanation of the accompanying drawings: 1-sample tank transmission module, 2-model training module, 3-packaging judgment module, 4-packaging execution module, 5-locking control module. DETAILED DESCRIPTION

[0017] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0018] Embodiment 1: like Figure 1 As shown, the present invention provides an intelligent packaging system for a fully sealed sample collection tank, comprising: Sample collecting can transmission module 1: used for one-way transmission of sample collecting cans, including waiting area, packaging area and completion area; the waiting area is used to wait for packaging, the packaging area performs packaging operations, and the completion area stores packaged sample collecting cans; the sample collecting cans are moved between each area through a multi-station rotary table; the first infrared sensor is configured at the edge of the packaging area close to the waiting area, and the second infrared sensor is configured at the edge of the packaging area close to the completion area.

[0019] Specifically, in this embodiment, the PLC uses the Siemens S7-200 series, extends the Modbus RTU communication module, and supports the RS-485 interface; the infrared sensor uses the Banner Engineering Q4X series; the multi-station rotary table uses a motor system driven by a frequency converter, and the frequency converter has a built-in Modbus RTU communication interface; an RS-485 communication cable is used to connect the communication interface of the PLC and the multi-station rotary table controller as well as the PLC and the infrared sensor.

[0020] Write the Modbus RTU communication initialization code in the PLC program, set the baud rate to 9600bps, the data bit to 8 bits, the stop bit to 1 bit, the parity bit to even parity, and establish a connection with the multi-station rotary table controller.

[0021] Every 1 second, use the Modbus RTU read command to read the current speed value from the holding register 40001 of the multi-station rotary table. When the multi-station rotary table needs to be stopped, set the value of the input register 30001 to a specific stop command code and send it to the multi-station rotary table controller via the Modbus RTU write command.

[0022] Connect the signal line of the infrared sensor to the digital input module of the PLC, write the PLC program in the TIA Portal, and determine whether the infrared sensor detects an object by reading the status of the digital input module.

[0023] When the first infrared sensor and the second infrared sensor are not blocked by an object, the PLC data is acquired as a low level, that is, 0; when they are blocked, the PLC data is acquired as a high level, that is, 1.

[0024] When the sample collecting tank triggers the first infrared sensor, that is, the data of the first infrared sensor changes from 1 to 0, the first infrared signal count in the PLC increases by 1. When the sample collecting tank triggers the second infrared sensor, that is, the data of the second infrared sensor changes from 1 to 0, the second infrared signal count in the PLC increases by 1. When the first infrared signal count is 1 more than the second infrared signal count, the PLC sends a station rotation stop command to the multi-station turntable controller, and the multi-station turntable stops transmission.

[0025] When the packaging operation starts, the first infrared signal count and the second infrared signal count in the PLC are set to 0.

[0026] When the first sample collecting tank enters the packaging area, the sample collecting tank triggers the first infrared sensor, and the first infrared signal count is increased by 1. At this time, the first infrared signal count is 1, and the second infrared signal count is 0. The first infrared signal count is 1 more than the second infrared signal count. The PLC sends a station rotation stop instruction to the multi-station rotary table controller at the moment when the first infrared signal count is 1 more than the second infrared signal count, and the multi-station rotary table stops transmission; For continuously transmitted sample collecting tanks, when the nth sample collecting tank enters and completes its task in the packaging area, it leaves the packaging area to trigger the second infrared sensor, and the second infrared signal count is increased by 1. This signal serves as a confirmation that the processing of the previous sample collecting tank has been completed. At this time, the second infrared signal count is n.

[0027] At this time, the n+1th sample collecting tank enters the packaging area, the first infrared signal count is increased by 1, the first infrared signal count is n+1, which is 1 more than the second infrared signal count. There is only one sample collecting tank in the current packaging area and the previous sample collecting tank has left the packaging area. The PLC sends a station rotation stop command to the multi-station rotary table controller when the first infrared signal count is 1 more than the second infrared signal count, and the multi-station rotary table stops transmission.

[0028] Model training module 2: Using the deep learning algorithm based on Vision Transformer, a packaging positioning model for the sample collection can is constructed by training historical data. The model can accurately identify the lock cover status and can mouth position of the sample collection can. After training, the model is deployed in the packaging judgment module and the packaging execution module.

[0029] Specifically, model training and model inference are carried out in the control center. The control center is a core facility that integrates data collection, storage, processing, analysis, mining and visualization, providing comprehensive, accurate and timely data support and decision-making basis for the intelligent packaging system.

[0030] The historical data is the historical sample collecting tank mouth image data, including the tank mouth image of the sample collecting tank transmitted on the multi-station rotating table; the historical data annotation includes whether the sample collecting tank in the image has a locked cover and the position of the sample collecting tank mouth; the sample collecting tank packaging positioning model is used to determine whether there is a locked cover and the position of the sample collecting tank mouth.

[0031] The network structure of the deep learning algorithm based on Vision Transformer specifically includes: The input layer first resizes the image to a preset size and splits it into blocks of a preset size, which is 224×224 in this example. The image is then split into multiple 16x16 blocks. Each block is then flattened into a one-dimensional vector and embedded into the specified embedding dimension through a linear layer. In addition, to support classification tasks, a classification tag is added to the embedded feature sequence, which is used for subsequent classification predictions.

[0032] Transformer Encoder layer, which is composed of multiple Transformer encoder layers stacked together. Each encoder layer contains the following main components: multi-head self-attention, which allows the model to simultaneously focus on information at different positions in the input sequence; layer normalization, which normalizes the output of the self-attention layer to speed up training and stabilize the model; a feedforward network, which is a network consisting of two linear layers with a ReLU activation function in the middle; the encoder layer also uses residual connections and dropout to prevent overfitting and help gradient propagation.

[0033] The task-specific layers include the classification task head and the regression task head. The classification task head extracts the embedding vector of the classification tag from the feature sequence output by the encoder and maps it to the number of categories through a fully connected layer. Finally, it is normalized using the softmax function to obtain the classification prediction; the regression task head ignores the classification tag and performs average pooling or similar operations on the remaining image block embedding vectors to reduce the spatial dimension. Then, the pooled features are mapped to the regression target through a fully connected layer.

[0034] The training process of the deep learning algorithm based on Vision Transformer specifically includes: data loading, using DataLoader to batch load images and corresponding labels from the training data set; forward propagation, sending a batch of images into the Vision Transformer model, and calculating the loss between the model output and the true label; backpropagation, using the optimizer to calculate the gradient of the loss with respect to the model parameters, and updating these parameters to minimize the loss; iteration, repeating the above steps until the preset training round is reached or other stopping conditions are met.

[0035] When building a deep learning model based on Vision Transformer, in addition to the network structure itself, it also includes loss function, optimizer, and data loader components.

[0036] The loss function used for classification tasks is cross entropy loss, which calculates the difference between the category probability distribution output by the model and the true category label. By minimizing this difference, we can learn how to better classify the input image into the correct category; the regression task uses mean square error loss. The average of the squared difference between the model's predicted value and the true value is calculated, and by minimizing this value, more accurate predictions of continuous outputs are achieved.

[0037] The optimizer is a tool that adjusts model parameters to minimize the loss function during model training. Common optimizers include stochastic gradient descent, Adam, etc. This embodiment uses the Adam optimizer, which combines the advantages of gradient descent and RMSprop optimizers, and adjusts the learning rate of each parameter by calculating the first-order moment estimate and second-order moment estimate of the gradient. The Adam optimizer is adaptive and can automatically adjust the learning rate according to the importance of the parameters, and can usually converge to the optimal solution faster.

[0038] The data loader is responsible for efficiently loading and preprocessing data so that the model can be trained or evaluated. This example uses torch.utils.data.DataLoader, which provides functions such as batch loading, shuffling data, and multi-process loading, which can significantly improve the efficiency and stability of data loading. Through the custom dataset class, the data preprocessing and loading logic can be easily implemented.

[0039] The training loop combines the above components to perform model training. In each training round, the data loader iterates over the entire dataset, feeds batches of data into the model for forward propagation, calculates the loss function value, and performs backpropagation to update the model parameters.

[0040] Packaging judgment module 3: Use the camera to capture the image of the sample collection tank in the packaging area, and use the deployed sample collection tank packaging positioning model to determine whether there is a lock cover on the sample collection tank. If the judgment result is that there is no lock cover, enter the packaging execution module to perform the lock cover packaging operation; if the judgment result is that there is a lock cover, directly enter the locking control module to perform the locking operation.

[0041] Specifically, in this embodiment, an RS-485 cable is used to connect the RS-485 interface of the PLC to the RS-485 interface of the control center. When the system starts, the PLC and the control center establish a Modbus RTU connection through the RS-485 interface; the control center periodically sends Modbus RTU read requests to the PLC, requesting data such as the packaging status of the sample collection tank; the PLC responds to these requests and sends the corresponding data back to the control center; when it is necessary to control the packaging process of the sample collection tank, the control center sends a ModbusRTU write request to the PLC; the PLC receives these instructions and performs corresponding control operations according to the instruction content.

[0042] A camera is deployed above the packaging area of ​​the sample collecting tank in the packaging area. The camera can completely capture the packaging area and the mouth of the sample collecting tank. In this embodiment, the camera adopts Basler acA1300-60gc GigE Vision, which provides a resolution of up to 1.3 million pixels to ensure that the details of the mouth of the sample collecting tank and the surrounding area can be clearly captured; it supports Gigabit Ethernet communication, has high bandwidth and low latency, and is suitable for real-time image transmission; it is suitable for use in industrial automation environments, and has dustproof, waterproof and other characteristics; it provides a rich software development kit to facilitate integration with various image processing software and control systems.

[0043] Since the Basler acA1300-60gc camera supports the GigE Vision interface, the control center communicates with the camera through the GigabitEthernet network. The control center sends instructions to the camera, sets the exposure time and gain parameters, and receives real-time image data transmitted by the camera.

[0044] When the control center receives information from the PLC that there is a sample collecting tank in the packaging area and the multi-station rotary table transmits a stop signal, the control center sends a shooting instruction to the camera, and the camera takes a picture to obtain the first photo.

[0045] The first photo is transmitted to the control center, and the sample collecting can packaging positioning model is input to obtain the probability that the sample collecting can has a locked cover. When the probability that the sample collecting can has a locked cover is greater than 90%, it is determined that the sample collecting can has a locked cover and is in a packaging state.

[0046] If the sample collecting tank is not in the sealing state, the sealing execution module is entered; if the sample collecting tank is in the sealing state, the locking control module is entered.

[0047] Packaging execution module 4: Use the sample collecting tank packaging positioning model to locate the tank mouth of the sample collecting tank, and use the robotic arm to transport the locking cover to the tank mouth to perform the packaging operation, so that the locking cover and the tank mouth are tightly fitted.

[0048] Specifically, in this embodiment, the robot arm adopts the Tecan Cavro Omni robot arm, and a gripper is configured at the front end of the robot arm. The gripper can firmly grasp the lock cover of the sample collection tank and has sufficient clamping force to prevent it from slipping during the packaging process. The gripper is designed with a communication interface, which can interact with the control center, receive the locking signal and send a feedback signal of the locking completion. It is fully automatically controlled by the control center, and is integrated with advanced control algorithms and sensor systems to ensure that the robot arm can accurately perform packaging operations.

[0049] When the control center receives information from PLC that there is a sample collecting can in the packaging area, the transmission of the multi-station rotary table stops and the sample collecting can is not in the packaging state, the control center sends a shooting command to the camera, and the camera takes a picture to obtain a second photo; the second photo is transmitted to the control center, and the sample collecting can packaging positioning model is input to obtain the first positioning information of the sample collecting can relative to the packaging area, that is, the coordinate value of the center point of the mouth of the sample collecting can.

[0050] According to the preset lock cover storage position, the robot arm first moves to the top of the lock cover area, and accurately adjusts the position through the X, Y, and Z axes of the robot arm to align the gripper with the lock cover, and starts the gripper to grab the lock cover with appropriate clamping force. According to the coordinate value of the center point of the mouth of the sample collection tank, the robot arm moves the lock cover smoothly to the top of the tank mouth through the movement of the X, Y, and Z axes. Make slight position adjustments to ensure that the lock cover is accurately aligned with the tank mouth.

[0051] According to the preset force, the robot arm performs a downward pressure operation to firmly fix the lock cover on the sample collection tank. The pressure sensor on the robot arm detects that the preset pressure has been reached and there is no abnormal slip or misalignment, and sends the information of packaging completion to the control center.

[0052] Locking control module 5: sends a locking signal through the communication interface between the mechanical arm and the lock cover, controls the lock cover to complete the locking action, and confirms the completion of the locking operation by receiving a feedback signal.

[0053] Specifically, in the present embodiment, the sample collecting tank is locked by a locking cover, and after receiving a locking instruction, the locking cover locks the sample collecting tank through a driving unit.

[0054] A communication interface is installed on the gripper of the robotic arm, and the interface is connected to the main control system of the robotic arm through a cable; a corresponding communication interface is installed on the lock cover, and a quick-plug connector is used.

[0055] When the packaging judgment module directly enters the locking control module, the control center sends a shooting command to the camera, and the camera takes a picture to obtain a third photo; the third photo is transmitted to the control center, and the packaging positioning model of the sample collecting can is input to obtain the second positioning information of the sample collecting can relative to the packaging area, that is, the coordinate value of the center point of the mouth of the sample collecting can.

[0056] The robotic arm grabs the locking cover on the sample collecting tank according to the second positioning information, connects the communication interface of the robotic arm with the communication interface of the cover, and the robotic arm transmits the locking instruction to the control drive unit in the locking cover through the communication protocol. After the locking cover completes the locking action, the robotic arm sends a locking completion signal to the control center, and the robotic arm disengages from the locking cover.

[0057] When the encapsulation execution module enters the locking control module, the robotic arm is in a state of grabbing the lock cover and connected to the communication interface of the cover. Through the communication protocol, the locking instruction is transmitted to the control drive unit in the lock cover. After the lock cover completes the locking action, the robotic arm sends a locking completion signal to the control center, and the robotic arm detaches from the lock cover.

[0058] When the control center receives the locking completion signal and the robotic arm is disengaged from the lock cover, it sends a station rotation start command to the multi-station rotary table controller to transfer the current sample collection tank to the completion area and the next sample collection tank to the packaging area.

[0059] In one embodiment, the system further includes a wireless power supply module for monitoring the power of the lock cover and performing a lock cover charging operation.

[0060] Specifically, the wireless power module first reads the remaining power of the lock cover through the communication interface between the front gripper of the robot arm and the lock cover. When it detects that the remaining power of the lock cover is less than 20%, the wireless power module triggers the charging process.

[0061] First, the robotic arm moves the lock cover to the charging area, which is equipped with a wireless charging platform that matches the lock cover. During the charging process, the wireless power supply module continuously monitors the charging status of the lock cover, including parameters such as charging current, voltage, and amount of power charged. At the same time, it maintains communication with the lock cover and receives the charging progress and status information sent by the lock cover so as to adjust the charging strategy in time or handle possible abnormal situations. When the lock cover is fully charged, the wireless power supply module stops charging and sends a charging completion signal to the robotic arm and control center through the communication interface. At this time, the robotic arm re-grasps the lock cover and continues to perform subsequent packaging tasks.

[0062] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.

Claims

1. An intelligent packaging system for a fully sealed sample collection tank, characterized in that: include: Sample collection tank transmission module: used for one-way transmission of sample collection tanks, including waiting area, packaging area and completion area; The waiting area is used to wait for packaging, the packaging area performs packaging operations, and the completion area stores packaged sample collection cans; the sample collection cans are moved between the various areas through a multi-station rotary table; the first infrared sensor is configured near the edge of the waiting area in the packaging area, and the second infrared sensor is configured near the edge of the completion area in the packaging area; Model training module: Using the deep learning algorithm based on Vision Transformer, the packaging positioning model of the sample collection tank is constructed by training historical data. The model is used to identify the lock cover status and tank mouth position of the sample collection tank. After training, the model is deployed in the packaging judgment module and the packaging execution module. Packaging judgment module: The camera is used to capture the image of the sample collection tank in the packaging area, and the deployed sample collection tank packaging positioning model is used to determine whether there is a lock cover on the sample collection tank; if the judgment result is that there is no lock cover, the sample collection tank enters the packaging execution module to perform the lock cover packaging operation; if the judgment result is that there is a lock cover, the sample collection tank directly enters the locking control module to perform the locking operation; Packaging execution module: Use the sample collection tank packaging positioning model to locate the can mouth of the sample collection tank, and use the robotic arm to transport the lock cover to the can mouth to perform the packaging operation, so that the lock cover and the can mouth are tightly fitted; Locking control module: sends a locking signal through the communication interface between the robotic arm and the lock cover, controls the lock cover to complete the locking action, and confirms the completion of the locking operation by receiving a feedback signal.

2. The system according to claim 1, characterized in that The PLC controls the rotation of the workstation according to the feedback signals from the first infrared sensor and the second infrared sensor to ensure that there is only one sample collection tank in the packaging area. The specific control logic is as follows: The PLC has a built-in first infrared signal counter and a second infrared signal counter. When the first infrared signal count is 1 more than the second infrared signal count, the PLC sends a station rotation stop command to the multi-station rotary table controller, and the multi-station rotary table stops transmitting. When the packaging operation starts, the first infrared signal count and the second infrared signal count are set to 0; When the first sample collecting tank enters the packaging area, the sample collecting tank triggers the first infrared sensor, and the first infrared signal count is increased by 1. At this time, the first infrared signal count is 1 more than the second infrared signal count. The PLC sends a station rotation stop instruction to the multi-station rotary table controller, and the multi-station rotary table stops transmission, so that the sample collecting tank is packaged in the packaging area; For the continuously transmitted sample collection tanks, each time a sample collection tank enters and completes its task in the packaging area, it leaves the packaging area to trigger the second infrared sensor, and the second infrared signal count is increased by 1. This signal serves as confirmation that the processing of the previous sample collection tank has been completed; When the sample collecting can to be packaged enters the packaging area, the first infrared signal count is increased by 1. At this time, the first infrared signal count is 1 more than the second infrared signal count. There is only one sample collecting can in the current packaging area and the previous sample collecting can has left the packaging area. The PLC sends a station rotation stop command to the multi-station rotary table controller, and the multi-station rotary table stops transmission.

3. The system according to claim 1, characterized in that The network structure of the deep learning algorithm based on Vision Transformer includes an input layer, a Transformer Encoder layer and a task-specific layer, which are used for image preprocessing, feature extraction, classification and positioning output respectively; Input layer: resizes the image to a preset size, splits it into blocks of preset size, flattens it and embeds it into the specified dimensions, and appends classification tags to support classification; Transformer Encoder layer: stacks multiple encoder layers, each of which contains multi-head self-attention, layer normalization, and feed-forward networks, combined with residual connections and dropout to optimize training and prediction; Task-specific layer: It is divided into classification and regression. The classification head extracts the classification label and maps it to the number of categories, and predicts it through softmax normalization. The regression head performs a pooling operation on the image block embedding vector and then maps it to the regression target.

4. The system according to claim 3, characterized in that The Vision Transformer-based deep learning algorithm also includes a loss function, an optimizer, and a data loader; Loss function: The classification task uses cross entropy loss for image classification; the regression task uses mean square error loss to approximate continuous value prediction; Optimizer: Adopt Adam optimizer to accelerate model convergence by adaptively adjusting the learning rate of each parameter; Data Loader: Use DataLoader to achieve batch processing, data shuffling, and multi-process acceleration.

5. The system according to claim 1, wherein: The historical data is historical sample collecting jar mouth data, including photos of sample collecting jar mouths with locked covers and photos of sample collecting jar mouths without locked covers.

6. The system according to claim 1, wherein: The encapsulation operation specifically includes: Control the camera to take photos of the mouth of the sample collecting tank; Use the sample collection tank packaging positioning model and the sample collection tank mouth photo to locate the tank mouth; The robot arm takes the lock cap and places it on the can mouth for pressurized packaging.

7. The system according to claim 1, characterized in that After the locking control module confirms that the locking operation is completed, it sends a station rotation start instruction to the multi-station rotary table controller to control the multi-station rotary table to transfer the sample collection tank to the completion area and simultaneously transfer the next sample collection tank to the packaging area.

8. The system of claim 1, wherein: A wireless power supply module is also included for monitoring the power level of the lock cover and performing a lock cover charging operation.

9. The system according to claim 8, characterized in that The wireless power supply module specifically includes: The remaining power of the current lock cover is read through the communication interface between the front gripper of the robot arm and the lock cover; When it is detected that the remaining power of the lock cover is lower than the preset power threshold, the charging process is triggered; The robotic arm moves the lock cover to the preset charging area for wireless charging; During the charging process, the lock cover maintains communication and receives charging status feedback; When the battery is fully charged, charging stops automatically and the robot arm resumes packaging operations.

Images