Burn personalized skin scaffold design method and system fusing 3D printing and adaptation algorithm

Abstract

The invention belongs to the technical field of burn personalized skin scaffold systems, and particularly relates to a burn personalized skin scaffold design method and system fusing 3D printing and an adaptation algorithm, and the method and system carry out multi-modal data synchronous collection, modal noise reduction and standardized safety storage. A high-precision clock module is additionally installed for OCT equipment, a temperature sensor, a pH biosensor and a fluorescence imaging module, unified acquisition software is established to set a corresponding sampling frequency, multi-equipment synchronous acquisition is realized through single triggering, and a fast expansion random tree and a particle swarm optimization algorithm accelerated by a GPU are adopted; an edge calculation module is integrated at a printing equipment end, a real-time data processing framework is adopted, a physical and digital twin system is established to optimize printing parameters to perform closed-loop control, and testing is performed on a bionic skin model containing a dynamic fluid circulation system and a mechanical stretching device.

Description

Technical Field

[0001] This invention belongs to the technical field of personalized skin scaffold systems for burns, and particularly relates to a design method and system for personalized skin scaffolds for burns that integrates 3D printing and adaptation algorithms. Background Technology

[0002] In the current field of burn wound repair, traditional skin scaffold design and fabrication methods have significant limitations and cannot meet personalized repair needs.

[0003] On the one hand, wound data acquisition often relies on single devices or discrete detection methods. Multimodal devices such as OCT devices, temperature sensors, and pH biosensors lack a unified time synchronization mechanism, making it difficult to align the collected wound structure, physicochemical environment, and other data in the time dimension. Furthermore, the data noise reduction and standardized storage process is imperfect, which can easily lead to data deviation or loss, directly affecting the accuracy of subsequent stent design.

[0004] On the other hand, existing 3D printing path planning mostly adopts a fixed mode, which cannot dynamically adapt to the irregular curved surface shape of the burn wound. In particular, the poor fit of the scaffold is prone to occur at the edge of the wound and in concave and convex areas. At the same time, the printing process lacks a real-time parameter adjustment mechanism, making it difficult to cope with the impact of dynamic changes in the printing equipment or the slight deformation of the wound. This results in insufficient scaffold forming accuracy and inability to effectively fit the wound to achieve the repair function. Summary of the Invention

[0005] In view of the aforementioned problems, and in conjunction with the first aspect of the present invention, embodiments of the present invention provide a method for designing personalized skin scaffolds for burns that integrates 3D printing and adaptation algorithms, the method comprising: Multimodal data synchronous acquisition, modal noise reduction and standardized secure storage are performed. High-precision clock modules are added to OCT equipment, temperature sensors, pH biosensors and fluorescence imaging modules. A unified acquisition software is established to set the corresponding sampling frequency, and multi-device synchronous acquisition is achieved through a single trigger. A decomposition conformal path planning algorithm is established, employing a fast-expanding random tree and a GPU-accelerated particle swarm optimization algorithm; an edge computing module is integrated into the printing device, and a real-time data processing framework is used to establish a physical and digital twin system to optimize printing parameters for closed-loop control; The above system path adjustment time, printing accuracy and vascular endothelialization efficiency were verified by testing on a biomimetic skin model containing a dynamic fluid circulation system and a mechanical stretching device. We constructed a self-sensing ink with integrated nanosensors to achieve in-situ monitoring of print quality and used optogenetics to precisely regulate cell behavior. Establish a dynamic medical device certification framework, formulate risk grading standards for the real-time decision-making capabilities of algorithms, and adopt federated learning technology to share data across centers while protecting patient privacy; adopt a modular design to integrate edge computing modules with printheads.

[0006] In another aspect, embodiments of the present invention also provide a personalized skin scaffold system for burns that integrates 3D printing and adaptation algorithms, including a processor and a machine-readable storage medium connected to the processor. The machine-readable storage medium is used to store programs, instructions, or code, and the processor is used to execute the programs, instructions, or code in the machine-readable storage medium to implement the above-described method.

[0007] The beneficial effects of this invention are: This significantly improves the accuracy and adaptability of personalized skin scaffolds for burns. In the data acquisition phase, the IEEE 1588PTP protocol enables nanosecond-level time synchronization across multimodal devices. Combined with modal noise reduction and standardized storage, this ensures the integrity and timeliness of data such as the three-dimensional structure and physicochemical environment of the wound. For path planning, a rapidly expanding random tree and GPU-accelerated particle swarm optimization algorithm are integrated, along with physical and digital twin closed-loop control. This allows for dynamic adaptation to irregular wound morphologies, maintaining printing accuracy at the micrometer level. Simultaneously, real-time parameter adjustments prevent excessive gaps between the scaffold and the wound, addressing the issues of rigid paths and insufficient accuracy in traditional printing.

[0008] The self-sensing ink, integrated with nanosensors, enables in-situ monitoring of printing quality, allowing for real-time correction of issues such as uneven material extrusion and interlayer bonding defects. Optogenetics technology precisely regulates cell proliferation and differentiation within the scaffold, promoting improved vascular endothelialization efficiency. The application of a federated learning and dynamic certification framework breaks down multi-center data barriers, optimizes model performance, and ensures clinical compliance through privacy protection mechanisms and risk-based verification. This comprehensive technical solution achieves personalized and intelligent management throughout the entire process, from data acquisition and scaffold design and printing to postoperative control, providing an efficient and safe solution for deep burn wound repair, significantly shortening the scaffold adaptation period and improving wound healing quality. Attached Figure Description

[0009] Figure 1 This is a schematic diagram of the execution flow of the personalized skin scaffold design method for burns that integrates 3D printing and adaptation algorithms, provided in an embodiment of the present invention.

[0010] Figure 2 This is a schematic diagram of exemplary hardware and software components of the personalized skin scaffold system for burns that integrates 3D printing and adaptation algorithms, provided in an embodiment of the present invention. Detailed Implementation

[0011] The present invention will now be described in detail with reference to the accompanying drawings. Figure 1 This is a flowchart illustrating a method for designing personalized skin scaffolds for burns that integrates 3D printing and adaptation algorithms, provided in one embodiment of the present invention. The wireless connection networking method for this deep-water mobile location will be described in detail below.

[0012] Step S110: Perform multimodal data synchronous acquisition, modal noise reduction and standardized secure storage, install high-precision clock modules for OCT equipment, temperature sensor, pH biosensor and fluorescence imaging module, establish unified acquisition software to set corresponding sampling frequencies, and achieve multi-device synchronous acquisition through a single trigger.

[0013] In this embodiment, a clinical example of burn wound repair is used. To comprehensively and accurately capture multi-dimensional information about the wound, it is necessary to conduct multi-modal data synchronous acquisition on these different types of devices. By adding high-precision clock modules to the OCT device, temperature sensor, pH biosensor, and fluorescence imaging module, and setting matching sampling frequencies for each device using unified acquisition software, synchronous acquisition by all devices can be achieved with a single trigger, ensuring strict alignment of different modal data in the time dimension. Subsequently, noise reduction processing is performed on each modal data to remove interference noise and improve data quality, and the data is securely stored through a standardized process.

[0014] Step S111: Select an industrial-grade clock module that supports the IEEE 1588 PTP protocol. Design a PCIe expansion card with PTP protocol parsing function for the OCT device, integrate the clock module into the expansion card, and communicate with the OCT device motherboard through the gold fingers; fabricate I2C to PTP protocol adapter boards for the temperature sensor and pH biosensor, and connect the sampling trigger terminals of the temperature sensor and pH biosensor through the pins of the clock module; connect a PTP synchronization module in series between the fluorescence imaging module and the imaging lens control board, and trigger fluorescence exposure using the hardware interrupt pin of the fluorescence imaging module.

[0015] To achieve high-precision time synchronization for multimodal wound monitoring equipment, a hardware adaptation scheme based on the IEEE 1588 PTP protocol needs to be designed, taking into account the communication interfaces and functional characteristics of different devices. Firstly, for OCT equipment, an industrial-grade clock module equipped with an OCXO temperature-controlled crystal oscillator, achieving a frequency stability of ±5ppb and an annual drift of <0.1ppm, is selected. This module supports the PTPv2 protocol and hardware timestamp function, and can output a synchronization clock with nanosecond-level accuracy. To achieve integration with the OCT equipment, a PCIe 3.0 expansion card is designed. In addition to reserving standard PCIe gold fingers, the card integrates a PTP protocol parsing chip. The clock module is fixed to the expansion card via a board-to-board connector. The module's time signal output pin is directly soldered to the synchronization clock input pin of the protocol parsing chip. Simultaneously, configuration data exchange between the module and the parsing chip is achieved through an SPI interface. After the expansion card is inserted into the OCT motherboard, a dedicated driver program needs to be written to enable the OCT system to read the PTP synchronization time output by the expansion card and bind it to the scan data of each layer, ensuring that the inter-layer timestamp error of the OCT 3D image is <100ns.

[0016] For temperature and pH biosensors, a customized I2C to PTP protocol adapter board needs to be fabricated. The adapter board uses an STM32H743 microcontroller as its core. The board is divided into a PTP signal processing area and an I2C signal conversion area. The PTP processing area uses an industrial-grade clock module of the same model as the OCT expansion card. The module's GPIO pins are connected to the microcontroller's external interrupt pins to receive the PTP synchronization trigger signal. The I2C conversion area integrates a TXS0108E level conversion chip. The microcontroller's I2C1 interface is connected to the SDA / SCL pins of the temperature and pH sensors respectively through the level conversion chip. Simultaneously, PTP-I2C protocol conversion logic is written in the adapter board firmware: when the microcontroller detects the PTP trigger signal from the clock module through the EXTI1 pin, it immediately sends a forced sampling command to the sensor via the I2C bus. After the sensor completes data acquisition, it packages the data with the PTP timestamp recorded by the microcontroller and sends it back, ensuring that the time deviation between the physicochemical parameter acquisition and the OCT scan is <1ms.

[0017] For the fluorescence imaging module, an independent PTP synchronization module needs to be connected in series between it and the imaging lens control board. This synchronization module uses a custom board equipped with a Xilinx Artix-7 FPGA. The module's input is connected to a PTP clock server via an RJ45 Ethernet port to receive global synchronization time. The output has two signal interfaces: one is a TTL level trigger signal interface, connected to the hardware interrupt pin of the fluorescence imaging module via a shielded cable; the other is an RS485 interface, connected to the shutter control pin of the lens control board. The synchronization module firmware is configured with timestamp comparison-trigger output logic: when the PTP time received by the module reaches the preset exposure time, it first sends a shutter open command to the lens control board via the RS485 interface, and after a 100ns delay, sends a trigger pulse to the hardware interrupt pin of the fluorescence imaging module, triggering the camera to start exposure. After exposure, the module sends a shutter close command to the lens control board, simultaneously recording the PTP timestamp of this exposure and uploading it to the data acquisition host via UDP protocol. This ensures that the time synchronization accuracy between fluorescence image acquisition and OCT scanning / physicochemical parameter acquisition is <500ns, meeting the alignment requirements of multimodal data in the time dimension.

[0018] Step S112: Read the native communication protocols of the OCT device, temperature sensor, pH biosensor, and fluorescence imaging module one by one, establish a protocol conversion intermediate layer program, embed the PTP protocol stack in the intermediate layer, and encapsulate the timestamp generated by the clock module into the sampling control instructions of each device. In the multimodal data acquisition for personalized burn skin scaffold design, the native communication protocols of each device are first captured and analyzed using tools such as protocol analyzers and oscilloscopes to clarify the protocol frame structure, instruction code definition, and data field format. For example, the meaning of the byte bits of the scan start instruction in the TCP data packet of the OCT device is deconstructed, the slave address and register address of the sampling request in the I2C protocol of the temperature sensor are recorded, and the timing sequence of the control signal triggered by exposure in the CameraLink protocol of the fluorescence imaging module is sorted out. Then, based on the analysis results, a protocol conversion intermediate layer program is established, and an independent protocol adaptation submodule is configured for each device. At the same time, a complete PTP protocol stack is embedded in the core logic of the intermediate layer, enabling it to receive nanosecond-level precision timestamps generated by the industrial-grade clock module in real time. When the unified acquisition software issues sampling control instructions, the instructions are first transmitted to the intermediate layer. The intermediate layer calls the corresponding adaptation submodule according to the target device type to convert the general instructions into the native protocol format that the device can recognize, and encapsulates the PTP timestamp into the reserved field of the instruction frame.

[0019] Step S113: The acquisition software is divided into a device management module, a parameter configuration module, and a trigger control module. The device management module scans and identifies connected OCT devices, temperature sensors, pH biosensors, and fluorescence imaging modules, reads the hardware IDs of the OCT devices, temperature sensors, pH biosensors, and fluorescence imaging modules, and associates them with the corresponding clock module IPs. In the multimodal data acquisition for personalized burn skin scaffold design, to ensure that the OCT device, temperature sensor, pH biosensor, and fluorescence imaging module can work collaboratively and achieve time synchronization, the acquisition software needs to complete the automatic identification and association configuration of the devices through the device management module. In actual operation, the device management module will simultaneously start a network scanning thread and a USB device enumeration thread: the network scanning thread traverses all IP addresses in the local area network by sending ARP requests, sending a device identification command to each IP. When the OCT device receives the command, it will return a response containing its own hardware ID and the IP of the associated clock module; the USB device enumeration thread calls the system API to read the VID / PID and device descriptor of the temperature sensor, pH biosensor, and fluorescence imaging module connected to the host. Based on this, it filters out the target devices, sends commands through the USB endpoint to obtain the hardware ID of each device, and reads the associated clock module IP pre-stored in the device firmware. After scanning is complete, the device management module will store the hardware ID, device type, and corresponding clock module IP of the four types of devices in a unified memory list and display it intuitively in the device list control on the software interface. This allows users to manually confirm whether the association is accurate. If there is an error in the clock module IP matching, users can directly modify and adjust it on the interface to ensure that each type of acquisition device can accurately connect to its dedicated high-precision clock module. This lays the foundation for subsequent multi-device synchronous acquisition and unified timestamp encapsulation, avoids sampling timing deviations caused by mismatched device and clock module associations, and ensures the consistency of multimodal wound data in the time dimension.

[0020] Step S1131: Initialize the network scanning thread and the USB device enumeration thread. The network scanning thread traverses all IPs in the local area network by sending ARP requests and sends a device identification command to each IP. After receiving the command, the OCT device returns a response containing the hardware ID and the clock module IP. The USB device enumeration thread calls the system API to read the VID / PID and descriptor of the temperature sensor, pH biosensor, and fluorescence imaging module, filters out the temperature / pH sensor and fluorescence module, and obtains the hardware ID by sending a command through the USB endpoint. At the same time, it reads the associated clock module IP pre-stored in the firmware of the temperature sensor, pH biosensor, and fluorescence imaging module. In the multimodal data acquisition preparation phase of personalized burn skin scaffold design, to accurately identify and associate all key acquisition devices with their corresponding clock modules, the acquisition software first needs to initialize a network scanning thread and a USB device enumeration thread to achieve device classification and detection. Specifically, the network scanning thread first sends ARP requests to the local area network, then traverses all online IP addresses by parsing the returned ARP response packets, and subsequently sends customized device identification commands to each IP address. Since the OCT device needs to communicate with the system through a network interface and has a pre-installed corresponding response program, it automatically returns a data packet containing its unique hardware ID and the IP address of its matching high-precision clock module upon receiving the identification command. Based on this, the software can clearly determine the correspondence between the OCT device and the clock module. Meanwhile, the USB device enumeration thread calls the underlying AP of the operating system to scan all devices connected to the host's USB ports, and reads the VID, PID and device descriptor information of each device. Through preset device filtering rules, the target acquisition device is accurately selected from a large number of USB devices. Then, the hardware ID reading command is sent to these devices through the USB endpoint to obtain the unique hardware identifier of each device. At the same time, the associated clock module IP pre-written in the device firmware is read. Finally, the initial association between all acquisition devices and clock modules is completed, which lays the foundation for subsequent unified configuration of sampling parameters and realization of multi-device synchronous acquisition, and avoids the timing disorder of acquisition data caused by device identification omission or clock module association error.

[0021] Step S1132: After scanning, the hardware ID, type, and clock module IP of the OCT device, temperature sensor, pH biosensor, and fluorescence imaging module are stored in the memory list and displayed in the device list control on the interface. The user can manually confirm or modify the association. Three numerical input boxes and an application button are placed. The input boxes have range limits and are automatically validated after input. If the input exceeds the range, a pop-up prompt will appear. After the user clicks "apply", the software calculates the trigger cycle of each device. The clock module receives the data and returns confirmation. At the same time, the software writes the configuration parameters to the local configuration file, recording the hardware IDs of the OCT device, temperature sensor, pH biosensor, and fluorescence imaging module, as well as their corresponding frequencies and cycles. These will be automatically loaded on the next startup. In the software operation for multimodal data acquisition using personalized skin scaffolds for burns, after the network scanning thread and USB device enumeration thread complete the identification of OUT devices, temperature sensors, pH biosensors, and fluorescence imaging modules, the software stores the core information of these devices—including their unique hardware IDs, device types, and associated clock module IPs—into a unified memory list. This information is then clearly displayed as entries in the device list control on the interface, allowing users to intuitively view the correspondence between each device and its clock module. Considering potential IP matching discrepancies in actual connections, users can manually verify the accuracy of the association on this interface. If errors are found, they can be directly modified and adjusted to ensure that each acquisition device accurately connects to its dedicated high-precision clock module. The interface also includes three numerical input boxes with range restrictions, corresponding to setting the sampling frequency of the OCT device, temperature / pH sensor, and fluorescence imaging module, respectively. These input boxes perform real-time validity checks on the user-entered values. Once the user confirms the parameters are correct and clicks the "Apply" button, the software will automatically calculate the clock trigger period for each device based on the input sampling frequency and send the period parameters to the corresponding clock module. After the clock module receives and returns a confirmation signal, the software will synchronously write all configuration parameters to the local configuration file. This way, the next time the software starts, it can automatically load these parameters without rescanning and configuring, directly restoring the previous device association and sampling settings. This significantly improves the operational efficiency of clinical multimodal data acquisition, avoids time loss and parameter errors caused by repeated configuration, and ensures the continuity and stability of the burn wound monitoring process.

[0022] Step S1133: Add a start acquisition button to the interface. After clicking, the control logic is triggered to check whether the OCT device, temperature sensor, pH biosensor, and fluorescence imaging module list are online. If there is an offline device, a pop-up window will be displayed and the process will be terminated. After confirming that all devices are online, a global trigger timestamp is generated, and a trigger command is constructed. A multi-threaded concurrent sending mechanism is adopted, with each device corresponding to a sending thread. The trigger command is sent to the IP of its associated clock module at the same time to ensure that all commands are sent within 1ms. In the software operation of multimodal data acquisition for personalized skin scaffolds for burns, to ensure that all devices can respond normally and achieve synchronous sampling when the acquisition process starts, a "Start Acquisition" button is added to the core operation area of ​​the software interface. This button is the key entry point for triggering the acquisition process. When the user clicks this button, the software immediately starts the preset trigger control logic. First, it checks the online status of the four core acquisition devices: OCT device, temperature sensor, pH biosensor, and fluorescence imaging module. By sending status detection commands to the corresponding communication interfaces of each device, if a device fails to return a normal response within a preset time, the software will immediately display a prominent pop-up prompt, clearly indicating the type and hardware ID of the offline device, and automatically terminate the subsequent acquisition process to avoid incomplete data acquisition due to the offline status of some devices. If all devices report being online, the software will generate a globally unified trigger timestamp based on the time of the current reference clock module. This timestamp will serve as the synchronization reference for all devices to start sampling. Subsequently, it constructs a dedicated trigger command according to the native communication protocol format of each device. The command encapsulates the global trigger timestamp and the sampling parameters required by the device. To ensure that all devices can start sampling at the same time, the software adopts a multi-threaded concurrent sending mechanism, allocating an independent instruction sending thread to each device. These threads simultaneously send trigger commands to the clock module IP associated with the corresponding device. By optimizing thread scheduling and network / USB communication priorities, all commands are strictly controlled to be sent within 1 millisecond, avoiding misalignment of device sampling timing due to instruction sending delays. Ultimately, this ensures that the OCT device's 3D scanning, the temperature / pH sensor's physicochemical parameter acquisition, and the fluorescence imaging module's cell signal capture can start synchronously, providing a precise timing basis for subsequent multimodal data time alignment and comprehensive wound analysis.

[0023] Step S1134: After sending, start timeout detection, receive responses from each clock module. If the OCT device, temperature sensor, pH biosensor, and fluorescence imaging module all respond, the interface will display "triggered successfully" and start receiving device sampling data. If any device times out and does not respond, immediately retry sending. If it still fails, mark the device as abnormal and continue the process for other devices.

[0024] In the multimodal data acquisition of personalized skin scaffolds for burns, after the software sends trigger commands to the clock modules associated with each device through a multi-threaded concurrent mechanism, it immediately initiates a timeout detection process. This process sets a uniform timeout threshold for the response of each device and monitors the response signals returned by each clock module in real time. For OCT devices, timeout detection monitors whether their network interface sends back a TCP data packet containing confirmation of trigger command reception; for temperature sensors and pH biosensors, it checks whether the USB port returns a status code indicating successful command execution; and the fluorescence imaging module needs to return an exposure trigger ready signal through the CameraLink interface. If all device clock modules successfully return responses within the timeout threshold, the software will display a green notification of successful triggering in the acquisition status area of ​​the interface and automatically switch to data receiving mode. It will then receive the 3D scan data from the OCT device, the physicochemical parameter data from the temperature and pH sensors, and the image data from the fluorescence imaging module through the corresponding communication interfaces, while temporarily storing these data according to timestamps. If a device fails to respond after exceeding the threshold, the software will immediately trigger a retry mechanism, resending the trigger command to the device's clock module. The number of retries is preset to 2 to eliminate temporary communication interference. If no response is received after 2 retries, the software will mark the device in red as abnormal in the device list on the interface and display an abnormal message in the status bar. At the same time, it will skip the subsequent triggering process of the device and continue to advance the data reception of other normal devices, ensuring that OCT scanning, temperature acquisition, fluorescence imaging and other processes unaffected by abnormal devices can proceed normally.

[0025] Step S114: The parameter configuration module provides a visual interface, allowing users to set the sampling frequency of the OCT device, the temperature sensor, the pH biosensor, and the fluorescence imaging module, converting the frequency parameters into the trigger cycle of the clock module and writing them into the module register; the trigger control module is designed with single-trigger logic. To adapt to different wound monitoring needs, the parameter configuration module provides users with an intuitive frequency adjustment interface through a visual interface. The interface features separate parameter adjustment areas for OCT devices, temperature sensors, pH biosensors, and fluorescence imaging modules. Each area includes a numerical input box and a drop-down menu, allowing users to directly input or select preset sampling frequencies based on their clinical monitoring priorities. During input, the interface continuously verifies whether the values ​​are within the device's supported range, ensuring that the parameter settings meet the hardware performance limits. Once the user confirms the parameters and clicks "Apply," the parameter configuration module automatically converts the sampling frequency into a trigger period recognizable by the clock module. This period parameter is then written to the corresponding clock module's register via the communication interface, completing the hardware-level timing configuration and establishing a unified time reference for synchronous acquisition from multiple devices. The single-trigger logic of the trigger control module is adapted to on-demand data acquisition in clinical practice. For example, after changing the dressing on a wound, a doctor needs to obtain a complete set of multimodal data immediately to evaluate the dressing effect. When the single-acquisition button on the interface is clicked, the trigger control module will first check the communication status of all devices and clock modules. After confirming that there are no offline devices, it will generate a globally unified trigger timestamp. Then, through a multi-threaded concurrency mechanism, it will synchronously send trigger commands to the clock modules associated with each device to ensure that the actions of OCT device starting scanning, temperature and pH sensors starting detection, and fluorescence imaging module performing exposure are triggered at the same time. This avoids data misalignment caused by differences in trigger timing and ensures that the single-acquisition data can accurately reflect the comprehensive status of the wound at that time point. It provides a time-consistent function for the immediate adjustment of skin scaffold printing parameters or the evaluation of healing effect.

[0026] Step S115: Embed a synchronization verification thread in the acquisition software. After acquiring data from a set period of time, automatically read the timestamp field from the data returned by the OCT device, temperature sensor, pH biosensor, and fluorescence imaging module, and calculate the deviation between the timestamp of the OCT device, temperature sensor, pH biosensor, and fluorescence imaging module and the timestamp of the reference clock module.

[0027] To avoid timestamp deviations caused by factors such as device clock drift and electromagnetic interference during long-term data acquisition, and to ensure the consistency of data from OCT devices, temperature sensors, pH biosensors, and fluorescence imaging modules in the time dimension, the acquisition software embeds a dedicated synchronization verification thread. In actual operation, users can preset verification cycle groups in the software interface according to their monitoring needs. When the set conditions are met, the synchronization verification thread automatically starts: first, it extracts the corresponding timestamp field from the acquisition data packets returned by each device; then, it calls the current standard timestamp from the reference clock module; and finally, it calculates the deviation between the timestamp of each type of device and the reference timestamp using the software's built-in logic. If the calculated deviation is within a preset threshold, the current synchronization status is considered normal, and the acquisition process continues. If the deviation of a certain type of device exceeds the threshold, the software will immediately display a pop-up indicating a synchronization error for that device and automatically trigger a calibration mechanism. It will send a time calibration command to the clock module of the corresponding device to correct the deviation to the allowable range, and simultaneously record this calibration event to a log file for subsequent tracing of the cause of the error. This periodic synchronous verification design can effectively avoid the cumulative error of time during long-term acquisition, ensuring that data from different modalities can always accurately correspond to the wound state at the same time point. For example, when doctors analyze wound data 6 hours after surgery, they can accurately match the OCT structural image, temperature distribution, pH value and fluorescence signal at that moment, clearly determine the correlation between the wound healing microenvironment and the progress of tissue repair, and dynamically optimize the printing parameters of the skin scaffold.

[0028] Step S120: Establish a split conformal path planning algorithm, using a fast-expanding random tree and a GPU-accelerated particle swarm optimization algorithm; integrate an edge computing module on the printing device, using a real-time data processing framework, and establish a physical and digital twin system to optimize printing parameters for closed-loop control.

[0029] In this embodiment, in the 3D printing of personalized skin scaffolds for burns, since the wound morphology is often an irregular curved surface, traditional path planning is prone to problems such as poor fit between the scaffold and the wound and low printing efficiency. Therefore, it is necessary to establish a decomposed conformal path planning algorithm: First, based on the three-dimensional point cloud data of the wound acquired in the early multimodal acquisition, the irregular wound curved surface is discretized into a three-dimensional mesh space adapted for printing. At the same time, path constraints are defined in combination with the scaffold material characteristics. In the global path planning stage, a fast expanding random tree algorithm is adopted. Through an adaptive sampling strategy, the sampling density is increased in areas with high curvature such as the edge of the wound and the sampling density is decreased in flat areas. An initial coarse path is generated with the printing start point as the root node. Subsequently, a GPU-accelerated particle swarm optimization algorithm is introduced to discretize the coarse path into multiple path control points and use them as particle dimension parameters. The fit error between the path and the wound curved surface and the smoothness of the path are used as optimization objectives. With the help of the parallel computing power of the GPU, the optimal conformal path is quickly iterated and screened to avoid the impact of excessive path planning time on the clinical printing timeliness.

[0030] Considering that the wound may experience parameter deviations during the printing process due to external environment or dynamic operating conditions of the printing equipment, it is necessary to integrate an edge computing module on the printing device and adopt a real-time data processing framework such as Pathway: divide the dynamic wound data collected by the sensor into real-time stream and near-real-time stream, allocate higher computing resources to the real-time stream to prioritize the processing of key parameters, and use an incremental computing engine to process the data through a sliding window to quickly capture differences in parameter changes. Based on this, a physical-digital twin system is established: a digital twin model of the skin scaffold is constructed using COMSOL Multiphysics, and a three-dimensional model of the wound is imported as a boundary condition. Solid mechanics and mass transfer coupled physical fields are selected to simulate the scaffold printing process, and the model parameters are calibrated using actual data collected by physical equipment. During printing, the digital twin model first predicts the optimal printing parameters based on the current wound data and sends them to the physical equipment for execution. The edge computing module collects sensor data from the physical system in real time and feeds it back to the digital twin model. If the model comparison finds that the deviation between the actual parameters and the predicted parameters exceeds the threshold, the optimized parameters are immediately recalculated and sent for adjustment, forming a closed-loop control of prediction, execution, feedback, and optimization. This ensures that the final printed skin scaffold can not only fit closely to the irregular wound but also meet the requirements of structural stability and biocompatibility, adapting to the personalized needs of burn wound repair for different patients.

[0031] Step S121: Construct a path planning environment model based on the 3D point cloud data of the wound surface, discretize the wound surface into a 3D mesh space, and define the constraints of the printing path; adopt an adaptive sampling strategy in the fast expanding random tree algorithm, increase the sampling density in the edge area of ​​the wound surface, and decrease the sampling density in the flat area; when the fast expanding random tree algorithm is started, take the printing start point as the root node, randomly generate sampling points and filter valid nodes through collision detection, connect adjacent nodes through a greedy strategy to form a path tree, and when the path tree extends to the printing end point, generate an initial coarse path through path pruning to complete the global planning; In the path planning of 3D printing personalized skin scaffolds for burns, a path planning environment model is first constructed based on the 3D point cloud data of the wound acquired through multimodal acquisition. A 3D modeling algorithm discretizes the irregular wound surface into a 3D mesh space adapted to printing accuracy. Simultaneously, path constraints are defined based on the actual needs of scaffold printing, such as matching the path spacing to the material curing characteristics, conforming the path direction to the curvature of the wound surface, and avoiding sensitive areas of the wound at the printing start and end points. In the application of the rapidly expanding random tree algorithm, an adaptive sampling strategy is adopted to balance planning accuracy and efficiency: for wound edges, the sampling point generation density is increased to ensure accurate capture of edge contour changes; for relatively flat areas inside the wound, the sampling density is reduced to decrease redundant computation. When the algorithm is started, a large number of sampling points are randomly generated in the three-dimensional grid space with the preset printing start point as the root node. The collision detection algorithm filters out invalid sampling points that exceed the wound area, conflict with the surrounding tissue model, or do not meet the physical limitations of the material printing. Then, the adjacent valid nodes are connected by a greedy strategy to gradually extend and form a path tree covering the wound. When the end node of the path tree extends to the preset printing end point, the path pruning process is started to delete redundant detour nodes in the path tree and optimize sharp angle turning nodes. Finally, an initial coarse path that fits the shape of the wound is generated, and the global path planning of the scaffold printing is completed.

[0032] Step S122: Discretize the global coarse path into N path control points, and use the three-dimensional coordinates of each control point as the dimension parameter of the particle to construct a particle swarm; define the fitness function as the fit error between the path and the wound surface and the path smoothness. The fit error is calculated by summing the shortest distances from the control points to the wound surface, and the path smoothness is calculated by weighting the cosine values ​​of the angles between adjacent control points. In the path optimization of 3D printing personalized skin scaffolds for burns, after completing the global coarse path planning, to further improve the fitting accuracy between the path and the irregular wound surface and the smoothness of the printing process, the global coarse path needs to be discretized into N path control points at preset intervals. The three-dimensional coordinates of each control point are directly used as the dimension parameter of a single particle in the particle swarm optimization algorithm, thereby constructing a particle swarm that matches the dimension of the path control points. Each particle represents a set of potential optimized path schemes. Subsequently, a fitness function adapted to the scaffold printing requirements is defined. The core of this function includes two key evaluation indicators: the fitting error between the path and the wound surface and the path smoothness. The fitting error is obtained by calculating and summing the shortest distance from each path control point to the wound surface; the path smoothness is calculated by calculating the cosine value of the angle between the lines connecting two adjacent path control points and then weighting the calculation according to the priority of the printing direction. Through such particle dimension construction and fitness function definition, a clear optimization goal and evaluation standard are provided for the subsequent selection of the optimal path by the GPU-accelerated particle swarm optimization algorithm, ensuring that the final generated printing path can closely fit the irregular shape of the burn wound.

[0033] Step S123: Transfer the particle swarm data to the GPU, configure the number of CUDA thread blocks to be 1 / 32 of the number of particles, and each block contains 32 threads; after each iteration, share the optimal particle information through the GPU global memory. In the path optimization stage of 3D printing personalized skin scaffolds for burns, after the particle swarm is constructed, to improve the iteration efficiency of the optimization algorithm and adapt to the timeliness requirements of clinical scaffold design, all data of the particle swarm is first transferred to the GPU via the PCIe high-speed interface. The GPU's parallel computing capabilities are then used to accelerate the particle swarm optimization process. During the CUDA kernel configuration stage, the allocation rules for thread blocks and threads are determined based on the number of particles. The number of thread blocks is set to 1 / 32 of the number of particles, with each thread block containing 32 threads. This configuration maximizes the utilization of the GPU's streaming multiprocessor resources, ensuring that each thread can independently handle the fitness calculation or position update of a single particle, avoiding wasted computing resources due to thread idleness. In each round of iteration, the threads within each thread block perform the fitness evaluation and velocity / position update of the corresponding particle in parallel. After the iteration ends, the optimal particle information obtained in the current iteration is stored and shared in real time through the GPU's global memory. All thread blocks can directly read this optimal information from the global memory for guiding particle position updates in subsequent iterations, eliminating the need for data transfer via the CPU and significantly reducing data transmission latency. This design, which combines GPU parallel computing with global memory sharing, can increase the iteration speed of the particle swarm optimization algorithm by several times, quickly selecting the optimal printing path with the smallest fitting error and the smoothest path. This ensures that even when faced with complex burn wound morphology, the path optimization of personalized scaffolds can be completed in a short time, meeting the needs of rapid design and timely printing in clinical practice.

[0034] Step S124: Select the NVIDIA Jetson AGX Orin module and connect it to the main control board of the 3D printing equipment via the PCIe 4.0 interface; at the same time, connect the wound sensor via the GPIO interface and configure the interface sampling rate to be consistent with that of the wound sensor. In the hardware integration of the personalized skin scaffold 3D printing system for burns, the NVIDIA Jetson AGX Orin module was selected as the core edge computing unit to achieve real-time processing of wound data and collaborative control of the equipment during the printing process. This module has powerful parallel computing capabilities and low power consumption, making it suitable for embedded deployment requirements of the printing equipment. Regarding hardware connectivity, a high-speed communication link is established between the edge computing module and the main control board of the 3D printing equipment via a PCIe 4.0 interface. The high bandwidth of the PCIe 4.0 interface ensures rapid transmission of printing parameters, path data, and equipment status information between the edge computing module and the main control board, avoiding printing accuracy deviations caused by data transmission delays. Furthermore, the module's GPIO interface is directly connected to the wound sensor. At the software level, the sampling rate of the GPIO interface is configured to match the output frequency of the wound sensor, ensuring accurate capture of real-time dynamic changes in the wound without data loss or redundancy due to sampling rate mismatch. This hardware connectivity design enables the NVIDIA Jetson AGX Orin module to efficiently handle 3D printing-related computational tasks and acquire wound status data in real time. This provides data transmission and computational support for subsequent closed-loop control based on digital twins, ensuring that the printed skin scaffold can dynamically adapt to subtle changes in the wound healing process and improve personalized repair results.

[0035] Step S125: Install the Ubuntu 20.04 system on the NVIDIA Jetson AGX Orin module, configure the Python 3.8 environment, and install the Pathway framework and its dependencies via pip; build a data processing pipeline based on Pathway, dividing the wound sensor data into real-time and near-real-time streams, allocating different computing resources to each; configure an incremental computing engine to process the wound sensor data using a sliding window, calculating the difference between newly added data and historical results within the window; and use the framework's data partitioning function to store multimodal data in partitions by timestamp. In edge computing for 3D printing of personalized skin scaffolds for burns, to achieve efficient processing and orderly management of wound sensor data, an Ubuntu 20.04 system was first installed on the NVIDIA Jetson AGX Orin module. Then, the Python 3.8 environment was configured using the system's built-in package manager to ensure consistency between the subsequent framework installation and code execution environment. Subsequently, the Pathway real-time data processing framework and related dependency libraries were installed using pip, laying the foundation for the data processing pipeline. When building the pipeline based on the Pathway framework, the wound sensor data is divided into real-time and near-real-time streams according to the application priority. The real-time stream corresponds to data that needs immediate use for printing parameter adjustment, allocating more CPU cores and memory resources to ensure zero-latency data processing. The near-real-time stream, on the other hand, allocates relatively fewer computing resources to non-urgent data used for subsequent wound healing analysis, achieving rational resource utilization. Simultaneously, the framework's incremental computing engine is configured to use a sliding window processing method for continuously collected data from the wound sensors, calculating only the difference between newly added data and historical calculation results within the window, eliminating the need to repeatedly process the entire dataset and significantly reducing computational overhead. In addition, by leveraging Pathway's data partitioning feature, multimodal data is stored in partitions based on timestamps. Each timestamp interval corresponds to an independent data partition. When multimodal data for a specific time period needs to be retrieved later, the corresponding partition can be located directly for quick reading, avoiding the time-consuming process of searching the entire dataset.

[0036] Step S126: Finally, establish a physical and digital twin system to construct closed-loop control.

[0037] Step S1261: Deploy multiple types of sensors, including pressure sensors, displacement sensors, and optical sensors, in the printing area of ​​the 3D printing bracket; transmit the data of each sensor to the edge computing module via RS485 bus; after data preprocessing, encapsulate the data according to each sensor ID, timestamp, and data value format, store it in the local database of the edge module, and send updated data to the digital twin model at regular intervals. In the closed-loop control of 3D printing personalized skin scaffolds for burns, multiple types of sensors are deployed in the printing area of ​​the 3D printed scaffold to capture the equipment's operating status and scaffold forming dynamics in real time: pressure sensors monitor the contact pressure between the nozzle and the scaffold surface, displacement sensors track the actual movement position of the print head in real time, and optical sensors provide feedback on the flatness of the scaffold layers through image acquisition. The data from these sensors is transmitted to the edge computing module via an RS485 bus. The RS485 bus has strong anti-interference capabilities, long transmission distances, and supports simultaneous communication from multiple devices, ensuring stable transmission of data from various sensors. After the data arrives at the edge computing module, it first undergoes a preprocessing process: removing outliers caused by electromagnetic interference, filtering and smoothing the displacement data, and converting the image data from the optical sensor into quantifiable flatness parameters. Then, it is packaged in a unified format, with each data entry containing the unique ID of the corresponding sensor, the timestamp of the acquisition time, and the processed valid data value. The packaged data is stored in the local database of the edge module on the one hand, and on the other hand, updated data is sent to the digital twin model at set time intervals. This provides data support for the digital twin model to restore the physical printing state in real time, ensuring that the digital twin model can accurately reflect the pressure changes, position deviations, and bracket forming conditions during the actual printing process. This lays the data foundation for subsequent printing parameter optimization and closed-loop control adjustment.

[0038] Step S1262: A digital twin model of the hydrogel scaffold is constructed using COMSOL Multiphysics. Solid mechanics and mass transfer coupled physical fields are selected, and scaffold material parameters are defined. The 3D model of the wound is imported as boundary conditions. The digital twin model is meshed, with a fine mesh for the scaffold region and a coarse mesh for the wound region, balancing computational accuracy and speed. The parameters of the digital twin model are calibrated using actual data collected by the physical system, and the material coefficients in the digital twin model are adjusted using the least squares method. In the digital twin modeling of personalized hydrogel skin scaffolds for burns, to accurately simulate the interaction between the scaffold printing process and the wound, COMSOL Multiphysics software is used to construct the digital twin model. Solid mechanics and mass transfer coupling physics fields are prioritized. The solid mechanics field simulates the deformation and stress distribution of the hydrogel scaffold during printing, while the mass transfer coupling field recreates the water diffusion and nutrient transfer processes within the hydrogel. The combination of these two fields comprehensively reflects the physical properties and biochemical environment of the scaffold. During modeling, specific parameters, including elastic modulus, Poisson's ratio, and diffusion coefficient, are first defined based on the selected hydrogel material. Then, the 3D model of the wound, acquired and processed using multimodal data, is imported into the software as boundary conditions to define the contact area and support range between the scaffold and the wound, ensuring the model closely matches the actual wound morphology.

[0039] Step S1263: Construct a printing parameter optimization module in the digital twin model, with the fit between the scaffold and the wound and the stability of the scaffold structure as optimization objectives. Use reinforcement learning training optimization strategy, with printing parameters as the action space and the fit and stability calculated by the digital twin model as reward values, and set the number of training iterations. When the closed-loop control is started, the digital twin model first predicts the optimal printing parameters based on the current wound data and sends them to the main control board of the 3D printing equipment for execution. In the digital twin modeling and printing control of personalized skin scaffolds for burns, a dedicated printing parameter optimization module needs to be constructed within the hydrogel scaffold digital twin model to achieve dynamic optimization of printing parameters to adapt to wound repair needs. This module optimizes two core indicators: scaffold-wound fit and scaffold structural stability. During the optimization strategy training phase, a reinforcement learning algorithm is employed. Key parameters in the 3D printing process are defined as the agent's action space, with each action corresponding to a specific set of parameters. Simultaneously, the scaffold fit and structural stability data calculated in real-time by the digital twin model are converted into reinforcement learning reward values. Higher fit and stronger structural stability result in higher reward values, and vice versa. This guides the agent to learn the optimal parameter adjustment direction, and a reasonable number of training iterations is set based on clinical printing needs and the model's computational capabilities. When the printing closed-loop control is initiated, the digital twin model first receives real-time data of the current wound from the edge computing module. Combined with the trained reinforcement learning optimization strategy, it quickly predicts the optimal combination of printing parameters to suit the current wound state. Then, these parameters are sent to the main control board of the 3D printing equipment through the PCIe 4.0 high-speed communication link. The main control board immediately adjusts the printing action according to the parameters to ensure that the printing of each layer of the scaffold can accurately match the current state of the wound, ensuring both a tight fit and reliable structural support, and realizing the dynamic and precise molding of personalized skin scaffolds.

[0040] Step S1264: During the execution of the 3D printing equipment main control board, the edge computing module collects data from various sensors of the physical system in real time and feeds it back to the digital twin model; the model compares the deviation between the actual data and the predicted data. If the deviation is >3%, the optimization parameters are recalculated and sent out, thus constructing a closed-loop process of prediction, execution, feedback and adjustment.

[0041] In the closed-loop control of 3D printing personalized skin scaffolds for burns, when the main control board of the 3D printing equipment executes the printing action according to the optimal parameters issued by the digital twin model, the edge computing module collects dynamic data from various sensors in the physical system in real time through GPIO interfaces, RS485 buses, etc. These data include the nozzle-scaffold contact pressure monitored by pressure sensors, the actual movement position of the print head recorded by displacement sensors, the interlayer flatness of the scaffold captured by optical sensors, and the local tissue deformation and temperature changes fed back by wound sensors. This collected actual data is transmitted to the digital twin model in real time, and the model compares it dimension by dimension with the simulation data predicted based on the optimized parameters (such as the predicted print head position trajectory, scaffold stress distribution, and wound contact pressure), calculating the deviation value between the two. If the deviation in any dimension exceeds 3% (e.g., the actual print head displacement deviates from the predicted displacement by 4%, or the actual fitting gap of the support deviates from the predicted gap by more than 5%), it is determined that the current printing parameters are no longer suitable for the real-time state of the physical system. The digital twin model will immediately start the parameter re-optimization process: combining the deviation data to adjust the printing parameters in reverse (e.g., fine-tuning the printing speed to correct the displacement deviation, adjusting the extrusion volume to improve the support fitting), recalculating the optimal parameters suitable for the current state, and quickly sending them to the main control board of the 3D printing equipment through a high-speed communication link. The main control board then updates the execution parameters. At the same time, the edge computing module continuously collects the adjusted actual data and feeds it back to the model for deviation verification.

[0042] Step S130: Test on a biomimetic skin model containing a dynamic fluid circulation system and a mechanical stretching device to verify the system path adjustment time, printing accuracy and vascular endothelialization efficiency.

[0043] In this embodiment, during the preclinical validation of personalized skin scaffolds for burns, to simulate the real physiological environment of human wounds (such as wound exudate circulation and the stress state of skin tissue), a biomimetic skin matrix containing a collagen and gelatin composite hydrogel was first prepared. The mechanical properties and biocompatibility of this type of matrix are close to those of human dermal tissue. At the same time, human umbilical vein endothelial cells were embedded in the matrix to simulate the vascular endothelial structure, providing a basis for subsequent validation of vascular endothelialization efficiency. A dynamic fluid circulation system was integrated at the bottom of the biomimetic skin model, which continuously delivered simulated wound exudate (containing nutrients such as glucose and amino acids and inflammatory factors, restoring the nutrient supply and metabolic state of the wound microenvironment) through a micro-peristaltic pump. Mechanical stretching devices driven by stepper motors were installed on both sides of the model, which can apply periodic or fixed deformation according to experimental needs (simulating the traction effect of human joint movement and the natural tension of the skin on the wound). During testing, the optimized 3D printing system was precisely integrated with the bionic skin model. First, the OCT device was used to collect real-time three-dimensional structural data of the surface and interior of the bionic skin model, which was then transmitted to the edge computing module to trigger the path planning process. Subsequently, a preset deformation (such as 10% stretching deformation simulating joint bending) was applied to the model through a mechanical stretching device. At the same time, the complete time from receiving the deformation data, recalculating the adaptation path, to outputting the new printing path was recorded. The test was repeated multiple times and the average value was taken to verify the system's path adjustment response speed in a dynamic wound environment, ensuring that the system can adapt to changes in wound morphology in a timely manner. After the scaffold is printed, the interface between the scaffold and the bionic skin model is scanned using a laser confocal microscope. The average gap between the interface and the scaffold is calculated using image analysis software (the average gap needs to be controlled within 50 micrometers) to verify whether the printing accuracy meets clinical requirements. The bionic skin model with the scaffold attached is then placed in a constant temperature incubator for continuous culture (usually 7-14 days, simulating the wound healing cycle). After culture, the model is subjected to CD31 immunofluorescence staining (CD31 is a specific marker for vascular endothelial cells). The number of tubular structures formed by endothelial cells is counted under a 200x microscope, and the endothelialization efficiency is assessed in conjunction with the integrity of the lumen. This determines whether the scaffold can provide a suitable microenvironment for angiogenesis, providing experimental evidence for the scaffold to promote wound vascularization and accelerate healing in subsequent clinical applications. The entire testing process not only replicates the dynamic physiological characteristics of human wounds but also comprehensively verifies the performance of the 3D printing system in path adjustment, molding accuracy, and biofunctional adaptability, ensuring that the personalized skin scaffold output by the system meets clinical repair needs.

[0044] Step S131: Prepare a biomimetic skin matrix containing collagen and gelatin composite hydrogel, embed human umbilical vein endothelial cells to simulate vascular endothelium, integrate a dynamic fluid circulation system at the bottom of the biomimetic skin model; use a micro peristaltic pump to deliver simulated wound exudate, and install mechanical stretching devices driven by stepper motors on both sides of the biomimetic skin model. In the biomimetic testing scenario of personalized skin scaffolds for burns, in order to reproduce the real physiological environment and mechanical state of the wound surface to verify the scaffold's adaptability, a biomimetic skin matrix of collagen and gelatin composite hydrogel was first prepared. The ratio of the two was adjusted according to the characteristics of clinical wound tissue, and the matrix was formed through a low-temperature cross-linking process so that the elastic modulus, water content and other parameters of the matrix were close to those of human dermal tissue. Subsequently, human umbilical vein endothelial cells cultured in vitro were uniformly embedded in the surface and superficial layers of the matrix to simulate the vascular endothelial structure of the wound, providing a basic model for subsequent evaluation of the scaffold's vascularization effect. A flow channel is pre-set at the bottom of the bionic skin model and a dynamic fluid circulation system is integrated. A micro peristaltic pump is selected as the power source. The pump body is connected to the flow channel through a silicone tube. The simulated wound exudate delivered is configured according to the composition of human wound exudate, and the flow rate is controlled at 0.5-1mL / h to simulate the nutrient supply and metabolic waste removal process of the wound microenvironment. At the same time, mechanical stretching devices driven by stepper motors are symmetrically installed on both sides of the bionic skin model. The motors are fixed to the edge of the model through elastic connectors. The stretching amplitude (0-20% deformation) and frequency (0.1-1Hz) can be set according to experimental needs. Periodic or static stretching is achieved through precise motor control to restore the stress state of wound tissue during human activity. This provides a bionic testing platform that fits clinical practice for subsequent testing of the path adjustment, printing accuracy and vascular endothelialization efficiency of the stent in a dynamic environment.

[0045] Step S132: Connect the optimized 3D printing equipment to the bionic skin model, collect the three-dimensional data of the bionic skin model surface in real time through the OCT device, and transmit it to the edge module for path calculation. After starting the test, apply the preset deformation through the mechanical stretching device, record the time from the edge module receiving the deformation data to outputting the new printing path, repeat the test and take the average value to verify the path adjustment time. In the dynamic adaptation performance test scenario of personalized skin scaffolds for burns, the 3D printing equipment with optimized parameters was first precisely docked with the bionic skin model. The initial distance between the printing nozzle and the model surface was adjusted (controlled within 0.1-0.3mm to ensure printing accuracy), and the relative position of the equipment and the model was calibrated so that the printing path could cover the simulated wound area of ​​the model. Subsequently, the OCT equipment was activated, and its high-resolution scanning function acquired real-time three-dimensional structural data of the surface and superficial layers of the bionic skin model (including the curvature of the model surface and the concavity and convexity of the simulated wound). This data was transmitted in real-time to the edge computing module (such as the NVIDIA Jetson AGX Orin module) via a data cable. The path planning algorithm within the module quickly calculated the initial printing path based on the latest three-dimensional data. After the test officially started, the mechanical stretching device was driven by stepper motors on both sides of the bionic skin model to apply deformation according to a preset plan (such as 5%-15% stretching deformation or local compression deformation simulating joint movement). At the same time, the timing function of the edge computing module was triggered. After receiving the deformation data, the edge module immediately started the path replanning process, re-analyzed the changes in wound morphology and generated a printing path adapted to the new morphology. When the new path calculation was completed and output to the main control board of the 3D printing equipment, the timing was stopped and the time taken for this path adjustment was recorded. To avoid the random error of a single test, the above test process was repeated (3-5 times). After removing outliers from the multiple recorded time data, the average value was taken to verify the path adjustment response speed of the entire system under dynamic wound changes, ensuring that the system can complete the path adaptation within a clinically acceptable time (such as milliseconds to seconds).

[0046] Step S133: After the scaffold is printed, use a laser confocal microscope to scan the interface between the scaffold and the bionic skin model, calculate the average bonding gap to verify the printing accuracy; after culturing for several more days, perform CD31 immunofluorescence staining on the bionic skin model, count the number of tubular structures formed by endothelial cells under 200x magnification, and calculate the vascular endothelialization efficiency.

[0047] After the 3D printing equipment completes the printing of the skin scaffold according to the optimized path, the accuracy of the fit between the scaffold and the bionic skin model is first verified. The scanning resolution of the laser confocal microscope is adjusted to an appropriate level, and multiple representative areas are selected along the bonding interface for layer-by-layer scanning to obtain the three-dimensional structure image of the interface. The scan data is processed by image analysis software to extract the gap value between the scaffold and the model surface in each area, and the arithmetic mean of the gaps in all areas is calculated. If the mean value is controlled within the preset threshold, it indicates that the printing accuracy meets the clinical requirements for scaffold fit. The biomimetic skin model with the scaffold was then transferred to a constant temperature incubator, and cultured under conditions set according to the wound healing simulation environment for several days. After culture, the model was subjected to CD31 immunofluorescence staining. First, the model tissue was fixed and permeable to enhance the permeability of the staining agent. Then, CD31-specific primary antibody and fluorescently labeled secondary antibody were added to make the vascular endothelial cells specifically visible. Finally, the stained model was placed under a fluorescence microscope, and multiple non-overlapping fields of view were randomly selected at 200x magnification. The number of tubular structures formed by endothelial cells in each field of view was counted. The vascular endothelialization efficiency was evaluated by calculating the average number of tubular structures in each field of view, thereby determining whether the scaffold can provide a good microenvironment for endothelial cell proliferation, migration and angiogenesis.

[0048] Step S140: Construct a self-sensing ink with integrated nanosensors to achieve in-situ monitoring of print quality and use optogenetics to precisely regulate cell behavior.

[0049] In this embodiment, in the clinical application of 3D printing of personalized skin scaffolds for burns, in order to control quality issues in the printing process in real time and ensure the subsequent biological function of the scaffold, it is necessary to construct a self-sensing ink with integrated nanosensors: carboxylated multi-walled carbon nanotubes are selected as the core sensing unit. Their excellent conductivity can accurately capture material state changes during the printing process. They are dispersed in 0.1 MME S buffer, and after removing agglomerates by probe ultrasonic treatment, they are magnetically stirred with a GelMA-sodium alginate mixture at 37°C. 0.5% w / v photoinitiator LAP is added to form a uniform and stable self-sensing ink. Simultaneously, a micro-flexible electrode array is embedded in the ink. One end of the electrode is electrically connected to the carbon nanotube network, and the other end is connected to the signal acquisition module built into the printhead via a flexible wire. During printing, the signal acquisition module captures the changes in resistance and impedance of the carbon nanotube network in real time. If uneven material extrusion occurs, the resistance will fluctuate abnormally. If the interlayer bonding is not strong, the impedance value will show a significant jump. These data are transmitted to the edge computing unit via Bluetooth Low Energy module. The quality assessment model based on gradient boosting tree in the unit will quickly analyze the data and output adjustment instructions, such as automatically increasing the nozzle temperature and adjusting the extrusion pressure to address clogging issues. This achieves a closed loop of in-situ monitoring of print quality, including sensing, evaluation, and adjustment, to avoid the scaffold failing to adhere to the wound or structural instability due to printing defects.

[0050] In the cell regulation phase after scaffold printing, optogenetics can precisely guide the proliferation and differentiation of cells within the scaffold: The seed cells on the scaffold are genetically modified beforehand to express channel proteins sensitive to specific wavelengths of light. Once the scaffold is transplanted to the wound, based on the healing stage, a preset wavelength and intensity of light signal is emitted to a specific area of ​​the scaffold via minimally invasively implanted micro-optical fibers. In the early stages of healing, blue light is emitted to activate the ChR2 protein, promoting calcium ion influx and accelerating epidermal stem cell proliferation to cover the wound. In the later stages of healing, the light signal wavelength is switched to inhibit excessive proliferation and guide fibroblasts to secrete appropriate amounts of collagen, preventing excessive scar formation. This precise regulation method can dynamically adjust cell behavior according to the patient's wound healing progress. Combined with real-time feedback from self-sensing ink on scaffold degradation status and cell activity data, it provides an integrated solution for personalized burn wound repair with controllable printing quality and adjustable cell behavior, especially suitable for repairing deep burn wounds (such as deep second- and third-degree burns).

[0051] Step S141: Select carboxylated multi-walled carbon nanotubes as sensing units and mix them with photocrosslinkable hydrogels to prepare bio-ink; Step S142: Disperse MWCNTs-COOH in 0.1M MES buffer, disperse by probe ultrasonication, and then remove undispersed aggregates by centrifugation; stir the treated MWCNTs solution with GelMA-sodium alginate mixture at 37°C with magnetic stirring, and add 0.5% w / v photoinitiator LAP to form a uniformly dispersed self-sensing ink. When preparing self-sensing ink for personalized skin scaffolds for burns, the carboxylated multi-walled carbon nanotubes (MWCNTs) must first be pretreated: they are slowly dispersed in 0.1M MES buffer and subjected to high-frequency ultrasonic treatment using a probe ultrasonic device to fully depolymerize MWCNTs-COOH; then, incompletely dispersed agglomerates in the solution are removed by centrifugation to avoid clogging the nozzles or affecting the sensing performance during subsequent printing. The pretreated MWCNTs solution needs to be compounded with a mixture of GelMA and sodium alginate: the two are magnetically stirred in a 37°C constant temperature water bath according to a preset ratio. During the stirring process, the speed needs to be controlled to prevent the generation of too many bubbles and ensure that the two solutions are fully mixed to form a homogeneous system. After the mixture is stable, 0.5% w / v of photoinitiator LAP is slowly added and stirring is continued until LAP is completely dissolved. Finally, a self-sensing ink with uniform dispersion, stable state and photocrosslinking properties is formed. This ink retains the conductive sensing ability of MWCNTs and has the biocompatibility and printability of GelMA and sodium alginate, which can meet the dual needs of quality monitoring and subsequent biological function during the printing of burn skin scaffolds.

[0052] Step S143: Embed a micro-flexible electrode array in the ink. One end of the electrode is electrically connected to the MWCNTs network, and the other end is connected to the signal acquisition module built into the printhead via a flexible wire. During printing, the signal acquisition module captures the resistance and impedance changes of the MWCNTs network in real time, and the data is transmitted to the edge computing unit via a Bluetooth Low Energy module. The edge unit runs a print quality evaluation model based on gradient boosting tree, outputs quality parameters in real time, and automatically sends adjustment commands to the printer main control board to form a closed loop of perception, evaluation, and adjustment. After printing, the uniformity of the distribution of nanosensors on the support surface is characterized by atomic force microscopy, and mechanical tests are performed.

[0053] Step S150: Establish a dynamic medical device certification framework, formulate risk grading standards for the real-time decision-making capabilities of algorithms, and use federated learning technology to share data across centers while protecting patient privacy; adopt a modular design to integrate edge computing modules with the printhead.

[0054] In the clinical promotion and multi-center collaborative scenarios of personalized skin scaffold 3D printing technology for burns, establishing a dynamic medical device certification framework is a core element in ensuring the safety and compliance of the technology. First, a cross-entity certification working group is formed, comprising regulatory agencies, third-party testing institutions, equipment R&D companies, and clinical experts in burn surgery. The responsibilities of each entity are clearly defined. Regulatory agencies formulate certification criteria based on clinical needs; third-party institutions evaluate device performance through simulation testing and in vitro validation; companies provide technical documentation; and clinical experts participate in risk assessment based on actual wound repair needs. The working group identifies core impact dimensions based on the algorithm's real-time decision-making capabilities, determining three categories of assessment indicators: decision relevance, data dependence, and error consequences. Based on this, a risk grading matrix is ​​constructed, classifying relevant algorithms into high, medium, and low levels: high-risk algorithms require three levels of validation—simulation testing, animal experiments, and clinical validation—to ensure reliability in complex wound scenarios; medium-risk algorithms require both simulation testing and in vitro biomimetic model validation; and low-risk algorithms only require simulation testing. At the same time, a dynamic certification mechanism is designed, requiring companies to establish algorithm update logs. Each iteration requires the submission of an update report to the regulatory agency. Low / medium risk algorithms can complete the certification update by passing online document review, while high risk algorithms need to supplement clinical data for re-evaluation, so as to avoid introducing new risks through algorithm iteration.

[0055] To break down data silos in multi-center collaboration and protect patient privacy, a federated learning technology was used to build a cross-center data sharing network. A central server was established by a hospital with leading technical capabilities, and a federated learning network comprised five to eight burn specialist hospitals from different regions. Each hospital deployed a local data preprocessing module to anonymize patient wound images and sensor monitoring data, removing direct identifiers such as names and ID numbers. Differential privacy techniques were used to add noise to wound images to protect patient privacy, and sensor data was standardized to ensure a consistent data format. During training, the central server first initialized a global deep learning model and distributed it to each hospital. Each hospital trained its model using its local anonymized data, calculating only the model gradients and processing them with homomorphic encryption before uploading them to the central server. The server decrypted the data, aggregated the gradients using a federated averaging algorithm, updated the global model parameters, and then distributed the updated model back to each hospital for iterative training until the model's loss function converged. Throughout the process, each hospital only uploaded encrypted gradients and did not share the original data. The central server used blockchain technology to record each parameter transmission, ensuring data traceability. The federated learning process was periodically audited by a third-party organization to further guarantee data security and model fairness.

[0056] Step S151: Establish a cross-entity certification working group, specifically including regulatory agencies, third-party testing institutions, equipment R&D companies, and clinical experts, clarifying the division of responsibilities. Regulatory agencies formulate certification criteria, third-party institutions are responsible for testing and verification, companies provide technical documentation, and clinical experts participate in risk assessment. The working group first identifies the core impact dimensions of real-time algorithm decision-making, determining three key assessment indicators: decision relevance, data dependence, and error consequences. Based on these indicators, a risk grading matrix is ​​constructed, classifying algorithms into three levels: high-risk, medium-risk, and low-risk. High-risk algorithms require three levels of verification: simulation testing, animal experiments, and clinical validation. Medium-risk algorithms require simulation testing and in vitro biomimetic model validation. Low-risk algorithms only require simulation testing. Simultaneously, a dynamic certification mechanism is designed, requiring companies to establish algorithm update logs and submit update reports to regulatory agencies for each iteration. Low / medium-risk algorithms only require online document review, while high-risk algorithms require supplementary clinical data for re-evaluation. A federated learning architecture is determined, adopting a horizontal federated learning model, selecting the federated averaging algorithm as the basic training framework, and building a system consisting of a central server and... A federated learning network consists of five to eight partner hospitals. Each participating hospital deploys a local data preprocessing module to anonymize patient data: removing direct identifiers such as names and ID numbers; employing differential privacy technology for wound images; standardizing various sensor data for wound condition monitoring; and conducting deep learning model training based on multimodal wound data. Specifically, the central server initializes the global model and distributes the initial model parameters to each hospital. Each hospital trains its model using its local anonymized data, calculating only the gradients of the deep learning model and adding homomorphic encryption to the gradients. Each hospital uploads the encrypted gradients to the central server, which decrypts them and aggregates the gradients using a federated averaging algorithm to update the global model parameters. The server then distributes the updated global model parameters back to each hospital, repeating iterative training until the deep learning model's loss function converges. It is explicitly stated that each center can only use the global model for local prediction and cannot download gradients or intermediate parameters from other centers. The central server uses blockchain technology to record each parameter transmission, ensuring data transmission traceability. The federated learning process is periodically audited by a third-party organization.

[0057] The edge computing module is designed as an independent unit, using a miniaturized edge computing board (such as the NVIDIA Jetson Nano 2GB, 69mm × 45mm), integrating a PCIe 2.0 interface (for high-speed data transmission), a GPIO interface (for sensor signal acquisition), and a USB 3.0 interface (for connecting OCT devices), and reserving a standardized power interface (12VDC, 15W). The printhead is divided into three independent units: a printhead module, a temperature control module, and a motor drive module. Each unit is designed with a standardized interface board: the printhead module interface board includes a PCIe slave interface (interfacing with the edge computing module's PCIe master interface) and a temperature sensor interface (NTC thermistor); the temperature control module interface board includes a PWM signal interface (receiving temperature control commands from the edge computing module); and the motor drive module interface board includes a stepper motor control interface (receiving path commands from the edge computing module).

[0058] The edge computing module and printhead interact via a PCIe interface. The edge computing module acts as the PCIe master, sending print path commands (G-code format, transmission rate ≥500Mbps) and temperature control commands (PWM signal, frequency 1kHz) to the printhead. The printhead acts as the slave, feeding back real-time status data (printhead temperature, motor position, extrusion pressure, sampling rate 100Hz) to the edge computing module via a GPIO interface. A signal isolation circuit (using an optocoupler isolation chip TLP181) is designed to prevent the printhead motor from interfering with the edge computing module's signals. Power supply integration is implemented using a 12VDC switching power supply (50W), which supplies power to the edge computing module (12V to 5V, 3A current) and the printhead module (12V direct supply, 3A current) via a splitter board, avoiding voltage fluctuations from affecting device stability.

[0059] A standardized driver program was developed on the edge computing module, compatible with the communication protocols of each module in the print head, enabling one-click driver loading; a data processing middleware was developed to convert the 3D wound data input from the OCT device and the printing status data fed back by the sensor into a format (unified to JSON format) and transmit it to the adaptation algorithm module; the printing path instructions and temperature control instructions generated by the algorithm module are sent to each module in the print head in real time through the driver program.

[0060] A compact mechanical housing (made of ABS plastic and fabricated via 3D printing) was designed to secure the edge computing module and other modules of the print head. The housing includes ventilation holes (a cooling fan with a speed of 30 CFM is designed above the edge computing module) and interface windows (exposing PCIe, USB, and power interfaces). Vibration testing was conducted (simulating printer operating vibration, frequency 50-200Hz, amplitude 0.1mm) to verify the stability of the module connection. Printing accuracy testing was also performed (printing a 5mm × 5mm grid support and measuring the grid spacing deviation, which must be <0.1mm) to ensure the integrated device meets performance standards.

[0061] Figure 2 The diagram illustrates exemplary hardware and software components of a personalized burn skin scaffold system 100 that integrates 3D printing and adaptation algorithms to realize the ideas of this application, according to some embodiments of this application.

[0062] For example, processor 120 can be used in a burn personalized skin scaffold system 100 that integrates 3D printing and adaptation algorithms, and to perform the functions in this application.

[0063] For example, a personalized burn skin scaffold system 100 based on fusion 3D printing and adaptation algorithms may include a network port 110 connected to a network, one or more processors 120 for executing program instructions, a communication bus 130, and various forms of storage media 140, such as a disk, ROM, or RAM, or any combination thereof. Exemplarily, the personalized burn skin scaffold system 100 may also include program instructions stored in ROM, RAM, or other types of non-transitory storage media, or any combination thereof. The methods of this application can be implemented according to these program instructions. The personalized burn skin scaffold system 100 also includes an I / O interface 150 between the computer and other input / output devices.

[0064] Furthermore, this embodiment of the invention also provides a readable storage medium, wherein computer-executable instructions are preset in the readable storage medium, and when the processor executes the computer-executable instructions, the above-mentioned fiber optic access network security protection and detection method based on AI abnormal pattern recognition is implemented.

[0065] It should be noted that, in order to simplify the description of the present invention and thus help to understand one or more embodiments of the invention, multiple features may sometimes be grouped into one embodiment, drawing or description thereof in the foregoing description of the embodiments of the present invention.

Claims

1. A method for designing personalized skin scaffolds for burns that integrates 3D printing and adaptation algorithms, characterized by: Includes the following steps: Multimodal data synchronous acquisition, modal noise reduction and standardized secure storage are performed. High-precision clock modules are added to OCT equipment, temperature sensors, pH biosensors and fluorescence imaging modules. A unified acquisition software is established to set the corresponding sampling frequency, and multi-device synchronous acquisition is achieved through a single trigger. A decomposition conformal path planning algorithm is established, employing a fast-expanding random tree and a GPU-accelerated particle swarm optimization algorithm; an edge computing module is integrated into the printing device, and a real-time data processing framework is used to establish a physical and digital twin system to optimize printing parameters for closed-loop control; The above system path adjustment time, printing accuracy and vascular endothelialization efficiency were verified by testing on a biomimetic skin model containing a dynamic fluid circulation system and a mechanical stretching device. We constructed a self-sensing ink with integrated nanosensors to achieve in-situ monitoring of print quality and used optogenetics to precisely regulate cell behavior. Establish a dynamic medical device certification framework, formulate risk grading standards for the real-time decision-making capabilities of algorithms, and adopt federated learning technology to share data across centers while protecting patient privacy; adopt a modular design to integrate edge computing modules with printheads.

2. The method for designing personalized skin scaffolds for burns that integrates 3D printing and adaptation algorithms according to claim 1, characterized in that: The method involves adding a high-precision clock module to the OCT device, temperature sensor, pH biosensor, and fluorescence imaging module, establishing unified acquisition software to set corresponding sampling frequencies, and achieving synchronous acquisition of multiple devices through a single trigger, including: An industrial-grade clock module supporting the IEEE 1588 PTP protocol was selected. The OCT equipment was designed with a PCIe expansion card with PTP protocol parsing function. The clock module was integrated into the expansion card and communicated with the OCT equipment motherboard through the gold fingers. I2C to PTP protocol adapter boards were made for the temperature sensor and pH biosensor. The sampling trigger terminals of the temperature sensor and pH biosensor were connected through the pins of the clock module. A PTP synchronization module was connected in series between the fluorescence imaging module and the imaging lens control board. The fluorescence exposure was triggered by the hardware interrupt pin of the fluorescence imaging module. Read the native communication protocols of OCT devices, temperature sensors, pH biosensors, and fluorescence imaging modules one by one, establish a protocol conversion intermediate layer program, embed the PTP protocol stack in the intermediate layer, and encapsulate the timestamp generated by the clock module into the sampling control instructions of each device. The acquisition software is divided into a device management module, a parameter configuration module, and a trigger control module. The device management module scans and identifies connected OCT devices, temperature sensors, pH biosensors, and fluorescence imaging modules, reads the hardware IDs of the OCT devices, temperature sensors, pH biosensors, and fluorescence imaging modules, and associates them with the corresponding clock module IPs. The parameter configuration module provides a visual interface that allows users to set the sampling frequency of the OCT device, the temperature sensor, the pH biosensor, and the fluorescence imaging module, converting the frequency parameters into the trigger period of the clock module and writing them into the module register; the trigger control module is designed with single-trigger logic. A synchronization verification thread is embedded in the acquisition software. After acquiring data for a set period of time, the timestamp field in the data returned by the OCT device, temperature sensor, pH biosensor, and fluorescence imaging module is automatically read, and the deviation between the timestamp of the OCT device, temperature sensor, pH biosensor, and fluorescence imaging module and the timestamp of the reference clock module is calculated.

3. The method for designing personalized skin scaffolds for burns that integrates 3D printing and adaptation algorithms according to claim 2, characterized in that: The acquisition software is divided into a device management module, a parameter configuration module, and a trigger control module. The device management module scans and identifies connected OCT devices, temperature sensors, pH biosensors, and fluorescence imaging modules, reads the hardware IDs of these devices, and associates them with the corresponding clock module IPs. The network scanning thread and the USB device enumeration thread are initialized. The network scanning thread traverses all IPs in the local area network by sending ARP requests and sends a device identification command to each IP. After receiving the command, the OCT device returns a response containing the hardware ID and the clock module IP. The USB device enumeration thread calls the system API to read the VID / PID and descriptor of the temperature sensor, pH biosensor, and fluorescence imaging module, filters out the temperature / pH sensor and fluorescence module, and obtains the hardware ID by sending a command through the USB endpoint. At the same time, it reads the associated clock module IP pre-stored in the firmware of the temperature sensor, pH biosensor, and fluorescence imaging module. After scanning, the hardware IDs, types, and clock module IPs of the OCT device, temperature sensor, pH biosensor, and fluorescence imaging module are stored in a memory list and displayed in the device list control on the interface. Users can manually confirm or modify the associations. Three numerical input boxes and an "Apply" button are placed. The input boxes have range limits and are automatically validated after input. If the input exceeds the range, a pop-up prompt will appear. After the user clicks "Apply," the software calculates the trigger cycle of each device. The clock module receives the data and returns confirmation. At the same time, the software writes the configuration parameters to a local configuration file, recording the hardware IDs of the OCT device, temperature sensor, pH biosensor, and fluorescence imaging module, as well as their corresponding frequencies and cycles. These are automatically loaded on the next startup. Add a "Start Acquisition" button to the interface. Clicking it triggers the control logic to check whether the OCT device, temperature sensor, pH biosensor, and fluorescence imaging module list are online. If any device is offline, a pop-up window will be displayed and the process will be terminated. After confirming that all devices are online, a global trigger timestamp will be generated, trigger commands will be constructed, and a multi-threaded concurrent sending mechanism will be adopted. Each device corresponds to a sending thread, and trigger commands will be sent to its associated clock module IP at the same time to ensure that all commands are sent within 1ms. After sending, a timeout detection is initiated, and responses are received from each clock module. If the OCT device, temperature sensor, pH biosensor, and fluorescence imaging module all respond, the interface will display "triggered successfully" and begin receiving device sampling data. If any device times out and does not respond, the sending will be retried immediately. If it still fails, the device will be marked as abnormal and the process for other devices will continue.

4. The method for designing personalized skin scaffolds for burns that integrates 3D printing and adaptation algorithms according to claim 1, characterized in that: A decomposition conformal path planning algorithm is established, employing a fast-expanding random tree and a GPU-accelerated particle swarm optimization algorithm. An edge computing module is integrated into the printing device, using a real-time data processing framework to establish a physical and digital twin system for closed-loop control of printing parameters, including: A path planning environment model is constructed based on the 3D point cloud data of the wound surface. The wound surface is discretized into a 3D mesh space, and the constraints of the printing path are defined. An adaptive sampling strategy is adopted in the fast expanding random tree algorithm, which increases the sampling density in the edge area of ​​the wound surface and decreases the sampling density in the flat area. When the fast expanding random tree algorithm is started, the printing start point is used as the root node, sampling points are randomly generated, and valid nodes are selected by collision detection. Adjacent nodes are connected to form a path tree through a greedy strategy. When the path tree extends to the printing end point, an initial coarse path is generated through path pruning to complete the global planning. The global coarse path is discretized into N path control points, and the three-dimensional coordinates of each control point are used as the dimension parameter of the particle to construct a particle swarm. The fitness function is defined as the fit error between the path and the wound surface and the path smoothness. The fit error is calculated by summing the shortest distances from the control points to the wound surface, and the path smoothness is calculated by weighting the cosine values ​​of the angles between adjacent control points. The particle swarm data is transferred to the GPU, and the number of CUDA thread blocks is configured to be 1 / 32 of the number of particles, with each block containing 32 threads; after each iteration, the optimal particle information is shared through the GPU's global memory. An NVIDIA Jetson AGX Orin module was selected and connected to the main control board of the 3D printing equipment via a PCIe 4.0 interface; at the same time, a wound sensor was connected via a GPIO interface, and the interface sampling rate was configured to be consistent with that of the wound sensor. An Ubuntu 20.04 system was installed on the NVIDIA Jetson AGX Orin module, and a Python 3.8 environment was configured. The Pathway framework and its dependencies were installed via pip. A data processing pipeline was built based on Pathway, dividing wound sensor data into real-time and near-real-time streams and allocating different computing resources to each. An incremental computing engine was configured to process wound sensor data using a sliding window, calculating the difference between new data and historical results within the window. Multimodal data was partitioned and stored according to timestamps using the framework's data partitioning function. Finally, a physical and digital twin system is established to construct a closed-loop control system.

5. The method for designing personalized skin scaffolds for burns that integrates 3D printing and adaptation algorithms according to claim 4, characterized in that: Establishing a physical and digital twin system to construct closed-loop control includes: Multiple types of sensors, including pressure sensors, displacement sensors, and optical sensors, are deployed in the printing area of ​​the 3D printing bracket. The data from each sensor is transmitted to the edge computing module via an RS485 bus. After data preprocessing, the data is encapsulated according to the sensor ID, timestamp, and data value format and stored in the local database of the edge module. At the same time, updated data is sent to the digital twin model periodically. A digital twin model of the hydrogel scaffold was constructed using COMSOL Multiphysics. Solid mechanics and mass transfer coupled physical fields were selected, and scaffold material parameters were defined. The 3D model of the wound was imported as boundary conditions. The digital twin model was meshed, with a fine mesh for the scaffold area and a coarse mesh for the wound area to balance computational accuracy and speed. The parameters of the digital twin model were calibrated using actual data collected by the physical system, and the material coefficients in the digital twin model were adjusted using the least squares method. A printing parameter optimization module is built in the digital twin model, with the fit between the scaffold and the wound and the stability of the scaffold structure as optimization objectives. A reinforcement learning training optimization strategy is adopted, with the printing parameters as the action space and the fit and stability calculated by the digital twin model as the reward value, and the number of training iterations is set. When the closed-loop control is started, the digital twin model first predicts the optimal printing parameters based on the current wound data and sends them to the main control board of the 3D printing equipment for execution. During the execution of the 3D printing equipment's main control board, the edge computing module collects data from various sensors of the physical system in real time and feeds it back to the digital twin model. The model compares the deviation between the actual data and the predicted data. If the deviation is greater than 3%, the optimization parameters are recalculated and sent out, thus constructing a closed-loop process of prediction, execution, feedback, and adjustment.

6. The method for designing personalized skin scaffolds for burns that integrates 3D printing and adaptation algorithms according to claim 5, characterized in that: The system was tested on a biomimetic skin model incorporating a dynamic fluid circulation system and a mechanical stretching device to verify the system's path adjustment time, printing accuracy, and vascular endothelialization efficiency, including: A biomimetic skin matrix containing collagen and gelatin composite hydrogel was prepared, and human umbilical vein endothelial cells were embedded to simulate vascular endothelium. A dynamic fluid circulation system was integrated at the bottom of the biomimetic skin model. A micro peristaltic pump was used to deliver simulated wound exudate, and mechanical stretching devices driven by stepper motors were installed on both sides of the biomimetic skin model. The optimized 3D printing equipment was docked with the bionic skin model. The OCT device was used to collect the three-dimensional data of the surface of the bionic skin model in real time and transmit it to the edge module for path calculation. After the test was started, the preset deformation was applied by the mechanical stretching device. The time from receiving the deformation data to outputting the new printing path was recorded. The test was repeated and the average value was taken to verify the path adjustment time. After the scaffold is printed, the interface between the scaffold and the bionic skin model is scanned using a laser confocal microscope, and the average bonding gap is calculated to verify the printing accuracy. After several more days of culture, the bionic skin model is subjected to CD31 immunofluorescence staining, and the number of tubular structures formed by endothelial cells is counted under 200x magnification to calculate the vascular endothelialization efficiency.

7. The method for designing personalized skin scaffolds for burns that integrates 3D printing and adaptation algorithms according to claim 1, characterized in that: Constructing self-sensing inks with integrated nanosensors to achieve in-situ monitoring of print quality, and utilizing optogenetics to precisely regulate cell behavior, including: Bio-ink was prepared by using carboxylated multi-walled carbon nanotubes as sensing units and mixing them with photocrosslinkable hydrogels. MWCNTs-COOH were dispersed in 0.1M MES buffer, ultrasonically dispersed by probe, and then the undispersed aggregates were removed by centrifugation. The treated MWCNTs solution was magnetically stirred with GelMA-sodium alginate mixture at 37°C, and 0.5% w / v photoinitiator LAP was added to form a uniformly dispersed self-sensing ink. A micro-flexible electrode array is embedded in the ink. One end of the electrode is electrically connected to the MWCNTs network, and the other end is connected to the signal acquisition module built into the printhead through a flexible wire. During the printing process, the signal acquisition module captures the resistance and impedance changes of the MWCNTs network in real time, and the data is transmitted to the edge computing unit via the Bluetooth Low Energy module. The edge unit runs a print quality evaluation model based on gradient boosting tree, outputs quality parameters in real time and automatically sends adjustment commands to the printer main control board, forming a closed loop of perception, evaluation and adjustment. After printing, the uniformity of the distribution of nanosensors on the scaffold surface was characterized by atomic force microscopy, and mechanical tests were performed.

8. The method for designing personalized skin scaffolds for burns that integrates 3D printing and adaptation algorithms according to claim 1, characterized in that: The establishment of a dynamic medical device certification framework, which formulates risk grading standards based on the real-time decision-making capabilities of algorithms, includes: A cross-entity certification working group was established, including regulatory agencies, third-party testing institutions, equipment R&D companies, and clinical experts. The division of responsibilities was clarified, with regulatory agencies formulating certification criteria, third-party institutions responsible for testing and verification, companies providing technical documentation, and clinical experts participating in risk assessment. The working group first identified the core impact dimensions of the algorithm's real-time decision-making, and determined three key assessment indicators: decision relevance, data dependence, and error consequences. Based on these indicators, a risk classification matrix was constructed, classifying the algorithm into three levels: high risk, medium risk, and low risk. High-risk algorithms need to complete three levels of validation: simulation testing, animal experiments, and clinical trials; medium-risk algorithms need to complete simulation testing and in vitro biomimetic model validation; low-risk algorithms only need to complete simulation testing. At the same time, the dynamic certification mechanism requires companies to establish algorithm update logs. Each iteration requires the submission of an update report to the regulatory agency. Low / medium risk algorithms can be reviewed online, while high risk algorithms need to be supplemented with clinical data for re-evaluation.

9. The method for designing personalized skin scaffolds for burns that integrates 3D printing and adaptation algorithms according to claim 1, characterized in that: The aforementioned federated learning technology enables cross-center data sharing while protecting patient privacy. The modular design integrates the edge computing module with the printhead, including: The federated learning architecture was determined, a horizontal federated learning model was adopted, the federated averaging algorithm was selected as the basic training framework, and a federated learning network consisting of a central server and five to eight cooperating hospitals was built. Each participating hospital deployed a data preprocessing module locally to de-identify patient data: removing direct identifiers such as names and ID numbers, using differential privacy technology for wound images, and standardizing various sensor data for wound status monitoring. The deep learning model training process based on multimodal wound data involves the following steps: The central server initializes the global model and distributes the initial model parameters to each hospital; each hospital trains its model using local anonymized data, calculating only the gradients of the deep learning model and adding homomorphic encryption to the gradients; each hospital uploads the encrypted gradients to the central server, which decrypts them and aggregates the gradients using a federated averaging algorithm to update the global model parameters; the server then distributes the updated global model parameters to each hospital, and the training is repeated iteratively until the loss function of the deep learning model converges. It is clearly stipulated that each center can only use the global model for local prediction and cannot download gradients or intermediate parameters of other centers' models; the central server uses blockchain technology to record each parameter transmission to ensure data transmission traceability; and the federated learning process is audited regularly by a third-party organization.

10. The personalized burn skin scaffold system integrating 3D printing and adaptation algorithms according to claim 1, characterized in that: include: processor; A machine-readable storage medium for storing machine-executable instructions of the processor; The processor is configured to execute the burn personalized skin scaffold design method that integrates 3D printing and adaptation algorithms as described in any one of claims 1 to 9 by executing the machine-executable instructions.

Images