A sewing technician online learning and practical operation evaluation method and system

Abstract

The present application belongs to the technical field of sewing technician online learning and practical operation evaluation system, and particularly relates to a sewing technician online learning and practical operation evaluation method and system, comprising the following steps: S1: initialization setting: starting a sewing device, a visual acquisition device, a man-machine interaction terminal and a server, completing self-checking of respective hardware, and establishing a multi-protocol layered communication submodule and the like. The present application cooperatively transmits three types of information of instructions, running state data and video streams through the multi-protocol layered communication submodule, realizes accurate time alignment of the video streams and the running state data by using a data fusion module and a time series interpolation compensation mechanism, and generates a structured analysis result locally by relying on an edge reasoning engine to load an adaptive recognition model matched with a process type and fabric characteristics, so as to support a high-precision, low-delay teaching and examination closed loop under the premise of guaranteeing communication reliability and real-time performance.

Description

Technical Field

[0001] This invention belongs to the technical field of online learning and practical assessment systems for sewing technicians, and specifically relates to a method and system for online learning and practical assessment of sewing technicians. Background Technology

[0002] In the existing field of online learning and practical assessment systems for sewing technicians, a common architecture is "intelligent sewing equipment + visual acquisition equipment + server," which uses communication protocols to achieve device interconnection and data interaction. To ensure the reliability of command issuance and status feedback, TCP is typically used to transmit the sewing equipment's operating status data (such as speed, stitch length, and presser foot pressure). To meet the real-time requirements of video streaming, UDP is typically used to transmit the operational video stream acquired by the visual acquisition equipment. However, this single-protocol approach leads to insufficient communication integration: while TCP can guarantee no packet loss of critical status data, its connection-oriented nature and complex retransmission mechanism significantly increase video stream transmission latency, affecting the timeliness of visual analysis; although UDP has the advantage of low latency, its connectionless and unacknowledged characteristics easily cause the loss of critical operating status data of the sewing equipment, leading to distorted assessment results. Meanwhile, existing systems lack a time alignment mechanism for cross-source data. Video streams and device operation status data are transmitted via different protocols and paths, making it difficult for the receiving end to accurately determine the specific device status corresponding to a certain video frame. For example, when an operator completes a corner sewing action, the video screen shows a trajectory deviation, but the synchronously received device data may be delayed due to UDP frame loss or TCP queuing delay, causing an "action-state" misalignment. This makes it impossible for AI to accurately attribute the error, directly affecting the accuracy of teaching error correction and assessment scoring. Summary of the Invention

[0003] (a) Technical problems to be solved To overcome the shortcomings of insufficient communication integration and cross-source data time misalignment in existing online learning and practical assessment systems for sewing technicians, this invention provides a method and system for online learning and practical assessment of sewing technicians. It utilizes a multi-protocol layered communication submodule to collaboratively transmit three types of information: instructions, operational status data, and video streams. A data fusion module and a time series interpolation compensation mechanism are used to achieve precise time alignment between the video stream and operational status data. Furthermore, an edge inference engine loads an adaptive recognition model that matches the process type and fabric characteristics, generating structured analysis results locally. This ensures high-precision, low-latency teaching and assessment closed loops while guaranteeing communication reliability and real-time performance.

[0004] (II) Technical Solution This invention is achieved through the following technical solution: This invention proposes an online learning and practical assessment method for sewing technicians, comprising the following steps: S1: Initialization settings: Start the sewing equipment, vision acquisition equipment, human-computer interaction terminal and server, complete their respective hardware self-tests, and establish a multi-protocol layered communication submodule; The multi-protocol layered communication submodule supports the coordinated transmission of three types of information: instructions, running status data, and video streams. These three types of information share a unified frame structure, which includes a device identification field, a data type field, and an integrity verification field. S2: Instruction Reception and Distribution: The server receives a mode selection instruction sent by the human-computer interaction terminal. The mode selection instruction includes a teaching mode or an assessment mode, and is accompanied by a process type identifier and a fabric characteristic identifier. Wherein, after the server parses the mode selection instruction, it sends the mode selection instruction to the sewing device through the multi-protocol layered communication submodule; S3: Data Acquisition and Fusion: The visual acquisition device acquires the operator's work video stream, and the sewing equipment acquires operating status data; The sewing equipment has a built-in data fusion module for receiving the operation video stream and the operation status data, and encapsulating the two based on a unified frame structure. The data fusion module operates a time series interpolation compensation mechanism to repair the missing timestamps caused by frame loss during video stream transmission and to achieve time alignment between the video stream and the running status data. S4: Edge Intelligent Analysis: First, the local edge inference engine of the sewing equipment loads an adaptive recognition model that matches the process type identifier and the fabric characteristic identifier, specifically including the following steps: S41: Model Index Location: The edge inference engine retrieves the corresponding model file path from the locally pre-stored model index table based on the received process type identifier and fabric characteristic identifier; the adaptive model index table is stored in key-value pair format, where the key is the combination identifier of process type and fabric characteristic, and the value is the storage address of the model file in the flash memory of the sewing equipment. S42: Model Loading and Memory Mapping: The edge inference engine calls the lightweight inference framework to read the retrieved model file from flash memory to runtime memory and establish a tensor memory mapping region; the lightweight inference framework supports integer quantization model parsing; S43: Parameter Dynamic Injection: The edge inference engine reads a preset parameter set that matches the adaptive model from the local configuration file. The preset parameter set includes image preprocessing parameters, trajectory deviation judgment threshold, and finished product quality score weight coefficient, and injects the preset parameter set into the model inference process. S44: Model Readiness Confirmation: The edge inference engine performs single-frame empty input inference to verify the integrity of model loading and operator compatibility; if the verification is successful, it returns a model loading success status code to the server and enters the pending analysis task state. Next, the edge inference engine performs at least one analytical task, including operational compliance judgment, trajectory deviation identification, or finished product quality assessment. The edge inference engine has a built-in task scheduling module, which is used to dynamically allocate computing power based on system resource utilization and task priority. The edge inference engine integrates an environment robustness enhancement module, which is used to perform illumination normalization, motion blur suppression and texture noise filtering on the input image sequence; Finally, based on the execution results of the operation compliance judgment, trajectory deviation identification, or finished product quality assessment, the edge inference engine generates teaching error correction instructions containing error type identifiers and spatial positioning coordinates, or structured assessment scoring results containing dimension scores, deviation positioning diagrams, and improvement suggestion texts. The teaching error correction instructions or structured assessment scoring results are compressed and uploaded to the cloud platform deployed on the server. S5: Result Feedback and Closed Loop: The cloud platform will send the teaching error correction guidance information or structured assessment scoring results generated from the analysis results to the sewing equipment and the human-computer interaction terminal through the multi-protocol layered communication submodule, driving the output of audio-visual prompts or graphic reports, and after the task is completed, the sewing equipment, visual acquisition equipment, human-computer interaction terminal and server will enter standby state.

[0005] Preferably, in step S3, after receiving the data packet sent by the vision acquisition device or the sewing device itself, the sewing device verifies the data packet according to the integrity verification field; if the verification fails, a retransmission instruction is returned to the corresponding sender.

[0006] Preferably, in step S3, the time series interpolation compensation mechanism specifically involves: the data fusion module extracting the timestamps of two adjacent valid video frames, constructing a linear time mapping relationship, and for video frames with missing timestamps, calculating their proper time sequence position based on the linear time mapping relationship and filling in the corresponding timestamps.

[0007] Preferably, in step S4, the adaptive recognition model is configured to automatically load the corresponding recognition parameter set in response to the process type identifier and the fabric characteristic identifier.

[0008] Preferably, in step S4, the task scheduling module is configured to: when it is detected that the video decoding task has a higher priority than the feature extraction task, prioritize the allocation of computing resources to complete the video frame decoding, and cache the decoding results for subsequent subtasks to call.

[0009] Preferably, in step S4, the environmental robustness enhancement module is configured to be coupled to the input layer of the adaptive recognition model to ensure that the enhanced image is directly used as the model input.

[0010] Preferably, in step S4, the strategy for uploading the analysis results is: only the analysis result summary, key event timestamps, and model running status logs are uploaded, and the original video stream or uncompressed running status data is not uploaded.

[0011] Preferably, in step S5: The teaching error correction guidance information is output in a multimodal form, including: synchronously displaying error type identifiers and corresponding standard action comparison diagrams on the human-computer interaction terminal, and synchronously outputting corrective voice prompts through an audio-visual prompting device; The structured assessment results are output in the form of a structured report, including: displaying dimensional scores, deviation location diagrams, and improvement suggestion text on the human-computer interaction terminal, and archiving the structured report to the cloud platform.

[0012] On the other hand, this application also provides an online learning and practical assessment system for sewing technicians, used to implement the online learning and practical assessment method for sewing technicians described in any of the above claims, including: The sewing equipment integrates an operating status sensing module, a data fusion module, an edge inference engine, and an audio-visual prompting device. A visual acquisition device is used to acquire video streams of the operator's work and send the video streams to the sewing equipment through the multi-protocol communication submodule; The human-computer interaction terminal is used to provide a mode selection interface and send mode selection instructions to the server and receive teaching error correction guidance information or structured assessment scoring results from the server through the multi-protocol communication submodule. The server is equipped with a cloud platform for parsing mode selection instructions, generating task instructions, and sending task instructions to the sewing equipment and receiving analysis results uploaded by the sewing equipment through the multi-protocol communication submodule. A multi-protocol communication submodule, integrated into the sewing equipment or the server, is used to establish and maintain a communication link between the sewing equipment, the vision acquisition device, the human-computer interaction terminal, and the server. It supports the coordinated transmission of three types of information: instructions, operating status data, and video streams, and ensures the time consistency of cross-source data.

[0013] (III) Beneficial Effects This invention constructs a multi-protocol layered communication submodule and adopts a unified frame structure to standardize the encapsulation and integrity verification of instructions, running status data and video streams. This achieves the effect of ensuring zero packet loss transmission of critical status data and meeting the requirement of low-latency real-time return of video streams, significantly improving the communication reliability and real-time performance of the system in complex industrial environments. This invention achieves millisecond-level precise time alignment between video stream and running status data by running a time series interpolation compensation mechanism in the data fusion module built into the sewing equipment. It constructs a linear mapping relationship based on the timestamps of adjacent valid video frames and dynamically fills in missing timestamps, thus providing an accurate action and status correlation benchmark for AI analysis and supporting high-precision trajectory deviation detection and error attribution. This invention loads an adaptive recognition model that matches the process type identifier and fabric characteristic identifier through an edge inference engine, and integrates an environment robustness enhancement module to perform illumination normalization, motion blur suppression and texture noise filtering on the input image. This achieves a recognition accuracy of over 85% under complex working conditions such as curved sewing and denim, and solves the problems of poor adaptability and weak generalization ability of traditional fixed models. This invention deploys core AI analysis tasks such as operation compliance judgment and trajectory deviation recognition on the edge inference engine of the sewing equipment, and the task scheduling module dynamically allocates computing power to prioritize high-priority tasks such as video decoding, achieving the effect of controlling the local analysis feedback latency within 130ms, realizing millisecond-level real-time error correction in the teaching process, and getting rid of the dependence on cloud computing power. This invention generates teaching error correction instructions containing error type identifiers and spatial positioning coordinates, as well as structured assessment scores containing dimensional scores, deviation positioning diagrams, and improvement suggestion texts. These scores are then displayed on the persistent interface of the human-computer interaction terminal, and the audio-visual prompts are driven to output feedback. This achieves the effects of highly intuitive teaching error correction and objective and fair assessment, significantly improving trainees' comprehension efficiency and the quality of the training loop. Attached Figure Description

[0014] Other features, objects, and advantages of the present invention will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings: Figure 1 This is a schematic diagram of the method flow of the present invention. Detailed Implementation

[0015] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description refers to the drawings, unless otherwise indicated, the same numbers in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present invention; rather, they are merely examples of apparatuses and methods consistent with some aspects of the invention as detailed in the appended claims. Existing online teaching systems for sewing technicians generally use a single communication protocol to transmit instructions, operational status data, and video streams. This makes it difficult to balance reliability and real-time performance. Furthermore, the lack of a cross-source data time alignment mechanism results in inaccurate correspondence between video footage and equipment status, affecting the accuracy of teaching error correction and assessment scoring. Therefore, referring to... Figure 1 As shown, this invention proposes an online learning and practical assessment method for sewing technicians, comprising the following steps: S1: Initialization settings: Start the sewing equipment, vision acquisition equipment, human-computer interaction terminal and server, complete their respective hardware self-tests, and establish a multi-protocol layered communication submodule; In practice, the sewing equipment uses an industrial-grade intelligent flatbed sewing machine with a built-in core control chip and various sensors (including: a speed sensor (for acquiring spindle rotation speed); a stitch length sensor (for detecting the displacement of the sewing needle per cycle); a presser foot pressure sensor (for monitoring the downward pressure of the presser foot on the fabric); a motor current sensor (for reflecting load changes and abnormal overloads); a bobbin thread tension sensor (for real-time monitoring of the bobbin thread tension and supporting thread breakage / missing thread detection); and a thread tension sensor (for monitoring the top thread tension and assisting in the control of stitch length and stitch stability)). The vision acquisition device uses a global shutter industrial camera with adjustable resolution. The human-machine interface terminal uses an industrial touch panel. The server uses a high-performance computing node deployed in the cloud. The multi-protocol layered communication submodule supports the coordinated transmission of three types of information: instructions, running status data, and video streams. These three types of information share a unified frame structure, which includes a device identification field, a data type field, and an integrity verification field. S2: Instruction Reception and Distribution: The server receives the mode selection instruction sent by the human-computer interaction terminal. The mode selection instruction includes teaching mode or assessment mode, and is accompanied by process type identifier (such as process type identifier code 0x01 representing flat sewing, 0x02 representing curved sewing, etc.) and fabric characteristic identifier (such as 0xA1 representing cotton and linen, 0xA2 representing denim, etc.). Wherein, after the server parses the mode selection instruction, it sends the mode selection instruction to the sewing device through the multi-protocol layered communication submodule; S3: Data Acquisition and Fusion: The visual acquisition device acquires the operator's work video stream, and the sewing equipment acquires operating status data; The sewing equipment has a built-in data fusion module for receiving the operation video stream and the operation status data, and encapsulating the two based on a unified frame structure. The data fusion module operates a time series interpolation compensation mechanism to repair the missing timestamps caused by frame loss during video stream transmission and to achieve time alignment between the video stream and the running status data. S4: Edge Intelligent Analysis: First, the local edge inference engine of the sewing equipment loads an adaptive recognition model that matches the process type identifier and the fabric characteristic identifier, specifically including the following steps: S41: Model Index Location: The edge inference engine retrieves the corresponding model file path from the locally pre-stored model index table based on the received process type identifier and fabric characteristic identifier; the adaptive model index table is stored in key-value pair format, where the key is the combination identifier of process type and fabric characteristic, and the value is the storage address of the model file in the flash memory of the sewing equipment. S42: Model Loading and Memory Mapping: The edge inference engine calls the lightweight inference framework to read the retrieved model file from flash memory to runtime memory and establish a tensor memory mapping region; the lightweight inference framework supports integer quantization model parsing; S43: Parameter Dynamic Injection: The edge inference engine reads a preset parameter set that matches the adaptive model from the local configuration file. The preset parameter set includes image preprocessing parameters, trajectory deviation judgment threshold, and finished product quality score weight coefficient, and injects the preset parameter set into the model inference process. S44: Model Readiness Confirmation: The edge inference engine performs single-frame empty input inference to verify the integrity of model loading and operator compatibility; if the verification is successful, it returns a model loading success status code to the server and enters the pending analysis task state. In the specific implementation, the model index table is stored in the external SPI Flash chip of the sewing device, using a hash table structure to improve retrieval speed. For example, the key value "Curve_Denim" points to address 0x08004000. The lightweight inference framework uses TensorFlow Lite Micro or ONNX Runtime Mobile. These frameworks are designed for embedded devices and have the advantages of small memory footprint and fast inference speed. In practical applications, other compatible lightweight engines can also be selected. This application embodiment does not limit this. The running memory is used to store model weights and intermediate activation values. The preset parameter set is stored in EEPROM, including normalized mean (e.g., 123.675), standard deviation (e.g., 58.395), and texture filtering threshold for specific fabrics.

[0016] Next, the edge inference engine performs at least one analytical task, including operational compliance judgment, trajectory deviation identification, or finished product quality assessment. The above-mentioned S41-S44 quickly locates the specific scene model through the index table, reduces data copying overhead by using memory mapping technology, dynamically injects parameters to adapt to different working conditions, and ensures model availability through no-load testing, thereby realizing on-demand loading and immediate effect of the model, thus achieving accurate matching between the model and the process fabric and greatly improving the recognition accuracy; through the localized loading and verification mechanism, it ensures that the system can still work normally in the offline environment, improving the robustness and response speed of the system.

[0017] The edge inference engine has a built-in task scheduling module, which is used to dynamically allocate computing power based on system resource utilization and task priority. The edge inference engine integrates an environment robustness enhancement module, which is used to perform illumination normalization, motion blur suppression and texture noise filtering on the input image sequence; Finally, based on the execution results of the operation compliance judgment, trajectory deviation identification, or finished product quality assessment, the edge inference engine generates teaching error correction instructions containing error type identifiers and spatial positioning coordinates, or structured assessment scoring results containing dimension scores, deviation positioning diagrams, and improvement suggestion texts. The teaching error correction instructions or structured assessment scoring results are compressed and uploaded to the cloud platform deployed on the server. S5: Result Feedback and Closed Loop: The cloud platform will send the teaching error correction guidance information or structured assessment scoring results generated by the analysis results to the sewing equipment and the human-computer interaction terminal through the multi-protocol layered communication submodule, drive the output of audio-visual prompts or graphic reports, and after the task is completed, the sewing equipment, visual acquisition equipment, human-computer interaction terminal and server will enter standby state. In the above, this application establishes a multi-protocol layered communication submodule to coordinate the transmission of three types of information: instructions, running status data, and video streams, and to share a unified frame structure; The video stream and operating status data are encapsulated by the data fusion module built into the sewing equipment, and a time series interpolation compensation mechanism is run to repair the missing timestamps caused by frame drops, so as to achieve time alignment between the two. Then, the edge inference engine loads an adaptive recognition model that matches the process type and fabric characteristics, performs analysis tasks and generates structured results. Finally, the cloud platform sends the results to the sewing equipment and human-machine interaction terminal to drive the output of audio-visual prompts or graphic reports. The above method can ensure that data of different natures are properly placed, which not only ensures that critical status data is not lost, but also meets the low latency requirements of video streams. Furthermore, time alignment ensures that operation actions are accurately associated with device status. In addition, edge-side intelligent analysis avoids cloud dependence and improves the real-time performance of feedback. Meanwhile, the structured output of results supports intuitive teaching and objective assessment.

[0018] In step S3, after receiving a data packet sent by the vision acquisition device or the sewing device itself, the sewing device verifies the data packet based on the integrity verification field. If the verification fails, a retransmission instruction is returned to the corresponding sender. That is, the cyclic redundancy check algorithm is used to detect bit flipping or packet loss errors during data transmission, and the sending end is triggered to retransmit through an instant feedback mechanism to ensure that the data obtained by the receiving end is complete and error-free. This effectively intercepts erroneous data transmission and avoids AI misjudgment caused by data anomalies. A reliable data transmission closed loop is established, which significantly improves the communication stability of the system in complex electromagnetic environments. In the specific implementation, the integrity check field is generated using the CRC8 algorithm, with the polynomial selected as 0x07. After receiving the data packet, the communication protocol stack of the sewing equipment immediately recalculates the CRC8 value of the payload and compares it with the integrity check field at the end of the data packet. If the two are inconsistent, the check fails. The communication protocol stack immediately constructs a retransmission request frame containing an error code (such as 0x01) and sends it back to the data sending end of the vision acquisition device or the sewing equipment through the multi-protocol layered communication submodule. After receiving the retransmission instruction, the sending end retrieves the original data packet from the sending buffer and retransmits it. In practical applications, the check algorithm can also be selected as CRC16 or MD5, and the retransmission limit can also be dynamically adjusted according to the network conditions. This application embodiment does not limit this.

[0019] In step S3, the time series interpolation compensation mechanism specifically involves the data fusion module extracting the timestamps of two adjacent valid video frames, constructing a linear time mapping relationship, and calculating the corresponding time position of video frames with missing timestamps based on the linear time mapping relationship. That is, based on the constant frame rate characteristics of the video stream, a linear model is constructed using the timestamps of known frames to deduce the theoretical time position of missing frames, thereby logically completing the time axis and achieving precise alignment between the video stream and continuously sampled device status data in the time dimension. This eliminates the problem of missing video frame timestamps caused by network fluctuations, ensures the accuracy of the "operation action - device status" association, and provides high-quality time reference data for subsequent trajectory deviation detection. In specific implementation, the data fusion module maintains a timestamp queue of size N (e.g., N=10). When a non-contiguous video frame sequence number is detected (e.g., a jump from Frame_100 to Frame_102), the module extracts the timestamp T_start of Frame_100 and the timestamp T_en of Frame_102. Assuming a frame rate of 30fps, the theoretical time interval Δt = 1 / 30 seconds. The module constructs a linear equation T_missing = T_start + k * Δt (where k is the offset of the missing frame sequence number), calculates the theoretical timestamp of the missing frame Frame_101, and fills it into the data structure. This process is executed by the DSP coprocessor inside the sewing device, using fixed-point arithmetic to improve efficiency. In practical applications, the interpolation algorithm can also be quadratic spline interpolation or Lagrange interpolation; this embodiment does not limit this choice.

[0020] In step S4, the adaptive recognition model is configured to automatically load the corresponding recognition parameter set in response to the process type identifier and the fabric characteristic identifier.

[0021] In step S4, the task scheduling module is configured to: when it is detected that the video decoding task has a higher priority than the feature extraction task, prioritize the allocation of computing resources to complete the video frame decoding, and cache the decoding results for subsequent subtasks to call.

[0022] In step S4, the environmental robustness enhancement module is configured to be coupled with the input layer of the adaptive recognition model to ensure that the enhanced image is directly used as the model input.

[0023] In step S4, the strategy for uploading the analysis results is as follows: only the analysis result summary, key event timestamps, and model running status logs are uploaded; the original video stream or uncompressed running status data is not uploaded.

[0024] In step S5: The teaching error correction guidance information is output in a multimodal form, including: synchronously displaying error type identifiers and corresponding standard action comparison diagrams on the human-computer interaction terminal, and synchronously outputting corrective voice prompts through an audio-visual prompting device; The structured assessment results are output in the form of a structured report, including: displaying dimensional scores, deviation location diagrams, and improvement suggestion text on the human-computer interaction terminal, and archiving the structured report to the cloud platform.

[0025] On the other hand, this application also provides an online learning and practical assessment system for sewing technicians, used to implement the online learning and practical assessment method for sewing technicians described in any of the above claims, including: The sewing equipment integrates an operating status sensing module, a data fusion module, an edge inference engine, and an audio-visual prompting device. A visual acquisition device is used to acquire video streams of the operator's work and send the video streams to the sewing equipment through the multi-protocol communication submodule; The human-computer interaction terminal is used to provide a mode selection interface and send mode selection instructions to the server and receive teaching error correction guidance information or structured assessment scoring results from the server through the multi-protocol communication submodule. The server is equipped with a cloud platform for parsing mode selection instructions, generating task instructions, and sending task instructions to the sewing equipment and receiving analysis results uploaded by the sewing equipment through the multi-protocol communication submodule. A multi-protocol communication submodule, integrated into the sewing equipment or the server, is used to establish and maintain a communication link between the sewing equipment, the vision acquisition device, the human-computer interaction terminal and the server, supporting the coordinated transmission of three types of information: instructions, running status data and video streams, and ensuring the time consistency of cross-source data. The above system embeds the method into hardware devices and integrates the dispersed sensing, computing, and interaction units into an organic whole through standardized communication interfaces and protocols, thereby realizing the closed-loop flow of data flow and control flow. In practice, the main body of the sewing equipment is an industrial flatbed sewing machine, and the operating status sensing module includes a Hall speed sensor and a pressure strain gauge, which are installed at the main shaft and presser foot bar. The data fusion module and edge inference engine run on an embedded motherboard built into the machine head; the audio-visual prompt device is embedded in the machine panel; The visual acquisition device is an industrial camera with a USB 3.0 interface, which is connected to the switch via a network cable; The human-computer interaction terminal is a wall-mounted touch screen that connects to the local area network via Wi-Fi; The server is an Alibaba Cloud ECS instance, deployed with Docker containerized services; The multi-protocol communication submodule is implemented by embedded gateway software, runs on the sewing equipment motherboard and server, and is configured with MQTT Broker and TCP / UDP listening ports; The devices are physically connected via gigabit industrial switches, forming a star topology network. In practical applications, the main control chip can also be NVIDIA Jetson Nano or Intel Core i series, and the network topology can also be bus or ring type. This application embodiment does not limit this.

[0026] Based on the same concept, corresponding to any of the above embodiments, the present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements any of the above evaluation methods and has the beneficial effects of the corresponding method embodiments, which will not be repeated here.

[0027] Based on the same concept, corresponding to any of the above embodiments, the present invention also provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements any of the above-described evaluation methods and has the beneficial effects of the corresponding method embodiments, which will not be repeated here.

[0028] Example 1: Real-time error correction closed loop for "uneven stitch spacing" in flat sewing teaching When students select "Teaching Mode - Flat Sewing - Cotton and Linen Fabric" on the human-computer interaction terminal, the system starts the sewing equipment, vision acquisition equipment and server, completes hardware self-check, and establishes a multi-protocol layered communication submodule. The visual acquisition device captures the trainee's operation video stream at a rate of 30 frames per second, and the sewing equipment simultaneously collects operating status data, including spindle speed (3200 stitches / minute), stitch pitch (standard value 2.8mm, tolerance ±0.3mm), and presser foot pressure (3.2N). The data fusion module completes the encapsulation of video frames and status data and timestamp interpolation compensation to ensure millisecond-level alignment between the two. The edge reasoning engine loads the "flat stitch - cotton and linen" adaptive recognition model, performs trajectory deviation recognition, and finds that the stitch distance increases to 3.6mm after the third needle insertion, generating a teaching error correction instruction containing the error type identifier "E03" and spatial positioning coordinates; After compression, the instruction is uploaded to the cloud platform from the edge via a multi-protocol layered communication submodule and then immediately distributed. Among them, AI inference takes 65ms, instruction distribution communication takes 22ms, human-computer interaction terminal receives and refreshes the teaching interface takes 25ms, and the total delay from the occurrence of the action to the display of the visual prompt is 112ms. At the same time, the sound and light prompt device triggers the red light to flash (response time <1ms), and the voice prompt starts after the visual display, playing "Please slow down" after about 420ms; After the task is completed, all devices enter standby mode.

[0029] Example 2: Structured scoring and archiving of "stitch deviation" in curved sewing assessment Trainees select "Assessment Mode - Curved Sewing - Denim" on the human-computer interaction terminal. After the server parses the instruction, it issues the assessment task. The visual acquisition device captures a video stream (lasting 42 seconds) of the trainee sewing along a preset quarter-circle path. The sewing equipment simultaneously acquires high-dynamic operating status data (speed 2500–4100 stitches / minute, stitch pitch 0.8–4.2mm, presser foot pressure peak 5.1N). The data fusion module completes encapsulation and timestamp interpolation compensation. The edge inference engine loads the "Curved Sewing - Denim" model, performs trajectory deviation identification and finished product quality assessment, and locates the... The trace at the beginning of the two arc segments deviated 2.3mm to the left (0.8mm out of tolerance), and the finished product was deemed qualified. The system generated a structured assessment score (trace accuracy 86 points, operation rhythm 92 points, equipment parameter compliance 98 points) and improvement suggestions. The report was uploaded after being compressed into ZIP. AI inference took 72ms, result upload and distribution communication took 28ms, and rendering of the assessment report interface on the human-computer interaction terminal took 30ms. The total delay from analysis completion to report display was 130ms. The report was simultaneously archived to the cloud platform, with the archiving operation taking 1.2 seconds. No original video stream was uploaded throughout the process. Example 2: Structured scoring and archiving of "stitch deviation" in curved sewing assessment Trainees select "Assessment Mode - Curved Sewing - Denim" on the human-computer interaction terminal, and the server parses the instructions and issues the assessment task. The visual acquisition device captures a video stream (lasting 42 seconds) of the trainee sewing along a preset quarter-circle path, while the sewing equipment simultaneously acquires high-dynamic operating status data (speed 2500–4100 stitches / minute, stitch pitch 0.8–4.2mm, presser foot pressure peak 5.1N). The data fusion module completes encapsulation and timestamp interpolation compensation; the edge inference engine loads the "curved sewing - denim" model, performs trajectory deviation identification and finished product quality assessment, locates the starting point of the second arc with a left deviation of 2.3mm (out of tolerance of 0.8mm), and determines that the finished product is qualified; The system generates structured assessment scores (wiring accuracy 86 points, operation rhythm 92 points, equipment parameter compliance 98 points) and improvement suggestions text; The report was uploaded after being compressed into ZIP format. The AI ​​inference took 72ms, the result upload and distribution communication took 28ms, and the rendering of the assessment report interface on the human-computer interaction terminal took 30ms. The total delay from the completion of the analysis to the display of the report was 130ms. The report was also archived to the cloud platform, with the archiving operation taking 1.2 seconds. No original video stream was uploaded at the end.

[0030] Those skilled in the art should understand that the discussion of any of the above embodiments is merely exemplary and is not intended to imply that the scope of this application is limited to these examples; under the concept of this application, the technical features of the above embodiments or different embodiments can also be combined, the steps can be implemented in any order, and there are many other variations of different aspects of the embodiments of this application as described above, which are not provided in detail for the sake of brevity.

[0031] Additionally, to simplify the description and discussion, and to avoid obscuring the embodiments of this application, the well-known power / ground connections to integrated circuit (IC) chips and other components may or may not be shown in the provided drawings. Furthermore, the apparatus may be shown in block diagram form to avoid obscuring the embodiments of this application, and this also takes into account the fact that the details of the implementation of these block diagram apparatuses are highly dependent on the platform on which the embodiments of this application will be implemented (i.e., these details should be fully understood by those skilled in the art). While specific details (e.g., circuits) have been set forth to describe exemplary embodiments of this application, it will be apparent to those skilled in the art that the embodiments of this application can be implemented without these specific details or with variations thereof. Therefore, these descriptions should be considered illustrative rather than restrictive.

[0032] Although this application has been described in conjunction with specific embodiments thereof, many substitutions, modifications, and variations of these embodiments will be apparent to those skilled in the art from the foregoing description. For example, other memory architectures (e.g., dynamic RAM (DRAM)) may be used with the embodiments discussed.

[0033] The embodiments of this application are intended to cover all such substitutions, modifications, and variations that fall within the broad scope of this application. Therefore, any omissions, modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the embodiments of this application should be included within the protection scope of this application.

Claims

1. A method for online learning and practical assessment of sewing technicians, characterized in that: Includes the following steps: S1: Initialization settings: Start the sewing equipment, vision acquisition equipment, human-computer interaction terminal and server, complete their respective hardware self-tests, and establish a multi-protocol layered communication submodule; The multi-protocol layered communication submodule supports the coordinated transmission of three types of information: instructions, running status data, and video streams. These three types of information share a unified frame structure, which includes a device identification field, a data type field, and an integrity verification field. S2: Instruction Reception and Distribution: The server receives a mode selection instruction sent by the human-computer interaction terminal. The mode selection instruction includes a teaching mode or an assessment mode, and is accompanied by a process type identifier and a fabric characteristic identifier. Wherein, after the server parses the mode selection instruction, it sends the mode selection instruction to the sewing device through the multi-protocol layered communication submodule; S3: Data Acquisition and Fusion: The visual acquisition device acquires the operator's work video stream, and the sewing equipment acquires operating status data; The sewing equipment has a built-in data fusion module for receiving the operation video stream and the operation status data, and encapsulating the two based on a unified frame structure. The data fusion module operates a time series interpolation compensation mechanism to repair the missing timestamps caused by frame loss during video stream transmission and to achieve time alignment between the video stream and the running status data. S4: Edge Intelligent Analysis: First, the sewing equipment's local edge inference engine loads an adaptive recognition model that matches the process type identifier and fabric characteristic identifier; Next, the edge inference engine performs at least one analytical task, including operational compliance judgment, trajectory deviation identification, or finished product quality assessment. The edge inference engine has a built-in task scheduling module, which is used to dynamically allocate computing power based on system resource utilization and task priority. The edge inference engine integrates an environment robustness enhancement module, which is used to perform illumination normalization, motion blur suppression and texture noise filtering on the input image sequence; Finally, based on the execution results of the operation compliance judgment, trajectory deviation identification, or finished product quality assessment, the edge inference engine generates teaching error correction instructions containing error type identifiers and spatial positioning coordinates, or structured assessment scoring results containing dimension scores, deviation positioning diagrams, and improvement suggestion texts. The teaching error correction instructions or structured assessment scoring results are compressed and uploaded to the cloud platform deployed on the server. S5: Result Feedback and Closed Loop: The cloud platform will send the teaching error correction guidance information or structured assessment scoring results generated from the analysis results to the sewing equipment and the human-computer interaction terminal through the multi-protocol layered communication submodule, driving the output of audio-visual prompts or graphic reports, and after the task is completed, the sewing equipment, visual acquisition equipment, human-computer interaction terminal and server will enter standby state.

2. The online learning and practical assessment method for sewing technicians according to claim 1, characterized in that: In step S4, the edge inference engine loads an adaptive recognition model that matches the process type identifier and the fabric characteristic identifier, specifically including the following steps: S41: Model Index Location: The edge inference engine retrieves the corresponding model file path from the locally pre-stored model index table based on the received process type identifier and fabric characteristic identifier; the adaptive model index table is stored in key-value pair format, where the key is the combination identifier of process type and fabric characteristic, and the value is the storage address of the model file in the flash memory of the sewing equipment. S42: Model Loading and Memory Mapping: The edge inference engine calls the lightweight inference framework to read the retrieved model file from flash memory to runtime memory and establish a tensor memory mapping region; The lightweight inference framework supports integer quantization model parsing; S43: Parameter Dynamic Injection: The edge inference engine reads a preset parameter set that matches the adaptive model from the local configuration file. The preset parameter set includes image preprocessing parameters, trajectory deviation judgment threshold, and finished product quality score weight coefficient, and injects the preset parameter set into the model inference process. S44: Model Readiness Verification: The edge inference engine performs single-frame empty input inference to verify the integrity of model loading and operator compatibility; If the verification is successful, a model loading success status code is returned to the server, and the process enters the pending analysis task state.

3. The online learning and practical assessment method for sewing technicians according to claim 1, characterized in that: In step S3, after receiving the data packet sent by the vision acquisition device or the sewing device itself, the sewing device verifies the data packet based on the integrity verification field. If the verification fails, a retransmission instruction is returned to the corresponding sender.

4. The online learning and practical assessment method for sewing technicians according to claim 1, characterized in that: In step S3, the time series interpolation compensation mechanism is specifically as follows: the data fusion module extracts the timestamps of two adjacent valid video frames, constructs a linear time mapping relationship, and for video frames with missing timestamps, calculates their proper time sequence position based on the linear time mapping relationship and fills in the corresponding timestamp.

5. The online learning and practical assessment method for sewing technicians according to claim 1, characterized in that: In step S4, the adaptive recognition model is configured to automatically load the corresponding recognition parameter set in response to the process type identifier and the fabric characteristic identifier.

6. The online learning and practical assessment method for sewing technicians according to claim 1, characterized in that: In step S4, the task scheduling module is configured to: when it is detected that the video decoding task has a higher priority than the feature extraction task, prioritize the allocation of computing resources to complete the video frame decoding, and cache the decoding results for subsequent subtasks to call.

7. The online learning and practical assessment method for sewing technicians according to claim 1, characterized in that: In step S4, the environmental robustness enhancement module is configured to be coupled with the input layer of the adaptive recognition model to ensure that the enhanced image is directly used as the model input.

8. The online learning and practical assessment method for sewing technicians according to claim 1, characterized in that: In step S4, the strategy for uploading the analysis results is as follows: only the analysis result summary, key event timestamps, and model running status logs are uploaded, and the original video stream or uncompressed running status data is not uploaded.

9. The online learning and practical assessment method for sewing technicians according to claim 1, characterized in that: In step S5: The teaching error correction guidance information is output in a multimodal form, including: synchronously displaying error type identifiers and corresponding standard action comparison diagrams on the human-computer interaction terminal, and synchronously outputting corrective voice prompts through an audio-visual prompting device; The structured assessment results are output in the form of a structured report, including: displaying dimensional scores, deviation location diagrams, and improvement suggestion text on the human-computer interaction terminal, and archiving the structured report to the cloud platform.

10. A sewing technician online learning and practical assessment system, used to implement the sewing technician online learning and practical assessment method as described in any one of claims 1-9, characterized in that: include: The sewing equipment integrates an operating status sensing module, a data fusion module, an edge inference engine, and an audio-visual prompting device. A visual acquisition device is used to acquire video streams of the operator's work and send the video streams to the sewing equipment through the multi-protocol communication submodule; The human-computer interaction terminal is used to provide a mode selection interface and send mode selection instructions to the server and receive teaching error correction guidance information or structured assessment scoring results from the server through the multi-protocol communication submodule. The server is equipped with a cloud platform for parsing mode selection instructions, generating task instructions, and sending task instructions to the sewing equipment and receiving analysis results uploaded by the sewing equipment through the multi-protocol communication submodule. A multi-protocol communication submodule, integrated into the sewing equipment or the server, is used to establish and maintain a communication link between the sewing equipment, the vision acquisition device, the human-computer interaction terminal, and the server. It supports the coordinated transmission of three types of information: instructions, operating status data, and video streams, and ensures the time consistency of cross-source data.

Images