An individualized employee education empowerment management system based on a capability map

Abstract

The application relates to the technical field of intelligent training, and discloses a kind of individualized employee education empowerment management system based on ability atlas, including the first visual angle video data obtained by intelligent glasses, hand action capture data collected by smart watch, which jointly constitute multi-modal data;Edge computing gateway receives multi-modal data for time alignment and fusion calibration to generate spatial trajectory data;Cloud server obtains spatial trajectory data and compares with standard operation trajectory model to calculate and generate operation proficiency numerical value, which is updated to ability atlas database, and dynamic control parameters are calculated by analyzing operation proficiency numerical value;Edge computing gateway calculates instantaneous spatial deviation according to dynamic control parameters, and generates tactile feedback trigger instruction when exceeding limit;Smart watch receives tactile feedback trigger instruction, and controls vibration motor to output physical vibration.The application realizes spatial quantitative evaluation and dynamic somatosensory interaction closed-loop management of operation skills.

Description

Technical Field

[0001] This invention relates to the field of intelligent training technology, specifically to a personalized employee education and empowerment management system based on competency graphs. Background Technology

[0002] In fields reliant on manual operation, the standardization and proficiency of operators' movements significantly impact the quality and efficiency of the production process. Existing employee training systems typically employ single-dimensional motion monitoring methods, such as relying solely on visual cameras or individual inertial sensors. Purely visual solutions are susceptible to physical obstruction and have limited sampling frequencies, while independent inertial sensors accumulate integral drift errors over extended periods, preventing the system from acquiring accurate spatial coordinates. This monitoring method struggles to output continuous and drift-free spatial trajectories, failing to eliminate the data drift defects of single-dimensional monitoring. Consequently, the system cannot perform spatial quantitative assessments of employees' actions and cannot provide objective data support for determining operational skills.

[0003] Meanwhile, when the existing system determines that an employee's movement has deviated spatially, it intervenes using a fixed spatial tolerance threshold and a uniform alarm feedback intensity. This judgment logic fails to adaptively adjust based on the employee's current skill level. For less skilled operators, the fixed tolerance boundary easily triggers frequent feedback, causing excessive interference and interactive fatigue; for more skilled operators, the uniform threshold cannot constrain their movement flaws, resulting in insufficient feedback. Due to the lack of a dynamic real-time interactive feedback mechanism that matches individual abilities, the existing system cannot automatically adjust the fault tolerance boundary and the vibration intensity of the haptic interaction based on the employee's skill level.

[0004] Furthermore, current education and training systems remain at the stage of single-point error correction for routine operations. After issuing an error correction command, the system directly switches back to normal monitoring mode, failing to capture the operator's action trajectory after receiving the alarm. This mechanism lacks secondary quantitative calculation of the employee's corrective behavior after receiving physical feedback, and cannot extract physical characteristics such as recovery time and smoothness of movement during the correction process. This missing data link means the system can only evaluate operational performance under normal conditions, unable to assess the employee's corrective ability under abnormal conditions, unable to form a closed-loop competency assessment, and unable to achieve closed-loop management of education empowerment based on competency mapping. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a personalized employee education and empowerment management system based on competency graphs. This system solves the problems in existing employee education and training, such as the difficulty in conducting high-precision spatial quantitative assessments of employees' practical actions and the lack of a dynamic real-time interactive feedback mechanism that matches employees' current competency levels. These issues result in poor education and empowerment effects and the inability to form a closed-loop competency evaluation system.

[0006] To achieve the above objectives, this invention provides a personalized employee education and empowerment management system based on capability graphs, comprising: Smart glasses are used to acquire first-person perspective video data; A smartwatch is used to collect hand motion capture data. The first-view video data and the hand motion capture data together constitute multimodal data. An edge computing gateway is used to receive the multimodal data, perform time-series alignment and fusion calibration to generate spatial trajectory data; A cloud server is used to acquire the spatial trajectory data and compare it with a preset standard operation trajectory model to calculate and generate an operation proficiency value; update the operation proficiency value to the competency graph database for personalized employee education and empowerment data management, and parse the operation proficiency value to calculate and generate dynamic control parameters; The edge computing gateway is also used to receive the dynamic control parameters and calculate the instantaneous spatial deviation between the current spatial trajectory data and the preset standard operation trajectory model. When the instantaneous spatial deviation exceeds the limit, a tactile feedback trigger command is generated based on the dynamic control parameters. The smartwatch is also used to receive the haptic feedback trigger command and control the vibration motor to output vibration.

[0007] Preferably, when the edge computing gateway receives the multimodal data and performs temporal alignment and fusion calibration to generate spatial trajectory data, it is specifically used for: using the time axis of the first-view video data in the multimodal data as a reference, performing timestamp-based matching and data alignment on the hand motion capture data in the multimodal data, extracting the pixel coordinates of key hand points to generate three-dimensional absolute coordinate anchor points; using the three-dimensional absolute coordinate anchor points as an external observation reference, employing an extended Kalman filter algorithm to correct errors in the hand motion capture data, and outputting continuous three-dimensional spatial coordinate vectors; combining the continuous three-dimensional spatial coordinate vectors to constitute the spatial trajectory data. This invention anchors the absolute spatial position of inertial data using visual data, combines a filtering algorithm to eliminate the integral drift error accumulated by the inertial sensor over time, solves the synchronization and calibration problem of multi-source heterogeneous sensor data in the spatiotemporal dimension, and outputs a continuous spatial trajectory reflecting the employee's hand movements.

[0008] Preferably, when the cloud server acquires the spatial trajectory data and compares it with the preset standard operation trajectory model to generate an operation proficiency value, it specifically performs the following steps: extracting the corresponding three-dimensional spatial coordinate vectors from the spatial trajectory data and the preset standard operation trajectory model to calculate the spatial Euclidean distance between each sampling point to construct a global cost matrix; using a dynamic programming algorithm to search for a continuous matching path in the global cost matrix and outputting a trajectory distance deviation value; and using a proficiency quantification mapping formula combined with the trajectory distance deviation value to calculate and generate the operation proficiency value. This invention addresses the problem of temporal misalignment between employee operation speed and the standard model by using dynamic programming logic to find the optimal matching path in the spatial distance dimension, avoiding misjudgments caused by simply relying on timeline comparisons, and achieving a quantitative evaluation of operational action characteristics.

[0009] Preferably, when the cloud server updates the operation proficiency value to the capability graph database for personalized employee education and empowerment data management, it specifically performs the following: calls an exponentially weighted moving average calculation to weight and fuse the newly acquired operation proficiency value with the historical proficiency parameters in the capability graph database, generating a baseline value for the corresponding process node after the update. This invention introduces a time-dimensional weight allocation mechanism to smooth out occasional fluctuations in an employee's performance in a single operation, establish a skill benchmark level, and achieve dynamic tracking of the evolution trend of employee capabilities.

[0010] Preferably, when the cloud server parses the operation proficiency value to generate dynamic control parameters, it specifically performs the following steps: 1) Calculates a dynamic position tolerance threshold using a dynamic threshold modulation formula based on the operation proficiency value; 2) Calculates a pulse width modulation duty cycle parameter using a duty cycle parameter calculation formula based on the operation proficiency value; 3) Encapsulates the dynamic position tolerance threshold and the pulse width modulation duty cycle parameter to form the dynamic control parameters; wherein, the dynamic position tolerance threshold is the dynamic boundary for exceeding limits. This invention establishes a reverse adjustment mapping relationship between proficiency and the system feedback mechanism, dynamically adjusting the fault tolerance space and vibration feedback intensity according to the employee's ability level, avoiding providing exceeding limits judgments to inexperienced employees or causing physical interference to skilled employees, thus achieving personalized empowerment.

[0011] Preferably, the edge computing gateway receives the dynamic control parameters and calculates the instantaneous spatial deviation between the current spatial trajectory data and the preset standard operation trajectory model. When the instantaneous spatial deviation exceeds the limit, the haptic feedback trigger command is generated according to the dynamic control parameters. Specifically, this is used to: extract the three-dimensional physical coordinates of the spatial trajectory data at the current timestamp; find the standard reference point with the shortest physical distance to the three-dimensional physical coordinates in the preset standard operation trajectory model; calculate the instantaneous spatial deviation between the three-dimensional physical coordinates and the standard reference point using the spatial Euclidean distance formula; and generate the haptic feedback trigger command according to the pulse width modulation duty cycle parameter in the dynamic control parameters when the instantaneous spatial deviation of multiple consecutive sampling frames within the time sliding window is greater than the dynamic position tolerance threshold in the dynamic control parameters. This invention comprehensively judges the deviation data of multiple consecutive frames through a time sliding window, filters out isolated anomalies caused by natural hand tremors or instantaneous sensor noise, and ensures the stability of the haptic feedback command trigger.

[0012] Preferably, after generating the haptic feedback trigger command, the edge computing gateway further generates a response trajectory feature sequence, which includes: extracting the system time record of the current system master clock as the starting absolute timestamp; calculating the intercepted time window length using a dynamic time window length calculation formula based on the instantaneous spatial deviation at the trigger time and the operation proficiency value; and intercepting the corresponding spatial trajectory data in the cache queue according to the intercepted time window length and the starting absolute timestamp to generate the response trajectory feature sequence. In this invention, after the system issues interactive feedback, it locks and intercepts the action trajectory segment immediately following the feedback, providing a data basis for subsequent evaluation of the employee's corrective behavior after receiving guidance.

[0013] Preferably, when the edge computing gateway sends the response trajectory feature sequence to the cloud server after generating it, it specifically performs the following steps: encapsulates the response trajectory feature sequence and the corresponding starting absolute timestamp into a data packet, sends it independently to the cloud server via asynchronous communication, and releases local cache resources. This invention employs an asynchronous upload and local cache release mechanism to separate real-time monitoring tasks from background data processing tasks, preventing network fluctuations from blocking the edge computing gateway's computing resources and ensuring the real-time performance of the system's front-end motion capture and haptic feedback links.

[0014] Preferably, when the cloud server acquires and parses the response trajectory feature sequence to generate multidimensional physical indicators, it specifically performs the following steps: extracting the time nodes in the response trajectory feature sequence where the instantaneous spatial deviation decreases back to and remains within the dynamic position tolerance threshold range of the dynamic control parameters; using the time difference between the time node and the corresponding starting absolute timestamp as the recovery time interval; calculating the third derivative of the three-dimensional spatial coordinates in the response trajectory feature sequence with respect to time, and performing square integration on the third derivative to obtain the smoothness attenuation; and combining the recovery time interval and the smoothness attenuation to construct the multidimensional physical indicators. This invention constructs an evaluation matrix from two physical dimensions—correction response speed and correction action stability—to analyze the employee's corrective capabilities after receiving physical feedback.

[0015] Preferably, when the cloud server completes the graph feature update after acquiring the multidimensional physical indicators, it specifically performs the following: It uses an evaluation value normalization formula to perform cross-penalty calculations on the recovery time interval and the smoothness decay in the multidimensional physical indicators to generate an abnormal state correction capability evaluation value, and records the abnormal state correction capability evaluation value in the corresponding high-order correction capability sub-node within the capability graph database to complete the graph feature update. This invention integrates speed and stability indicators through penalty calculations, extending the evaluation of employees' routine operations to the dimension of abnormal state response evaluation, improving the hierarchical structure of the capability graph database, and realizing a closed loop of educational empowerment management.

[0016] The present invention provides a personalized employee education and empowerment management system based on capability graphs, which has the following beneficial effects: This invention acquires first-person perspective video data and hand motion capture data, performs temporal alignment and fusion calibration at an edge computing gateway to generate spatial trajectory data, and compares it with a standard operation trajectory model to determine the operator's proficiency level. This solution utilizes visual data to anchor spatial position and combines it with filtering algorithms to correct accumulated sensor errors, overcoming the data drift defects of single-dimensional monitoring. It also solves the problem of difficulty in spatially quantifying practical actions in existing employee education and training, providing objective data support for determining operational skills.

[0017] This invention parses the operational proficiency value generated by the cloud server into dynamic control parameters, including dynamic position tolerance thresholds and pulse width modulation duty cycle parameters. The edge computing gateway uses these parameters to determine instantaneous spatial deviations and triggers physical vibration of the smartwatch when these deviations exceed limits. This mechanism automatically adjusts the system's tolerance boundaries and the vibration intensity of the haptic interaction based on the employee's current actual operational level, solving the problem of existing systems lacking a dynamic real-time interactive feedback mechanism that matches individual abilities, and avoiding excessive interference or insufficient feedback caused by fixed, uniform thresholds.

[0018] This invention, after issuing a tactile feedback trigger command, extracts the response trajectory feature sequence based on the initial absolute timestamp and the length of the intercepted time window. It then extracts the recovery time interval and smoothness decay of the correction process to generate an abnormal state correction capability evaluation value, which is updated to the capability graph database. This process performs secondary quantitative calculations on employees' action correction behaviors after receiving physical feedback, evaluating not only normal operations but also responses under abnormal conditions. It solves the problem that traditional training models cannot form a closed-loop capability evaluation, realizing closed-loop management of education empowerment based on capability graphs. Attached Figure Description

[0019] Figure 1 This is a physical topology diagram of the personalized employee education empowerment management system based on capability graphs according to the present invention. Figure 2 This is a flowchart illustrating the overall system workflow of the present invention. Figure 3 This is a flowchart of the multimodal operation data perception and edge fusion calibration process of the present invention; Figure 4 This is a flowchart of the high-precision spatial trajectory comparison and map update based on edge-cloud collaboration of the present invention; Figure 5 This is a flowchart of the dynamic control parameter calculation and backflow process based on the spectrum state of the present invention. Figure 6 This is a flowchart of the edge-side closed-loop motion sensing interaction and correction control execution of the present invention; Figure 7 This is a flowchart of the timing slice extraction of the haptic feedback response trajectory of the present invention. Figure 8 This is a flowchart illustrating the quantification and high-order graph feature mapping of the abnormal state correction capability of the present invention. Figure 9 This is a comparison chart showing the numerical evolution of operational proficiency during the training period of this invention. Figure 10 This is a comparison chart of recovery time intervals under abnormal conditions according to the present invention.

[0020] Among them, 100 is smart glasses; 200 is smartwatches; 300 is edge computing gateways; and 400 is cloud servers. Detailed Implementation

[0021] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0022] See Figure 1 This invention provides a personalized employee education and empowerment management system based on capability graphs. The personalized employee education and empowerment management system based on capability graphs includes: smart glasses 100, smartwatch 200, edge computing gateway 300, and cloud server 400.

[0023] The smart glasses 100 are equipped with an image acquisition module, which is used to acquire first-person view video data of the operator.

[0024] The smartwatch 200 incorporates an inertial measurement unit (IMU) and a vibration motor. The IMU collects acceleration and angular velocity data of the operator's hand and sends this data as inertial input for hand motion capture to the edge computing gateway 300. The vibration motor is configured to receive pulse-width modulation (PWM) signals and output physical vibrations to provide tactile feedback.

[0025] Both the smart glasses 100 and the smartwatch 200 establish a data connection with the edge computing gateway 300 via a local area wireless communication protocol. The edge computing gateway 300 is deployed locally in the operating environment. The edge computing gateway 300 establishes a communication connection with the cloud server 400 via a wide area network, thereby constructing an edge-cloud collaborative architecture.

[0026] The edge computing gateway 300 is used to receive multimodal data collected by the smart glasses 100 and the smartwatch 200. The edge computing gateway 300 performs spatiotemporal alignment of the data and fusion calculations of visual and inertial data. The edge computing gateway 300 is also used to perform spatial position comparison and send control commands to the smartwatch 200 to trigger motion-sensing interaction.

[0027] The cloud server 400 receives spatial motion trajectory data uploaded by the edge computing gateway 300 and performs trajectory feature comparison calculations. The cloud server 400 also constructs and maintains a capability graph database, calculates dynamic control parameters, and distributes these parameters to the edge computing gateway 300. The capability graph database stores employee capability-related data in a graph-based relational structure. This data includes at least employee identity information, job category information, work process / task information, skill element information, historical operational performance data, training record data, operational proficiency assessment results, and corresponding standard operating requirements. These data objects are organized through job requirement associations, skill mapping associations, work process correspondence associations, training evolution associations, and capability assessment associations to represent the correspondence between an employee's current capability status, target job capability requirements, and capability improvement paths.

[0028] See Figure 2 This invention provides a personalized employee education and empowerment management method based on capability graphs, which includes the following steps: S1, the image acquisition module of the smart glasses 100 acquires first-view video data, the inertial measurement unit of the smartwatch 200 acquires hand motion capture data, and the edge computing gateway 300 performs time-series alignment and fusion calibration of the first-view video data and hand motion capture data to generate spatial trajectory data. S2, cloud server 400 acquires spatial trajectory data and compares it with the preset standard operation trajectory model. Based on the comparison results, it generates operation proficiency values ​​and updates them to the competency graph database for personalized employee education and empowerment data management. S3, cloud server 400 analyzes the operation proficiency value to generate dynamic control parameters, and sends the dynamic control parameters to edge computing gateway 300; S4, the edge computing gateway 300 receives dynamic control parameters and calculates the instantaneous spatial deviation between the current spatial trajectory data and the preset standard operation trajectory model. When the instantaneous spatial deviation exceeds the limit, it sends a tactile feedback trigger command to the vibration motor of the smartwatch 200 according to the dynamic control parameters to realize somatosensory interaction. S5, the edge computing gateway 300 intercepts the response trajectory feature sequence within the time window and sends it to the cloud server 400 while issuing the haptic feedback trigger command; S6, the cloud server 400 analyzes the response trajectory feature sequence to calculate the abnormal state correction capability evaluation value, and records the abnormal state correction capability evaluation value into the capability map database.

[0029] The steps of the method provided in the embodiments of the present invention will be described in detail below with reference to specific scenarios.

[0030] See Figure 3 In the personalized employee education and empowerment management system based on capability graphs provided in this embodiment of the invention, step S1 specifically includes the following sub-steps: S101, the image acquisition module of the smart glasses 100 acquires the operator's first-person perspective video data, while the inertial measurement unit of the smartwatch 200 acquires the operator's hand motion capture data.

[0031] First-person perspective video data includes a continuous sequence of video frames within the operator's field of view, with each frame containing visual feature information about the hand and the operating tool. Hand motion capture data includes high-frequency three-axis acceleration and three-axis angular velocity data along the local hand coordinate system. To ensure the system captures subtle hand movements and high-frequency vibrations without omission, the inertial measurement unit's data sampling frequency is higher than the video frame rate of the image acquisition module, thus providing sufficient data granularity for subsequent instantaneous deviation comparison model building.

[0032] S102, due to the difference in sampling frequency of the above heterogeneous devices, the multi-source data is in a discrete state on the time axis. In order to solve this synchronization problem in physical space, the edge computing gateway 300 receives the first-view video data and hand motion capture data and performs time alignment processing.

[0033] The edge computing gateway 300 parses the hardware timestamps attached to the two sets of underlying data and uses a local area network time synchronization protocol to unify the system clocks of heterogeneous devices. Based on this, the edge computing gateway 300 uses the timeline of the low-frequency first-person perspective video data as a reference, and selects the high-frequency hand motion capture data frame closest to the visual frame time node for matching and data alignment according to the principle of closest timestamp, thereby outputting a synchronized multimodal data pair with a unified time label. To ensure the reliability of the operational data fusion, the edge computing gateway 300 presets a maximum allowable time deviation threshold before performing the above timing alignment calculation. The maximum allowable time deviation threshold is set to 1.5 to 2 times the ideal time interval between two adjacent frames of the image acquisition module. When the actual time interval between adjacent valid visual frames exceeds this maximum allowable time deviation threshold, the edge computing gateway 300 immediately blocks the visual observation update calculation for the current time period, avoiding spatial coordinate distortion caused by excessive reliance on mathematical extrapolation of underlying inertial data when severe frame loss occurs due to local area network signal jitter. Through the above timing constraint mechanism, the system effectively aligns the two sets of heterogeneous data to a unified time reference plane.

[0034] S103, after completing the timing alignment, the edge computing gateway 300 performs fusion calibration on the synchronized data stream and finally outputs spatial trajectory data.

[0035] In this computational phase, the system constructs a state observation model based on the general technical principle of fusion of visual and inertial data. It utilizes visual coordinates with global absolute attributes to continuously constrain and calibrate the inertial integral drift, which has highly dynamic relative attributes. The edge computing gateway 300 extracts the pixel coordinates of key hand points from the first-person view video data and combines them with the camera intrinsic parameter matrix of the image acquisition module. This camera intrinsic parameter matrix is ​​an internal parameter matrix pre-calibrated from the image acquisition module. This process generates low-frequency three-dimensional absolute coordinate anchor points. For the computational process of image feature point extraction and camera intrinsic parameter calibration mapping, those skilled in the art can employ mature computer vision geometric algorithms. The feature matching and matrix transformation logic are well-known technologies in this field and will not be elaborated upon here.

[0036] Using the calculated 3D absolute coordinate anchor points as external observation benchmarks, and calculating the observation residuals of the system state based on the state observation model, the edge computing gateway 300 calls the extended Kalman filter algorithm to correct errors in the inertial data. Considering that hand motion capture data will generate inherent physical cumulative drift when performing quadratic integration to obtain spatial displacement, the edge computing gateway 300 periodically uses low-frequency 3D absolute coordinate anchor points to calculate the observation residuals of the system state. In the standard extended Kalman filter logic, a novel covariance matrix needs to be constructed by combining the observation residuals and their corresponding uncertainties, and the inverse matrix of the novel covariance matrix needs to be calculated to solve for the Kalman gain. In the above solution process, the system has a built-in matrix singularity detection mechanism. When it is detected that the determinant of the novel covariance matrix approaches zero, indicating singularity, the edge computing gateway 300 dynamically injects a positive definite regularization coefficient into the main diagonal of the novel covariance matrix. The value range of this positive definite regularization coefficient is a constant 10. 6 Up to 10 4 This avoids the systemic risks of zero overflow or filter divergence.

[0037] After introducing this regularization mechanism, the edge computing gateway 300 calculates the corrected Kalman gain and uses this Kalman gain to robustly update the state covariance matrix, which characterizes the system's state uncertainty. This covariance update process physically realizes multi-dimensional dynamic weighting based on visual observation confidence and inertial integral confidence. Based on this weighting logic, the system forcibly resets the cumulative drift error generated during the integration process of the hand motion capture data. The edge computing gateway 300 outputs spatial trajectory data, which eliminates drift components and high-frequency noise, through iterative updates of the extended Kalman filter algorithm. The spatial trajectory data consists of continuous and drift-free three-dimensional spatial coordinate vectors in time series, serving as the physical basis for subsequent edge-cloud collaborative evaluation.

[0038] See Figure 4 In this embodiment, step S2 specifically includes the following sub-steps: S201, based on an edge-cloud collaborative architecture, involves the edge computing gateway 300 continuously uploading spatial trajectory data to the cloud server 400 after generating it, thereby offloading high-computing-power-demand analysis tasks to the cloud. The cloud server 400 then acquires the spatial trajectory data and compares it with a preset standard operation trajectory model for calculation.

[0039] Considering the objective physical differences in the action rates of different operators performing the same process, directly aligning distance coordinates on the same time axis can easily lead to phase misalignment. Based on this phenomenon, the cloud server 400 invokes a dynamic time warping algorithm to find the alignment path with the minimum global cumulative distance between two time series of different lengths.

[0040] The specific implementation process of this dynamic time warping algorithm is as follows: The system extracts the three-dimensional spatial coordinate vectors corresponding to the spatial trajectory data and the preset standard operation trajectory model, calculates the spatial Euclidean distance between each sampling point to construct a global cost matrix (the global cost matrix is ​​a two-dimensional cost matrix composed of sampling points of spatial trajectory data as row indices, sampling points of the standard operation trajectory model as column indices, and spatial Euclidean distances between corresponding sampling points as matrix element values); then, using dynamic programming logic, a continuous matching path from the starting point to the ending point with the minimum cumulative cost is searched in the global cost matrix. Through this nonlinear time axis mapping mechanism, the system effectively eliminates the time distortion error caused by motion rate fluctuations, and thus focuses on the deviation calculation of spatial physical pose. The final output is a trajectory distance deviation value that can quantitatively characterize the spatial difference between the current actual action and the standard model.

[0041] S202, In order to transform the above parameters reflecting the physical space span into an evaluation baseline with educational management business significance, the cloud server 400 generates an operational proficiency value based on the comparison results.

[0042] Before proceeding to the specific calculation steps, the system constructs evaluation logic based on the general mathematical principle of inverse decay. In this embodiment, the proficiency measurement mapping formula is used to calculate the operational proficiency value. The proficiency measurement mapping formula is as follows: ; In the formula, Indicates the process sequence number is The user's proficiency level at that time; Represents the decay mapping constant; Indicates the process sequence number is The trajectory distance deviation value at time; the constant in the denominator of the formula. It is the basic bias term.

[0043] The purpose of introducing this proficiency measurement mapping formula is to construct an adaptive negative correlation logic for the evaluation baseline. As a preferred approach, the attenuation mapping constant is set to a range of 0.1 to 0.5, with its specific value dynamically configured by the system based on the spatial fault tolerance requirements of different business processes. The physical meaning of the attenuation mapping constant is the rate at which the control capability score decreases with increasing operational deviation. Since the trajectory distance deviation is a non-negative physical distance quantity, and it is superimposed with the basic bias term constant... The denominator of this formula is always greater than or equal to 1, effectively avoiding the systematic risk of the denominator approaching zero in division operations. When the trajectory distance deviation value approaches zero, the operator proficiency value smoothly approaches the upper threshold of 1, indicating a high degree of spatial motion matching of the operator. This mapping logic normalizes the originally divergent physical error data to a standard range, providing dimensionless data support for the subsequent dynamic parameter modulation of the edge actuator.

[0044] S203, after completing feature quantization and mapping, the system enters the stage of persistent storage of personalized data and graph evolution. The cloud server 400 updates the operational proficiency values ​​to the competency graph database for data management of personalized employee education and empowerment.

[0045] The capability graph database is internally structured with detailed basic execution capability nodes based on practical operation procedures. To prevent drastic changes in the evaluation baseline due to a single operational error, the cloud server 400 abandons the logic of directly overwriting historical data when updating node data. As a more robust data processing method, the system introduces a multi-dimensional historical data weighting logic. The cloud server 400 calls an exponentially weighted moving average calculation to weight and fuse the newly acquired operational proficiency value with the existing historical proficiency parameters in the capability graph database.

[0046] The data processing logic for this index-weighted moving average calculation is as follows: A set time decay weight is assigned to the newly acquired operational proficiency value, and complementary weights are given to historical proficiency parameters. These two weights are linearly added to generate the baseline value after the update of this process node. This dynamic update mechanism based on weighted logic ensures that the final data model objectively reflects the recent evolution trend of employees' action levels while effectively smoothing out occasional data fluctuations caused by environmental physical interference. This ensures that the operational proficiency values ​​ultimately deposited in the competency graph database can robustly represent the actual practical skill base of employees, thereby effectively supporting subsequent motion-sensing interactive control logic.

[0047] See Figure 5 In this embodiment, step S3 specifically includes the following sub-steps: S301, based on the continuous evolution of the above-mentioned map nodes, in order to realize the logical closed loop from data evaluation to underlying hardware control, the cloud server 400 parses the operation proficiency values ​​extracted from the capability map database and calculates and generates the corresponding dynamic position tolerance threshold.

[0048] Considering the differences in physiological responses at different proficiency levels, using a fixed and uniform physical space tolerance could easily lead to excessively high feedback trigger frequencies at lower proficiency levels, causing interaction fatigue, or excessively wide spatial constraints at higher proficiency levels, reducing motion correction sensitivity. Based on these physical and interactive conditions, the system constructs a threshold modulation logic that is negatively correlated with proficiency. In this embodiment, a dynamic threshold modulation formula is used to calculate the dynamic position tolerance threshold. The dynamic threshold modulation formula is: ; In the formula, Indicates the dynamic position tolerance threshold; Indicates the basic tolerance benchmark value; Indicates the tolerance scaling factor; Indicates the process sequence number is The user's proficiency level at that time; This indicates a margin for complementary proficiency.

[0049] Introducing this dynamic threshold modulation formula helps to give the system adaptive fault tolerance. As a preferred approach, the range of the basic tolerance benchmark value is set to the lower limit of the physical space dimensional tolerance that meets the minimum quality requirements of the process. To ensure the completeness of the algorithm implementation, the specific value of the tolerance scaling factor is set to the spatial difference between the physical safety boundary of the workbench and the basic tolerance benchmark value. When the operator's proficiency value is at a low level, the corresponding proficiency complementarity margin is large, and the system outputs a wider dynamic position tolerance threshold to encourage operators to prioritize mastering the overall process continuity; as practical skills improve, this value converges, and the threshold narrows smoothly, thereby guiding the operation to gradually approach the preset standard operation trajectory model.

[0050] S302, in conjunction with the dynamic scaling of the aforementioned spatial boundary determination, the physical execution intensity of the haptic interaction also needs to be adjusted using corresponding nonlinear modulation. Based on this operational proficiency value, the cloud server 400 calculates the pulse width modulation duty cycle parameters for the vibration motor built into the smartwatch 200.

[0051] The vibration intensity of the physical actuator is positively correlated with the duty cycle of the pulse drive signal. To construct a haptic feedback mechanism with stepped discrimination, the pulse width modulation duty cycle parameter is calculated using the duty cycle parameter calculation formula: ; In the formula, This represents the pulse width modulation duty cycle parameter; This indicates the lower limit of the base duty cycle; Indicates the feedback strength gain coefficient; Indicates the process sequence number is The user's proficiency level at that time; This indicates a margin for complementary proficiency.

[0052] The physical meaning of this formula lies in establishing a dynamic attenuation model for the intensity of the correction prompt. In a preferred embodiment, the lower limit of the basic duty cycle is set as the minimum level of drive percentage that can induce basic tactile perception on the skin of the human wrist, and the specific value range of this parameter is configured within the range of 15% to 20%. The specific calculation logic of the feedback intensity gain coefficient is the difference between the upper limit of the rated safe duty cycle of the motor (limited to 80% in this embodiment) and the lower limit of the basic duty cycle. Through the constraints of the above upper and lower limit logic, the system avoids the risk of hardware overheating or lifespan degradation caused by continuous operation under full load, and also realizes differentiated tactile stimulation for employees with different skill levels. For operators with lower skill levels, the system sends a higher proportion of vibration to form a clear physical prompt; while for operators with higher skill levels, only a lower intensity pulse vibration is needed to achieve closed-loop motion fine-tuning.

[0053] S303, to ensure the low-latency characteristics of high-frequency physical control, the system employs an asynchronous backflow mechanism to process the generated control parameters. The cloud server 400 packages the calculated dynamic position tolerance threshold and pulse width modulation duty cycle parameters into dynamic control parameters.

[0054] The cloud server 400 continuously sends the dynamic control parameters to the edge computing gateway 300 deployed locally in the operating environment via a wide area network channel. The edge computing gateway 300 receives the dynamic control parameters and overwrites and updates them in the execution configuration table in its local memory. This edge-cloud collaborative parameter backflow mechanism (i.e., the cloud asynchronously sends and overwrites the high-order control parameters calculated based on global data to the local cache of the edge node) enables the edge node to independently complete high-frequency instantaneous spatial deviation comparison tasks based on the latest personalized state parameters cached locally, without relying on the real-time computing power of the cloud. This network topology architecture, which combines macro-level offloading of computing tasks with asynchronous parameter updates, effectively overcomes the physical defects of traditional cloud-directly-sent control commands, which cause lag in haptic interaction due to network transmission delays.

[0055] See Figure 6 In this embodiment, step S4 specifically includes the following sub-steps: Based on the pre-configured personalized baseline on the edge side, the S401 system initiates a high-frequency physical status monitoring mechanism. The edge computing gateway 300 reads the dynamic control parameters asynchronously updated by the cloud server 400 from its local memory. Leveraging the high concurrency processing performance of the edge nodes, the edge computing gateway 300 extracts the three-dimensional physical coordinates of the spatial trajectory data at the current timestamp in real time. To achieve quantitative comparison of spatial pose, the edge computing gateway 300 performs spatial node traversal in the preset standard operation trajectory model, searching for the standard reference point with the shortest physical distance to the actual three-dimensional physical coordinates in the current spatial trajectory data.

[0056] In this embodiment, the instantaneous spatial deviation between the two points is calculated using the spatial Euclidean distance formula, which is: ; In the formula, Indicates instantaneous spatial deviation; , , These represent the three-dimensional coordinate components of the spatial trajectory data at the current timestamp; , , These represent the three-dimensional coordinate components of the corresponding standard reference point in the preset standard operation trajectory model. This comparison calculation process is implemented in a closed loop locally in the operating environment, effectively avoiding network latency caused by unreliable wide area network transmission, thereby ensuring the continuity of motion monitoring at the physical level.

[0057] S402, after acquiring continuous pose deviation data, the system needs to determine whether the action constitutes a substantial deviation from the limit by combining the employee's current capability map status. The edge computing gateway 300 calls the dynamic position tolerance threshold included in the dynamic control parameters as the dynamic boundary for the deviation judgment. Considering the natural physiological jitter of the human body and the occasional high-frequency drift of the underlying sensors in industrial field operations, if hardware feedback is triggered solely by the extreme value exceeding the limit in a single frame, it is easy to cause frequent false alarms and interfere with the employee's normal construction work. Based on this working condition characteristic, the system introduces a fault-tolerant filtering mechanism based on a time sliding window.

[0058] As a preferred approach, the length of this time-sliding window is dynamically set based on the sampling frequency of the underlying sensor, configured to include 3 to 5 consecutive valid sampling frames, whose corresponding time span precisely covers the basic physiological jitter cycle of the human body. The edge computing gateway 300 extracts multiple consecutive sampling frames within the current observation time window. Only when the instantaneous spatial deviation of all consecutive frames within this time window is greater than the dynamic position tolerance threshold does the system determine that the current action is in a true out-of-limit state. This multi-dimensional time and space joint determination logic effectively avoids one-sided decisions caused by relying on a single outlier. When the instantaneous spatial deviation exceeds the limit, the edge computing gateway 300 generates a tactile feedback trigger command based on dynamic control parameters and encapsulates the calculated pulse width modulation duty cycle parameters into the underlying data payload of the command.

[0059] In step S403, in conjunction with the data commands generated above, the system drives the edge execution device to complete the physical mapping of the haptic interaction. The edge computing gateway 300, through a local wireless communication protocol, sends a low-latency haptic feedback trigger command, encapsulated with pulse width modulation duty cycle parameters, to the smartwatch 200 worn by the operator. The microcontroller built into the smartwatch 200 receives and parses the command payload, extracting the pulse width modulation duty cycle parameters to configure the internal motor drive amplifier circuit.

[0060] After the circuit parameters are modulated, the vibration motor built into the smartwatch 200 receives the corresponding pulse width modulation level signal and outputs physical vibrations of appropriate intensity to achieve haptic interaction. Through this closed-loop control execution logic, the system can provide differentiated corrective prompts that match the operator's current skill level the instant their actions deviate substantially. This closed-loop mechanism not only realizes the mapping from digital space state assessment to physical space action constraints, but also helps operators gradually establish standardized operating habits through obvious tactile feedback.

[0061] Meanwhile, to ensure the physical continuity of the subsequent edge-cloud collaborative evaluation logic, the edge computing gateway 300 simultaneously uses the issued haptic feedback trigger command as a trigger anchor point to open a data listening window for the response action, which is beneficial for subsequent response trajectory feature extraction.

[0062] See Figure 7 In this embodiment, step S5 specifically includes the following sub-steps: S501, in conjunction with the issuance of the aforementioned hardware correction command, the system simultaneously activates a quantitative monitoring mechanism for the employee's corrective action response. In this embodiment, during the hardware interrupt cycle of the edge computing gateway 300 outputting the tactile feedback trigger command to the outside, the system time of the current system master clock is extracted and recorded as the starting absolute timestamp. This starting absolute timestamp serves as the starting physical anchor point for subsequent action capture in the time domain. To effectively record the operator's actual muscle response and action correction process after receiving physical vibration, the system constructs a forward-covering interception time window based on this starting absolute timestamp.

[0063] Considering the objective physiological differences in the response delays to physical stimuli among employees of varying skill levels, and the varying physical correction cycles required for actions of different degrees of deviation, using a fixed and uniform truncation time could easily lead to premature truncation of correction data from less skilled employees, or the introduction of a large amount of redundant resting data from skilled workers after they have completed their corrections. Based on these multi-source differentiated work conditions, the system constructs an adaptive window logic that is positively correlated with the deviation value and negatively correlated with skill level. The truncation time window length is calculated using a dynamic time window length calculation formula, which is: ; In the formula, Indicates the length of the time window being captured; Indicates the fundamental response time constant; Indicates the time scaling factor for correction; This represents the instantaneous spatial deviation value at which the system generates the haptic feedback trigger command; Indicates the process sequence number is The user's proficiency level at that time; This indicates the minimum value of the zero bias protection.

[0064] Introducing this dynamic time window length calculation formula is beneficial for endowing edge computing nodes with adaptive time-series slicing capabilities. In the preferred method, the basic response time constant is set based on the lower limit of the physiological delay of basic human nerve conduction and muscle response, and its actual value range is configured between 300 milliseconds and 500 milliseconds. To ensure the dimensionality consistency of the formula calculation, the physical unit of the time scaling correction coefficient is defined as time / distance (e.g., milliseconds / millimeters), and its value is statically calibrated based on the spatial complexity of the monitored process, representing the average physical correction time required per unit spatial deviation. The minimum value of the zero-bias protection is set to a small positive real number such as 0.01 to maintain the stability of mathematical operations when the operation proficiency value approaches zero in division operations, preventing the system from generating software faults such as division-by-zero overflow.

[0065] The physical execution logic of this formula is that when the instantaneous spatial deviation value of the trigger moment is large, or when the employee is at a low proficiency level, the system dynamically extends the observation span to fully capture the long-cycle action adjustment trajectory; conversely, it narrows the window to ensure the memory compactness of subsequent data processing from a physical perspective.

[0066] S502, based on the aforementioned dynamic timing boundaries, the system needs to extract valid data segments at the underlying level. The edge computing gateway 300 allocates an independent first-in-first-out (FIFO) cache queue in its local memory, continuously writing newly incoming spatial trajectory data into this queue, starting from the initial absolute timestamp. When the step size of the system master clock reaches the length of the truncation time window for the calculation output, the edge computing gateway 300 triggers a data truncation instruction at the underlying level, stopping the continuous writing operation to the cache queue. Through the mechanism combining the aforementioned initial absolute timestamp tracking and dynamic window constraints, the edge computing gateway 300 reliably strips the originally continuous physical motion data without clear business boundaries in the time domain, generating a response trajectory feature sequence containing a complete closed loop of correction features.

[0067] Based on the limited computing resources at the edge, the S503 system, to avoid the high-frequency real-time monitoring computing power at the edge being squeezed out by subsequent complex feature analysis tasks, immediately initiates an asynchronous communication process after completing the time-series slice extraction. The edge computing gateway 300 encapsulates the generated response trajectory feature sequence with the corresponding starting absolute timestamp into a data packet and independently sends it to the cloud server 400 via a wide area network channel. This independent cloud upload mechanism allows the edge computing gateway 300 to quickly release local cache resources and smoothly switch to regular high-frequency spatial location comparison monitoring after extracting data snapshots of abnormal states. Simultaneously, this asynchronously uploaded response trajectory feature sequence provides the cloud server 400 with an objective data foundation to support subsequent in-depth analysis and evaluation of employee status.

[0068] See Figure 8 In this embodiment, step S6 specifically includes the following sub-steps: S601, based on the asynchronous communication mechanism between the edge and cloud, after receiving the uploaded continuous trajectory segments, the system enters the feature analysis stage for employee emergency adjustment efficiency. In this embodiment, the cloud server 400 acquires and parses the response trajectory feature sequence sent by the edge computing gateway 300, and extracts multi-dimensional physical indicators characterizing the action correction process from it.

[0069] To objectively quantify the efficiency and stability of operators in correcting spatial deviations, the system extracts two key evaluation parameters: The first is the recovery time interval. The calculation logic of this indicator is as follows: extract the time node in the response trajectory feature sequence where the instantaneous spatial deviation drops back to and remains within the dynamic position tolerance threshold, and calculate the time difference between the time node and the corresponding starting absolute timestamp of the sequence. This time difference represents the comprehensive physiological delay of human nerve reflex and action execution. Secondly, there is the smoothness attenuation. For the spatial discretization extraction of this parameter, those skilled in the art can use a well-known trajectory smoothness calculation method based on jerk, that is, to solve by square integral of the third derivative of the three-dimensional spatial coordinates with time. The calculation logic is a well-known technique in this field and will not be elaborated here. Through the joint extraction of the above two indicators, the system constructs a set of evaluation parameters covering time efficiency and spatial steady state.

[0070] S602, after obtaining the aforementioned underlying physical characteristic parameters, the system needs to combine multi-dimensional weighted logic to transform the physical parameters into a comprehensive evaluation result. Considering that relying solely on the time consumption of correction or solely on the smoothness of movement can easily lead to a one-sided judgment of the employee's actual corrective skills, the system constructs a multi-dimensional cross-penalty evaluation logic.

[0071] In this embodiment, the evaluation value of the abnormal state correction capability is calculated using an evaluation value normalization formula. The evaluation value normalization formula is as follows: ; In the formula, This represents the evaluation value of the ability to correct abnormal states; Indicates the recovery time interval; Indicates the amount of smoothness decay; This represents the time penalty weighting coefficient; This represents the smoothness penalty weighting coefficient; a constant in the denominator of the formula. This is a base-partial configuration item.

[0072] As a preferred approach, the specific values ​​of the time penalty weighting coefficient and the smoothness penalty weighting coefficient are configured between 0.1 and 0.9, and the sum of their values ​​is always equal to 1. The specific numerical allocation is dynamically adjusted by the system based on the current process's emphasis on time sensitivity and quality sensitivity. To ensure dimensional consistency in mathematical operations at the underlying level, the time penalty weighting coefficient is assigned a physical unit that is the reciprocal of the recovery time interval, and the smoothness penalty weighting coefficient is assigned a physical unit that is the reciprocal of the smoothness decay amount.

[0073] Since both the extracted recovery time interval and the smoothness decay are non-negative physical quantities, the addition of a constant of 1 ensures that the denominator is always greater than or equal to 1, effectively avoiding the risk of division-by-zero overflow caused by the denominator approaching zero in the cloud server's 400-level calculation logic. The physical meaning of this formula is that the longer the employee's correction time or the more severe the accompanying spasms and tremors, the greater the accumulated penalty value, and the smoother the decay of the system's output abnormal state correction capability evaluation value, thus objectively reflecting the operator's weaknesses in responding to sudden abnormal states.

[0074] S603, after completing the calculation and normalization of the above high-dimensional features, the system needs to store the personalized evaluation label in the corresponding system database to complete the final evaluation process. The cloud server 400 records the calculated abnormal state correction capability evaluation value in the capability graph database.

[0075] To achieve multi-dimensional quantification of employee skills, the capability graph database, under the basic execution capability node, extends and constructs independent higher-order correction capability sub-nodes for each specific process. In this embodiment, the cloud server 400 uses a well-known exponentially weighted moving average algorithm to update the graph nodes. Specifically, the system sets a smoothing factor (e.g., 0.2), multiplies the newly generated abnormal state correction capability evaluation value by this smoothing factor, and adds the product of the original historical evaluation value and the corresponding complementary factor (e.g., 0.8), thereby completing the weighted fusion update of the sub-node's state.

[0076] This time-series weighted mechanism, based on multi-source historical data, effectively filters out localized data noise caused by single-time correction anomalies, ensuring that the accumulated values ​​within each node can reliably reflect the true evolution trend of operators' capabilities. Thus, the system not only records employees' basic operational proficiency under normal conditions but also assesses their dynamic recovery capabilities under disrupted conditions, providing effective data support for the subsequent system to push relevant virtual reality training courses or supplementary practical guidance strategies to operators.

[0077] To further illustrate the execution logic and physical control process of the technical solution of this invention in actual industrial scenarios, we will now take the application scenario of wire harness insertion of a precision electromechanical component as an example to specifically deduce the quantitative calculation and graph mapping process under the above-mentioned end-cloud collaborative architecture.

[0078] Operators wearing smart glasses 100 and smartwatches 200 enter the workbench to perform wiring harness connection work. At the beginning of the operation, the edge computing gateway 300 receives and fuses multimodal data to generate spatial trajectory data, which is then uploaded to the cloud server 400. The cloud server 400 calculates the trajectory distance deviation between the current action and the preset standard operation trajectory model. Assuming the currently acquired trajectory distance deviation is 1.5 mm, the system uses a proficiency measurement mapping formula to calculate the operation proficiency value. In this calculation, the attenuation mapping constant is configured as 0.2. Substituting into the formula, the operation proficiency value equals 1 divided by (1 plus the product of 0.2 and 1.5), i.e., 1 divided by 1.3, resulting in an operation proficiency value of approximately 0.77.

[0079] After updating the operational proficiency value to the capability graph database, the cloud server 400 further calculates the dynamic control parameters of the underlying hardware. The system uses a dynamic threshold modulation formula to calculate the dynamic position tolerance threshold. A base tolerance value of 2.0 mm and a tolerance scaling factor of 10.0 mm are set. Substituting the operational proficiency value of 0.77, the dynamic position tolerance threshold is calculated as 2.0 plus the product of 10.0 and (1 minus 0.77), resulting in a dynamic position tolerance threshold of 4.3 mm. Simultaneously, the system uses a duty cycle parameter calculation formula to calculate the pulse width modulation duty cycle parameter. A base duty cycle lower limit of 20% and a feedback strength gain factor of 60% are set. The calculated pulse width modulation duty cycle parameter is 20% plus the product of 60% and 0.23, resulting in 33.8%. The cloud server 400 encapsulates the 4.3 mm dynamic position tolerance threshold and the 33.8% pulse width modulation duty cycle parameter and sends them to the edge computing gateway 300.

[0080] In subsequent practical monitoring, the edge computing gateway 300 compared spatial coordinates in real time. When it detected a significant deviation in the operator's hand movements, with the instantaneous spatial deviation reaching 5.0 mm across multiple consecutive sampling frames—a value exceeding the set dynamic position tolerance threshold of 4.3 mm—the edge computing gateway 300 immediately sent a haptic feedback trigger command to the smartwatch 200 based on dynamic control parameters. The vibration motor built into the smartwatch 200 then drove physical vibrations at a duty cycle of 33.8%, providing the operator with appropriate haptic cues.

[0081] At the moment the command is issued, the edge computing gateway 300 records the starting absolute timestamp and calculates the intercepted time window length using a dynamic time window length calculation formula. The base response time constant is set to 400 milliseconds, the correction time scaling factor is 50 milliseconds / mm, the trigger instantaneous spatial deviation is 5.0 mm, and the minimum zero offset is 0.01. Substituting these values ​​into the formula, the intercepted time window length equals 400 plus the product of 50 and (5.0 divided by 0.78), resulting in approximately 720.5 milliseconds. The edge computing gateway 300 intercepts the response trajectory feature sequence based on this window length and uploads it. The cloud server 400 parses this sequence, extracting a recovery time interval of 0.45 seconds and a smoothness decay of 1.2.

[0082] The system uses a normalized evaluation formula to calculate the anomaly state correction capability evaluation value. The time penalty weight coefficient is set to 0.6, and the smoothness penalty weight coefficient is set to 0.4. Substituting the parameters, the anomaly state correction capability evaluation value equals 1 divided by (1 plus the product of 0.6 and 0.45 plus the product of 0.4 and 1.2), resulting in an anomaly state correction capability evaluation value of approximately 0.57. The cloud server 400 ultimately records this value in the higher-order correction capability sub-node of the capability graph database, completing a single business loop.

[0083] To verify the actual operational effectiveness of the technical solution of this invention, multiple comparative experiments were conducted in a real industrial training environment. Two groups of operators with the same initial skill level were introduced into the experiments, and trained for 20 cycles on assembly processes using the scheme of this embodiment and a traditional fixed-threshold scheme, respectively. In the traditional fixed-threshold scheme, the spatial tolerance judgment boundary and tactile feedback intensity were both set as fixed constants, not dynamically evolving with the operator's skill map state. The experiment continuously recorded key parameters deposited in the skill map database to quantitatively evaluate the actual impact of the two system architectures on employee training and empowerment.

[0084] See Figure 9 , Figure 9 The solid black line in the graph represents the scheme of this embodiment, while the dark gray dashed line represents the traditional fixed threshold scheme. As can be seen from the data trend in the graph, in the initial stage of training, the operational proficiency values ​​corresponding to both schemes show an upward trend. However, as the training cycle progresses, the operational proficiency value corresponding to the dark gray dashed line shows a significant stagnation in growth around the 10th cycle, and even exhibits a localized decline. This physical phenomenon stems from the fact that the traditional fixed threshold scheme uses a single judgment boundary. Once employees' skills have initially improved, the fixed tolerance cannot further constrain their minor movement flaws, and the continuous high-intensity fixed tactile feedback easily leads to human nerve fatigue, resulting in a failure to continuously improve movement accuracy.

[0085] This embodiment incorporates a dynamic control parameter backflow mechanism based on graph state. As employee skills increase, the system adaptively narrows the dynamic position tolerance threshold and smoothly reduces the pulse width modulation duty cycle parameter. The operational proficiency value, corresponding to the black solid line, maintains a robust and continuous upward trend throughout the 20 cycles, converging to a higher level in the later stages. The evolution of the above multidimensional data indicates that the negative correlation threshold modulation logic constructed in this scheme effectively overcomes the interaction fatigue problem, more reasonably guiding operators towards the standard operation trajectory model and improving the conversion efficiency of education and training.

[0086] See Figure 10 , Figure 10 The solid black line represents the scheme in this embodiment, while the dark gray dashed line represents the traditional fixed threshold scheme. This chart records the total time required for the operator's motion coordinates to converge back to the safe space when faced with randomly introduced operational disturbances. The data shows that the recovery time intervals of the traditional scheme, represented by the dark gray dashed line, are more dispersed across cycles, and the overall time consumption is relatively high, indicating that a single physical intervention is insufficient to help operators establish stable muscle memory and emergency corrective reflexes.

[0087] In contrast, the solid black line representing the solution in this embodiment not only falls within a lower time interval overall, but also shows a gradually narrowing fluctuation range in its recovery time interval as the training cycle increases. This physical performance is attributed to the solution's acquisition of response trajectory feature sequences in closed-loop control and the quantification of abnormal state correction capability evaluation values, enabling the system to provide differentiated guidance and feedback intensity in subsequent training based on these high-order characteristics. Under the guidance of dynamic haptic interaction matched to their own abilities, operators experience reduced neural reflex delays in responding to sudden spatial deviations, and the continuity of action execution is strengthened. The above results verify, from a physical data perspective, the capability map quantification and edge-cloud collaborative closed-loop control mechanism proposed in this embodiment of the invention, which can robustly improve employees' comprehensive operational skills and emergency recovery capabilities in complex operating environments.

[0088] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in this application, and these modifications or substitutions should all be covered within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A personalized employee education and empowerment management system based on competency graphs, characterized in that, include: Smart glasses (100) are used to acquire first-person view video data; A smartwatch (200) is used to collect hand motion capture data, wherein the first-view video data and the hand motion capture data together constitute multimodal data; An edge computing gateway (300) is used to receive the multimodal data, perform time-series alignment and fusion calibration to generate spatial trajectory data; A cloud server (400) is used to acquire the spatial trajectory data and compare it with a preset standard operation trajectory model to calculate and generate an operation proficiency value; update the operation proficiency value to the capability graph database for personalized employee education and empowerment data management, and parse the operation proficiency value to calculate and generate dynamic control parameters; The edge computing gateway (300) is also used to receive the dynamic control parameters and calculate the instantaneous spatial deviation between the current spatial trajectory data and the preset standard operation trajectory model. When the instantaneous spatial deviation exceeds the limit, a tactile feedback trigger command is generated based on the dynamic control parameters. The smartwatch (200) is also used to receive the haptic feedback trigger command and control the vibration motor to output vibration.

2. The personalized employee education and empowerment management system based on capability graphs according to claim 1, characterized in that, When the edge computing gateway (300) receives the multimodal data and performs time-series alignment and fusion calibration to generate spatial trajectory data, it is specifically used for: Using the timeline of the first-view video data in the multimodal data as a reference, the hand motion capture data in the multimodal data is matched and aligned with the nearest timestamp, and the pixel coordinates of key hand points are extracted to generate three-dimensional absolute coordinate anchor points. Using the three-dimensional absolute coordinate anchor point as the external observation reference, the extended Kalman filter algorithm is used to correct the error of the hand motion capture data and output a continuous three-dimensional spatial coordinate vector. The spatial trajectory data is formed by combining the continuous three-dimensional spatial coordinate vectors.

3. The personalized employee education and empowerment management system based on capability graphs according to claim 2, characterized in that, When the cloud server (400) acquires the spatial trajectory data and compares it with a preset standard operation trajectory model to calculate and generate an operation proficiency value, it is specifically used for: Extract the spatial trajectory data and calculate the spatial Euclidean distance between each sampling point by comparing it with the corresponding three-dimensional spatial coordinate vector in the preset standard operation trajectory model to construct a global cost matrix; The dynamic programming algorithm is used to search for continuous matching paths in the global cost matrix and output the trajectory distance deviation value. The operational proficiency value is calculated by combining the proficiency metric mapping formula with the trajectory distance deviation value.

4. The personalized employee education and empowerment management system based on capability graphs according to claim 1, characterized in that, When the cloud server (400) updates the operational proficiency value to the competency graph database for personalized employee education and empowerment data management, it is specifically used for: The newly acquired operation proficiency value is weighted and fused with the historical proficiency parameters in the capability graph database to generate the baseline value after the corresponding process node is updated.

5. The personalized employee education and empowerment management system based on capability graphs according to claim 1, characterized in that, The process by which the cloud server (400) parses the operation proficiency value to generate dynamic control parameters includes: Based on the operational proficiency value, the dynamic position tolerance threshold is calculated using the dynamic threshold modulation formula; Based on the aforementioned operational proficiency value, the pulse width modulation duty cycle parameter is calculated using the duty cycle parameter calculation formula. The dynamic position tolerance threshold and the pulse width modulation duty cycle parameter are encapsulated to form the dynamic control parameter; The dynamic position tolerance threshold is the dynamic boundary for determining whether an error exceeds the limit.

6. The personalized employee education and empowerment management system based on capability graphs according to claim 5, characterized in that, The edge computing gateway (300) receives the dynamic control parameters and calculates the instantaneous spatial deviation between the current spatial trajectory data and the preset standard operation trajectory model. When the instantaneous spatial deviation exceeds the limit, it generates a tactile feedback trigger command based on the dynamic control parameters, specifically for: Extract the three-dimensional physical coordinates of the spatial trajectory data at the current timestamp; In the preset standard operation trajectory model, find the standard reference point with the shortest physical distance to the three-dimensional physical coordinates; The instantaneous spatial deviation between the three-dimensional physical coordinates and the standard reference point is calculated using the spatial Euclidean distance formula; When the instantaneous spatial deviation of multiple consecutive sampling frames within the time sliding window is greater than the dynamic position tolerance threshold in the dynamic control parameters, the haptic feedback trigger command is generated based on the pulse width modulation duty cycle parameter in the dynamic control parameters.

7. The personalized employee education and empowerment management system based on capability graphs according to claim 6, characterized in that, After generating the haptic feedback trigger command, the edge computing gateway (300) further generates a response trajectory feature sequence, which includes: Extract the current system time record from the system master clock as the starting absolute timestamp; The length of the intercepted time window is calculated using the dynamic time window length calculation formula based on the instantaneous spatial deviation at the trigger time and the operation proficiency value. Based on the interception time window length and the starting absolute timestamp, the corresponding spatial trajectory data is intercepted from the cache queue to generate the response trajectory feature sequence.

8. The personalized employee education and empowerment management system based on capability graphs according to claim 7, characterized in that, The edge computing gateway (300) is also used to encapsulate the response trajectory feature sequence and the corresponding starting absolute timestamp into a data packet after generating the response trajectory feature sequence, send it independently to the cloud server (400) through asynchronous communication, and release local cache resources.

9. A personalized employee education and empowerment management system based on capability graphs as described in claim 8, characterized in that, The process by which the cloud server (400) acquires and parses the response trajectory feature sequence to generate multidimensional physical indicators includes: Extract the time points in the response trajectory feature sequence where the instantaneous spatial deviation decreases again and remains within the dynamic position tolerance threshold range of the dynamic control parameters; The time difference between the time node and the corresponding starting absolute timestamp is used as the recovery time interval; The third derivative with respect to time is calculated for the three-dimensional spatial coordinates in the response trajectory feature sequence, and the smoothness attenuation is obtained by square integration of the third derivative. The recovery time interval and the smoothness decay amount are combined to form the multidimensional physical index.

10. A personalized employee education and empowerment management system based on capability graphs according to claim 9, characterized in that, When the cloud server (400) completes the map feature update after acquiring the multidimensional physical indicators, it is specifically used for: An evaluation value for abnormal state correction capability is generated by cross-penalizing the recovery time interval and the smoothness decay in the multidimensional physical index using an evaluation value normalization formula. The evaluation value for abnormal state correction capability is then recorded in the corresponding high-order correction capability sub-node in the capability map database to complete the map feature update.

Images