A joint wear prediction method and system based on artificial intelligence

Abstract

The present application provides a joint wear prediction method and system based on artificial intelligence, wherein the method includes: obtaining three-dimensional mechanical fluctuation trajectory data and target gradient data of a dynamic load joint during a motion cycle of the dynamic load joint; using a pre-trained reinforcement learning model to output a weight distribution result for correcting the joint wear prediction path based on the three-dimensional mechanical fluctuation trajectory data and the target gradient data; extracting friction coefficient mutation characteristics from the target gradient data when the dynamic load joint meets the preset joint material fatigue critical condition; adjusting the composite mechanical characterization parameters according to the weight distribution result to obtain adjusted composite mechanical characterization parameters; generating a joint wear prediction result including a multi-dimensional wear accumulation trend based on the nonlinear superposition result of the adjusted composite mechanical characterization parameters and the friction coefficient mutation characteristics. The present application improves the accuracy and real-time performance of joint wear prediction.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence technology, and in particular to a joint wear prediction method and system based on artificial intelligence. Background Art

[0002] During postoperative rehabilitation after artificial joint replacement surgery or during the operation of high-load mechanical joints (such as those in industrial robots), early prediction of joint wear is crucial for extending service life and avoiding sudden failures. The core challenge facing existing technologies is how to capture the microscopic trends of joint wear in real time under complex dynamic load conditions using multi-source sensor data (such as mechanical and friction coefficient data) and predict its long-term cumulative effects. This requires prediction methods that can not only process high-frequency, nonlinear mechanical fluctuation data but also integrate tribological characteristics to establish a dynamic correlation model of wear evolution.

[0003] A representative approach currently involves a prediction method based on frequency-domain feature analysis and static weight fusion. This method uses accelerometers and torque sensors to collect vibration signals and load data from joint motion. It then extracts the frequency-domain energy distribution characteristics of the vibration signals and linearly weights them with the statistical characteristics of the load data to generate a wear risk score. The model uses historical data to train static weight coefficients, triggering an alert when the score exceeds a threshold.

[0004] The limitations of this method are that static weights lead to short-term wear prediction deviations; frequency domain features make it difficult to capture the cumulative path of asymmetric wear; linear fusion ignores the nonlinear superposition effect of wear accumulation, and the error between long-term prediction results and actual detection values ​​gradually increases. Summary of the Invention

[0005] The present application provides a joint wear prediction method and system based on artificial intelligence to solve the problems of low accuracy and poor real-time performance of joint wear prediction in the prior art.

[0006] In a first aspect, the present application provides a joint wear prediction method based on artificial intelligence, comprising:

[0007] During a motion cycle of the dynamic load joint, three-dimensional mechanical fluctuation trajectory data and target gradient data of the dynamic load joint are obtained, wherein the target gradient data is gradient data of a friction coefficient of the dynamic load joint changing with an angular velocity of motion;

[0008] Based on the three-dimensional mechanical fluctuation trajectory data and the target gradient data, a pre-trained reinforcement learning model is used to output a weight distribution result for correcting the joint wear prediction path;

[0009] Extracting friction coefficient mutation characteristics from the target gradient data when the dynamically loaded joint meets a preset joint material fatigue critical condition;

[0010] Adjusting a composite mechanical characterization parameter according to the weight distribution result to obtain an adjusted composite mechanical characterization parameter, wherein the composite mechanical characterization parameter is generated according to the three-dimensional mechanical fluctuation trajectory data and the target gradient data;

[0011] According to the nonlinear superposition result of the adjusted composite mechanical characterization parameters and the friction coefficient mutation characteristics, a joint wear prediction result including a multi-dimensional wear accumulation trend is generated.

[0012] Optionally, the outputting of a weight distribution result for correcting the joint wear prediction path using a pre-trained reinforcement learning model based on the three-dimensional mechanical fluctuation trajectory data and the target gradient data includes:

[0013] Calculating, based on the asymmetric stress distribution characteristics in the target gradient data, the frequency band overlap ratio between the instantaneous load impact amplitude and the energy distribution interval in the three-dimensional mechanical fluctuation trajectory data, where the energy distribution interval is the energy distribution interval of the mechanical vibration signal generated by the dynamically loaded joint during the current motion cycle;

[0014] Combined with the frequency band overlap ratio, a pre-trained reinforcement learning model is used to output a weight distribution result.

[0015] Optionally, the combining the frequency band overlap ratio and using a pre-trained reinforcement learning model to output a weight distribution result includes:

[0016] Constructing a state space of a pre-trained reinforcement learning model, wherein the state space includes: frequency band overlap ratio, local deviation distribution, and gradient direction consistency coefficient;

[0017] Inputting the state space into a pre-trained reinforcement learning model and outputting a dynamic correction strategy, wherein the action space of the reinforcement learning model includes a dynamic correction strategy for adjusting the contribution coefficient, and the reward function is generated based on the error between the joint wear prediction result and the actual detection value and the fluctuation range of the weight distribution;

[0018] In a plurality of consecutive motion cycles, dynamically smoothing constraints are applied to the contribution coefficients of the asymmetric stress distribution characteristics and the instantaneous load impact amplitude in the composite mechanical characterization parameters in the dynamic correction strategy;

[0019] The contribution coefficient after dynamic smoothing constraint is used as the weight distribution result.

[0020] Optionally, the state space of the pre-trained reinforcement learning model is constructed, wherein the state space includes: frequency band overlap ratio, local deviation distribution and gradient direction consistency coefficient, including:

[0021] According to the historical motion cycle of the dynamically loaded joint, a reference sequence is extracted from the frequency band overlap ratio corresponding to the historical motion cycle;

[0022] Matching the reference sequence with the frequency band overlap ratio corresponding to the current motion cycle to generate a local deviation distribution of the frequency band overlap ratio in a continuous time interval;

[0023] The consistency coefficient between the change direction of the adjacent period fluctuation amplitude in the deviation sequence of the asymmetric stress distribution feature and the gradient direction of the deviation distribution is calculated.

[0024] Optionally, the dynamically smoothing constraint on the contribution coefficients of the asymmetric stress distribution characteristics and the instantaneous load impact amplitude in the composite mechanical characterization parameters in the dynamic correction strategy over a plurality of consecutive motion cycles includes:

[0025] In the plurality of consecutive motion cycles, respectively extracting a first contribution coefficient corresponding to the asymmetric stress distribution feature and a second contribution coefficient corresponding to the instantaneous load impact amplitude in each motion cycle;

[0026] For each motion cycle, calculating a first change in the first contribution coefficient and a second change in the second contribution coefficient between a current cycle and an adjacent previous cycle;

[0027] Determining a maximum allowable amplitude of the first variation and a maximum allowable amplitude of the second variation according to the first contribution coefficient sequence and the second contribution coefficient sequence in the plurality of consecutive motion cycles;

[0028] If the first variation of the current period exceeds the maximum allowable amplitude of the first variation, adjusting the first contribution coefficient of the current period to the superposition value of the first contribution coefficient of the previous period and the maximum allowable amplitude;

[0029] If the second variation of the current cycle exceeds the maximum allowable amplitude of the second variation, adjusting the second contribution coefficient of the current cycle to the superposition value of the second contribution coefficient of the previous cycle and the maximum allowable amplitude;

[0030] The adjusted first contribution coefficient and the second contribution coefficient are respectively used as the contribution coefficients after dynamic smoothing constraints in the current period.

[0031] Optionally, generating composite mechanical characterization parameters according to the three-dimensional mechanical fluctuation trajectory data and the target gradient data includes:

[0032] Obtaining a friction coefficient change rate corresponding to a timestamp in the target gradient data;

[0033] Filtering out a fluctuation trajectory data segment that coincides with the acquisition time of the target gradient data from the three-dimensional mechanical fluctuation trajectory data, wherein the fluctuation trajectory data segment includes an instantaneous load impact amplitude corresponding to each time stamp;

[0034] Marking the energy release direction of the instantaneous load impact amplitude according to the friction coefficient change rate;

[0035] Performing spatial correlation mapping between the energy release direction and the deviation of the three-dimensional mechanical fluctuation trajectory data in the direction of the normal vector of the joint rotation plane to generate an asymmetric stress distribution feature;

[0036] According to the target gradient data, the instantaneous load impact amplitude and the asymmetric stress distribution characteristics are weightedly fused to generate composite mechanical characterization parameters.

[0037] Optionally, generating a joint wear prediction result including a multi-dimensional wear accumulation trend based on the nonlinear superposition result of the adjusted composite mechanical characterization parameter and the friction coefficient mutation characteristic includes:

[0038] Inputting the adjusted composite mechanical characterization parameters and the friction coefficient mutation characteristics into a pre-constructed joint influence relationship structure;

[0039] determining the superposition priority of each mechanical component in the adjusted composite mechanical characterization parameter according to the change direction of the friction coefficient mutation characteristic within the continuous motion cycle;

[0040] Based on the superposition priority, the adjusted composite mechanical characterization parameters and the friction coefficient mutation characteristics are superimposed and calculated in sections to obtain the cumulative wear amount in different superposition directions within each motion cycle;

[0041] Dividing the accumulated wear amount into multiple independent dimensions according to the superposition direction;

[0042] Generating a trend prediction curve for each independent dimension based on the direction of change of the cumulative amount of each independent dimension in the continuous motion cycle;

[0043] The trend prediction curves of the multiple independent dimensions are combined according to the preset wear stage division conditions to output the joint wear prediction result.

[0044] In a second aspect, the present application provides an artificial intelligence-based joint wear prediction system, comprising:

[0045] an acquisition module, configured to acquire three-dimensional mechanical fluctuation trajectory data and target gradient data of the dynamic load joint during a motion cycle of the dynamic load joint, wherein the target gradient data is gradient data of a friction coefficient of the dynamic load joint changing with a motion angular velocity;

[0046] an output module, configured to output a weight distribution result for correcting the joint wear prediction path using a pre-trained reinforcement learning model based on the three-dimensional mechanical fluctuation trajectory data and the target gradient data;

[0047] An extraction module is used to extract friction coefficient mutation characteristics from the target gradient data when the dynamic load joint meets a preset joint material fatigue critical condition;

[0048] an adjustment module, configured to adjust a composite mechanical characterization parameter according to the weight distribution result to obtain an adjusted composite mechanical characterization parameter, wherein the composite mechanical characterization parameter is generated according to the three-dimensional mechanical fluctuation trajectory data and the target gradient data;

[0049] A generation module is used to generate a joint wear prediction result including a multi-dimensional wear accumulation trend based on the nonlinear superposition result of the adjusted composite mechanical characterization parameters and the friction coefficient mutation characteristics.

[0050] In a third aspect, the present application provides a computing device comprising a processor and a memory, wherein the memory stores a computer program, and the processor is configured to run the computer program to execute an artificial intelligence-based joint wear prediction method as described in any one of the first aspects.

[0051] In a fourth aspect, the present application provides a computer storage medium having computer program instructions stored thereon, wherein the computer program instructions, when executed by a processor, implement an artificial intelligence-based joint wear prediction method as described in any one of the first aspects.

[0052] In the present application, a joint wear prediction method based on artificial intelligence is provided, which includes: obtaining three-dimensional mechanical fluctuation trajectory data and target gradient data of the dynamic load joint during the movement cycle of the dynamic load joint, wherein the target gradient data is the gradient data of the friction coefficient of the dynamic load joint changing with the angular velocity of movement; according to the three-dimensional mechanical fluctuation trajectory data and the target gradient data, using a pre-trained reinforcement learning model to output a weight distribution result for correcting the joint wear prediction path; when the dynamic load joint meets the preset joint material fatigue critical condition, extracting the friction coefficient mutation feature from the target gradient data; adjusting the composite mechanical characterization parameter according to the weight distribution result to obtain the adjusted composite mechanical characterization parameter, wherein the composite mechanical characterization parameter is generated according to the three-dimensional mechanical fluctuation trajectory data and the target gradient data; and generating a joint wear prediction result including a multi-dimensional wear accumulation trend according to the nonlinear superposition result of the adjusted composite mechanical characterization parameter and the friction coefficient mutation feature.

[0053] The technical solution provided by this application has the following beneficial effects:

[0054] This application establishes a dynamic correlation between the complete mechanical characteristics and friction characteristics of joint motion, providing a multi-dimensional data foundation for wear analysis. It realizes the adaptive dynamic fusion of mechanical data and friction gradient data to optimize the accuracy of the prediction path. It captures the key tribological change characteristics of joint materials when they are close to fatigue and improves the sensitivity of critical state identification. It optimizes the expression of mechanical-friction coupling parameters through dynamic weights to enhance the physical meaning integrity of the characterization parameters. It outputs a multi-dimensional prediction that includes spatial distribution characteristics and time accumulation trends to realize the visualization of full-cycle wear evolution.

[0055] Furthermore, this application also analyzes the asymmetric stress characteristics in the target gradient data, calculates the degree of frequency band overlap between the instantaneous load impact and the joint vibration energy distribution in the three-dimensional mechanical data, and dynamically generates the optimal weight distribution strategy based on the overlap ratio using the reinforcement learning model.

[0056] In addition, this method achieves precise spatiotemporal matching of mechanical impact characteristics and vibration energy distribution. Through dynamic weight adjustment of reinforcement learning, it improves the accuracy of quantitative evaluation of wear influencing factors under different motion states, and solves the problem of disconnection between frequency domain characteristics and transient mechanical responses in traditional methods.

[0057] These and other aspects of the present application will become more readily apparent from the description of the following embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, a brief introduction will be given below to the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0059] Figure 1 A flowchart of a joint wear prediction method based on artificial intelligence provided in an embodiment of the present application;

[0060] Figure 2 A schematic structural diagram of an artificial intelligence-based joint wear prediction system provided in an embodiment of the present application;

[0061] Figure 3 A schematic diagram of the structure of a computing device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0062] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application.

[0063] In some of the processes described in the specification and claims of this application and the above-mentioned figures, multiple operations that appear in a specific order are included, but it should be clearly understood that these operations may not be executed in the order in which they appear in this document or may be executed in parallel. The serial numbers of the operations, such as 101, 102, etc., are only used to distinguish between different operations, and the serial numbers themselves do not represent any order of execution. In addition, these processes may include more or fewer operations, and these operations may be executed in sequence or in parallel. It should be noted that the descriptions of "first", "second", etc. in this document are used to distinguish different messages, devices, modules, etc., and do not represent a sequential order, nor do they limit "first" and "second" to being different types.

[0064] Researchers have found that existing joint wear prediction methods are difficult to accurately capture the dynamic coupling relationship between mechanical properties and tribological properties under dynamic load conditions, and are unable to effectively identify the mutation characteristics under the critical fatigue state of the material, resulting in deviations between the prediction results and the actual wear conditions. Based on this, an embodiment of the present application provides a multi-dimensional joint wear prediction method based on reinforcement learning. This method synchronously collects three-dimensional mechanical fluctuation trajectory data and friction coefficient gradient data, uses a reinforcement learning model to dynamically optimize weight distribution, and combines the friction mutation characteristics under critical fatigue conditions to achieve nonlinear coupling analysis of mechanical-friction characteristics, thereby generating multi-dimensional wear prediction results that include spatial distribution and time evolution trends. The technical solution of the present application can be applied to scenarios such as artificial joint implant life assessment and industrial robot joint health monitoring that require accurate prediction of dynamic load joint wear.

[0065] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without making creative efforts are within the scope of protection of this application.

[0066] Figure 1 A flowchart of a joint wear prediction method based on artificial intelligence provided in an embodiment of the present application is shown in FIG. Figure 1 As shown, the method includes:

[0067] Step 101: During a motion cycle of a dynamic load joint, three-dimensional mechanical fluctuation trajectory data and target gradient data of the dynamic load joint are obtained. The target gradient data is gradient data of a friction coefficient of the dynamic load joint changing with a motion angular velocity.

[0068] In this step, 3D mechanical fluctuation trajectory data refers to a continuous data sequence collected by multi-axis force sensors mounted on joints, reflecting the force changes in the joint within three dimensions during movement. Target gradient data, measured by friction sensors, quantifies the rate of change of the joint friction coefficient with angular velocity, reflecting the dynamic friction characteristics of the joint contact surface.

[0069] In this embodiment, a six-dimensional force sensor built into the joint collects force data in real time in the X, Y, and Z directions at a sampling rate of 1000Hz. After noise removal through a Kalman filter, a three-dimensional mechanical fluctuation trajectory is generated. Simultaneously, a high-precision friction sensor measures the instantaneous friction coefficient during joint rotation. Combined with the angular velocity data obtained by the encoder, a friction coefficient-angular velocity gradient curve is calculated through numerical differentiation. These two types of data are synchronized through hardware clock synchronization to ensure time alignment, ultimately forming a joint mechanical-friction dataset with strictly matched timestamps.

[0070] For example, in the AI ​​monitoring system of intelligent knee prostheses, a six-dimensional force sensor collects the patient's three-dimensional dynamic load (such as a sagittal plane peak force of 800N) in real time when walking at a frequency of 1kHz, and simultaneously measures the friction coefficient of the femoral-tibia contact surface through an embedded friction sensor. Combined with the acquired joint angular velocity data (0-120° / s), the edge computing unit calculates the friction coefficient gradient curve (such as 0.08→0.15 / (rad / s)). All data are uploaded to the cloud AI platform through the 5G module, with a time alignment accuracy of 0.1ms.

[0071] Step 102: Based on the three-dimensional mechanical fluctuation trajectory data and the target gradient data, a pre-trained reinforcement learning model is used to output a weight distribution result for correcting the joint wear prediction path.

[0072] In this step, the joint wear prediction path refers to the evolution trajectory of the wear accumulation process in the feature space. The generation process of the joint wear prediction path includes: within the initial motion cycle of the dynamic load joint, extracting the mechanical fluctuation extreme value of each motion phase point based on the three-dimensional mechanical fluctuation trajectory data, and calculating the mechanical fluctuation change rate between adjacent phase points; establishing an initial wear influence relationship table according to the correspondence between the mechanical fluctuation change rate and the target gradient data, and the initial wear influence relationship table contains the mapping relationship between the mechanical fluctuation change rate and the friction coefficient gradient under different motion phases; marking the mechanical fluctuation change rate in the initial wear influence relationship table that meets the preset change rate threshold as a key influence node, and connecting each key influence node in the order of motion phase to form an initial prediction path; based on the historical wear data of the dynamic load joint in the continuous motion cycle, calculating the path offset of each key influence node, and dynamically correcting the initial prediction path according to the path offset to obtain the joint wear prediction path. A pre-trained reinforcement learning model is an intelligent decision-making model that has completed multiple stages of training before being deployed in an actual joint wear prediction system. This model analyzes multi-source sensor data under dynamic loads in real time and autonomously determines the contribution of each parameter to wear. For example, it automatically increases the friction gradient weight (from 0.4 to 0.7) during sudden acceleration to address the risk of abnormal wear under boundary lubrication conditions. The weight assignment result is the quantized coefficient matrix output by the reinforcement learning model, which is used to dynamically adjust the contribution of mechanical and friction parameters to wear prediction.

[0073] In an embodiment of the present application, the acquired joint data set is input into a pre-trained reinforcement learning model. The model first extracts the spatial characteristics of the mechanical trajectory through a 3D convolutional network, and analyzes the time dependence of the gradient data. It then outputs a weight allocation strategy (such as a mechanical amplitude weight of 0.6 and a friction gradient weight of 0.4) to evaluate the degree to which the strategy improves the historical wear prediction error. Finally, the model gradually optimizes the weight allocation through policy gradient updates, and outputs a dynamic weight matrix that minimizes the prediction error.

[0074] For example, after the cloud-based AI platform receives the data, the pre-trained reinforcement learning model analyzes the current gait phase (such as the mid-stance phase): based on the spatiotemporal correlation between the main frequency of mechanical fluctuations (15 Hz) and the sudden change in friction gradient (0.12 to 0.18 / (rad / s)), the weight matrix is ​​dynamically adjusted - the axial impact force weight is increased from 0.5 to 0.7, while the lateral shear force weight is reduced to 0.3.

[0075] Step 103: When the dynamic load joint meets the preset joint material fatigue critical condition, extract the friction coefficient mutation feature from the target gradient data.

[0076] In this step, the critical fatigue condition for the joint material refers to the stress-friction combined threshold, determined in advance through finite element analysis, at which microscopic damage to the joint material begins. The friction coefficient mutation feature refers to an abnormal change in the gradient data where the slope exceeds a preset threshold (e.g., 0.15 / (rad / s)).

[0077] In an embodiment of the present application, the second-order derivative of the target gradient data is monitored in real time, and feature extraction is triggered when it is detected that the derivative change of three consecutive sampling points exceeds a threshold; a sliding window is used to analyze the frequency domain components of the mutation segment, and the dominant frequency component (such as 80-120Hz) is extracted; at the same time, the mechanical fluctuation data at the corresponding moment are correlated, and the average stress amplitude (such as 45MPa) and the main direction (such as a 22° angle with the rotation plane) during the mutation period are calculated, and finally a three-dimensional feature vector containing frequency, amplitude, and direction is formed.

[0078] For example, when the AI ​​system detects that the friction gradient slope continuously exceeds the threshold (0.2 / (rad / s)) when the patient walks up the stairs, the fatigue analysis module is triggered: the 88Hz high-frequency vibration component of the mutation period (1.2-1.5 seconds) is extracted through wavelet transform, and the three-dimensional mechanical data of this period is correlated to find that the contact pressure is concentrated on the outer platform (peak pressure 6.2MPa), and the characteristic vector [88Hz, 6.2MPa, outer side] is generated, which is marked as the polyethylene liner fatigue initiation signal.

[0079] Step 104: adjusting the composite mechanical characterization parameters according to the weight distribution result to obtain adjusted composite mechanical characterization parameters, wherein the composite mechanical characterization parameters are generated according to the three-dimensional mechanical fluctuation trajectory data and the target gradient data.

[0080] In this step, the composite mechanical characterization parameters are comprehensive indicators that integrate the spatial distribution characteristics of dynamic loads and friction gradient changes, and include three sub-parameters: the main frequency of mechanical fluctuations, equivalent stress, and friction work density.

[0081] In an embodiment of the present application, a weighted principal component analysis is first performed on the mechanical trajectory data based on the weight matrix, and the first three principal components (with a contribution rate of more than 85%) are extracted as spatial features; then the friction power density (gradient × angular velocity) is calculated according to the friction gradient weight; finally, the weighted principal component amplitude, maximum shear stress and friction power density are fused through a fully connected neural network to output an adjusted composite parameter vector.

[0082] For example, AI reconstructs composite parameters based on the weight matrix (axial 0.7 / lateral 0.3): the original axial pressure features and mutation features are fused through the attention mechanism to generate an optimized three-dimensional representation vector.

[0083] Step 105: Generate a joint wear prediction result including a multi-dimensional wear accumulation trend according to the nonlinear superposition result of the adjusted composite mechanical characterization parameter and the friction coefficient mutation characteristic.

[0084] In this step, nonlinear superposition involves the cross-scale coupling of composite parameters and mutation characteristics using a hyperbolic tangent function. The joint wear prediction results are three-dimensional, encompassing spatial distribution (contact area), temporal evolution (wear rate), and sensitivity to operating conditions (load-velocity combinations).

[0085] In an embodiment of the present application, the adjusted composite parameter vector and the mutation feature vector are input into a pre-trained deep cross network. First, a second-order interaction term (such as mechanical main frequency × mutation frequency) is generated through the feature cross layer, and then a gated cyclic unit is used to model the dynamic coupling relationship between the parameters; the final output includes the predicted results of the isotropic wear depth change curve in the next N cycles (such as 1 million cycles).

[0086] For example, AI integrates historical gait data (2 million steps in the past three months) with current composite parameters to output multi-dimensional predictions: 1. The spatial heat map shows that the wear depth of the outer platform will reach 0.28mm (0.11mm on the inner side) in the next six months; 2. The time curve predicts that it will enter a period of rapid wear after 80,000 steps; 3. Working condition sensitivity analysis shows that climbing stairs contributes 56% of the wear increase.

[0087] This solution simultaneously collects real-time mechanical fluctuations and friction gradient data from joint motion, and uses reinforcement learning to dynamically optimize a multi-source parameter fusion strategy to accurately capture the characteristic mutations of critical material fatigue states. Ultimately, it generates wear predictions that reflect both spatial distribution differences and temporal evolution patterns. Compared to traditional methods, this approach can identify abnormal wear patterns earlier and provide comprehensive wear development assessments in dynamic load scenarios such as artificial joints and industrial robots.

[0088] To address the issue of inaccurate coupled analysis of mechanical and frictional characteristics in dynamic load joint wear prediction, in some embodiments, step 102: outputting a weight distribution result for correcting the joint wear prediction path using a pre-trained reinforcement learning model based on the three-dimensional mechanical fluctuation trajectory data and the target gradient data, includes:

[0089] Step 201: Based on the asymmetric stress distribution characteristics in the target gradient data, calculate the frequency band overlap ratio of the instantaneous load impact amplitude and the energy distribution range in the three-dimensional mechanical fluctuation trajectory data, and the energy distribution range is the energy distribution range of the mechanical vibration signal generated by the dynamic load joint in the current motion cycle.

[0090] In step 201, the asymmetric stress distribution characteristic refers to the spatial stress difference pattern formed by uneven force on the joint contact surface during movement, specifically manifested as the difference in the friction coefficient gradient on both sides of the joint rotation plane exceeding a set threshold (e.g., 0.05 / (rad / s)). The transient load impact amplitude refers to the peak value of the transient mechanical response generated by sudden external forces (such as stepping impact and sudden stop) during the movement of a dynamically loaded joint. The energy distribution range refers to the energy distribution characteristics of the mechanical vibration signal in each frequency band within a single complete motion cycle of a dynamically loaded joint (such as the 0-100% gait phase of the walking cycle). The frequency band overlap ratio refers to the degree of overlap between the main vibration frequency of the transient load impact and the frequency band where the overall vibration energy of the joint is concentrated, and is used to quantify the coupling strength between local impact and overall vibration.

[0091] In an embodiment of the present application, the stress asymmetry index is first extracted from the target gradient data by Hilbert-Huang transform (such as the gradient mean 0.12 / (rad / s) on the left and 0.07 / (rad / s) on the right), and then a short-time Fourier transform is performed on the three-dimensional mechanical data to identify the dominant frequency band of the instantaneous load impact. Finally, the proportion of the frequency band overlapping area with the full-cycle vibration energy spectrum is calculated to obtain the frequency band overlap ratio.

[0092] Step 202: Based on the frequency band overlap ratio, a pre-trained reinforcement learning model is used to output a weight distribution result.

[0093] In step 202 , the weight distribution result refers to the contribution adjustment coefficient matrix of the mechanical parameters and friction parameters autonomously generated by the reinforcement learning model according to the dynamic working conditions.

[0094] In this embodiment, the frequency band overlap ratio is compared with a reference sequence in a historical motion cycle database to generate a deviation index. Simultaneously, the gradient direction consistency coefficient between the current asymmetric stress signature and the historical data is calculated. These two parameters form the state inputs of the reinforcement learning model. The policy network outputs three sets of weights: axial force weight, tangential force weight, and friction gradient weight. After evaluation, the weight matrix is ​​ultimately determined.

[0095] Here is a specific example;

[0096] In the AI ​​health monitoring system of the intelligent artificial hip joint, when the patient walks daily, the system obtains the friction coefficient change data of the joint contact surface in real time. Through analysis, it is found that there is a significant difference in the friction coefficient change between the outside and inside of the joint, forming an asymmetric stress characteristic; at the same time, the three-dimensional mechanical sensor captures the impact vibration signal generated by the joint at the moment the heel touches the ground. The spectrum analysis determines the degree of overlap between the main vibration frequency band of the impact and the energy concentration frequency band of the joint vibration during the entire gait cycle; the system inputs this frequency band overlap feature into the pre-trained reinforcement learning decision model. Based on the current motion state and historical wear data, the model automatically generates a dynamic weight combination for axial pressure, lateral shear force and friction coefficient.

[0097] In this application, by quantifying the frequency-domain coupling relationship between mechanical impact and friction characteristics and combining it with the dynamic decision-making capabilities of reinforcement learning, we achieve precise weighting of wear-influencing factors. This allows the prediction model to automatically adapt to the changing importance of parameters under different motion conditions, improving the timeliness of identifying abnormal wear conditions. Furthermore, a feature extraction method based on the overlap ratio of frequency bands effectively overcomes the technical obstacle of traditional time-domain analysis, which makes it difficult to distinguish between transient impact and sustained vibration.

[0098] To address the adaptability of weight allocation strategies in dynamic load joint wear prediction, in some embodiments, step 202 : combining the frequency band overlap ratio and outputting a weight allocation result using a pre-trained reinforcement learning model includes:

[0099] Step 301: Construct a state space of a pre-trained reinforcement learning model, wherein the state space includes: frequency band overlap ratio, local deviation distribution, and gradient direction consistency coefficient.

[0100] In step 301, the local deviation distribution refers to the change pattern of the degree of difference between the frequency band overlap ratio of the current motion cycle and the historical normal reference sequence on the time axis. The local deviation is used to characterize the deviation characteristics of the current motion cycle relative to the historical reference state in a specific phase interval. Generation process: within the continuous motion cycle of the dynamic load joint, the frequency band overlap ratio sequence of the current motion cycle is extracted and matched with the reference frequency band overlap ratio sequence of the historical motion cycle; the numerical difference between each data point in the current cycle sequence and the corresponding position data point in the reference sequence is calculated to obtain the original deviation set; the original deviation set is divided into a preset phase window, and the statistical distribution characteristics of the deviation in each phase window are calculated, and the statistical distribution characteristics include the central tendency and discrete degree of the deviation in the window; the statistical distribution characteristics of each phase window are quantified as local deviation. The gradient direction consistency coefficient refers to the degree of matching between the evolution direction of the asymmetric stress feature and the change direction of the local deviation.

[0101] In an embodiment of the present application, a sliding time window algorithm is used to extract the frequency band overlapping ratio sequence of the most recent several motion cycles, and its dynamic time regularization distance with the standard health reference curve is calculated to generate a local deviation curve that changes with time; at the same time, the cosine value of the angle between the changing trajectory of the asymmetric stress characteristic vector and the slope of the deviation curve is analyzed to obtain a quantitative index of directional consistency.

[0102] Step 302: Input the state space into a pre-trained reinforcement learning model and output a dynamic correction strategy. The action space of the reinforcement learning model includes a dynamic correction strategy for adjusting the contribution coefficient. The reward function is generated based on the error between the joint wear prediction result and the actual detection value and the fluctuation amplitude of the weight distribution.

[0103] In step 302, the dynamic correction strategy refers to the mechanical parameter weight adjustment scheme generated in real time by the reinforcement learning model based on the current state. The contribution coefficient refers to the quantitative value of the influence of each mechanical characteristic on the final wear prediction result. The expression of the reward function is

[0104] ;

[0105] Where R represents the instantaneous reward value during the current exercise cycle. E represents the mean absolute error between the predicted and actual joint wear values. ΔW represents the fluctuation in the weight distribution during the current exercise cycle compared to the previous cycle. Wmax represents the maximum allowable threshold for weight distribution fluctuation. α, β, and γ represent adjustable reward coefficients, which balance the effects of prediction error and weight stability on the reward.

[0106] In an embodiment of the present application, the three-dimensional features of the state space are input into a reinforcement learning network based on a policy gradient, and a preliminary weight adjustment amount is output through a policy network comprising multiple fully connected layers, which is then passed through an action space constraint layer to ensure that the output value is within a reasonable range; the reward mechanism simultaneously considers the degree of reduction of the prediction error and the smoothness of the weight changes in adjacent cycles, and uses a dual network to evaluate these two goals separately.

[0107] Step 303: In a plurality of consecutive motion cycles, dynamically smoothing constraints are applied to the contribution coefficients of the asymmetric stress distribution characteristics and the instantaneous load impact amplitude in the composite mechanical characterization parameters in the dynamic correction strategy.

[0108] In step 303, dynamic smoothness constraint refers to imposing a limit on the weight change rate within a continuous motion cycle to avoid sudden changes in the prediction results.

[0109] In an embodiment of the present application, a sliding average calculation window for the contribution coefficient is established. When it is detected that the change in the coefficient of adjacent cycles exceeds a preset threshold, an exponentially weighted moving average algorithm is used for smoothing, and the number of times the limit is exceeded is recorded as a trigger condition for model retraining.

[0110] Step 304: The contribution coefficient after dynamic smoothing constraint is used as the weight distribution result.

[0111] In the embodiment of the present application, the contribution coefficients of the smoothed mechanical parameters are normalized to generate a standardized weight matrix that can be used for wear prediction calculations.

[0112] Here's a specific example:

[0113] In the AI ​​health monitoring system for intelligent artificial hip joints, when a patient transitions from daily walking to stair climbing, the system first detects an increase in the friction coefficient on the lateral side of the joint, resulting in a more pronounced asymmetric stress signature. Simultaneously, the three-dimensional mechanical sensor detects a new overlap between the impact vibration frequency band generated by stepping and the energy distribution range during normal gait. These features are fed into a reinforcement learning model, which analyzes the data trends over multiple recent cycles and finds a continuous increase in local deviation and a decrease in gradient consistency. The model then gradually adjusts the contribution coefficients of various mechanical parameters. Initially, it recommends significantly increasing the weight of the friction coefficient. However, as the regularity of this movement pattern is continuously monitored, the system automatically smooths the weight changes, ultimately forming a stable weight distribution scheme suitable for stair climbing and descending. This scheme moderately reduces the weight of axial pressure while increasing the consideration of lateral shear force. This optimized weight combination is then synchronized and updated in the patient's personalized wear prediction model, providing doctors with a more accurate basis for decision-making in rehabilitation training plans.

[0114] In this application, a reinforcement learning framework incorporating multidimensional state features, combined with a dynamic smoothing constraint mechanism, ensures both the sensitivity of the weight allocation strategy to abnormal operating conditions and the stability of the prediction results. This method adaptively adjusts the emphasis on different mechanical parameters, enabling the wear prediction model to maintain optimal performance and providing a reliable basis for clinical decision-making.

[0115] To address the accuracy and reliability issues of constructing the state space of the reinforcement learning model, in some embodiments, step 301: constructing the state space of the pre-trained reinforcement learning model, wherein the state space includes: the frequency band overlap ratio, the local deviation distribution, and the gradient direction consistency coefficient, including:

[0116] Step 401: According to the historical motion cycle of the dynamic load joint, a reference sequence is extracted from the frequency band overlap ratio corresponding to the historical motion cycle.

[0117] In step 401, the historical motion cycle refers to a past motion cycle that is earlier than the current cycle and is not continuously included, which may be data from hours, days or longer ago. The current motion cycle refers to the latest complete motion cycle of the joint being analyzed or processed. Multiple consecutive motion cycles refer to multiple motion cycles that are continuous and without intervals in time, usually the current cycle and several adjacent cycles before it (such as the previous 5 cycles). The logical relationship between the three: time range: historical cycle ≫ multiple consecutive cycles include the current cycle. Data usage: historical cycles are used for long-term regularity mining, multiple consecutive cycles are used for short-term dynamic adjustment, and the current cycle is used for real-time decision-making. The reference sequence refers to the typical change pattern of the frequency band overlap ratio extracted from historical normal motion data, reflecting the mechanical-vibration coupling characteristics of the joint under standard working conditions.

[0118] In an embodiment of the present application, a large amount of motion cycle data is collected from healthy subjects or under normal working conditions, and a cluster analysis method is used to identify the typical change curve of the frequency band overlap ratio. After removing the outliers, the median value of each phase point is taken as the reference benchmark to form a standard reference sequence containing a complete motion cycle.

[0119] Step 402: Match the reference sequence with the frequency band overlap ratio corresponding to the current motion cycle to generate a local deviation distribution of the frequency band overlap ratio in a continuous time interval.

[0120] In an embodiment of the present application, a dynamic time warping algorithm is used to match the frequency band overlap ratio curve of the current period with the reference sequence point by point, calculate the relative deviation value of each matching point, and then obtain the deviation distribution histogram of different motion stages through sliding window statistics.

[0121] Step 403: Calculate the consistency coefficient between the change direction of the adjacent period fluctuation amplitude in the deviation sequence of the asymmetric stress distribution feature and the gradient direction of the deviation distribution.

[0122] In an embodiment of the present application, a time series vector of asymmetric stress deviation and a gradient vector of frequency band overlap deviation are first constructed, and then the cosine value of the angle between these two vectors in the phase space is calculated. Finally, a smoothed directional consistency coefficient curve is obtained by moving average processing.

[0123] Here are some specific examples:

[0124] In the AI ​​health monitoring system for intelligent artificial hip joints, when a patient transitions from ground walking to stair climbing, the system first retrieves a frequency band overlap reference sequence established during a historical normal walking cycle as a baseline. As the patient continues ascending and descending stairs, the system dynamically matches the vibration frequency band characteristics generated by the current stepping motion with the reference sequence in real time. Deviations during the stance phase are detected, generating a local deviation distribution over time. Analysis also reveals a gradual decrease in the coordination between the evolution direction of the asymmetric stress characteristics and the vibration characteristics. Based on these state characteristics, the reinforcement learning model identifies this as a new regular motion pattern rather than a temporary anomaly. Therefore, after an initial increase in the friction coefficient weight, as the deviation distribution continues to stabilize and the directional consistency remains at a new equilibrium state, the system ultimately confirms this as a stable characteristic pattern suitable for stair climbing. This new pattern is then added to the patient's personalized reference sequence library, providing a more comprehensive comparison benchmark for subsequent motion pattern recognition and wear prediction. The system also generates a weight allocation scheme tailored to this type of motion, enabling accurate and adaptive prediction in different activity scenarios.

[0125] In this application, a dynamic state assessment system based on historical benchmarks is established, enabling a reinforcement learning model to accurately identify abnormal evolution patterns in joint mechanical characteristics. This approach not only considers the static differences between current data and the standard reference, but also captures the dynamic coordination relationship between multiple parameter changes, providing a reliable state perception foundation for adaptive decision-making in intelligent joint systems.

[0126] In order to solve the stability problem of the weight distribution strategy in the dynamic load joint wear prediction, in some embodiments, step 303: dynamically smoothing the contribution coefficients of the asymmetric stress distribution characteristics and the instantaneous load impact amplitude in the composite mechanical characterization parameters in the dynamic correction strategy over multiple consecutive motion cycles includes:

[0127] Step 501: extracting, in the plurality of consecutive motion cycles, a first contribution coefficient corresponding to the asymmetric stress distribution feature and a second contribution coefficient corresponding to the instantaneous load impact amplitude in each motion cycle.

[0128] In step 501, the first contribution coefficient is the weight of the influence of the asymmetric stress distribution characteristics on the composite mechanical characterization parameters, reflecting the contribution of the uneven stress on the joint contact surface to wear. The second contribution coefficient is the weight of the influence of the instantaneous load impact amplitude on the composite mechanical characterization parameters, representing the intensity of the sudden mechanical impact on material damage.

[0129] In an embodiment of the present application, the weight component corresponding to the asymmetric stress feature is separated from the original weight matrix output by the reinforcement learning model as the first contribution coefficient, and the weight component related to the impact amplitude is extracted as the second contribution coefficient, forming two independent weight sequences according to the time sequence of the motion cycle.

[0130] Step 502: For each motion cycle, calculate a first change in the first contribution coefficient and a second change in the second contribution coefficient between the current cycle and the previous cycle.

[0131] In step 502 , the first variation and the second variation respectively represent the relative variation amplitudes of the contribution coefficients between adjacent motion cycles.

[0132] In the embodiment of the present application, the first-order difference method is used to calculate the absolute difference between the contribution coefficient of the current period and the previous period, and then the coefficient value of the previous period is divided to obtain the normalized percentage of change to eliminate the influence of different parameter dimensions.

[0133] Step 503: Determine the maximum allowable amplitude of the first variation and the maximum allowable amplitude of the second variation according to the first contribution coefficient sequence and the second contribution coefficient sequence in the plurality of consecutive motion cycles.

[0134] In step 503, the maximum allowable amplitude refers to the upper limit of the reasonable range of contribution coefficient variation derived from historical data statistics. The first contribution coefficient sequence is a chronologically ordered set of numerical values ​​representing the weights of the influence of asymmetric stress distribution characteristics on the composite mechanical characterization parameters over multiple consecutive motion cycles. The second contribution coefficient sequence is a chronologically ordered set of numerical values ​​representing the weights of the influence of instantaneous load impact amplitude on the composite mechanical characterization parameters over the same time range.

[0135] In the embodiment of the present application, a Gaussian distribution is fitted to the contribution coefficient change sequence of several past motion cycles, and the upper limit of the confidence interval is taken as the maximum allowable amplitude. At the same time, a minimum guarantee threshold is set to prevent excessive constraints.

[0136] Step 504: If the first variation of the current cycle exceeds the maximum allowable amplitude of the first variation, the first contribution coefficient of the current cycle is adjusted to the superposition value of the first contribution coefficient of the previous cycle and the maximum allowable amplitude.

[0137] In step 504, the superimposed value refers to a new coefficient value obtained by increasing or decreasing the maximum allowable amplitude based on the contribution coefficient of the previous cycle.

[0138] In the embodiment of the present application, the superposition method is determined according to the direction of change: if the current coefficient increases beyond the limit, the previous cycle value plus the maximum amplitude is used; if it decreases beyond the limit, the previous cycle value minus the maximum amplitude is used, maintaining the change trend but limiting the amplitude.

[0139] Step 505: If the second variation of the current cycle exceeds the maximum allowable amplitude of the second variation, the second contribution coefficient of the current cycle is adjusted to the superposition value of the second contribution coefficient of the previous cycle and the maximum allowable amplitude.

[0140] In the embodiment of the present application, the superposition method is determined according to the direction of change: if the current coefficient increases beyond the limit, the previous cycle value plus the maximum amplitude is used; if it decreases beyond the limit, the previous cycle value minus the maximum amplitude is used, maintaining the change trend but limiting the amplitude.

[0141] Step 506: Using the adjusted first contribution coefficient and the second contribution coefficient as the contribution coefficients after dynamic smoothing constraints in the current period.

[0142] In step 506 , the contribution coefficient after dynamic smoothing constraint is the final weight value after stability processing.

[0143] In an embodiment of the present application, the adjusted first and second contribution coefficients are renormalized to ensure that the sum of the coefficients remains unchanged, and the weight output module of the reinforcement learning model is updated at the same time.

[0144] Here's a specific example:

[0145] In the AI ​​health monitoring system for intelligent artificial hip joints, when a patient transitions from daily walking to continuous stair climbing training, the system first records the raw contribution coefficients of asymmetric stress signatures and step impact during multiple consecutive stair climbing cycles. As training progresses, the system detects that the stress signature coefficient suddenly increases significantly in certain cycles, exceeding the reasonable variation range calculated based on historical data, while the step impact coefficient remains stable. At this point, the system automatically activates a dynamic smoothing mechanism: the suddenly changing stress signature coefficient is adjusted to the coefficient value of the previous cycle plus a preset maximum allowable variation, while the step impact coefficient remains unchanged. After multiple cycles of adaptive adjustment, the system confirms that this step-by-step increase in stress signature weights is an inherent characteristic of stair climbing, rather than an abnormal fluctuation. Ultimately, an optimized weighting scheme is developed that both reflects the characteristics of the new movement pattern and maintains predictive stability. The stress signature weight is moderately increased but changes gradually, while the impact signature weight is fine-tuned. This weighting combination, adapted for stair climbing, is then integrated into the patient's personalized model. It can be quickly recalled when similar movement patterns are detected again, providing a more reliable basis for assessing rehabilitation progress.

[0146] In the present embodiment, a dynamic smoothing constraint mechanism for the contribution coefficient is established to effectively suppress abnormal fluctuations in weight distribution while retaining the adaptive capabilities of the reinforcement learning model. This method can not only capture changes in feature weights caused by changes in movement patterns, but also ensure the stability of prediction results, making clinical decisions both timely and reliable.

[0147] To address the issue of insufficient integration of dynamic load joint mechanical characteristics and friction characteristics, in some embodiments, step 104: generating composite mechanical characterization parameters based on the three-dimensional mechanical fluctuation trajectory data and the target gradient data, includes:

[0148] Step 601: Obtain the friction coefficient change rate corresponding to the time stamp in the target gradient data.

[0149] In step 601, the friction coefficient change rate refers to the rate at which the friction coefficient changes with the angular velocity of movement per unit time, reflecting the dynamic characteristics of the lubrication state of the joint contact surface.

[0150] In an embodiment of the present application, the difference in friction coefficients between adjacent sampling points in the target gradient data is calculated by numerical differentiation, and then divided by the corresponding time interval to obtain the instantaneous rate of change, and a sliding average filter is used to eliminate measurement noise.

[0151] Step 602: Filter out the fluctuation trajectory data segments that coincide with the acquisition time of the target gradient data from the three-dimensional mechanical fluctuation trajectory data, wherein the fluctuation trajectory data segments include the instantaneous load impact amplitude corresponding to each timestamp.

[0152] In step 602, the acquisition moment specifically refers to the time at which the 3D mechanical wave trajectory data acquisition device actually records the data. This moment must coincide with or be synchronized with the "corresponding timestamp" of the target gradient data. The corresponding timestamp is the precise time stamp recorded for each data point during the continuous acquisition of the target gradient data (friction coefficient gradient data). The relationship between the two is that the "corresponding timestamp" is used to locate the critical moment of friction coefficient change, and then the data segment with the same "acquisition moment" is locked in the mechanical data. The wave trajectory data segment is a slice of mechanical data that is strictly synchronized with the friction data.

[0153] In an embodiment of the present application, a hardware clock synchronization signal is used to align the two types of data acquisition moments, and a mechanical data segment matching the friction sampling point timestamp is extracted from the three-dimensional mechanical data, including the impact amplitude, duration and spatial vector.

[0154] Step 603: Mark the energy release direction of the instantaneous load impact amplitude according to the friction coefficient change rate.

[0155] In step 603, the energy release direction refers to the main transmission direction of the load impact energy on the joint contact surface.

[0156] In the embodiment of the present application, the energy input / output state is determined based on the positive and negative signs of the friction coefficient change rate, and the three-dimensional direction cosines of the energy release are calculated in combination with the spatial vector direction of the impact amplitude.

[0157] Step 604: spatially correlate and map the energy release direction and the deviation of the three-dimensional mechanical fluctuation trajectory data in the direction of the normal vector of the joint rotation plane to generate an asymmetric stress distribution feature.

[0158] In step 604, the deviation is obtained by calculating the average length of the projection of the wave trajectory data segment in the normal vector direction. Spatial correlation mapping refers to establishing a topological relationship between mechanical direction characteristics and joint geometric properties.

[0159] In an embodiment of the present application, the energy release direction vector is projected into the normal space of the joint rotation plane through coordinate transformation, the deviation angle and distance are calculated, and an asymmetric stress characteristic map containing regional distribution and intensity information is generated.

[0160] Step 605: Based on the target gradient data, perform weighted fusion on the instantaneous load impact amplitude and the asymmetric stress distribution characteristics to generate composite mechanical characterization parameters.

[0161] In step 605 , weighted fusion refers to adaptively integrating mechanical features based on the importance of friction gradients.

[0162] In the embodiment of the present application, the weighting coefficient is determined according to the overall fluctuation amplitude of the target gradient data, the impact amplitude characteristics and the stress distribution characteristics are subjected to principal component analysis and linear combination, and a low-dimensional composite parameter vector is output.

[0163] Here's a specific example:

[0164] In the AI ​​health monitoring system for intelligent artificial hip joints, when a patient performs stair climbing training, the system first obtains the rapid change characteristics of the friction coefficient of the lateral side of the hip joint during the stepping action in real time, and simultaneously extracts the impact signal generated by the heel strike in the three-dimensional mechanical data at the corresponding moment. By analyzing the changing trend of the friction coefficient, the system determines that the impact energy is mainly released to the posterosuperior area of ​​the acetabular cup, and spatially compares this energy release direction with the normal rotation plane of the prosthesis, and finds a significant posterosuperior deviation feature. Based on this, the system dynamically weights and fuses this specific mechanical direction deviation pattern with the friction gradient amplitude to generate a comprehensive characteristic parameter that reflects the concentrated wear risk in the posterosuperior area. This composite parameter is then input into a prediction model with optimized weight distribution, accurately predicting the cumulative wear trend of the posterosuperior area of ​​the acetabular cup under continuous stair climbing training, providing a key basis for doctors to formulate stair training intensity and prosthesis position adjustment plans. At the same time, the characteristic parameters of this specific movement pattern are updated in the patient's personalized database to achieve intelligent prediction of similar activities in the future.

[0165] In this application, a multidimensional unified representation of joint dynamic load characteristics is achieved through a spatiotemporally aligned mechanics-friction feature fusion method. This composite parameter preserves the physical meaning of the original data while highlighting key wear influencing factors, providing high-information input features for subsequent prediction models.

[0166] In order to solve the problem of insufficient fusion of multi-source features in joint wear prediction, in some embodiments, step 105: generating a joint wear prediction result including a multi-dimensional wear accumulation trend based on the nonlinear superposition result of the adjusted composite mechanical characterization parameter and the friction coefficient mutation feature, includes:

[0167] Step 701: inputting the adjusted composite mechanical characterization parameters and the friction coefficient mutation characteristics into a pre-constructed joint influence relationship structure.

[0168] In step 701, the joint influence relationship structure refers to a topological network that characterizes the interaction between the composite mechanical parameters and the friction mutation characteristics, and includes multiple sets of parameter interval division rules and corresponding superposition weight distribution methods.

[0169] In an embodiment of the present application, a neural network structure including a multi-layer perceptron is constructed, wherein the input layer receives composite parameters and mutation features, the hidden layer stores the nonlinear relationship between parameters under different motion states, and the output layer generates a preliminary superposition result.

[0170] Step 702: Determine the superposition priority of each mechanical component in the adjusted composite mechanical characterization parameter according to the change direction of the friction coefficient mutation characteristic in the continuous motion cycle.

[0171] In step 702 , the superposition priority is used to characterize the response strength of different mechanical components to the friction coefficient mutation feature.

[0172] In the embodiment of the present application, by analyzing the gradient change trend of the friction mutation characteristics, the mutual information between each mechanical component and the mutation characteristics is calculated, and the superposition weight coefficient is dynamically allocated according to the correlation strength.

[0173] Step 703: Based on the superposition priority, perform segmented superposition calculation on the adjusted composite mechanical characterization parameters and the friction coefficient mutation characteristics to obtain the cumulative wear amount in different superposition directions in each motion cycle.

[0174] In step 703, segmented superposition calculation refers to a feature fusion method based on the phase of the motion cycle. The cumulative wear amount is the wear increment calculated through nonlinear superposition in a specific superposition direction (such as axial, radial, or tangential) within a single motion cycle, reflecting the instantaneous damage accumulation in that direction.

[0175] In an embodiment of the present application, a single motion cycle is divided into multiple phase intervals, and a different superposition function (such as a hyperbolic tangent function for the support phase and an S-type function for the swing phase) is used to calculate the wear increment in each interval.

[0176] Step 704: Divide the accumulated wear amount into multiple independent dimensions according to the superposition direction.

[0177] In step 704 , the independent dimensions refer to different development directions of wear in the joint space.

[0178] In the embodiment of the present application, the joint is divided into three basic dimensions: axial, radial and tangential according to the anatomical structure, and the cumulative wear in each direction is counted separately.

[0179] Step 705: Generate a trend prediction curve for each independent dimension based on the direction of change of the cumulative amount of each independent dimension in the continuous movement cycle.

[0180] In step 705 , the trend prediction curve refers to a dynamic process reflecting the evolution of wear.

[0181] In the embodiment of the present application, a time series prediction algorithm is used to fit the cumulative amount of each dimension to generate a prediction curve containing short-term fluctuations and long-term trends.

[0182] Step 706: Combine the trend prediction curves of the multiple independent dimensions according to the preset wear stage division conditions, and output the joint wear prediction result.

[0183] In step 706 , the wear stage division is set based on the fatigue characteristics of the material, including an initial wear period, a stable wear period, and an accelerated wear period.

[0184] In the embodiment of the present application, three stages are determined according to the curve of the joint material: the initial wear period, the stable wear period and the accelerated wear period, and the curves of each dimension are segmented and combined according to the critical points.

[0185] Here's a specific example:

[0186] In an AI-powered health monitoring system for intelligent artificial hip joints, as patients continue stair climbing rehabilitation training, the system feeds optimized composite mechanical parameters (including enhanced lateral shear characteristics) and friction mutation characteristics detected during stepping into a pre-trained joint analysis network. The system first prioritizes the lateral mechanical component based on the increasing trend of the friction mutation characteristics during successive stepping cycles. It then calculates the cumulative wear of the lateral region of the acetabular cup during each movement cycle, segmented by phase (e.g., pedal contact phase and propulsion phase) of the stair climbing motion. This wear is spatially decomposed into three independent dimensions: axial compression, lateral shear, and anteversion torque. The system then generates curves for the evolution of each dimension over future training cycles. The results show that the wear rate in the lateral shear dimension is significantly higher than in the other dimensions. Finally, the system compares these curves with clinical wear stage standards and outputs a prediction that the lateral region of the acetabular cup will enter the accelerated wear phase first during continuous stair climbing training. It then recommends adjusting training intensity to avoid the risk of premature liner failure, providing intelligent decision support for personalized rehabilitation programs.

[0187] In this application's examples, by establishing nonlinear coupling relationships between parameters and conducting multidimensional trend analysis, we achieve accurate prediction of joint wear from microscopic characteristics to macroscopic evolution. This method not only reflects the spatial distribution of wear but also captures the temporal evolution of wear across different movement phases, providing comprehensive decision-making support for personalized rehabilitation programs.

[0188] Figure 2 A schematic diagram of the structure of a joint wear prediction system based on artificial intelligence provided in an embodiment of the present application is shown in FIG. Figure 2 As shown, the system includes:

[0189] The acquisition module 21 is used to acquire the three-dimensional mechanical fluctuation trajectory data and target gradient data of the dynamic load joint during the motion cycle of the dynamic load joint. The target gradient data is the gradient data of the friction coefficient of the dynamic load joint changing with the motion angular velocity.

[0190] The output module 22 is used to output a weight distribution result for correcting the joint wear prediction path based on the three-dimensional mechanical fluctuation trajectory data and the target gradient data using a pre-trained reinforcement learning model.

[0191] The extraction module 23 is used to extract the friction coefficient mutation feature from the target gradient data when the dynamic load joint meets the preset joint material fatigue critical condition.

[0192] The adjustment module 24 is configured to adjust the composite mechanical characterization parameters according to the weight distribution result to obtain adjusted composite mechanical characterization parameters, wherein the composite mechanical characterization parameters are generated according to the three-dimensional mechanical fluctuation trajectory data and the target gradient data.

[0193] The generation module 25 is used to generate a joint wear prediction result including a multi-dimensional wear accumulation trend according to the nonlinear superposition result of the adjusted composite mechanical characterization parameters and the friction coefficient mutation characteristics.

[0194] Figure 2 The artificial intelligence-based joint wear prediction system can perform Figure 1 The implementation principles and technical effects of the artificial intelligence-based joint wear prediction method described in the illustrated embodiment are not further elaborated. The specific manner in which the various modules and units perform operations in the artificial intelligence-based joint wear prediction system described in the aforementioned embodiment have been described in detail in the related embodiments of the method and will not be further elaborated here.

[0195] In one possible design, Figure 2 An artificial intelligence-based joint wear prediction system of the illustrated embodiment can be implemented as a computing device, such as Figure 3 As shown, the computing device may include a storage component 31 and a processing component 32;

[0196] The storage component 31 stores one or more computer instructions, wherein the one or more computer instructions are called and executed by the processing component 32 .

[0197] The processing component 32 is as follows Figure 1 The embodiment provides a joint wear prediction method based on artificial intelligence.

[0198] The processing component 32 may include one or more processors to execute computer instructions to complete all or part of the steps in the above method. Of course, the processing component may also be implemented as one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors, or other electronic components to perform the above method.

[0199] The storage component 31 is configured to store various types of data to support operations on the terminal. The storage component can be implemented by any type of volatile or non-volatile memory device, or a combination thereof, such as static random-access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk, or optical disk.

[0200] Of course, a computing device may also include other components, such as input / output interfaces, display components, communication components, etc.

[0201] The input / output interface provides an interface between the processing component and the peripheral interface module, which can be an output device, an input device, etc.

[0202] The communication component is configured to facilitate, among other things, wired or wireless communications between the computing device and other devices.

[0203] Among them, the computing device can be a physical device or an elastic computing host provided by a cloud computing platform, etc. In this case, the computing device can refer to a cloud server, and the above-mentioned processing components, storage components, etc. can be basic server resources rented or purchased from the cloud computing platform.

[0204] The present application also provides a computer storage medium storing a computer program, wherein the computer program can achieve the above-mentioned Figure 1 The illustrated embodiment is an artificial intelligence-based joint wear prediction method.

[0205] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0206] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, i.e., they may be located in one location or distributed across multiple network units. Some or all of the modules may be selected based on actual needs to achieve the objectives of the present embodiment. Persons of ordinary skill in the art will be able to understand and implement the present invention without inventive effort.

[0207] Through the above description of the embodiments, those skilled in the art will clearly understand that each embodiment can be implemented using software plus a necessary general-purpose hardware platform, or of course, hardware. Based on this understanding, the essence of the above technical solution, or the portion that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a magnetic disk, or an optical disk, and includes a number of instructions for causing a computer device (such as a personal computer, server, or network device) to execute the methods described in each embodiment or certain portions of the embodiments.

[0208] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A joint wear prediction method based on artificial intelligence, characterized in that: include: During a motion cycle of the dynamic load joint, three-dimensional mechanical fluctuation trajectory data and target gradient data of the dynamic load joint are obtained, wherein the target gradient data is gradient data of a friction coefficient of the dynamic load joint changing with an angular velocity of motion; Based on the three-dimensional mechanical fluctuation trajectory data and the target gradient data, a pre-trained reinforcement learning model is used to output a weight distribution result for correcting the joint wear prediction path; Extracting friction coefficient mutation characteristics from the target gradient data when the dynamically loaded joint meets a preset joint material fatigue critical condition; Adjusting a composite mechanical characterization parameter according to the weight distribution result to obtain an adjusted composite mechanical characterization parameter, wherein the composite mechanical characterization parameter is generated according to the three-dimensional mechanical fluctuation trajectory data and the target gradient data; According to the nonlinear superposition result of the adjusted composite mechanical characterization parameters and the friction coefficient mutation characteristics, a joint wear prediction result including a multi-dimensional wear accumulation trend is generated.

2. The method according to claim 1, characterized in that The method of outputting a weight distribution result for correcting the joint wear prediction path using a pre-trained reinforcement learning model based on the three-dimensional mechanical fluctuation trajectory data and the target gradient data includes: Calculating, based on the asymmetric stress distribution characteristics in the target gradient data, the frequency band overlap ratio between the instantaneous load impact amplitude and the energy distribution interval in the three-dimensional mechanical fluctuation trajectory data, where the energy distribution interval is the energy distribution interval of the mechanical vibration signal generated by the dynamically loaded joint during the current motion cycle; Combined with the frequency band overlap ratio, a pre-trained reinforcement learning model is used to output a weight distribution result.

3. The method according to claim 2, characterized in that The method of combining the frequency band overlap ratio and using a pre-trained reinforcement learning model to output a weight distribution result includes: Constructing a state space of a pre-trained reinforcement learning model, wherein the state space includes: frequency band overlap ratio, local deviation distribution, and gradient direction consistency coefficient; Inputting the state space into a pre-trained reinforcement learning model and outputting a dynamic correction strategy, wherein the action space of the reinforcement learning model includes a dynamic correction strategy for adjusting the contribution coefficient, and the reward function is generated based on the error between the joint wear prediction result and the actual detection value and the fluctuation range of the weight distribution; In a plurality of consecutive motion cycles, dynamically smoothing constraints are applied to the contribution coefficients of the asymmetric stress distribution characteristics and the instantaneous load impact amplitude in the composite mechanical characterization parameters in the dynamic correction strategy; The contribution coefficient after dynamic smoothing constraint is used as the weight distribution result.

4. The method according to claim 3, characterized in that The state space of the pre-trained reinforcement learning model is constructed, and the state space includes: the frequency band overlap ratio, the local deviation distribution and the gradient direction consistency coefficient, including: According to the historical motion cycle of the dynamically loaded joint, a reference sequence is extracted from the frequency band overlap ratio corresponding to the historical motion cycle; Matching the reference sequence with the frequency band overlap ratio corresponding to the current motion cycle to generate a local deviation distribution of the frequency band overlap ratio in a continuous time interval; The consistency coefficient between the change direction of the adjacent period fluctuation amplitude in the deviation sequence of the asymmetric stress distribution feature and the gradient direction of the deviation distribution is calculated.

5. The method according to claim 3, characterized in that The step of dynamically smoothing the contribution coefficients of the asymmetric stress distribution characteristics and the instantaneous load impact amplitude in the composite mechanical characterization parameters in the dynamic correction strategy over a plurality of consecutive motion cycles includes: In the plurality of consecutive motion cycles, respectively extracting a first contribution coefficient corresponding to the asymmetric stress distribution feature and a second contribution coefficient corresponding to the instantaneous load impact amplitude in each motion cycle; For each motion cycle, calculating a first change in the first contribution coefficient and a second change in the second contribution coefficient between a current cycle and an adjacent previous cycle; Determining a maximum allowable amplitude of the first variation and a maximum allowable amplitude of the second variation according to the first contribution coefficient sequence and the second contribution coefficient sequence in the plurality of consecutive motion cycles; If the first variation of the current period exceeds the maximum allowable amplitude of the first variation, adjusting the first contribution coefficient of the current period to the superposition value of the first contribution coefficient of the previous period and the maximum allowable amplitude; If the second variation of the current cycle exceeds the maximum allowable amplitude of the second variation, adjusting the second contribution coefficient of the current cycle to the superposition value of the second contribution coefficient of the previous cycle and the maximum allowable amplitude; The adjusted first contribution coefficient and the second contribution coefficient are respectively used as the contribution coefficients after dynamic smoothing constraints in the current period.

6. The method according to claim 1, characterized in that Generating composite mechanical characterization parameters according to the three-dimensional mechanical fluctuation trajectory data and the target gradient data includes: Obtaining a friction coefficient change rate corresponding to a timestamp in the target gradient data; Filtering out a fluctuation trajectory data segment that coincides with the acquisition time of the target gradient data from the three-dimensional mechanical fluctuation trajectory data, wherein the fluctuation trajectory data segment includes an instantaneous load impact amplitude corresponding to each time stamp; Marking the energy release direction of the instantaneous load impact amplitude according to the friction coefficient change rate; Performing spatial correlation mapping between the energy release direction and the deviation of the three-dimensional mechanical fluctuation trajectory data in the direction of the normal vector of the joint rotation plane to generate an asymmetric stress distribution feature; According to the target gradient data, the instantaneous load impact amplitude and the asymmetric stress distribution characteristics are weightedly fused to generate composite mechanical characterization parameters.

7. The method according to claim 1, characterized in that The method generates a joint wear prediction result including a multi-dimensional wear accumulation trend based on the nonlinear superposition result of the adjusted composite mechanical characterization parameter and the friction coefficient mutation characteristic, including: Inputting the adjusted composite mechanical characterization parameters and the friction coefficient mutation characteristics into a pre-constructed joint influence relationship structure; determining the superposition priority of each mechanical component in the adjusted composite mechanical characterization parameter according to the change direction of the friction coefficient mutation characteristic within the continuous motion cycle; Based on the superposition priority, the adjusted composite mechanical characterization parameters and the friction coefficient mutation characteristics are superimposed and calculated in sections to obtain the cumulative wear amount in different superposition directions within each motion cycle; Dividing the accumulated wear amount into multiple independent dimensions according to the superposition direction; Generating a trend prediction curve for each independent dimension based on the direction of change of the cumulative amount of each independent dimension in the continuous motion cycle; The trend prediction curves of the multiple independent dimensions are combined according to the preset wear stage division conditions to output the joint wear prediction result.

8. An artificial intelligence-based joint wear prediction system, characterized in that: include: an acquisition module, configured to acquire three-dimensional mechanical fluctuation trajectory data and target gradient data of the dynamic load joint during a motion cycle of the dynamic load joint, wherein the target gradient data is gradient data of a friction coefficient of the dynamic load joint changing with a motion angular velocity; an output module, configured to output a weight distribution result for correcting the joint wear prediction path using a pre-trained reinforcement learning model based on the three-dimensional mechanical fluctuation trajectory data and the target gradient data; An extraction module is used to extract friction coefficient mutation characteristics from the target gradient data when the dynamic load joint meets a preset joint material fatigue critical condition; an adjustment module, configured to adjust a composite mechanical characterization parameter according to the weight distribution result to obtain an adjusted composite mechanical characterization parameter, wherein the composite mechanical characterization parameter is generated according to the three-dimensional mechanical fluctuation trajectory data and the target gradient data; A generation module is used to generate a joint wear prediction result including a multi-dimensional wear accumulation trend based on the nonlinear superposition result of the adjusted composite mechanical characterization parameters and the friction coefficient mutation characteristics.

9. A computing device, characterized in that It includes a processing component and a storage component; the storage component stores one or more computer instructions; the one or more computer instructions are used to be called and executed by the processing component to implement an artificial intelligence-based joint wear prediction method as described in any one of claims 1 to 7.

10. A computer storage medium, characterized in that A computer program is stored, and when the computer program is executed by a computer, the method for predicting joint wear based on artificial intelligence as described in any one of claims 1 to 7 is implemented.

Images