Interactive design method and device for ankle-foot orthosis based on digital twinning

Abstract

The invention belongs to the technical field of ankle joint rehabilitation, and relates to an interactive design method and device for an ankle foot orthosis based on digital twinning, and the method comprises the following steps: obtaining structural data of an ankle foot and the ankle foot orthosis, and building a three-dimensional model; acquiring plantar real-time pressure distribution data, and mapping the plantar real-time pressure distribution data to the three-dimensional model to form a digital twinborn model containing dynamic behavior information; plantar pressure features are extracted from the digital twinborn model, a gait evaluation model and a gait anomaly detection model are constructed by using the plantar pressure features, the gait evaluation model is used for reflecting the stability and pressure distribution condition of gaits, and the gait anomaly detection model is used for detecting abnormal gaits in real time; when the gait evaluation model is lower than a preset threshold value or the gait anomaly detection model detects gait anomaly, feedback information which is sent to the ankle-foot orthosis and used for adjusting parameters of the ankle-foot orthosis is generated. According to the ankle foot orthosis, the ankle foot orthosis can better adapt to individual differences of users, and personalized rehabilitation support is provided for the users.

Description

Technical Field

[0001] The present invention belongs to the technical field of ankle rehabilitation, and particularly relates to an interactive design method and device for an ankle-foot orthosis based on digital twin. Background Art

[0002] Ankle injury is a common health problem, associated with weakened strength, muscle dysfunction, spasm or limited range of motion. Rehabilitation is required to improve the user's quality of life and motor function. An ankle-foot orthosis is a common medical device used to correct the ankle-foot area, provide support, improve gait and promote rehabilitation. According to different design principles, the ankle-foot orthoses currently on the market are mainly divided into three categories: passive, quasi-passive and active. Among them, a passive ankle-foot orthosis refers to an ankle-foot orthosis that can generate assistance without any external power source, and usually relies on passive components such as springs, dampers or elastic bands to provide auxiliary functions. Its main advantages are light weight and low price. However, in most cases, the design parameters of passive components can only be optimized under some special circumstances. A quasi-passive ankle-foot orthosis uses pressure sensors and low-power small motors to adjust the stiffness or damping of passive components to improve the functional efficiency of the ankle-foot orthosis, which enhances the adaptability of the passive ankle-foot orthosis to gait or environmental changes. Limited by the limited adaptability and control ability of passive components, passive components cannot achieve certain motion control and assistance conditions, so control strategies such as rigid transmission devices, cable drives and pneumatic muscles are used for the active control of ankle-foot orthoses. The main advantage of an active ankle-foot orthosis is that it can provide a large amount of assistance with completely controllable amplitude and time, enabling it to adapt to different speeds or ground conditions, but its efficiency depends on the performance of the controller and actuator.

[0003] Although there have been considerable improvements in the design of existing ankle-foot orthoses, since they are designed to help users recover, the ability to assist walking is often given top priority. Their structure and function mainly focus on supporting and restricting the movement of the ankle joint to assist or correct abnormalities in the ankle-foot area. Most passive and quasi-passive ankle-foot orthoses can only perform dorsiflexion and plantarflexion, which may limit the natural gait of users, making the user's walking less natural and smooth. At the same time, in order to reduce costs, an ankle-foot orthosis is provided to a large number of users, and only simple adjustments are made to the height and tightness of the ankle-foot orthosis, resulting in poor sole fit of the ankle-foot orthosis and difficulty in fully adapting to the individual needs of each user, with a poor user experience. Summary of the Invention

[0004] In view of this, the present invention provides an interactive design method and device for an ankle-foot orthosis based on digital twin, which can improve the sole fit of the ankle-foot orthosis, enable the ankle-foot orthosis to better adapt to the individual differences and personalized needs of users, and has good comfort for long-term use by users, with good correction effect and user experience.

[0005] The technical solution of the present invention is as follows:

[0006] An interactive design method for an ankle-foot orthosis based on digital twin, comprising the following steps:

[0007] Obtain the structural data of the ankle-foot and the ankle-foot orthosis;

[0008] Based on the structural data of the ankle-foot and the ankle-foot orthosis, establish a three-dimensional model;

[0009] Obtain the real-time plantar pressure distribution data and map it onto the three-dimensional model to form a digital twin model containing dynamic behavior information;

[0010] Extract the plantar pressure features in the digital twin model, and use the plantar pressure features to construct a gait evaluation model y and a gait abnormality detection model p c , the gait evaluation model y is used to reflect the stability and pressure distribution of the gait, and the gait abnormality detection model p c is used to detect abnormal gaits in real time;

[0011] When the gait evaluation model y is lower than the preset threshold or the gait abnormality detection model detects an abnormal gait, generate feedback information sent to the ankle-foot orthosis for adjusting the parameters of the ankle-foot orthosis.

[0012] Preferably, based on the structural data of the ankle-foot and the ankle-foot orthosis, establishing a three-dimensional model includes the following steps:

[0013] Obtain the point cloud data of the ankle-foot and the point cloud data of the ankle-foot orthosis on the three-dimensional scanner;

[0014] Based on the point cloud data of the ankle-foot and the point cloud data of the ankle-foot orthosis, perform three-dimensional reconstruction to obtain a three-dimensional model.

[0015] Preferably, obtaining the real-time plantar pressure distribution data and mapping it onto the three-dimensional model to form a digital twin model containing dynamic behavior information includes the following steps:

[0016] Preprocess the real-time plantar pressure distribution data;

[0017] Map the preprocessed real-time plantar pressure distribution data onto the three-dimensional model to form a digital twin model.

[0018] Preferably, the preprocessing includes the following steps:

[0019] Use the Kalman filtering method to denoise the real-time plantar pressure distribution data;

[0020] Perform standardization processing on the denoised real-time plantar pressure distribution data.

[0021] Preferably, the plantar pressure characteristics include the pressure center point (x c (t), y c (t)), the maximum pressure value P max (t), the pressure distribution area A(t), the gait cycle pressure change rate ΔP(t), and the pressure distribution variance

[0022] Preferably, a gait evaluation model y and a gait abnormality detection model p are constructed using the plantar pressure characteristics c , including the following steps:

[0023] The plantar pressure characteristics are formed into a vector X,

[0024] The vector X is preprocessed to obtain the plantar pressure feature vector Z;

[0025] A neural network is constructed. The neural network includes an input layer, a first hidden layer, a second hidden layer, and an output layer. The input layer is used to input the plantar pressure feature vector Z. The first hidden layer is a fully connected layer with 128 neurons, and the activation function is the rectified linear unit ReLU. The calculation formula is: h1 = ReLU(W1Z + b1), where W1 is a weight matrix with 128 rows and 6 columns, b1 is a bias vector with 128 rows and 1 column, and h1 is the output of the first hidden layer of the neural network model. The second hidden layer is a fully connected layer with 64 neurons, and the activation function is the rectified linear unit ReLU. The calculation formula is: h2 = ReLU(W2h1 + b2), where W2 is a weight matrix with 64 rows and 128 columns, b2 is a bias vector with 64 rows and 1 column, h1 is the output of the first hidden layer of the neural network model, and h2 is the output of the second hidden layer of the neural network model. When the output layer is used for gait evaluation, the number of neurons in the output layer is 1, and the activation function is linear. The gait evaluation model y = W0h2 + b0, where y is the gait evaluation model, representing the evaluation score situation, W0 is the weight matrix, h2 is the output of the second hidden layer of the neural network model, and b0 is the bias matrix. When the output layer is used for gait abnormality detection, the number of neurons in the output layer is 2, and the activation function is Softmax. The gait abnormality detection model p c = Softmax(W c h2 + b c ), where p c is the gait abnormality detection model, representing the gait detection result, W c is the weight matrix, h2 is the output of the second hidden layer of the neural network model, and b c is the bias vector, and Softmax is the activation function of the output layer of the neural network model.

[0026] Preferably, the feedback information is determined based on the following formula

[0027]

[0028] Among them, score(t) is the anomaly level at time t, α and β are adjustment coefficients, P base is the maximum pressure reference value, A base is the pressure distribution area reference value, P max (t) is the maximum pressure value, and A(t) is the pressure distribution area.

[0029] An electronic device, including a memory, a processor, and a computer program stored on the memory and capable of running on the processor, where the processor executes the computer program to implement the digital-twin-based interactive design method for ankle-foot orthoses as described above.

[0030] A computer-readable storage medium, on which a computer program is stored, and the computer program is executed by a processor to implement the digital-twin-based interactive design method for ankle-foot orthoses as described above.

[0031] Compared with the prior art, a digital-twin-based interactive design method and device for ankle-foot orthoses provided by the present invention realizes real-time collection, analysis, and feedback of user foot data by integrating pressure sensor technology, digital twin technology, and human-computer interaction design methods, can improve the sole fitting degree of the ankle-foot orthosis, enable the ankle-foot orthosis to better adapt to the individual differences of users, provide personalized rehabilitation support for users, have good comfort even when used by users for a long time, have good correction effects and user experiences, and thus improve the functionality and user experience of the ankle-foot orthosis. Specifically, the solution of the present invention precisely matches the individual differences of users through fine modeling, optimizes the design of the ankle-foot orthosis, makes full use of the real-time monitoring and feedback functions of digital twin technology, meets the personalized rehabilitation needs, improves the user experience, has strong practicability, and is worthy of promotion. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] Figure 1 is the main flowchart of the present invention.

[0033] Figure 2 is the sub-flowchart Figure 1 .

[0034] Figure 3 is the sub-flowchart Figure 2 .

[0035] Figure 4 is the sub-flowchart Figure 3 .

[0036] Figure 5 is the layout diagram of the pressure sensors of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0037] At present, ankle-foot orthoses on the market are mainly divided into three categories: passive, quasi-passive, and active. Among them, passive ankle-foot orthoses refer to those that can generate assistance without any external power source, usually relying on passive components such as springs, dampers, or elastic bands to provide auxiliary functions. Its main advantages are light weight and low price. However, in most cases, the design parameters of passive components can only be optimized under some special circumstances. Quasi-passive ankle-foot orthoses use pressure sensors and small low-energy motors to adjust the stiffness or damping of passive components to improve the functional efficiency of ankle-foot orthoses, which enhances the adaptability of passive ankle-foot orthoses to gait or environmental changes. Limited by the limited adaptability and control ability of passive components, passive components cannot achieve certain motion control and assistance conditions, so control strategies such as rigid transmission devices, cable drives, and pneumatic muscles are adopted for the active control of ankle-foot orthoses. The main advantage of active ankle-foot orthoses is that they can provide a large amount of assistance with completely controllable amplitude and time, enabling them to adapt to different speeds or ground conditions, but their efficiency depends on the performance of the controller and actuator.

[0038] Although there have been quite significant improvements in the design of existing ankle-foot orthoses, since they are designed to help users recover, the ability to assist walking is often given top priority. Their structure and function mainly focus on supporting and restricting the movement of the ankle joint to assist or correct abnormalities in the ankle and foot areas. Most passive and quasi-passive ankle-foot orthoses can only perform dorsiflexion and plantarflexion, which may limit the natural gait of users, making the user's walking less natural and smooth. At the same time, in order to reduce costs, one type of ankle-foot orthosis is provided to a large number of users, and only simple adjustments are made to the height and tightness of the ankle-foot orthosis, resulting in poor fitting of the ankle-foot orthosis and difficulty in fully adapting to the personalized needs of each user. When users use it for a long time, they feel uncomfortable or the correction effect is not good, and the user experience is poor.

[0039] In order to solve the technical problems that the current ankle-foot orthoses will limit the natural gait of users, making it difficult for users to walk naturally and smoothly, the fitting and comfort of the ankle-foot orthoses are poor, it is difficult to fully adapt to the personalized needs of each user, users feel uncomfortable or the correction effect is not good when using it for a long time, and the user experience is poor, the technical solution of this application is specifically designed.

[0040] In order to enable those skilled in the art to better understand and implement the technical solution of the present invention, the present invention will be further described below in conjunction with specific embodiments and drawings.

[0041] The present invention provides an interactive design method and device for an ankle-foot orthosis based on digital twin. The following combines Figures 1 to 5 schematic diagrams to illustrate the present invention.

[0042] Embodiment 1

[0043] AsFigure 1 As shown in the flowchart, an interactive design method and device for an ankle-foot orthosis based on digital twin provided by the present invention include the following steps:

[0044] Obtain the structural data of the ankle-foot and the ankle-foot orthosis.

[0045] Based on the structural data of the ankle-foot and the ankle-foot orthosis, establish a three-dimensional model.

[0046] Obtain the real-time plantar pressure distribution data and map it onto the three-dimensional model to form a digital twin model containing dynamic behavior information.

[0047] Extract the plantar pressure features in the digital twin model, and use the plantar pressure features to construct a gait evaluation model y and a gait abnormality detection model p c , where the gait evaluation model y is used to reflect the stability and pressure distribution of the gait, and the gait abnormality detection model p c is used to detect abnormal gaits in real time. When the gait evaluation model y is lower than a preset threshold or the gait abnormality detection model detects an abnormal gait, feedback information for adjusting the parameters of the ankle-foot orthosis is generated and sent to the ankle-foot orthosis.

[0048] In the interactive design of the ankle-foot orthosis, each component has a physical entity and its corresponding virtual model. The physical entity is the actual hardware part of the component, and the virtual model is a digital mirror established through pressure sensors, data acquisition, and modeling tools, which consists of a geometric model and a dynamic behavior model and is used to simulate, analyze, and adjust the performance of the barefoot ankle-foot orthosis.

[0049] Combined with Figures 1 to 4 As shown in the flowchart, the specific implementation methods for the key steps of implementing the interactive design method of the ankle-foot orthosis are as follows:

[0050] 1. Obtain the structural data of the user's ankle-foot and the ankle-foot orthosis

[0051] Specifically, the structural data of the user's ankle-foot and the ankle-foot orthosis can be collected by scanning the user's ankle-foot and the ankle-foot orthosis respectively with a 3D scanner.

[0052] 2. Based on the structural data of the ankle-foot and the ankle-foot orthosis, establish a three-dimensional model

[0053] Obtain the point cloud data of the ankle-foot on the 3D scanner and the point cloud data of the ankle-foot orthosis as the original data. Then, import the point cloud data of the ankle-foot and the ankle-foot orthosis into Rhino software, and perform 3D reconstruction with the aid of Rhino software to generate the geometric models of the ankle-foot and the ankle-foot orthosis. The geometric models of the ankle-foot and the ankle-foot orthosis are high-precision 3D models created according to the actual size and structure of the ankle-foot and the ankle-foot orthosis, and are precisely corresponding to the physical entities in terms of shape and size.

[0054] 3. Obtain the real-time plantar pressure distribution data and map it onto the 3D model to form a digital twin model containing dynamic behavior information

[0055] Use multiple pressure sensors arranged on the plantar surface to collect the real-time plantar pressure distribution data of the plantar surface under different gaits respectively.

[0056] The pressure sensors used are flexible pressure sensors. The flexible pressure sensors in a rows and b columns are arranged as shown, and the distribution of the pressure data of the pressure sensors at time t is represented as a vector F(t). Figure 5 As shown, where i takes integer values between 1 and a, j takes integer values between 1 and b, a and b are the number of rows and columns of the pressure sensors respectively. Since the sensing range and measurement accuracy of each model of flexible pressure sensor are different, a and b are determined based on the model of the flexible pressure sensor adopted.

[0057]

[0058] Among them, i takes integer values between 1 and a, j takes integer values between 1 and b, a and b are the number of rows and columns of the pressure sensors respectively. Since the sensing range and measurement accuracy of each model of flexible pressure sensor are different, a and b are determined based on the model of the flexible pressure sensor adopted.

[0059] To facilitate data processing using the Kalman filter, convert F(t) into a matrix with a*b rows and 1 column, denoted as vector P(t).

[0060]

[0061] Among them, p i,j (t) represents the plantar pressure value of the pressure sensor in the i-th row and j-th column at time t.

[0062] Set the state equation of the state space model as P(t) = A·P(t - 1) + w(t).

[0063] Among them, P(t) is the plantar pressure distribution vector at time t, P(t - 1) is the plantar pressure distribution vector at time t - 1, A is the state transition matrix, which is the identity matrix, and w(t) is a Gaussian process noise with a mean of 0 and its covariance is Q.

[0064] The measurement equation of the state space model is z(t) = H·P(t) + v(t).

[0065] Among them, z(t) is the noisy measurement value at time t, that is, the real-time plantar pressure distribution data collected by the pressure sensor, P(t) is the plantar pressure distribution vector at time t, H is the measurement matrix, which is the identity matrix, and v(t) is Gaussian measurement noise with a mean of 0 and a covariance of R.

[0066] In order to eliminate noise, preprocessing is required for the real-time plantar pressure distribution data. The preprocessing of the real-time plantar pressure distribution data mainly includes denoising and normalization.

[0067] Among them, denoising uses the Kalman filtering method, and the filtering process is as follows;

[0068] (1) Initialization:

[0069] X(0|0) = I, Q = 0.1·I, R = 0.5·I,

[0070] Among them, the number of elements in P′(0|0), X(0|0), Q, and R is all a*b, and I is the identity matrix.

[0071] (2) Prediction:

[0072] 1) Prediction of plantar pressure state:

[0073] P′(t|t - 1) = A·P′(t - 1|t - 1),

[0074] Among them, P′(t|t - 1) is the predicted state at time t, A is the state transition matrix, which is the identity matrix, and P′(t - 1|t - 1) is the pressure state estimate at time t - 1, that is, the estimated plantar pressure distribution after denoising at time t - 1.

[0075] 2) Prediction of error covariance:

[0076] X(t|t - 1) = A·X(t - 1|t - 1)·A T +Q,

[0077] Among them, X(t|t - 1) is the predicted covariance matrix, A is the state transition matrix, which is the identity matrix, X(t - 1|t - 1) is the covariance matrix at the updated time t - 1, and Q is the covariance matrix of the process noise.

[0078] (3) Update:

[0079] 1) Calculation of Kalman gain:

[0080] K(t) = X(t|t - 1)·H T ·[H·X(t|t - 1)·H T +R] -1 ,

[0081] Among them, K(t) is the Kalman gain, reflecting the influence of the measurement value on the state estimation, R is the covariance matrix of the measurement noise, H is the measurement matrix, and X(t|t - 1) is the predicted covariance matrix.

[0082] 2) Plantar pressure state update:

[0083] P′(t|t) = P′(t|t - 1) + K(t)·[z(t) - H·P′(t|t - 1)],

[0084] Among them, z(t) is the noisy measurement value, that is, the real - time plantar pressure distribution data collected by the pressure sensor, P′(t|t) is the updated state estimation, that is, the denoised plantar pressure distribution estimation, P′(t|t - 1) is the pressure prediction state at time t, H is the measurement matrix, and K(t) is the Kalman gain.

[0085] 3) Error covariance update:

[0086] X(t|t) = (I - K(t)·H)·X(t|t - 1),

[0087] Among them, X(t|t) is the covariance matrix at the updated time t, I is the identity matrix, H is the measurement matrix, K(t) is the Kalman gain, and X(t|t - 1) is the predicted covariance matrix.

[0088] Among them, the standardization process is as follows:

[0089]

[0090] Among them, is the standardized pressure value of the pressure sensor at the i - th row and j - th column at time t, p′ i,j (t|t) is the pressure value of the pressure sensor at the i - th row and j - th column after denoising at time t, P max (t) is the maximum value of the pressures of all pressure sensors after denoising at time t, P min (t) is the minimum value of the pressures of all pressure sensors after denoising at time t.

[0091] P max (t) and P min (t) are calculated as follows:

[0092] P max (t) = max{P′(t|t)},

[0093] P min (t) = min{P′(t|t)}.

[0094] To map the pressure sensor data into the digital twin model, a layout identical to that of the physical pressure sensor is established, and linear mapping is used to directly project onto the corresponding points of the digital twin model, thereby enabling the digital twin model to receive and process the pressure sensor data and reflect the dynamic behaviors of each component in real time, such as the pressure change rate during the gait cycle, pressure distribution, etc.

[0095] The physical entity of the pressure sensor and its corresponding virtual model constitute a digital twin unit. Through the interaction between the physical entity and the virtual model, the digital twin unit can achieve intelligent feedback and adaptive adjustment of the barefoot ankle-foot orthosis.

[0096] 4. Extract plantar pressure features in the digital twin model and construct a gait evaluation model y and a gait abnormality detection model p using the plantar pressure features c , where the gait evaluation model y is used to reflect the stability and pressure distribution of the gait, and the gait abnormality detection model p c is used to detect abnormal gaits in real time. When the gait evaluation model y is lower than a preset threshold or the gait abnormality detection model detects an abnormal gait, feedback information for adjusting the parameters of the ankle-foot orthosis is generated and sent to the ankle-foot orthosis, and wearing suggestions are synchronized on the mobile terminal.

[0097] Extracting plantar pressure features mainly includes the pressure center point (x c (t), y c (t)), the maximum pressure value P max (t), the pressure distribution area A(t), the pressure change rate ΔP(t) during the gait cycle, and the pressure distribution variance The calculation methods of each feature are as follows:

[0098] (1) Pressure center point (x c (t), y c (t)):

[0099] After denoising, the position of the pressure sensor at the i-th row and j-th column on the plantar pressure sensor array is (x i , y j ), and the pressure value at time t is Then the calculation method of the pressure center point (x c (t), y c (t)) is:

[0100]

[0101] where is the normalized pressure value of the pressure sensor at the i-th row and j-th column at time t, y j is the column coordinate of the pressure sensor at the i-th row and j-th column, that is, the position in the horizontal direction, x iis the row coordinate of the pressure sensor at the \(i\)-th row and \(j\)-th column, that is, the position in the vertical direction.

[0102] (2) Maximum pressure value \(P\) max (t):

[0103] Maximum pressure value

[0104] Among them, represents the plantar pressure value after denoising at time \(t\).

[0105] (3) Pressure distribution area \(A(t)\):

[0106] The pressure distribution area \(A(t)\) is used to analyze the distribution of the pressure concentration area. Given the plantar pressure threshold \(P\) threshhold , calculate the area \(A(t)\) of the region exceeding this threshold:

[0107]

[0108] Among them, is the plantar pressure at the \(i\)-th row and \(j\)-th column after denoising at time \(t\), \(\alpha\) i,j is the area covered by each pressure sensor, \(\Pi()\) is the indicator function, and it takes 1 when and 0 otherwise.

[0109] (4) Gait cycle pressure change rate \(\Delta P(t)\):

[0110] The gait cycle pressure change rate \(\Delta P(t)\) represents the pressure change between two time points. The time point when the heel of the left or right foot first touches the ground detected by the pressure sensor is the starting point \(t\) of the gait cycle s , and the time point when the heel touches the ground again is the ending point \(t\) of the gait cycle e , then the gait cycle is \(\Delta t = t\) e - \(t\) s , and the calculation method of the pressure change rate within the gait cycle is

[0111] Among them,

[0112]

[0113] \(P\) s (t e ) is the total pressure on the sole at time \(t\) e ,

[0114] \(P\) s (t s ) is the total pressure on the sole at time \(t\) s .

[0115] (5) Variance of pressure distribution

[0116] Variance of pressure distribution Evaluate the dispersion degree of the distribution of pressure values at various parts of the sole of the foot to identify pressure concentration areas and dispersion conditions.

[0117] Among them,

[0118]

[0119] is the average pressure at time t, is the plantar pressure at the i-th row and j-th column after denoising at time t.

[0120] Use the above-mentioned extracted plantar pressure features to form a vector X, and the vector X is expressed as:

[0121]

[0122] Preprocess the vector X, including feature Z-score standardization to eliminate the dimension difference. After preprocessing, the plantar pressure feature vector Z is obtained. The i-th element in the plantar pressure feature vector Z is represented by z[i], and the preprocessing formula of z[i] is:

[0123]

[0124] Among them, X[i] represents the i-th element in the vector X. In the given time period S, μ[i] and σ[i] respectively represent the mean and standard deviation of the i-th plantar pressure feature, and their calculation formulas are as follows:

[0125]

[0126] Among them, j is an integer between 0 and S.

[0127] Neural network architecture design, including an input layer, a first hidden layer, a second hidden layer, and an output layer.

[0128] Among them, the input layer only transmits data, and the input is the preprocessed plantar pressure feature vector Z. The input layer includes 6 neurons, corresponding to 6 plantar pressure features respectively.

[0129] The first hidden layer is a fully connected layer with 128 neurons, and the activation function is the rectified linear unit ReLU. The calculation formula is: h1 = ReLU(W1Z + b1), where W1 is a weight matrix with 128 rows and 6 columns, b1 is a bias vector with 128 rows and 1 column, and h1 is the output of the first hidden layer of the neural network model.

[0130] The second hidden layer is a fully connected layer with 64 neurons. The activation function is the same as that of the first hidden layer, which is also ReLU. The calculation formula is: h2 = ReLU(W2h1 + b2), where W2 is a weight matrix with 64 rows and 128 columns, b2 is a bias vector with 64 rows and 1 column, h1 is the output of the first hidden layer of the neural network model, and h2 is the output of the second hidden layer of the neural network model.

[0131] The output layer is used for gait evaluation and anomaly detection. When used for gait evaluation, the number of neurons in the output layer is 1, and the activation function is linear. The calculation formula for the gait evaluation model y is: y = W0h2 + b0, where W0 is a weight matrix with 1 row and 64 columns, b0 is a bias matrix with 1 row and 1 column, h2 is the output of the second hidden layer of the neural network model, and y is the gait evaluation model, representing the evaluation score.

[0132] For gait anomaly detection, the number of neurons in the output layer is 2, and the activation function is Softmax. The calculation formula for the gait anomaly detection model p c is p c = Softmax(W c h2 + b c ), where W c is a weight matrix with 2 rows and 64 columns, b c is a bias vector with 2 rows and 1 column, h2 is the output of the second hidden layer of the neural network model, and p c is the gait anomaly detection model, representing the gait detection result.

[0133] Model training: For gait evaluation, the Huber loss is used as the loss function to quantify the prediction error and guide the update of model parameters. The formula for the Huber loss is:

[0134]

[0135] where L e is the loss function for gait evaluation, y is the gait evaluation model, representing the evaluation score, which ranges from 0 to 100, is the predicted value of the gait model, and δ is the threshold parameter that controls the sensitivity to outliers, with a value of 1.

[0136] For gait anomaly detection, the weighted cross-entropy loss is used as the loss function to guide the update of model parameters. The formula is:

[0137]

[0138] where L cla is the loss function for gait anomaly detection, C is the total number of categories, which is 2 in this case, divided into two situations: abnormal and normal. p Ti is the true label of the sample. Here, p Ti∈ {0, 1}. p ci is the probability of class i predicted by the model, w i is the weight of the i-th class, N toal is the total number of samples, N i is the number of samples of the i-th class.

[0139] Before real-time detection, first collect the plantar pressure distribution data of this user for 30 minutes, and slice the collected data according to the minimum time unit supported by the pressure sensor model used to form the model training data set. Then, extract 70% of the data from the model training data set to train the above-built neural network model. After training, extract 30% of the data from the model training data set to test the above-built neural network model. After the test results meet the requirements, input the real-time collected plantar pressure distribution data into the neural network model for gait evaluation and abnormal gait detection. In gait evaluation, when the predicted value is less than 70, start the feedback mechanism.

[0140] The gait evaluation and abnormal detection results are used to guide the adjustment of the pressure distribution of the ankle-foot orthosis. The above-measured plantar pressure characteristics, gait evaluation results, and abnormal gait detection results are timely fed back to the software. When the gait evaluation is lower than 70 or there is an abnormal gait, the software feeds back to the user through vibration, ringing, etc., and gives suggestions for adjusting the ankle-foot orthosis, facilitating the user to make adjustments according to their actual needs.

[0141] The digital twin model has a real-time feedback function. When the gait evaluation value is lower than 70 or an abnormal gait is detected, it generates feedback information sent to the ankle-foot orthosis for adjusting the parameters of the ankle-foot orthosis and wearing suggestions synchronized on the mobile terminal, enabling the ankle-foot orthosis to increase the support in the low-pressure area and reduce the support in the high-pressure area to adjust the support of the ankle-foot orthosis and change the pressure distribution state.

[0142] The feedback information is used for abnormal level determination, thus providing a basis for the adjustment of the orthosis. The abnormal level score is calculated based on the maximum pressure P max (t) and the pressure distribution area A(t). The maximum pressure reference value P base and the pressure distribution area reference value A base are related to the user's individual. P base takes the average pressure of normal gait, and A base takes the effective contact area of the sole under normal gait. The abnormal level calculation uses a two-parameter combined scoring model, and the formula is as follows:

[0143]

[0144] Among them, score(t) is the anomaly level at time t, and α and β are adjustment coefficients. According to the clinical harmfulness of pressure overload, α and β are respectively set to 0.7 and 0.3 here. If score(t) < 0.5, it indicates that the range is normal and no adjustment is required; if 0.5 ≤ score(t) < 1.0, it indicates a first-level anomaly and mild adjustment is required; if score(t) ≥ 1.0, it indicates a second-level anomaly and severe adjustment is required.

[0145] After the user completes the support adjustment of the ankle-foot orthosis according to the feedback information and wearing suggestions, the gait at time t + 1 is recalculated according to the gait evaluation model and the current pressure distribution is evaluated to verify the adjustment effect. Continuously monitor the change in the pressure distribution after adjustment to form a closed-loop feedback, so that the ankle-foot orthosis can adaptively adjust the support and pressure distribution every time the gait changes.

[0146] The above-mentioned interactive design method of the ankle-foot orthosis based on digital twin can be implemented by means of an electronic device. Specifically, the electronic device includes a storage, a processor, and a computer program stored on the storage and capable of running on the processor. The processor executes the computer program to implement the above-mentioned interactive design method of the ankle-foot orthosis based on digital twin.

[0147] This embodiment also provides a computer-readable storage medium, which stores a computer program. The computer program is executed by the processor to implement the above-mentioned interactive design method of the ankle-foot orthosis based on digital twin.

[0148] Those of ordinary skill in the art can understand that all or part of the processes in the methods of the above embodiments can be completed by instructing relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above methods.

[0149] Among them, any reference to the memory, storage, database, or other media used in the various embodiments provided in this application can include at least one of non-volatile and volatile memories. The non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, or optical memory, etc. The volatile memory can include random access memory (RAM) or an external cache. By way of illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc.

[0150] Based on the modeling and simulation analysis results of the digital twin model, it can provide feedback on the actual usage of users, so as to intervene in a timely manner, adjust the working state of the ankle-foot orthosis, and provide personalized rehabilitation support for users. Through the mobile application, users can also view the working state of the ankle-foot orthosis in real time, adjust parameters, set personalized usage patterns, and obtain rehabilitation suggestions based on the modeling and simulation analysis results of the digital twin model, realizing user interaction and control.

[0151] An interactive design method and device for an ankle-foot orthosis based on digital twins provided by the present invention can enable the ankle-foot orthosis to better adapt to the individual differences of users, provide personalized rehabilitation support for users, have good comfort even for long-term use by users, have good correction effects and user experiences, thereby enhancing the functionality and user experience of the ankle-foot orthosis. Specifically, the solution of the present invention precisely matches the individual differences of users through fine modeling, optimizes the design of the ankle-foot orthosis, makes full use of the real-time monitoring and feedback functions of digital twin technology, meets personalized rehabilitation needs, improves user experiences, has strong practicability, and is worthy of promotion.

[0152] The above-disclosed are only the preferred specific embodiments of the present invention. However, the embodiments of the present invention are not limited thereto, and any changes that can be conceived by those skilled in the art should fall within the protection scope of the present invention.

Claims

1. An interactive design method for an ankle-foot orthosis based on digital twin, characterized in that, Including the following steps: Obtain the structural data of the ankle-foot and the ankle-foot orthosis; Based on the structural data of the ankle-foot and the ankle-foot orthosis, establish a three-dimensional model; Obtain the real-time plantar pressure distribution data and map it onto the three-dimensional model to form a digital twin model containing dynamic behavior information; Extract plantar pressure features in the digital twin model, and construct a gait evaluation model y and a gait abnormality detection model p using the plantar pressure features c , the gait evaluation model y is used to reflect the stability and pressure distribution of the gait, and the gait abnormality detection model p c is used to detect abnormal gaits in real time; When the gait evaluation model y is lower than a preset threshold or the gait abnormality detection model detects a gait abnormality, generate feedback information sent to the ankle-foot orthosis for adjusting the parameters of the ankle-foot orthosis.

2. The interactive design method of an ankle-foot orthosis based on digital twin according to claim 1, wherein, Based on the structural data of the ankle-foot and the ankle-foot orthosis, establish a three-dimensional model, including the following steps: Obtain the point cloud data of the ankle-foot and the point cloud data of the ankle-foot orthosis on the three-dimensional scanner; Perform three-dimensional reconstruction based on the point cloud data of the ankle-foot and the point cloud data of the ankle-foot orthosis to obtain a three-dimensional model.

3. The interactive design method of an ankle-foot orthosis based on digital twins according to claim 1, wherein, Obtain the real-time plantar pressure distribution data and map it onto the three-dimensional model to form a digital twin model containing dynamic behavior information, including the following steps: Preprocess the real-time plantar pressure distribution data; Map the preprocessed real-time plantar pressure distribution data onto the three-dimensional model to form a digital twin model.

4. The interactive design method of an ankle-foot orthosis based on digital twin according to claim 3, characterized in that The preprocessing includes the following steps: Denoise the real-time plantar pressure distribution data using the Kalman filtering method; Perform normalization processing on the denoised real-time plantar pressure distribution data.

5. The interactive design method of an ankle-foot orthosis based on digital twin according to claim 1, wherein The plantar pressure characteristics include the pressure center point (x c (t), y c (t)), the maximum pressure value P max (t), the pressure distribution area A(t), the gait cycle pressure change rate ΔP(t), and the pressure distribution variance 6. The interactive design method of an ankle-foot orthosis based on digital twin according to claim 5, characterized in that Construct a gait assessment model y and a gait abnormality detection model p using plantar pressure characteristics c , including the following steps: Construct the plantar pressure feature into a vector X, Preprocess the vector X to obtain the plantar pressure feature vector Z; Construct a neural network, which includes an input layer, a first hidden layer, a second hidden layer, and an output layer. The input layer is used to input the plantar pressure feature vector Z. The first hidden layer is a fully connected layer with 128 neurons, and the activation function is the rectified linear unit ReLU. The calculation formula is: h1 = ReLU(W1Z + b1), where W1 is a weight matrix with 128 rows and 6 columns, b1 is a bias vector with 128 rows and 1 column, and h1 is the output of the first hidden layer of the neural network model. The second hidden layer is a fully connected layer with 64 neurons, and the activation function is the rectified linear unit ReLU. The calculation formula is: h2 = ReLU(W2h1 + b2), where W2 is a weight matrix with 64 rows and 128 columns, b2 is a bias vector with 64 rows and 1 column, h1 is the output of the first hidden layer of the neural network model, and h2 is the output of the second hidden layer of the neural network model. When the output layer is used for gait evaluation, the number of neurons in the output layer is 1, and the activation function is linear. The gait evaluation model is y = W0h2 + b0, where y is the gait evaluation model, representing the evaluation score situation, W0 is the weight matrix, h2 is the output of the second hidden layer of the neural network model, and b0 is the bias matrix. When the output layer is used for gait abnormality detection, the number of neurons in the output layer is 2, and the activation function is Softmax. The gait abnormality detection model p c = Softmax(W c h2 + b c ), p c is the gait abnormality detection model, representing the gait detection result, W c is the weight matrix, h2 is the output of the second hidden layer of the neural network model, and b c is the bias vector, and Softmax is the activation function of the output layer of the neural network model.

7. The interactive design method of an ankle-foot orthosis based on digital twin according to claim 1, characterized in that The feedback information is determined based on the following formula Among them, score(t) is the anomaly level at time t, α and β are adjustment coefficients, P base is the maximum pressure reference value, A base is the pressure distribution area reference value, P max (t) is the maximum pressure value, and A(t) is the pressure distribution area.

8. An electronic device, characterized in that, including: A memory, a processor, and a computer program stored on the memory and capable of running on the processor. The processor executes the computer program to implement the digital twin-based interactive design method of the ankle-foot orthosis as claimed in claim 1.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, The computer program is executed by the processor to implement the digital twin-based interactive design method of the ankle-foot orthosis as claimed in claim 1.

Images