Method for identifying and predicting high-rigidity collision mode in knee joint osteotomy

Abstract

The invention belongs to the field of robot control, and particularly relates to a method for identifying and predicting a high-rigidity collision mode in knee joint osteotomy, which comprises the following steps of: acquiring force data measured by a force sensor, and extracting force change characteristics by acquiring a derivative of the force data; estimating the force derivative data in real time, eliminating measurement noise, and calculating a residual error between a real force derivative and an estimated value; dynamically calculating a standard deviation and an average value of the residual error, and generating a dynamic detection threshold adaptive to different bone stiffness; a force change mode is divided into a collision state, a non-collision state and a transition state, a high-rigidity collision mode is accurately identified, and prediction of future collision, optimization of admittance control parameters and reduction of rebound caused by high-rigidity collision are realized by analyzing a residual trend; and finally, a double-window mechanism is designed to filter misjudgment and update the collision modal state in real time. The adaptability of a flexible bone cutting strategy is enhanced, the rebound effect is remarkably reduced, and the safety and accuracy of the bone cutting process are improved.

Description

Technical Field

[0001] The present invention belongs to the field of robot control, and specifically relates to a method for identifying and predicting high-stiffness collision modes during knee joint osteotomy. Background Art

[0002] In knee joint surgery, osteotomy is a common and crucial step, especially in knee joint replacement surgery. To ensure the accuracy and safety of the osteotomy process, existing technologies usually adopt a compliant osteotomy strategy based on admittance control. Admittance control can provide a certain degree of compliance for the osteotomy tool, enabling it to adapt to complex bone environments and effectively avoid harm to the patient's soft tissues. However, when faced with high-stiffness bone, existing admittance control strategies often produce significant rebound phenomena. The collision characteristics of high-stiffness bone are quite different from those of soft tissue contact. When the cutting tool collides with high-stiffness bone, the rebound effect can lead to inaccurate cutting and may even cause damage to the surgical tool or the patient's tissue. In actual operation, the force changes when the osteotomy tool contacts the bone are often complex, and the rapid force fluctuations and instantaneous collisions with high-stiffness substances increase the difficulty of identification, resulting in system response lag or misjudgment.

[0003] Most existing collision detection methods rely on simple threshold settings or relatively crude models to judge the collision state, lacking precise identification and prediction of collision modes. This makes it often impossible for the system to adjust cutting parameters in a timely manner in the case of high-stiffness bone collision, resulting in rebound phenomena. Therefore, how to accurately identify high-stiffness collision modes and effectively predict their trends has become a key technical problem in improving the cutting accuracy and safety of knee joint osteotomy surgery. Summary of the Invention

[0004] The purpose of the present invention is to provide an algorithm suitable for identifying and predicting high-stiffness collision modes during knee joint osteotomy, aiming to solve the problem of significant rebound caused by insufficient modal identification in existing compliant osteotomy strategies based on admittance control when encountering high-stiffness bone collision, thereby improving the accuracy and safety of osteotomy surgery.

[0005] The technical solution adopted by the present invention to achieve the above purpose is: a method for identifying and predicting high-stiffness collision modes during knee joint osteotomy, including the following steps:

[0006] S1: Data preprocessing: Collect the force data measured by the force sensor, and extract the force change characteristics by obtaining its derivative.

[0007] S2: Kalman filter estimation: Perform real-time estimation on the force derivative data, eliminate measurement noise, and calculate the residual between the true force derivative and the estimated value as a key indicator for collision detection.

[0008] S3: Adaptive Threshold Generation: Based on the statistical characteristics of the residuals within the sliding window, dynamically calculate the standard deviation and mean of the residuals to generate a dynamic detection threshold adapted to different bone stiffnesses;

[0009] S4: Collision Mode Identification and Prediction: Divide the force change pattern into collision, non - collision, and transition states, accurately identify the high - stiffness collision mode, and through analyzing the residual trend, achieve the prediction of future collisions, optimize the admittance control parameters, and reduce the rebound caused by high - stiffness collisions;

[0010] S5: Collision Mode Update: Design a dual - window mechanism to filter misjudgments and update the collision mode status in real - time.

[0011] The specific steps of step S1 are as follows:

[0012] S11: Force Data Acquisition: Real - time acquire the force data during the knee osteotomy process through a force sensor. The data format is a discrete time series, i.e.,

[0013] F(t1),F(t2),...,F(t n )

[0014] where F(t) represents the measured value of the force signal at time t;

[0015] S12: Force Data Convolution Smoothing Processing: Perform triangular convolution smoothing processing on the force data, and use a window of size w for filtering. Here, w is an odd number to ensure that the smoothing center point is the current time t;

[0016] For each time point t, calculate its smoothed value F smooth (t) as:

[0017]

[0018] where F(t) represents the original force data, i represents the distance between the current position and the window center position, and w - |i| is the triangular weight factor;

[0019] S13: Calculate the force derivative F′(t) using the discretized difference method, which represents the force change rate at the time step Δt, i.e.,

[0020]

[0021] The specific steps of step S2 are as follows:

[0022] S21: Initialize the Kalman filter: Set the initial state estimate value The initial error covariance matrix P 0∣0 , the state transition matrix A, the observation matrix H, the process noise covariance Q, and the measurement noise covariance R;

[0023] S22: Prediction stage: At each time step t, based on the estimated value predict the force derivative estimate at the current moment. Meanwhile, update the error covariance:

[0024]

[0025] S23: Update stage: After receiving the new observation F′(t), update the estimated value and the error covariance:

[0026]

[0027] P k∣k =(I - K k H)P k∣k-1

[0028] where K k is the Kalman gain, which adjusts the weight of the estimated value according to the current error covariance and the observation noise;

[0029] Calculate the residual, and the calculated residual value is used for subsequent collision detection and modal identification:

[0030]

[0031] The step S3 is specifically as follows:

[0032] S31: Calculate the statistical characteristics of the residual: Analyze the residual in real time through a sliding window, and calculate the average value and standard deviation of the residual. For each window w k , calculate its statistical characteristics:

[0033]

[0034] where |w k | is the window size, and r(i) is the residual at the i-th time point.

[0035] S32: Dynamically generate a threshold: Use the average value μ r and the standard deviation σ r of the residual, combined with a sensitivity number k, to dynamically generate a detection threshold: μ lim = μ r + k * σ r .

[0036] The step S4 is specifically as follows:

[0037] S41: According to the comparison between the residual and the threshold, judge the current collision mode. If the residual exceeds the threshold, it is determined to be in a collision state; otherwise, it is in a non-collision state:

[0038]

[0039] S42: Predict future collisions based on continuous residual values and their changing trends. Predict future residual values through linear extrapolation, i.e.:

[0040]

[0041] where r′(t) is the change rate of the residual;

[0042] S43: Update and optimize the collision state using the current residual and the prediction result.

[0043] The specific steps of step S5 are as follows:

[0044] The dual-window mechanism includes a collision detection window and a non-collision detection window to filter misjudgments;

[0045] S51: Define two detection windows, a short-term window and a long-term window;

[0046] Among them, the short-term window is used to detect instantaneous collisions, store recent force data, and monitor the force changes therein. If the change in force data exceeds the set threshold, a collision determination will be triggered;

[0047] The long-term window: is used to confirm and filter misjudgments, store force data within a set time range, and judge the trend of force changes; if the force changes in the non-collision detection window are stable for a long time, it is confirmed that no collision has occurred at that moment;

[0048] S52: Combine the results of the short-term and long-term windows and update the final collision mode state through a majority voting mechanism;

[0049] S53: Collect six-dimensional force data at the end of the robotic arm and perform collision mode identification and prediction verification according to the calibration parameters.

[0050] The calibration parameters include: Kalman filter parameters, adaptive threshold parameters, and collision window parameters;

[0051] Among them, for the Kalman filter parameters: the process noise covariance Q = 1e-5, and the measurement noise covariance R = 9e-3;

[0052] For the adaptive threshold parameters: the window size w = 71, and the sensitivity factor is 30;

[0053] For the collision window parameters: the collision window size collision_window_size = 7, and the non-collision window size non_collision_window_size = 14.

[0054] The present invention has the following beneficial effects and advantages:

[0055] 1. The present invention can effectively identify and predict the mode of high-stiffness bone collision during knee osteotomy, enhance the adaptability of the compliant osteotomy strategy, significantly reduce the rebound effect, improve the safety and precision of the osteotomy process, and has high clinical application value.

[0056] 2. Through residual dynamic threshold calculation, the present invention can adaptively adjust the collision recognition sensitivity to adapt to different bone stiffnesses; combined with residual trend analysis, it can predict high-stiffness collisions in advance and optimize admittance control parameters, thereby reducing transient impacts and improving the stability of osteotomy. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] Figure 1 is the flowchart of the method of the present invention;

[0058] Figure 2 is the osteotomy force data plus collision prediction result of the present invention;

[0059] Figure 3 is the comparison between the Kalman prediction of the osteotomy force derivative and the true residual of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0060] The present invention will be further described in detail below with reference to the drawings and embodiments.

[0061] As Figure 1 shown, it is the flowchart of the method of the present invention. The present invention proposes a method for identifying high-stiffness collision modes based on Kalman filtering and adaptive thresholds, aiming to solve the rebound problem caused by high-stiffness bone collisions in the admittance control strategy during knee osteotomy. The specific core steps are as follows:

[0062] S1: Data preprocessing: Collect the force data measured by the force sensor, calculate its derivative to extract the force change characteristics, and establish a basic data framework for collision mode identification to ensure the accuracy and real-time performance of subsequent calculations.

[0063] S2: Kalman filter estimation: Introduce Kalman filtering technology to perform real-time estimation on the force derivative data, filter out sensor noise, and obtain a more accurate force derivative estimation value. At the same time, calculate the residual between the true force derivative and the Kalman filter estimation value, and the residual is used as a key indicator for detecting high-stiffness collision modes.

[0064] S3: Adaptive threshold generation: Adopt a sliding window method to dynamically adjust the collision detection threshold based on the statistical characteristics (such as standard deviation and mean) of the residual data. By adaptively adjusting the threshold range, the detection sensitivity and accuracy of the algorithm in different bone stiffness environments are improved.

[0065] S4: Collision Mode Identification and Prediction: Using a continuous window detection mechanism, the force change pattern is divided into collision, non - collision, and transition states to accurately identify the high - stiffness collision mode. By analyzing the residual trend change, the future collision situation is predicted, and the admittance control parameters (such as stiffness, damping, etc.) are optimized to effectively reduce the rebound phenomenon caused by high - stiffness collisions.

[0066] S5: Collision Mode Update: Set up dedicated collision detection windows and non - collision detection windows. Through the dual - window mechanism, misjudgments are filtered, and the collision state information is updated in real time to ensure the stability and robustness of the detection results.

[0067] Example 1:

[0068] Combined with the embodiments of the present invention Figure 1 , the algorithm for identifying and predicting high - stiffness collision modes during knee osteotomy will be described in detail and clearly. The specific steps are as follows:

[0069] Step 1: Data pre - processing

[0070] Collect force data: Real - time collect the force data during knee osteotomy through a force sensor. The data format is a discrete time series.

[0071] F(t1), F(t2),..., F(t n )

[0072] where F(t) represents the measured value of the force signal at time t.

[0073] Convolution smoothing process for force data: Through triangular convolution smoothing, the force data is made smoother to reduce the influence of sudden noise in a short time, thereby providing a more accurate input signal for subsequent derivative calculation, Kalman filtering, and collision detection. A window of size w is used for filtering, where w is an odd number to ensure that the smoothing center point is the current time t.

[0074] For each time point t, calculate its smoothed value F smooth (t)

[0075]

[0076] where F(t) represents the original force data, and i represents the distance between the current position and the window center position. The triangular weight factor w - |i| makes the weight at the window center the largest and the weight at the edge positions smaller.

[0077] To capture the change trend of the force, the discrete difference method is used to calculate the force derivative F′(t), which represents the force change rate at the time step Δt. The specific calculation formula is as follows:

[0078]

[0079] Step 2: Kalman filter estimation

[0080] Use Kalman filtering technology to remove measurement noise, accurately estimate the force derivative, and calculate the residual value as an important basis for collision detection.

[0081] Initialize the Kalman filter: Set the initial state estimate Initial error covariance matrix P 0∣0 , state transition matrix A, observation matrix H, process noise covariance Q, and measurement noise covariance R. These parameters can be adjusted according to experimental data.

[0082] Prediction stage: At each time step t, predict the force derivative estimate at the current moment based on the estimate of the previous step. At the same time, update the error covariance::

[0083]

[0084] Update stage: After receiving the new observation value F′(t), update the estimate and error covariance::

[0085] K k = P k∣k-1 H T (HP k∣k-1 H T + R) -1

[0086]

[0087] P k∣k =(I - K k H)P k∣k-1

[0088] where K k is the Kalman gain, which adjusts the weight of the estimate according to the current error covariance and observation noise.

[0089] Calculate the residual. The calculated residual value is used for subsequent collision detection and modal identification::

[0090]

[0091] Step 3: Adaptive threshold generation

[0092] Dynamically adjust the threshold according to real-time data to adapt to the collision characteristics of different bone stiffnesses and improve the sensitivity and accuracy of detection. Calculate the statistical characteristics of the residual: Perform real-time analysis of the residual through a sliding window and calculate the average value and standard deviation of the residual. For each window w k , calculate its statistical characteristics::

[0093]

[0094] where |w k | is the window size and r(i) is the residual at the i-th time point.

[0095] Dynamically generate a threshold: Use the mean μ r and standard deviation σ r of the residuals, combined with a sensitivity number k, to dynamically generate a detection threshold that adapts to different bone stiffness conditions and ensures sensitive identification during high-stiffness collisions

[0096] μ lim = μ r + k * σ r

[0097] Step 4: Collision mode identification and prediction

[0098] Based on the comparison between the residual and the threshold, determine the current collision mode and predict the future collision trend.

[0099]

[0100] If the residual exceeds the threshold, it is determined to be in a collision state; otherwise, it is in a non-collision state.

[0101] Through continuous residual values and their trend of change, predict future collisions. Use simple linear extrapolation to predict future residual values:

[0102]

[0103] where r′(t) is the rate of change of the residual. The predicted value can be used as a basis for adjusting operation parameters. Use the current residual and the prediction result to update and optimize the collision state:

[0104] Step 5: Collision mode update

[0105] Filter misjudgments through a dual-window mechanism to ensure the stability and robustness of the collision mode state. Windows: Define two detection windows, a short-term window and a long-term window: Short-term window: Used for quickly detecting instantaneous collisions. This window is used to store recent force data and strictly monitor the force changes therein. If the change in force data exceeds a certain threshold, a collision determination will be triggered. Long-term window: Used to confirm and filter misjudgments and avoid overly frequent state switches. This window is used to store force data within a longer time range and judge the trend of force changes. If the force change in the non-collision detection window is stable over a long time, it can be confirmed that no collision has occurred at that moment. Mode update and stability: Combine the results of the short-term and long-term windows and update the final collision mode state through a "majority voting" mechanism.

[0106] Collect the six-dimensional force data at the end of the robotic arm for algorithm verification. Kalman filter parameters: process noise covariance Q = 1e-5, measurement noise covariance R = 9e-3. Adaptive threshold parameters: window size w = 71, sensitivity factor 30. Collision window parameters: collision window size collision_window_size = 7, non-collision window size non_collision_window_size = 14. The prediction results are as shown in Figure 2 and the residual estimation is as shown in Figure 3 . The time series analysis of the binary output results in the figure (1 represents a collision event, 0 represents normal operating force) shows that the collision prediction algorithm proposed in this study achieves accurate discrimination between human-machine interaction force and sudden collision force during the dynamic response process.

[0107] The present invention proposes an innovative high-stiffness collision mode identification and prediction algorithm. By real-time monitoring the change of force, combining residual analysis and dynamic threshold adjustment technology, it can accurately identify the characteristics of high-stiffness collisions, and optimize control parameters through trend prediction, effectively avoiding the rebound phenomenon and improving the robustness and safety of the bone cutting process.

[0108] Those skilled in the art can understand that the above are only the preferred embodiments of the present invention. The features described in each embodiment and / or claim of the present disclosure can be combined or combined in various ways, even if such combinations or combinations are not explicitly described in the present disclosure. It is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements for some of the technical features. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

[0109] Although the preferred embodiments of the present invention have been described, those skilled in the art can make additional changes and modifications once they learn the basic creative concepts. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications falling within the scope of the present invention. Obviously, those skilled in the art can make various changes and variations to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalent technologies, the present invention is also intended to include these modifications and variations.

Claims

1. A method for identifying and predicting high-stiffness collision modes during knee joint osteotomy, characterized in that, It includes the following steps: S1: Data preprocessing: Collect the force data measured by the force sensor, and extract the force change characteristics by obtaining its derivative; S2: Kalman filter estimation: Perform real-time estimation on the force derivative data, eliminate measurement noise, and calculate the residual between the true force derivative and the estimated value as the key index for collision detection; S3: Adaptive threshold generation: Based on the statistical characteristics of the residuals within the sliding window, dynamically calculate the standard deviation and mean of the residuals to generate a dynamic detection threshold adapted to different bone stiffnesses; S4: Collision mode identification and prediction: Divide the force change patterns into collision, non-collision, and transition states, accurately identify the high-stiffness collision mode, and through analyzing the residual trend, realize the prediction of future collisions, optimize the admittance control parameters, and reduce the rebound caused by high-stiffness collisions; S5: Collision mode update: Design to filter misjudgments through a dual-window mechanism to update the collision mode status in real time.

2. The high-rigidity collision mode identification and prediction method during knee joint osteotomy according to claim 1, wherein, The specific content of step S1 is as follows: S11: Collect force data: Real-time collect the force data during the knee osteotomy process through the force sensor. The data format is a discrete time series, that is: F(t1), F(t2),..., F(t n ) where F(t) represents the measured value of the force signal at time t; S12: Triangular convolution smoothing process for force data: Perform triangular convolution smoothing on the force data, and use a window of size w for filtering, where w is an odd number to ensure that the smoothing center point is the current time t; For each time point t, calculate its smoothed value F smooth (t) as follows: where F(t) represents the original force data, i represents the distance between the current position and the window center position, and w - |i| is the triangular weight factor; S13: Calculate the force derivative F′(t) using the discretized difference method, which represents the force change rate at the time step Δt, that is:

3. A method for identifying and predicting high-stiffness collision modes during knee joint osteotomy according to claim 1, characterized in that, The specific content of step S2 is as follows: S21: Initialize the Kalman filter: Set the initial state estimate value The initial error covariance matrix P 0∣0 , the state transition matrix A, the observation matrix H, the process noise covariance Q, and the measurement noise covariance R; S22: Prediction stage: At each time step t, based on the estimated value predict the force derivative estimate at the current moment. Meanwhile, update the error covariance: S23: Update stage: After receiving the new observation value F′(t), update the estimated value and the error covariance: K k = P k∣k-1 H T (HP k∣k-1 H T + R) -1 P k∣k = (I - K k H)P k∣k-1 Among them, K k is the Kalman gain, which adjusts the weight of the estimated value according to the current error covariance and the observation noise; Calculate the residual, and the calculated residual value is used for subsequent collision detection and mode identification:

4. A method for identifying and predicting high-stiffness collision modes in knee osteotomy according to claim 1, characterized in that, The specific content of step S3 is as follows: S31: Calculate the statistical characteristics of the residuals: Perform real-time analysis on the residuals through a sliding window, calculate the mean and standard deviation of the residuals, and for each window w k , calculate its statistical characteristics: where |w k | is the window size and r(i) is the residual at the i-th time point. S32: Dynamically generate a threshold: Use the average value μ of the residuals r and the standard deviation σ r , combined with a sensitivity number k, to dynamically generate a detection threshold: μ lim = μ r + k * σ r .

5. A method for identifying and predicting high-stiffness collision modes in knee osteotomy according to claim 1, characterized in that, The specific content of step S4 is as follows: S41: According to the comparison between the residual and the threshold, judge the current collision mode. If the residual exceeds the threshold, it is determined to be in the collision state, otherwise it is in the non-collision state: S42: Through continuous residual values and their change trends, predict future collisions, and predict the future residual values through linear extrapolation, that is: where r′(t) is the change rate of the residual; S43: Use the current residual and the prediction result to update and optimize the collision state.

6. A method for identifying and predicting high-stiffness collision modes during knee joint osteotomy according to claim 1, characterized in that, The specific content of step S5 is as follows: The dual-window mechanism includes: a collision detection window and a non-collision detection window to filter misjudgments; S51: Define two detection windows, a short-term window and a long-term window; Among them, the short-term window is used to detect instantaneous collisions, store recent force data, and monitor the force changes therein. If the force data changes exceed the set threshold, a collision determination will be triggered; The long-term window: is used to confirm and filter misjudgments, store the force data within a set time range, and judge the trend of force changes; if the force changes in the non-collision detection window are stable for a long time, it is confirmed that no collision has occurred at that moment; S52: Update the final collision mode state through a majority voting mechanism by combining the results of short-term and long-term windows; S53: Collect six-dimensional force data at the end of the robotic arm, and perform collision mode identification and prediction verification according to the verification parameters.

7. A method for identifying and predicting high-stiffness collision modes during knee joint osteotomy according to claim 1, characterized in that The verification parameters include: Kalman filter parameters, adaptive threshold parameters, and collision window parameters; Among them, for the Kalman filter parameters: the process noise covariance Q = 1e-5, and the measurement noise covariance R = 9e-3; For the adaptive threshold parameters: the window size w = 71, and the sensitivity factor 30; For the collision window parameters: the collision window size collision_window_size = 7, and the non-collision window size non_collision_window_size = 14.

Images