Pressure error compensation method for knee joint spreader

Abstract

The invention discloses a knee joint spreader pressure error compensation method, which comprises the steps of error compensation model construction: performing dynamic modeling on a knee joint spreader through finite element software ANSYS, performing test simulation on an established dynamic model to obtain a plurality of groups of real pressure values at the bottom of a motor, meanwhile, pressure values measured by pressure sensors at the bottoms of motors of a plurality of groups of corresponding knee joint distraction devices, namely measurement data, are obtained, the data serve as training data to train a BP neural network model to form an error compensation model, and model prediction is carried out: the knee joint distraction devices measure knee joint soft tissue pressure data in real time; inputting the measured knee joint soft tissue pressure data into an error compensation model, performing error compensation on the knee joint soft tissue pressure data by the error compensation model according to a previous training result, and combining a compensation value with the measured knee joint soft tissue pressure data to obtain an accurate pressure value; the problem of pressure measurement errors caused by the influence of the structure of the knee joint dilator is effectively solved.

Description

Technical Field

[0001] The invention relates to the technical field of pressure measurement, and in particular to a pressure error compensation method for a knee spreader. Background Art

[0002] During knee surgery, pressure measurement of soft tissues (such as ligaments, tendons, etc.) is crucial to the success of the surgery and postoperative recovery. Accurately measuring the pressure of the knee soft tissue not only helps to ensure the quality of the surgery and restore joint function, but also effectively avoids postoperative complications caused by excessive pressure. Currently, the pressure of the knee soft tissue is measured through the pressure sensor of the motor at the bottom of the knee spreader during surgery.

[0003] Theoretically, according to Newton's third law, the force measured by the pressure sensor of the motor at the bottom of the knee spreader should be equal to the force acting on the external actuator. However, due to the various mechanical parts between the external actuator and the motor, when the external actuator is under pressure, the knee spreader will generate a bending moment to drive the internal mechanical parts to shake, and the external actuator may also produce a slight deformation. In addition, the mechanical parts inside the knee spreader (such as flanges, motion mechanisms, etc.) will also generate friction, resulting in energy loss. For the above reasons, the force measured by the pressure sensor of the motor at the bottom of the knee spreader is not completely equal to the pressure applied by the external actuator. From the above analysis, it can be seen that even if the mechanical device of the knee spreader is more precise, due to the structural problems of the knee spreader itself, the measurement error of the knee spreader during surgery is inevitable. Therefore, if you want to accurately measure the pressure applied to the external actuator through the motor at the bottom of the knee spreader, you need to improve the measurement accuracy through error compensation.

[0004] At present, traditional error solutions include hardware improvement and static error compensation, etc. Among them, static error compensation measures include establishing a simple linear compensation model based on known sensor error characteristics to correct the measurement results. For example, Chinese patent CN117442396A discloses a data processing method for a knee spreader, which solves the measurement error problem caused by the soft tissue of the knee joint.

[0005] Although these methods can reduce the measurement error of knee spreaders to a certain extent, they are often unable to respond to the dynamic changes of complex surgical environments in real time, resulting in limitations in measurement accuracy and surgical outcomes. Summary of the invention

[0006] In order to solve the above technical problems, the present invention provides a knee spreader pressure error compensation method, which effectively solves the pressure measurement error problem caused by the influence of the knee spreader structure, thereby improving the accuracy and reliability of the measurement result.

[0007] The present invention provides a method for compensating pressure errors of a knee spreader, comprising: error compensation model construction: performing dynamic modeling of a knee spreader through finite element software ANSYS, testing and simulating the established dynamic model, obtaining multiple groups of real pressure values ​​at the bottom of a motor, and simultaneously obtaining multiple groups of corresponding pressure values ​​measured by pressure sensors at the bottom of a knee spreader motor, i.e., measurement data, and using the measurement data and the real pressure values ​​as training data to train a BP neural network model to form an error compensation model; model prediction: during surgery, a knee spreader measures the knee joint soft tissue pressure data in real time, and the measured knee joint soft tissue pressure data is input into the error compensation model, and the error compensation model automatically performs error compensation on the knee joint soft tissue pressure data according to previous training results, combines the compensation value with the real-time measured knee joint soft tissue pressure data, and outputs a corrected accurate pressure value.

[0008] Optionally, the test simulation of the dynamic model specifically includes: applying forces of different sizes on each grid of the dynamic model, i.e., the external actuator, the force range is from 0N to 100N, the interval of each force applied is 0.1N, and each force value is tested and simulated multiple times. Through the test simulation, the actual pressure value at the bottom of the motor is obtained.

[0009] Optionally, error compensation verification is performed after the error compensation model is constructed.

[0010] Optionally, the BP neural network model error compensation verification includes: directly collecting the pressure applied by the knee joint, i.e., the actual pressure value, through a high-precision pressure sensor, and inputting the measurement data corresponding to the actual pressure value into the BP neural network model, which captures the nonlinear characteristics of the knee joint soft tissue and the pattern of measurement errors, continuously optimizes the model weights and parameters, generates a predicted pressure value, and compares the generated predicted pressure value with the actual pressure value to verify the accuracy of the BP neural network model.

[0011] Optionally, the measured data and the actual pressure value are normalized respectively before using the measured data and the actual pressure value as training data to train the BP neural network model to form the error compensation model. The normalization formula is:

[0012]

[0013] Where μ i is the data to be normalized, μ min is the minimum value in the data to be normalized, μ max is the maximum value of the data to be normalized, x i is the normalized result.

[0014] Optionally, building a BP neural network model includes weight initialization. The formula for weight initialization is as follows:

[0015]

[0016] Where w is the connection weight, n in is the number of neural units in the previous layer, n out is the number of neural units in the next layer, and Var(w) is the variance of the weight.

[0017] Optionally, network regularization is performed during the BP neural network model training process, and the network regularization adopts a Dropout method to perform network regularization.

[0018] Optionally, the BP neural network model uses a linear rectification function as an activation function.

[0019] Optionally, the variance cost function is used as the loss function to calculate the error during the BP neural network model training process. The loss function expression is as follows:

[0020]

[0021] Where y is the expected output, a is the actual output of the neuron, and C is the function value.

[0022] The technical solution provided by the embodiment of the present invention has the following advantages compared with the prior art:

[0023] A method for compensating a pressure error of a knee spreader provided by an embodiment of the present invention solves the measurement error problem of the existing knee spreader during surgery by introducing an error compensation technology based on a BP neural network. The finite element software ANSYS mathematical modeling is used to accurately simulate the different pressure behaviors of the knee spreader, i.e., the external actuator, to obtain a large number of real pressure values ​​at the bottom of the motor, and then obtain the corresponding pressure value measured by the pressure sensor at the bottom of the knee spreader motor, i.e., the measurement data. A large number of real pressure values ​​and measurement data are transmitted to the BP neural network model for training. The task of the error compensation model is to adjust the measured pressure to the real pressure value. By establishing a BP neural network model, nonlinear relationships can be captured, and the problem of inaccurate pressure measurement caused by the structure of the knee spreader can be compensated in real time, ensuring that high-precision measurement results are obtained in a dynamic environment. Through the continuous self-learning mechanism of the neural network, the aging of the equipment and the wear and tear during surgery can be adapted to ensure the measurement accuracy of the knee spreader after long-term use. At the same time, the need for manual adjustment based on the doctor's experience during surgery is greatly reduced, thereby improving the efficiency and safety of surgery. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 A schematic diagram of the structure of a knee spreader provided by an embodiment of the present invention;

[0025] Figure 2 A schematic diagram of the structure of a knee spreader for measuring a knee joint provided by an embodiment of the present invention;

[0026] Figure 3 A simplified model and force analysis diagram of the knee spreader provided in an embodiment of the present invention;

[0027] Figure 4 A flowchart of the pressure error compensation process of the knee spreader provided by the embodiment of the present invention;

[0028] Figure 5 A schematic diagram of modeling an external actuator using the ANSYS software provided in an embodiment of the present invention;

[0029] Figure 6 A comparison diagram of the actual force measured by the motor provided in an embodiment of the present invention and the force measured after compensation.

[0030] Description of reference numerals:

[0031] 1. External actuator; 2. Flange; 3. Motion mechanism; 4. Motor; 5. Pressure sensor. DETAILED DESCRIPTION

[0032] A specific implementation of the present invention is described in detail below in conjunction with the accompanying drawings, but it should be understood that the protection scope of the present invention is not limited by the specific implementation.

[0033] In the description of the present invention, it is to be understood that the terms “center”, “longitudinal”, “lateral”, “length”, “width”, “thickness”, “up”, “down”, “front”, “back”, “left”, “right”, “vertical”, “horizontal”, “top”, “bottom”, “inside”, “outside”, “axial”, “radial”, “circumferential”, etc., indicating orientations or positional relationships based on the orientations or positional relationships shown in the accompanying drawings, are only for the convenience of describing the technical solutions of the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as a limitation on the present invention.

[0034] According to the simplified model of the knee spreader, Figure 1 As shown, 1 is the external actuator, 2 is the flange, 3 is the motion mechanism, 4 is the motor, and 5 is the pressure sensor. The force analysis of the knee spreader shows that, Figure 2 As shown, when the external actuator 1 is inserted into the knee joint, the knee spreader will be subjected to the pressure F exerted by the knee joint, which can be decomposed into the force Q measured by the motor. y And two bending moments M x and M z ,like Figure 3As shown in the figure, the pressure F on the external actuator and the force Q measured by the motor y The two bending moments have the following relationship:

[0035]

[0036] At present, traditional error solution methods include hardware improvement and static error compensation. Specific measures for hardware improvement include the use of higher-precision force sensors or strain gauges, such as micro strain gauges and micro force sensors; the use of more stable materials and optimized sensor structure design to reduce sensor drift and noise caused by changes in external pressure, temperature, etc.; and the use of physical shock absorption design and shielding materials to reduce the interference of mechanical vibration and electromagnetic noise on sensors in the surgical environment. For example, Chinese patents CN116035633A and CN116350461A both designed new knee spreader structures, which can effectively achieve the stable contact effect between the spreader and the human body when used, and reduce the measurement errors caused by environmental factors. Static error compensation measures include establishing a simple linear compensation model based on known sensor error characteristics to correct the measurement results. For example, Chinese patent CN117442396A discloses a data processing method for a knee spreader, which solves the measurement error problem caused by soft tissue of the knee joint.

[0037] Because soft tissue has complex mechanical properties, traditional knee spreaders are prone to errors when dealing with these complex mechanical properties. First, the soft tissue of the knee joint is not as simple as a linear spring. When the force changes, its elastic modulus will change. Most traditional knee spreaders are designed based on simplified linear models, ignoring the nonlinear behavior of soft tissue under different pressures, resulting in measurement errors; secondly, even high-precision sensors may have certain measurement errors in practical applications. For example, the sensor may fail to accurately reflect complex pressure changes, especially under different pressure conditions, the sensor's response may not be completely linear. Finally, due to temperature, humidity, sensor aging and improper human operation, deviations in pressure measurement data can also occur. In summary, there are many reasons why traditional knee spreaders are prone to errors, mainly including nonlinear deformation of soft tissue, sensor accuracy problems, operational errors, equipment aging and environmental factors.

[0038] To this end, an embodiment of the present invention provides a method for compensating a knee spreader pressure error, which effectively solves the problem of pressure measurement error caused by the influence of the knee spreader structure, thereby improving the accuracy and reliability of the measurement result.

[0039] At least one embodiment of the present invention provides a method for compensating pressure errors of a knee spreader, comprising: error compensation model construction: performing dynamic modeling of the knee spreader through finite element software ANSYS, testing and simulating the established dynamic model, obtaining multiple groups of real pressure values ​​at the bottom of the motor, and simultaneously obtaining multiple groups of corresponding pressure values ​​measured by the pressure sensor at the bottom of the knee spreader motor, i.e., measurement data, and using the measurement data and the real pressure values ​​as training data to train a BP neural network model to form an error compensation model. The model continuously optimizes the weights and parameters of the model by capturing the nonlinear characteristics of the soft tissue of the knee joint and the pattern of measurement errors. Training the BP neural network model is actually as follows: collecting the pressure data measured by the pressure sensor at the bottom of the motor of the knee spreader, and performing external actuator modeling of the ANSYS software. On each grid of the robot, different forces are applied to obtain the true pressure value of the motor, and the measured pressure data and the true pressure value are input into the BP neural network model as training data. The model is trained to capture the nonlinear characteristics of the knee soft tissue and the pattern of measurement errors, and the weights and parameters of the model are continuously optimized. The neural network can learn the rules between the measured pressure data and the true pressure value, so as to achieve more accurate pressure compensation. The model predicts: During the operation, the knee spreader measures the knee soft tissue pressure data in real time, and the measured knee soft tissue pressure data is input into the error compensation model. The error compensation model automatically compensates the knee soft tissue pressure data according to the previous training results, combines the compensation value with the real-time measured knee soft tissue pressure data, and outputs the corrected precise pressure value.

[0040] In the knee spreader pressure error compensation method provided by the above embodiment of the present invention, the error compensation technology based on BP neural network is introduced to solve the measurement error problem encountered by the existing knee spreader during surgery.

[0041] The present invention is described below by several specific embodiments. In order to keep the following description of the embodiments of the present invention clear and concise, the detailed description of known functions and known components may be omitted. When any component of the embodiments of the present invention appears in more than one accompanying drawings, the component may be represented by the same reference numeral in each of the accompanying drawings.

[0042] The embodiment of the present invention provides a method for compensating pressure errors of a knee spreader, comprising: error compensation model construction: performing dynamic modeling of the knee spreader through finite element software ANSYS, testing and simulating the established dynamic model, and obtaining multiple groups of real pressure values ​​at the bottom of the motor, that is, modeling and training through ANSYS software. Different from directly using standard measuring equipment to collect pressure data for training, the simulation software can generate training data more accurately and efficiently, thereby saving time and labor costs, and simultaneously obtaining multiple groups of corresponding pressure values ​​measured by the pressure sensor at the bottom of the knee spreader motor, that is, measurement data, using the measurement data and the real pressure value as training data to train a BP neural network model to form an error compensation model, the model continuously optimizes the weights and parameters of the model by capturing the nonlinear characteristics of the knee soft tissue and the pattern of the measurement error; model prediction: during the operation, the knee spreader measures the knee soft tissue pressure data in real time, and the measured knee soft tissue pressure data is input into the error compensation model. The error compensation model automatically performs error compensation on the knee soft tissue pressure data according to the previous training results, combines the compensation value with the real-time measured knee soft tissue pressure data, and outputs the corrected accurate pressure value.

[0043] A method for compensating a pressure error of a knee spreader provided by an embodiment of the present invention solves the measurement error problem of the existing knee spreader during surgery by introducing an error compensation technology based on a BP neural network. The finite element software ANSYS mathematical modeling is used to accurately simulate the different pressure behaviors of the knee spreader, i.e., the external actuator, to obtain a large number of real pressure values ​​at the bottom of the motor, and then obtain the corresponding pressure value measured by the pressure sensor at the bottom of the knee spreader motor, i.e., the measurement data. A large number of real pressure values ​​and measurement data are transmitted to the BP neural network model for training. The task of the error compensation model is to adjust the measured pressure to the real pressure value. By establishing a BP neural network model, nonlinear relationships can be captured, and the problem of inaccurate pressure measurement caused by the structure of the knee spreader can be compensated in real time, ensuring that high-precision measurement results are obtained in a dynamic environment. Through the continuous self-learning mechanism of the neural network, the aging of the equipment and the wear and tear during surgery can be adapted to ensure the measurement accuracy of the knee spreader after long-term use. At the same time, the need for manual adjustment based on the doctor's experience during surgery is greatly reduced, thereby improving the efficiency and safety of surgery.

[0044] Specifically, the test simulation of the dynamic model includes: applying different forces on each grid of the dynamic model, that is, the external actuator. For comprehensive testing, the force ranges from 0N to 100N. Since the motor accuracy is 0.1N, the force interval applied each time is 0.1N. Each force value is tested and simulated multiple times. Specifically, 500 simulations can be performed. Through the test simulation, the real pressure value at the bottom of the motor is obtained. As a powerful finite element analysis software, ANSYS can accurately simulate the different pressure behaviors of the knee spreader (specifically the external actuator) through mathematical modeling. For example, by applying a force of 30N to the external actuator through this software, it is believed that the pressure measured at the bottom of the motor is 30N, which is a real pressure value. The real pressure and the measured actual pressure are transmitted to the BP neural network model training, and the task of the error compensation model is to adjust the measured pressure to the real pressure value.

[0045] In order to ensure the accuracy of the BP neural network model, error compensation verification is performed after the error compensation model is constructed.

[0046] Specifically, in this embodiment, after the error compensation model is constructed, error compensation verification is performed including: directly collecting the pressure applied by the knee joint, i.e., the actual pressure value, through a high-precision pressure sensor. At this time, the collected pressure will not produce a large error due to the structural problem of the knee spreader. The pressure obtained at this time is considered to be the actual pressure, and the measurement data corresponding to the actual pressure value is input into the BP neural network model. The model captures the nonlinear characteristics of the soft tissue of the knee joint and the pattern of measurement errors, continuously optimizes the weights and parameters of the model, generates a predicted pressure value, and compares the generated predicted pressure value with the actual pressure value to verify the accuracy of the BP neural network model. If the difference between the predicted pressure value and the actual pressure value is large, the BP neural network model needs to be retrained.

[0047] Generally speaking, the value of the strain response obtained by the finite element method is very small, and compared with the force applied to the knee spreader structure, the magnitude difference between the two is relatively large, which will cause the network to be very sensitive to the input strain response. In order to solve this problem, in the knee spreader pressure error compensation method provided in the embodiment of the present invention, the measured data and the true pressure value are used as training data to train the BP neural network model to form an error compensation model. The measured data and the true pressure value are first normalized respectively. The normalization processing formula is:

[0048]

[0049] Where μ i is the data to be normalized, μ min is the minimum value in the data to be normalized, μ max is the maximum value of the data to be normalized, xi is the normalized result.

[0050] The normalized data training set is usually normally distributed. In order to achieve good training results, the weight initialization should be as close to the peak or valley as possible. The ideal interval [-1, 1] used in the embodiment of the present invention is 0 at the valley. The neural network used is the BP neural network, and its basic structure is as follows: Figure 5 As shown, Figure 5 The schematic diagram of the modeling of the external actuator by the ANSYS software provided in the embodiment of the present invention, if the weight is set to 0, the results calculated by all neurons in the network will be the same, and the gradient results of back propagation will be completely consistent, so the changes after iteration are also all the same, which means that the neural network cannot learn from the training set. Therefore, the construction of the BP neural network model includes weight initialization, and the formula for weight initialization is as follows:

[0051]

[0052] Where w is the connection weight, n in is the number of neural units in the previous layer, n out is the number of neural units in the next layer, and Var(w) is the variance of the weights, which reflects the distribution range of these weight values. For example, if the variance of the weights is small, then all weights are very close to 0, which may cause the network to fail to learn effective features; if the variance is too large, the weights may produce very large values ​​during the training process, resulting in gradient explosion or gradient disappearance problems. Weight initialization refers to setting the initial weights of each connection in the network before the neural network training begins. It is the starting point of the training process. Optimizing the weights of the model refers to continuously adjusting and updating the weights through back propagation and optimization algorithms (such as gradient descent, Adam, etc.) during the training process to minimize the loss function, thereby improving the model's predictive ability. Simply put, weight initialization is an initial setting step to provide a reasonable starting point for the model. Optimizing the weights of the model is a dynamic adjustment process that adjusts the weights through training data and back propagation to achieve optimal performance. By setting the initial value of the weight reasonably, the problem of gradient disappearance or gradient explosion is avoided, making the forward propagation and back propagation processes more stable, reducing training time, improving training efficiency, and accelerating model convergence; it can maintain the stability and efficiency of deep neural networks and is suitable for deep networks; it ensures that the output and gradient updates of each neuron are different, thereby improving learning ability and avoiding symmetry problems, thereby improving network training stability.

[0053] In the process of deep learning, in order to avoid overfitting during the training process, the network is regularized during the training process of the BP neural network model. The network regularization adopts the Dropout method for network regularization. Dropout means that in the training process of the deep learning network, for the neural network unit, a certain proportion of nodes are randomly selected and temporarily discarded from the network in each round of weight update. During the training process, the model may overfit to the training data, resulting in poor performance on unseen data (i.e., test data). One of the reasons for overfitting is that the model is too dependent on the specific patterns and noise of the training set. The role of Dropout: By randomly discarding neurons during the training process, Dropout effectively reduces the model's dependence on specific features in the training data, forcing the model to learn more general features. In this way, the features learned by the network can perform better on different input data and will not be limited to specific patterns in the training data. Summary: Improve generalization ability: Dropout reduces overfitting and improves the model's adaptability to test data; Enhance robustness: By randomly discarding neurons, Dropout makes the model less likely to rely on noise or specific patterns in training data, thereby improving robustness to potential noise or changes in test data; Better adaptability: Dropout forces the model to learn more general features, giving it better predictive power for test data (even unseen data).

[0054] In order to improve the learning ability of the neural network, an activation function needs to be added. In the embodiment of the present invention, the BP neural network model uses a linear rectifier function as the activation function. The pressure data of the knee spreader involves complex nonlinear relationships. The linear rectifier function introduces nonlinear characteristics, so that the neural network can effectively capture these complex nonlinear relationships. The calculation of the linear rectifier function is very simple, which can greatly improve the calculation efficiency when processing the pressure data of the knee spreader in real time.

[0055] The process of neural network training is the process of minimizing the loss function. In order to make the prediction results of the neural network more accurate, it is necessary to select a suitable loss function. In the embodiment of the present invention, the variance cost function is used as the loss function to calculate the error during the BP neural network model training process. The loss function expression is as follows:

[0056]

[0057] Where y is the expected output, a is the actual output of the neuron, and C is the function value. The smaller the value of the function, the more accurate the prediction result. The variance cost function can quantify the prediction error and emphasize the impact of large errors. It is smooth and differentiable, suitable for optimization methods such as gradient descent. It is sensitive to outliers and is very effective in some tasks that need to pay attention to outliers. It is simple and easy to understand and has good mathematical properties, which helps the training to converge stably.

[0058] refer to Figure 4 The knee spreader pressure error compensation process provided by the embodiment of the present invention is as follows:

[0059] Step 1: Sample data collection

[0060] The finite element software ANSYS was used to perform dynamic modeling of the knee spreader structure. Different forces were applied to each grid of the model. For comprehensive testing, the force ranged from 0N to 100N. Since the motor accuracy was 0.1N, the force interval applied each time was 0.1N, and each force value was tested 500 times. Through these test simulations, the pressure data at the bottom of the motor was obtained as the training data and test data of the neural network.

[0061] Step 2: Normalization of sample data

[0062] Generally speaking, the value of the strain response obtained by the finite element method is very small, and the magnitude difference between the two is relatively large compared to the force applied to the knee spreader structure, which will cause the network to be very sensitive to the input strain response. In order to solve this problem, the sample data is normalized. The specific normalization formula is:

[0063]

[0064] Where μ i is the data to be normalized, μ min is the minimum value in the data to be normalized, μ max is the maximum value of the data to be normalized, x i is the normalized result.

[0065] Step 3: Weight initialization

[0066] The normalized data training set is usually normally distributed. In order to achieve good training results, the weights should be initialized as close to the peak or valley as possible. The ideal interval [-1,1] used in the present invention is 0 at the valley. The neural network used is the BP neural network, and its basic structure is as follows Figure 5 As shown:

[0067] If the weight is set to 0, the results calculated by all neurons in the network will be the same, and the gradient results of back propagation will be exactly the same, so the changes after iterations will also be the same, which means that the neural network cannot learn from the training set. Therefore, the present invention adopts the following weight initialization method:

[0068]

[0069] Where w is the connection weight, nin and n out is the number of neural units in the previous and next layers.

[0070] Step 4: Network Regularization

[0071] In the process of deep learning, in order to avoid overfitting during training, the Dropout method is used to regularize the network during the neural network learning process.

[0072] Dropout refers to the process of training a deep learning network. For a neural network unit, a certain proportion of nodes are randomly selected and temporarily discarded from the network in each round of weight update.

[0073] Step 5: Neural Network Activation Function

[0074] In order to improve the learning ability of the neural network, an activation function needs to be added, and the activation function of the present invention adopts a linear rectifier function (RELU).

[0075] Step 6: Neural Network Loss Function

[0076] The process of neural network training is the process of minimizing the loss function. In order to make the prediction results of the neural network more accurate, it is necessary to select a suitable loss function. The present invention selects the variance cost function, and the function expression is as follows:

[0077]

[0078] Where y is the expected output and a is the actual output of the neuron. The smaller the value of this function, the more accurate the prediction result.

[0079] Step 7: After error compensation, the force measured by the motor at the bottom of the knee spreader is almost consistent with the actual force applied. The test data is shown in Table 1. This test data is only the data obtained by applying different forces to a single grid. The force range is between 30N and 40N. The plotted image is shown in Figure 6 shown.

[0080] After error compensation, the force measured by the motor at the bottom of the knee spreader is almost consistent with the actual force applied. The test data is shown in Table 1. This test data is only the data obtained by applying different forces to a single grid. The force range is between 30N and 40N. The plotted image is shown in Figure 6 As shown in the figure, the force error measured by the motor is large when not compensated, and it presents an irregular and complex nonlinear relationship. Ideally, it should be a linear relationship after compensation, that is, the force applied is equal to the force measured by the motor. After compensation by the neural network, it will be close to a linear relationship. At this time, the measured force is close to the applied force, as shown in the figure. Figure 6 Near a straight line.

[0081] Table 1 Actual force measured by the motor and the force measured after compensation

[0082]

[0083]

[0084] Example 1

[0085] Soft tissue pressure measurement and error compensation in knee replacement surgery

[0086] Step 1: Installation and preliminary commissioning of the knee spreader

[0087] In knee replacement surgery, a knee spreader is first used to measure the pressure on the soft tissue of the patient's knee joint. At this time, the initial data of the knee spreader will be affected by the structure of the knee spreader, resulting in measurement errors.

[0088] Step 2: Data collection and BP neural network model training

[0089] Before surgery, actual measurement data of the knee spreader were collected through a series of experiments and compared with the actual pressure values ​​obtained by standard measurement equipment. These data were used as training data to input into the BP neural network model, which was trained to capture the nonlinear characteristics of the soft tissue of the knee joint and the pattern of measurement errors. By continuously optimizing the weights and parameters of the model, the neural network was trained to identify the relationship between the measurement error and the true pressure value.

[0090] Step 3: Model prediction

[0091] During surgery, the knee spreader measures the soft tissue pressure of the knee joint in real time. The measured data is input into the BP neural network model, and the network automatically compensates the measured value for errors based on the previous training results. The compensation value is combined with the real-time measured pressure data to output the corrected accurate pressure value.

[0092] For example, in a knee replacement surgery, the initial measurement showed that the pressure of the medial ligament of the knee joint was 32.6N, but due to the structure of the knee spreader, the actual pressure had a large deviation. Through real-time compensation of the neural network, the final adjusted accurate pressure measurement value was 37.1N, helping doctors make accurate judgments and adjust surgical strategies.

[0093] A knee spreader pressure error compensation method provided by an embodiment of the present invention has the following advantages:

[0094] 1. The measurement accuracy is significantly improved

[0095] By introducing BP neural network for error compensation, the present invention can effectively handle the measurement error of knee spreader pressure. Since soft tissue has complex mechanical properties, traditional knee spreaders are prone to errors when dealing with these complex mechanical properties, while the neural network model of the present invention can capture these nonlinear relationships, provide more accurate measurement results, and significantly improve the accuracy of knee spreaders in surgery.

[0096] 2. Reduce human intervention during surgery and improve surgical efficiency

[0097] Traditional knee spreaders require doctors to manually adjust them based on their experience during surgery, which is complicated to operate and prone to human errors. The present invention reduces the reliance on manual adjustment by doctors during surgery through an automated error compensation mechanism, improves the automation of the surgical process, thereby improving surgical efficiency and reducing surgical risks.

[0098] 3. Adapt to equipment aging and extend equipment life

[0099] Through the self-learning function of the neural network, the present invention can dynamically adjust according to the performance changes of the knee spreader during long-term use, and compensate for the measurement errors caused by the aging of the equipment. This not only improves the stability of the equipment in long-term use, but also effectively extends the service life of the knee spreader and reduces the cost of frequent calibration and equipment replacement.

[0100] 4. Improve postoperative recovery effect and surgical safety

[0101] In knee surgery, accurate measurement of soft tissue pressure is crucial to the recovery of joint function after surgery. The present invention improves postoperative recovery and reduces the incidence of postoperative complications by providing more accurate pressure measurement. This also helps doctors make more scientific decisions during surgery and improves the safety of surgery.

[0102] The above inventions are only several specific embodiments of the present invention. However, the embodiments of the present invention are not limited thereto. Any changes that can be conceived by those skilled in the art should fall within the protection scope of the present invention.

Claims

1. A method for compensating a pressure error of a knee spreader, characterized in that: include: Error compensation model construction: The finite element software ANSYS is used to perform dynamic modeling of the knee spreader, and the established dynamic model is tested and simulated to obtain multiple sets of real pressure values ​​at the bottom of the motor. At the same time, multiple sets of corresponding pressure values ​​measured by the pressure sensors at the bottom of the knee spreader motors are obtained, i.e., measurement data. The measurement data and the real pressure values ​​are used as training data to train the BP neural network model to form an error compensation model. Model prediction: During the operation, the knee spreader measures the knee soft tissue pressure data in real time, and inputs the measured knee soft tissue pressure data into the error compensation model. The error compensation model automatically compensates for the error of the knee soft tissue pressure data based on previous training results, combines the compensation value with the real-time measured knee soft tissue pressure data, and outputs the corrected precise pressure value.

2. The knee spreader pressure error compensation method according to claim 1, characterized in that: The test simulation of the dynamic model specifically includes: applying forces of different sizes on each grid of the dynamic model, i.e., the external actuator, the force range is from 0N to 100N, the interval of each force applied is 0.1N, and each force value is tested and simulated multiple times. Through the test simulation, the real pressure value at the bottom of the motor is obtained.

3. The knee spreader pressure error compensation method according to claim 1, characterized in that: After the error compensation model is constructed, error compensation verification is performed.

4. The knee spreader pressure error compensation method according to claim 3, characterized in that: The error compensation verification includes: directly collecting the pressure applied by the knee joint, i.e., the actual pressure value, through a high-precision pressure sensor, and inputting the measurement data corresponding to the actual pressure value into an error compensation model. The model continuously optimizes the weights and parameters of the model by capturing the nonlinear characteristics of the soft tissue of the knee joint and the pattern of the measurement error, and generates a predicted pressure value. The generated predicted pressure value is compared with the actual pressure value to verify the accuracy of the error compensation model.

5. The knee spreader pressure error compensation method according to claim 1, characterized in that: Before using the measured data and the real pressure value as training data to train the BP neural network model to form the error compensation model, the measured data and the real pressure value are first normalized respectively. The normalization formula is: Where μ i is the data to be normalized, μ min is the minimum value in the data to be normalized, μ max is the maximum value of the data to be normalized, x i is the normalized result.

6. The knee spreader pressure error compensation method according to claim 1, characterized in that: Constructing the BP neural network model includes weight initialization, and the formula for weight initialization is as follows: Where w is the connection weight, n in is the number of neural units in the previous layer, n out is the number of neural units in the next layer, and Var(w) is the variance of the weight.

7. The knee spreader pressure error compensation method according to claim 1, characterized in that: During the training process of the BP neural network model, network regularization is performed, and the network regularization adopts the Dropout method to perform network regularization.

8. The knee spreader pressure error compensation method according to claim 1, characterized in that: The BP neural network model adopts a linear rectification function as an activation function.

9. The knee spreader pressure error compensation method according to claim 1, characterized in that: The variance cost function is used as the loss function to calculate the error during the BP neural network model training process. The loss function expression is as follows: Where y is the expected output, a is the actual output of the neuron, and C is the function value.

Images