Total knee arthroplasty dilator deformation error compensation method

Abstract

The invention discloses a deformation error compensation method for a dilator in total knee arthroplasty, and aims to solve the problem of measurement errors caused by deformation of a cantilever structure. The method comprises the following steps of: firstly, performing parameterized dynamic modeling and simulation on an external execution mechanism through finite element software, calibrating a model by combining actual measurement data of a physical prototype, and constructing a high-precision deformation data set; then, the data set is used for training a BP neural network optimized through a genetic algorithm, and the network takes the applied force and the displacement measurement value which are collected in real time as input and outputs a real displacement value after compensation. And finally deploying the trained lightweight network model in an embedded processing unit of the spreader to realize real-time error compensation. According to the method, complex nonlinear deformation errors are effectively compensated, the average absolute error is reduced to 0.056 mm, the real-time response delay is smaller than or equal to 5 ms, the method is remarkably superior to a traditional linear method, more accurate gap measurement is provided for an operation, and the method can be further used for forming closed-loop control and improving the intellectualization and precision of the operation.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to a method for compensating for deformation error of a spreader in total knee arthroplasty. Background Technology

[0002] Total knee arthroplasty is a technique that replaces a damaged joint with an artificial prosthesis, commonly used to treat conditions such as arthritis, trauma, and rheumatoid arthritis. During the procedure, it's crucial to maintain a rectangular joint space in both flexion and extension of the knee joint. Two main osteotomy techniques achieve this: one is the measurable osteotomy method, also known as the equal-volume osteotomy method, which determines the osteotomy location and amount based on the anatomical structure of the femur and tibia; the other is the space-balanced osteotomy method. This method involves creating a certain gap when the knee is extended to balance the soft tissues before performing the osteotomy. Using this gap as a reference, the same opening force is applied to the knee joint, and then the femoral condyle is osteotomized to achieve soft tissue tension balance in both extension and flexion positions.

[0003] With technological advancements, patent CN 114847887A proposes a novel total knee replacement surgery spreader. This device comprises an upper base plate 1, an extension rod 2, a flange 3, a motion mechanism 4, a motor 5, and a displacement sensor. Figure 1 As shown; for subsequent research, the upper base plate and the extension rod extending beyond the flange are collectively referred to as the external actuator. The simplified model and force analysis of the knee joint spreader are as follows: Figure 2 , Figure 3 As shown.

[0004] Through mechanical analysis, the following formula is obtained:

[0005]

[0006] in, It represents the deformation of the plate on the expander, and F is the applied force. It is the length of the upper film. It is the elastic modulus of the upper film. It is the moment of inertia of the cross section of the upper film; The length of the moving mechanism exceeding the flange is... It is its elastic modulus. It is its moment of inertia.

[0007] As can be seen from the above formula, the deformation of the upper plate of the spreader is mainly affected by the applied force and the length of the moving mechanism exceeding the flange, and this effect is nonlinear. Patent CN117442396A mentions a linear regression error compensation method, but this method has the following technical defects:

[0008] 1. The model assumptions do not match the actual structure. This technology uses the assumption that the stress values ​​are all within the elastic deformation range to establish a force and deformation model of the spreader under different force values. It does not take into account that the upper plate and the long rod structure are not rigidly connected, and because there is a gap between the flange and the motion mechanism, the running mechanism tilts slightly under the action of bending moment, resulting in a relatively large downward displacement of the upper plate's force-bearing end, which leads to inaccurate measurement of the gap parameters in total knee replacement surgery.

[0009] 2. The factors affecting deformation are not fully considered. This technology only considers the relationship between the deformation of the actuator and the force value, but does not take into account the influence of the change in the length of the long rod structure of the total knee arthroplasty spreader during use. Summary of the Invention

[0010] The technical solution of this invention to solve the above-mentioned technical problems is to provide a method for compensating for deformation error of a retractor in total knee arthroplasty, comprising the following steps:

[0011] S1. The external actuator of the total knee arthroplasty device is dynamically modeled and simulated using finite element software to obtain the simulated displacement under different cantilever lengths and different applied forces. The simulation model is calibrated based on the measured displacement of the physical prototype to ensure that the relative error between the simulated deformation value and the measured deformation value is less than a set threshold.

[0012] S2. Divide the dataset obtained from the calibrated finite element simulation into a training set and a test set, and normalize the data.

[0013] S3. Construct a BP neural network as an error compensation model. The input of the BP neural network is the applied force and the displacement measurement value of the spreader, and the output is the compensated true displacement value.

[0014] S4. The BP neural network model is trained using the training set. The training process includes forward propagation to calculate the predicted value, calculation of the error through the loss function, backpropagation to update the network parameters, and iteration until the error meets the requirements.

[0015] S5. Deploy the trained error compensation model to the embedded processing unit of the expander to perform error compensation on the real-time collected sensor data and obtain the corrected displacement value.

[0016] Further, step S1 includes:

[0017] S101, Establish a model of the external actuator in the finite element software, and perform parametric simulation with cantilever length L1 and applied force F as variables, where F ranges from 1N to 100N with an interval of 1N; L1 ranges from 1mm to 20mm with an interval of 1mm, and obtain the simulation displacement dataset.

[0018] S102, on the physical prototype, under the selected cantilever length and applied force, the measured displacement is obtained and the measured deformation value is calculated; the simulated deformation value under the same working condition is compared with the measured deformation value to calculate the relative error;

[0019] S103, if the relative error is greater than 5%, adjust the material parameters of the simulation model for calibration, and repeat steps S101 and S102 until the relative error is less than 5%.

[0020] Furthermore, the measured deformation value δz is obtained through the formula Calculation, where For theoretical displacement, The actual displacement; the relative error RE is expressed by the formula calculate.

[0021] Furthermore, in step S2, the formula for normalizing the data is:

[0022] ;

[0023] in, For data to be normalized, and These are the minimum and maximum values ​​in the dataset, respectively. This is the result after normalization.

[0024] Further, in step S3, the construction of the BP neural network includes:

[0025] S301, Set the network topology: The input layer has 2 neurons, corresponding to the applied force and displacement measurement values ​​respectively; at least one hidden layer; the output layer has 1 neuron, corresponding to the actual displacement value;

[0026] S302 uses the Tanh function as the activation function, and the specific formula is as follows:

[0027] ;

[0028] in, It is the input of the function, and the output range of the function is fixed between (-1, 1);

[0029] S303 uses the Xavier method to initialize network weights, with the specific formula as follows:

[0030] ;

[0031] And regularization is performed using the Dropout method;

[0032] S304 uses the variance cost function as the loss function, and its expression is:

[0033] ;

[0034] in This is the expected output. This is the actual output of the neuron. The function value;

[0035] S305, the hyperparameters of the BP neural network are optimized using a genetic algorithm, including the hidden layer structure, learning rate, and regularization strength.

[0036] Furthermore, the genetic algorithm optimization steps include: encoding hyperparameters to generate an initial population; using the performance on the test set as the fitness evaluation of individuals; and iteratively optimizing through selection, crossover, and mutation operations to finally obtain the optimal hyperparameter combination and construct the BP neural network.

[0037] Furthermore, step S5 specifically includes:

[0038] S501, Load the trained error compensation model into the embedded processing unit and calibrate the displacement sensor and pressure sensor;

[0039] S502 collects displacement measurements and applied forces in real time and performs normalization processing;

[0040] S503, input the normalized data into the error compensation model to obtain the normalized compensation displacement value, and then perform inverse normalization to obtain the final true displacement value.

[0041] S504, output the actual displacement value to the display interface.

[0042] Furthermore, it also includes step S505: generating a control signal based on the deviation between the actual displacement value and the target gap to adjust the drive motor of the spreader, forming a closed-loop control.

[0043] This invention, by integrating finite element simulation, intelligent algorithms, and embedded technology, not only significantly improves the accuracy and reliability of deformation error compensation for total knee arthroplasty retractors, but also realizes the clinical application potential of real-time closed-loop control, while possessing good economic efficiency. It provides an effective technical solution for the intelligent and precise development of this type of surgical instrument. Attached Figure Description

[0044] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the structures shown in these drawings without creative effort.

[0045] Figure 1 A schematic diagram of the structure of a prior art knee joint spreader;

[0046] Figure 2 A structural diagram of the knee joint measured using a knee joint spreader in the prior art;

[0047] Figure 3 A simplified model and force analysis diagram of existing knee joint spreaders;

[0048] Figure 4 This is a flowchart of the method for compensating for deformation error of the spreader in total knee arthroplasty disclosed in this invention.

[0049] Figure 5 This is a force simulation diagram of the external actuator of the present invention;

[0050] Figure 6 The comparison diagrams provided by this invention show the displacement before and after compensation, as well as the actual displacement. Detailed Implementation

[0051] This invention proposes a method for compensating for deformation error of a knee joint retractor in total knee arthroplasty, aiming to solve the technical problem that the external actuator of the knee joint retractor has deformation error during use, resulting in low accuracy of knee joint space parameter measurement.

[0052] The deformation error compensation method for the total knee arthroplasty expander proposed in this invention will be described below in specific embodiments:

[0053] Example 1:

[0054] A method for compensating for deformation error of a retractor in total knee arthroplasty includes the following steps:

[0055] S1. The external actuator of the total knee arthroplasty device is dynamically modeled and simulated using finite element software to obtain the simulated displacement under different cantilever lengths and different applied forces. The simulation model is calibrated based on the measured displacement of the physical prototype to ensure that the relative error between the simulated deformation value and the measured deformation value is less than a set threshold.

[0056] S2. Divide the dataset obtained from the calibrated finite element simulation into a training set and a test set, and normalize the data.

[0057] S3. Construct a BP neural network as an error compensation model. The input of the BP neural network is the applied force and the displacement measurement value of the spreader, and the output is the compensated true displacement value.

[0058] S4. The BP neural network model is trained using the training set. The training process includes forward propagation to calculate the predicted value, calculation of the error through the loss function, backpropagation to update the network parameters, and iteration until the error meets the requirements.

[0059] S5. Deploy the trained error compensation model to the embedded processing unit of the expander to perform error compensation on the real-time collected sensor data and obtain the corrected displacement value.

[0060] Further, step S1 includes:

[0061] S101, Establish a model of the external actuator in the finite element software, and perform parametric simulation with cantilever length L1 and applied force F as variables, where F ranges from 1N to 100N with an interval of 1N; L1 ranges from 1mm to 20mm with an interval of 1mm, and obtain the simulation displacement dataset.

[0062] S102, on the physical prototype, under the selected cantilever length and applied force, the measured displacement is obtained and the measured deformation value is calculated; the simulated deformation value under the same working condition is compared with the measured deformation value to calculate the relative error;

[0063] S103, if the relative error is greater than 5%, adjust the material parameters of the simulation model for calibration, and repeat steps S101 and S102 until the relative error is less than 5%.

[0064] Furthermore, the measured deformation value δz is obtained through the formula Calculation, where For theoretical displacement, The actual displacement; the relative error RE is expressed by the formula calculate.

[0065] Furthermore, in step S2, the formula for normalizing the data is:

[0066] ;

[0067] in, For data to be normalized, and These are the minimum and maximum values ​​in the dataset, respectively. This is the result after normalization.

[0068] Further, in step S3, the construction of the BP neural network includes:

[0069] S301, Set the network topology: The input layer has 2 neurons, corresponding to the applied force and displacement measurement values ​​respectively; at least one hidden layer; the output layer has 1 neuron, corresponding to the actual displacement value;

[0070] S302 uses the Tanh function as the activation function, and the specific formula is as follows:

[0071] ;;

[0072] in, It is the input of the function, and the output range of the function is fixed between (-1, 1);

[0073] S303 uses the Xavier method to initialize network weights, with the specific formula as follows:

[0074] ;

[0075] And regularization is performed using the Dropout method;

[0076] S304 uses the variance cost function as the loss function, and its expression is:

[0077] ;

[0078] in This is the expected output. This is the actual output of the neuron. The function value;

[0079] S305, the hyperparameters of the BP neural network are optimized using a genetic algorithm, including the hidden layer structure, learning rate, and regularization strength.

[0080] Furthermore, the genetic algorithm optimization steps include: encoding hyperparameters to generate an initial population; using the performance on the test set as the fitness evaluation of individuals; and iteratively optimizing through selection, crossover, and mutation operations to finally obtain the optimal hyperparameter combination and construct the BP neural network.

[0081] Furthermore, step S5 specifically includes:

[0082] S501, Load the trained error compensation model into the embedded processing unit and calibrate the displacement sensor and pressure sensor;

[0083] S502 collects displacement measurements and applied forces in real time and performs normalization processing;

[0084] S503, input the normalized data into the error compensation model to obtain the normalized compensation displacement value, and then perform inverse normalization to obtain the final true displacement value.

[0085] S504, output the actual displacement value to the display interface.

[0086] Furthermore, it also includes step S505: generating a control signal based on the deviation between the actual displacement value and the target gap to adjust the drive motor of the spreader, forming a closed-loop control.

[0087] Example 2:

[0088] A method for compensating for deformation error of a retractor in total knee arthroplasty includes the following steps:

[0089] S1. The simulated displacement of the external mechanism under different working conditions is obtained by simulating with the finite element software ANSYS. The measured displacement is compared and analyzed with the simulated displacement, and the elastic modulus is calibrated based on the comparison and analysis results.

[0090] Step S1 includes the following steps:

[0091] S101. The external actuator of the total knee arthroplasty retractor was dynamically modeled using the finite element method ANSYS. Boundary constraints were set, and the outermost point of maximum deformation of the external actuator was selected as the stress point. During the simulation, the cantilever L2 and the applied force F were used as variables. Specifically, different magnitudes of force were applied to the stress point, ranging from 1N to 100N, with a 1N interval between each application. The movement of the bottom of the external actuator to different positions was simulated by extending the bottom of the actuator, with displacements ranging from 1mm to 20mm, and a 1mm interval between each displacement.

[0092] An automation script was written, and the main steps of the script are shown in Table 1. Running the automation script yielded 2000 samples, as shown in Table 2.

[0093] Table 1:

[0094]

[0095] Table 2:

[0096]

[0097] S102. To verify the accuracy of the finite element simulation model, a precise calibration test was performed on the physical prototype of the total knee arthroplasty retractor. The force simulation diagram of the external actuator is shown below. Figure 5As shown, the force ranges from 20N to 100N, with a force interval of 20N per application; the displacement of the external actuator ranges from 5mm to 20mm, with a displacement interval of 5mm per application. A total of 20 sets of measured data were obtained, and the actual deformation value was calculated using the following formula.

[0098] ;

[0099] in, It is the theoretical displacement of the external actuator. It is the actual displacement of the external actuator.

[0100] The actual deformation value is compared with the simulated deformation value to calculate the relative error. The formula for calculating the relative error is as follows:

[0101] ;

[0102] in, It is the deformation error obtained from simulation. It is the deformation error measured by actual equipment.

[0103] S103. If the relative error is >5%, start the calibration. Adjust the elastic modulus to calibrate the parameters. Repeat steps S101 and S102 until the relative error is less than 5%.

[0104] The results of the comparative analysis after calibration are shown in Table 3, and the relative errors are all less than 5%.

[0105] Table 3:

[0106]

[0107] S2. Divide the dataset obtained from S1 through finite element analysis into two parts: 80% for model training and 20% for model testing. Before training the error compensation model, normalize the data. The normalization formula is as follows:

[0108] ;

[0109] In the formula, For data to be normalized, The minimum value in the data to be normalized. The maximum value in the data to be normalized. This is the result after normalization.

[0110] S3. A backpropagation (BP) neural network is used as the error compensation model. First, the network topology and initialization are selected; second, a genetic algorithm is used to optimize the network parameters; finally, the BP neural network is trained.

[0111] Step S3 includes the following steps:

[0112] S301. The network topology includes: input layer neurons (2 neurons, corresponding to stress value and displacement value respectively), hidden layer neurons, and output layer (1 neuron, corresponding to the actual displacement value), with full connectivity between neurons in each layer.

[0113] S302. After each layer of neurons, a Tanh activation function is configured, with the specific formula as follows:

[0114] ;;

[0115] in, It is the input of the function, and the output range of the function is fixed between (-1, 1).

[0116] S303. Neural network initialization configuration includes weight initialization and network regularization configuration. Weight initialization uses a classic adaptive weight initialization method, Xavier initialization. The specific formula for weight initialization is as follows:

[0117] ;

[0118] The Dropout method is used for network regularization to prevent model overfitting.

[0119] S304. The variance cost function is selected as the loss function to calculate the error. The function expression is as follows:

[0120] ;

[0121] in This is the expected output. This is the actual output of the neuron. This is the function value.

[0122] S305. A genetic algorithm is used to optimize the network parameters. First, the topology (number of hidden layers and number of neurons) and training hyperparameters (learning rate, regularization strength) of the BP network are encoded to generate an initial population. Then, for each individual in the population, the activation function and loss function are configured and the weights are initialized. The predicted value is calculated through forward propagation of the BP network, the error is calculated through the loss function, and the weights are updated through backpropagation. The performance on the test set is used as the fitness evaluation of the individual. Then, the next generation of the population is generated through selection, crossover, and mutation operations of GA. The fitness evaluation and genetic operations are repeated until the GA termination condition is met. Finally, the optimal parameter combination with the highest fitness in the population is selected to build the BP neural network.

[0123] S4. The normalized training data is input into the BP neural network model for training. First, forward propagation is performed, following the preset network topology, sequentially passing through the weighted summation of hidden layers (combining weights and biases), and nonlinear mapping of activation functions, finally obtaining the predicted value at the output layer. Simultaneously, the error between the predicted value and the true label is calculated using the loss function, providing a basis for subsequent parameter optimization. The model's weights are continuously optimized to capture the nonlinear characteristics of the knee joint soft tissue and the pattern of measurement errors. Then, backpropagation is performed, starting with the error of the loss function, using the chain rule to calculate the gradient of the error with respect to the weights and biases of each layer layer in reverse order. Based on the gradient direction and learning rate, all weights and biases are updated in the direction of "reducing error," completing one round of parameter optimization. Afterward, forward propagation is restarted, using the updated parameters to calculate new predicted values ​​and errors. This process is iterated until the error drops to a threshold or the required number of training rounds is reached.

[0124] S5. Initialize the embedded processing unit of the expander using the error compensation model, and perform real-time error compensation on the data collected by the sensor.

[0125] Step S5 includes the following steps:

[0126] S501. Burn the trained improved GA-BP neural network model into the embedded processing unit of the expander; calibrate the displacement sensor and pressure sensor to ensure that the input signal reference is consistent.

[0127] S502. Obtain the displacement measurement value l of the spreader through the displacement sensor; obtain the applied force F through the pressure sensor.

[0128] Convert the real-time collected (l,F) data into the [0,1] interval:

[0129] ;

[0130] In the formula For data to be normalized, The minimum value in the data to be normalized. The maximum value in the data to be normalized. This is the result after normalization.

[0131] S503. Input the normalized data into the embedded GA-BP neural network model; the model outputs the compensated displacement value. Inverse normalization yields the true displacement. :

[0132] ;

[0133] in, The minimum value of the true displacement in the training data. This represents the maximum value of the actual displacement in the training data.

[0134] S504, the compensated displacement Displayed in real-time on the surgical navigation interface;

[0135] Optionally, according to The deviation from the target gap automatically adjusts the motor drive signal of the spreader (forming a closed-loop control).

[0136] To demonstrate the superiority of the method of this invention, the linear regression method from patent CN 117442396 A was selected for a comparative experiment. Different displacements measured by the sensor, different forces measured by the sensor, and the actual displacements were measured and recorded, as shown in Table 4. Then, the linear regression method was used to reconcile the measured displacements. , measure strength and actual displacement value Fit it to a surface.

[0137] Table 4:

[0138]

[0139] The displacement error compensation model of the expander obtained by fitting the linear regression method is as follows: .

[0140] A comparative experiment was conducted. Using the linear regression equation and the displacement error compensation method of this invention, the compensated displacement was obtained from the data measured by the sensor. This represents the displacement after compensation by the linear regression displacement error compensation model. The displacement after compensation by the displacement compensation model of this invention is shown in Table 5.

[0141] Table 5:

[0142]

[0143] As shown in Table 5, the displacement after compensation by the method proposed in this invention is closer to the actual displacement, indicating that the method proposed in this invention is superior to the linear regression method. To further demonstrate that the method proposed in this invention is superior to the linear regression method, the average absolute error between the displacement after compensation and the displacement before compensation by these two methods was calculated. The average absolute errors are as follows:

[0144] ;

[0145] in, Indicates the total number of samples. This represents the displacement of the i-th sample after compensation. This represents the actual displacement of the i samples.

[0146] Furthermore, the real-time response delay of deploying the method of this invention and the linear regression method into the embedded unit is recorded and then compared. The comparison graphs before and after compensation with the actual displacement are shown below. Figure 6 As shown in Table 6, the comparison results are as follows.

[0147] Table 6:

[0148]

[0149] As shown in Table 6, the average absolute error of the method proposed in this invention is 0.056 mm, and the real-time response delay is ≤5 ms. In contrast, the average absolute error of the linear regression method used in CN 117442396 A is 0.097 mm, and the real-time response delay is ≤1 ms. Although the real-time response delay of the method proposed in this invention is slightly higher, it still meets the clinical real-time requirements, and the average absolute error is significantly smaller, indicating that its displacement compensation effect is significantly better than that of the linear regression method.

[0150] The above description is merely a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for compensating for deformation error of a retractor in total knee arthroplasty, characterized in that, Includes the following steps: S1. The external actuator of the total knee arthroplasty device is dynamically modeled and simulated using finite element software to obtain the simulated displacement under different cantilever lengths and different applied forces. The simulation model is calibrated based on the measured displacement of the physical prototype to ensure that the relative error between the simulated deformation value and the measured deformation value is less than a set threshold. S2, divide the dataset obtained from the calibrated finite element simulation into training and test sets, and normalize the data; S3, Construct a BP neural network as an error compensation model. The input of the BP neural network is the applied force and the displacement measurement value of the spreader, and the output is the compensated true displacement value. S4, the BP neural network model is trained using the training set. The training process includes forward propagation to calculate the predicted value, calculation of the error through the loss function, backpropagation to update the network parameters, and iteration until the error meets the requirements. S5, the trained error compensation model is deployed to the embedded processing unit of the expander to perform error compensation on the real-time collected sensor data and obtain the corrected displacement value.

2. The method according to claim 1, characterized in that, Step S1 includes: S101, Establish a model of the external actuator in the finite element software, and perform parametric simulation with cantilever length L1 and applied force F as variables, where F ranges from 1N to 100N with an interval of 1N; L1 ranges from 1mm to 20mm with an interval of 1mm, and obtain the simulation displacement dataset. S102, on the physical prototype, under the selected cantilever length and applied force, the measured displacement is obtained and the measured deformation value is calculated; the simulated deformation value under the same working condition is compared with the measured deformation value to calculate the relative error; S103, if the relative error is greater than 5%, adjust the material parameters of the simulation model for calibration, and repeat steps S101 and S102 until the relative error is less than 5%.

3. The method according to claim 2, characterized in that, The measured deformation value δz is obtained through the formula Calculation, where For theoretical displacement, The actual displacement; the relative error RE is expressed by the formula calculate.

4. The method according to claim 1, characterized in that, In step S2, the formula for normalizing the data is: ； in, For data to be normalized, and These are the minimum and maximum values ​​in the dataset, respectively. This is the result after normalization.

5. The method according to claim 1, characterized in that, In step S3, the construction of the BP neural network includes: S301, Set the network topology: The input layer has 2 neurons, corresponding to the applied force and displacement measurement values ​​respectively; at least one hidden layer; the output layer has 1 neuron, corresponding to the actual displacement value; S302 uses the Tanh function as the activation function, and the specific formula is as follows: ；； in, It is the input of the function, and the output range of the function is fixed between (-1, 1); S303 uses the Xavier method to initialize network weights, with the specific formula as follows: ； And regularization is performed using the Dropout method; S304 uses the variance cost function as the loss function, and its expression is: ； in This is the expected output. This is the actual output of the neuron. The function value; S305, the hyperparameters of the BP neural network are optimized using a genetic algorithm, including the hidden layer structure, learning rate, and regularization strength.

6. The method according to claim 5, characterized in that, The genetic algorithm optimization steps include: encoding hyperparameters to generate an initial population; using the performance on the test set as the fitness evaluation of individuals; iteratively optimizing through selection, crossover, and mutation operations to finally obtain the optimal hyperparameter combination and construct the BP neural network.

7. The method according to claim 1, characterized in that, Step S5 specifically includes: S501, Load the trained error compensation model into the embedded processing unit and calibrate the displacement sensor and pressure sensor; S502 collects displacement measurements and applied forces in real time and performs normalization processing; S503, input the normalized data into the error compensation model to obtain the normalized compensation displacement value, and then perform inverse normalization to obtain the final true displacement value. S504, output the actual displacement value to the display interface.

8. The method according to claim 7, characterized in that, It also includes step S505: generating a control signal based on the deviation between the actual displacement value and the target gap to adjust the drive motor of the spreader, forming a closed-loop control.

Images