Auxiliary bone identification system based on vertebral plate cutting robot

Abstract

The invention discloses a bone auxiliary recognition system based on a vertebral plate cutting robot, and belongs to the technical field of artificial intelligence, the bone auxiliary recognition system comprises a robot subsystem and a bone auxiliary recognition subsystem, and the bone auxiliary recognition subsystem is integrated with an acquisition module, a feature vector extraction module, a data dimension reduction module and a bone recognition prediction module; the robot subsystem comprises a six-degree-of-freedom mechanical arm and a compensation module; the compensation module is used for performing gravity compensation on the force sensor; the acquisition module is used for acquiring cutting force signals under different bones; the feature vector extraction module is used for establishing a vertebral plate cutting unit energy consumption function to extract feature vectors; the data dimension reduction module is used for carrying out secondary extraction and noise reduction processing on the feature vectors; and the bone identification prediction module is used for building a bone identification prediction model by using a support vector machine, training the model by using the data after dimension reduction, and performing parameter optimization on the model by using a particle swarm optimization algorithm. According to the invention, the robot can accurately identify the bone layer cutting state.

Description

Technical Field

[0001] The invention belongs to the technical field of artificial intelligence, and in particular relates to a bone auxiliary identification system based on a lamina cutting robot. Background Art

[0002] The structure of the vertebral lamina is somewhat complex. According to the definition, it is mainly composed of cortical bone and cancellous bone. The cortical bone tissue is wrapped around the surface of the vertebral lamina and is more delicate in microstructure, while the cancellous bone composed of trabecular structure is surrounded by cortical bone in the lamina. There is a difference in density between the two different bone tissues, and the mechanical properties are significantly different. Therefore, during the cutting process, the doctor needs to rely on personal experience to perform the cutting operation. The identification of bone and the amount of cutting depend on the operator's feel, which is difficult and time-consuming. Due to its complexity, even experienced operators find it difficult to guarantee the quality of each cutting, and the imaging radiation dose received by the operator will accumulate with the number of cutting times.

[0003] With the expansion and application of artificial intelligence technology, computer technology, image processing technology, surgical navigation technology, robot control and other technical fields, orthopedic robots are constantly developing. Robot systems with various robot operation forms such as active, semi-active, passive, collaborative and teleoperated have gradually been developed and started to perform certain complex operations, which has good development prospects. However, with the deepening of practice, it is found that orthopedic robots cannot have the same "feel" as operators during the cutting process, resulting in orthopedic robots being unable to match experienced operators during the cutting process. Therefore, orthopedic robots are in urgent need of improving their dynamic perception ability during cutting in the process of vertebral plate cutting for different patients. Summary of the invention

[0004] In view of the above problems existing in the prior art, the present invention proposes a bone auxiliary identification system based on a lamina cutting robot, which has a reasonable design, solves the shortcomings of the prior art and has good effects.

[0005] A bone-assisted identification system based on robot laminectomy, comprising a robot subsystem and a bone-assisted identification subsystem, wherein the bone-assisted identification subsystem integrates an acquisition module, a feature vector extraction module, a data dimension reduction module, and a bone-assisted identification prediction module;

[0006] The robot subsystem includes a six-degree-of-freedom mechanical arm and a compensation module. The six-degree-of-freedom mechanical arm is provided with a force sensor. The compensation module is used to perform gravity compensation on the force sensor to eliminate the influence of errors.

[0007] The acquisition module is used to design an orthogonal cutting test, collect cutting force signals under different bone qualities through a robot subsystem, and form a data set;

[0008] The feature vector extraction module is used to establish a unit energy consumption function for lamina cutting and extract feature vectors from a data set;

[0009] The data dimension reduction module is used to perform secondary extraction and noise reduction processing on the extracted feature vectors using the principal component analysis method to achieve data dimension reduction;

[0010] The bone identification prediction module is used to construct a bone identification prediction model using a support vector machine, train the model using dimension-reduced data, and optimize the model parameters using a particle swarm optimization algorithm.

[0011] Furthermore, in the robot subsystem, an end cutting device is provided at the end of the robotic arm, and the end cutting device includes a bone drill and a six-dimensional force sensor. The bone drill is composed of a DC brushless motor, a connector and a ball-end milling cutter. The connector is installed on the measuring end of the six-dimensional force sensor. The six-dimensional force sensor is fixed on the flange at the end of the robot. The speed of the ball-end milling cutter is controlled by setting the motor speed. The six-dimensional force sensor stores the force signal data collected during the cutting process in an acquisition card and transmits it to the bone auxiliary identification subsystem.

[0012] Furthermore, in the compensation module, performing gravity compensation on the six-dimensional force sensor includes:

[0013] Place the sensor under a known weight for testing, and use the tested reference value to calibrate the displayed value to eliminate the sensor error;

[0014] The center of gravity of the bone drill and the magnitude of the gravity it is subjected to are calculated through coordinate detection and the measurement values ​​of the three-axis accelerometer;

[0015] According to the measured gravity value, accurate force measurement results can be obtained through coordinate conversion.

[0016] Furthermore, calculating the center of gravity of the bone drill and the magnitude of the gravity it receives specifically includes:

[0017] During the bone cutting process, the force F measured by the six-dimensional force sensor includes the bone cutting force f y , the gravity of the bone drill f g , sensor zero point f0, the expression is:

[0018] F=f y +f g +f0 (1)

[0019] The center of gravity coordinates of the bone drill in the six-dimensional force sensor coordinate system are (x, y, z), and the bone drill gravity is set at the same time. S The components of G in the directions of the three coordinate axes are G x , G y , G zThe acting moments are M gx 、M gy 、M gz , we get the following formula:

[0020]

[0021] The feedback data measured from the six-dimensional force sensor is set as Fx, Fy, Fz, and the three torque components are set as M x 、M y 、M z , when the tool is unloaded, only the tool gravity and zero value are affected. Combining equations (1) and (2), we get:

[0022]

[0023] Substituting formula (2) into formula (3), we can obtain the relationship between the torque generated by the tool gravity and the torque data measured by the six-dimensional force sensor coordinate system:

[0024]

[0025] Among them: F x0 、F y0 、F z0 、M x0 、M y0 、M z0 All are the force data of the six-dimensional force sensor under no-load condition, which are constants;

[0026] Substitute the constant term in equation (4) to obtain:

[0027]

[0028] Combining equations (4) and (5), we can get the following equation in matrix form:

[0029]

[0030] Take k different cutting postures to operate the robot, k ≥ 3, and at least 3 positions and postures must have sensor end pointing vectors that are not coplanar, so as to obtain the following k groups of six-dimensional sensor data, denoted as M xi 、M yi 、M zi , i=1,...,k, the expression is:

[0031]

[0032] Solve equation (7) to calculate the coordinates (x, y, z) of the bone drill's center of gravity in the six-dimensional force sensor coordinate system and the constant terms c1, c2, c3.

[0033] Furthermore, according to the measured gravity value, accurate force measurement results are obtained through coordinate transformation, including:

[0034] In the robot base coordinate system, the negative direction of the z-axis is set as the direction of gravity, and the bone drill gravity is expressed as B G. Since the robot base is fixed, B G is a constant, expressed as:

[0035]

[0036] Wherein, g is the gravity value of the bone drill;

[0037] According to the conversion relationship, the bone drill gravity in the user coordinate system is U G is converted into the six-dimensional force sensor coordinate system to obtain S G, the expression is:

[0038]

[0039] in, is the rotation matrix between the six-dimensional force sensor coordinate system and the user coordinate system, is the rotation matrix between the robot base coordinate system and the user coordinate system; and It only involves rotation around the X-axis and Y-axis. Based on the definition of Euler angles, let the rotation angle around the X-axis be U and the rotation angle around the Y-axis be V, then the rotation matrix is The expression is:

[0040]

[0041] Substituting equations (8) and (10) into equation (9), we obtain:

[0042]

[0043] Combining equation (3), substituting equation (11) into equation (11) yields:

[0044]

[0045] make:

[0046]

[0047] Combining equations (12) and (13), we can change the equation to:

[0048]

[0049] Where I is a 3×3 identity matrix;

[0050] Take k different postures in equation (14), and record the feedback data measured by the six-dimensional force sensor under different postures as F xj 、F yj 、F zj , j=1,...,k, the constructed equation group is:

[0051]

[0052] Where: represents the rotation matrix between the six-dimensional force sensor coordinate system and the user coordinate system in the jth posture, j = 1,...,k;

[0053] Solving equation (15), we get b x 、b y 、b z and F x0 、F y0 、F z0 ;

[0054] Combining formula (5), we get the zero point of the sensor torque as:

[0055]

[0056] According to formula (13), the tool gravity g is obtained as:

[0057]

[0058] According to formula (13), the rotation angles U and V between the coordinate systems are:

[0059]

[0060] Furthermore, the acquisition module includes:

[0061] The vertebrae of several different pigs were selected as experimental objects and separated into different lamina bones; the lamina bones are divided into three parts: outer cortical bone, cancellous bone, and inner cortical bone;

[0062] In the bone cutting test, an orthogonal test variation level table with feed speed, motor spindle speed, bone layer cutting depth, and bone density parameters as factors was set;

[0063] Cutting force signals under different cutting parameters and different bone density layers are collected to form a data set.

[0064] Furthermore, in the feature vector extraction module, the energy consumption function in the cutting process is established as:

[0065] E c =E y -Q fy (19)

[0066] Among them, E c is the total energy consumed in transverse cutting, E y is the lateral feed energy of the milling cutter, Q fy The energy consumed by the friction of the transverse feed;

[0067] The volume of bone layer cut during the interval is V y , energy consumption function per unit volume W y for:

[0068]

[0069] but:

[0070] E y =T·F ya ·v (22)

[0071] Q fy =μ0·T·F za ·v (24)

[0072] Where: F ya is the average value of the lateral cutting force in the cutting cycle T, v is the lateral feed speed of the tool, F za is the average longitudinal cutting force within the cutting cycle T, μ0 is the friction coefficient;

[0073] The volume of bone layer cut in a cycle V y :

[0074]

[0075] Where d0 is the diameter of the tool, d a is the cutting depth when cutting the bone layer, α is the included angle,

[0076] Based on the above formula, the calculation formula of the unit energy consumption function within the cutting cycle T is as follows:

[0077]

[0078] Extract two independent eigenvalues ​​of the unit energy consumption function to form a new eigenvector E T =(E T1 , E T2 ) T , as shown below:

[0079]

[0080] Furthermore, in the data dimension reduction module, the lateral cutting force F is introduced yThe cutting parameters including feed speed v, cutting depth d and the two-dimensional feature vector constructed based on the unit energy consumption function form a five-dimensional vector for principal component analysis. By controlling the size of the cumulative contribution rate and determining the value of the number of principal components m, the features are combined and extracted.

[0081] Furthermore, in the bone identification prediction module, the data after dimension reduction is normalized, and the expression is:

[0082]

[0083] Where: x l ' is the normalized value, x l is the lth value in the data sequence, x min is the minimum value of the data sequence, x max is the maximum value of the data series.

[0084] The normalized feature vector is used as the training data of the bone recognition model, and the bone recognition model training based on the support vector machine is carried out, and the kernel function is selected as the RBF kernel function;

[0085] The particle swarm optimization algorithm is used to optimize the penalty factor C1 and the kernel function γ.

[0086] Beneficial technical effects brought by the present invention:

[0087] In order to improve the dynamic perception ability of the orthopedic robot, the present invention takes into account the differentiated effects of bone density of different individuals, collects signals through orthogonal cutting experiments, performs gravity compensation and calibration calculations based on the signal acquisition problem of the six-dimensional force sensor, eliminates the influence of its own gravity, and lays the foundation for the subsequent acquisition of real force signals. A stable unit energy consumption function eigenvector is constructed, and the particle swarm optimization algorithm is used to optimize the machine learning model to improve the accuracy of bone identification. BRIEF DESCRIPTION OF THE DRAWINGS

[0088] Figure 1 It is a schematic diagram of a bone auxiliary identification system based on a laminectomy robot in the present invention;

[0089] Figure 2 This is a diagram showing the effect of gravity compensation in the present invention;

[0090] Figure 3 It is a schematic diagram of the six-dimensional force sensor coordinate system and the user coordinate system in the present invention;

[0091] Figure 4 It is a schematic diagram of the contact of the cutting tool in the present invention;

[0092] Figure 5 It is a schematic diagram of the bone feature extraction and processing process in the present invention;

[0093] Figure 6 This is a recognition effect diagram based on feature extraction in the present invention;

[0094] Figure 7 This is a diagram of the model recognition effect after optimization and improvement in the present invention; DETAILED DESCRIPTION

[0095] The specific implementation of the present invention is further described below in conjunction with specific embodiments:

[0096] A bone-assisted identification system based on a lamina cutting robot, such as Figure 1 As shown, it includes a robot subsystem and a control subsystem, and the control subsystem integrates an acquisition module, a feature vector extraction module, a data dimension reduction module, and a bone identification and prediction module;

[0097] The robot subsystem includes a six-degree-of-freedom robotic arm and a compensation module. The six-degree-of-freedom robotic arm is provided with a force sensor. The compensation module is used to perform gravity compensation on the force sensor to eliminate the influence of errors.

[0098] The acquisition module is used to design orthogonal cutting experiments, collect cutting force signals under different bone qualities through the robot subsystem, and form a data set;

[0099] A feature vector extraction module is used to establish a unit energy consumption function for lamina cutting and extract feature vectors from a data set;

[0100] The data dimension reduction module is used to perform secondary extraction and noise reduction on the extracted feature vectors using the principal component analysis method to achieve data dimension reduction;

[0101] The bone identification prediction module is used to construct a bone identification prediction model using a support vector machine, train the model using dimensionality-reduced data, and optimize the model parameters using a particle swarm optimization algorithm.

[0102] Specifically, in the robot subsystem, an end cutting device is provided at the end of the robot arm, and the end cutting device includes a bone drill and a six-dimensional force sensor. The bone drill is composed of a DC brushless motor, a connector and a ball-end milling cutter. The connector is installed at the measuring end of the six-dimensional force sensor. The six-dimensional force sensor is fixed on the flange at the end of the robot. The speed of the ball-end milling cutter is controlled by setting the motor speed. The six-dimensional force sensor stores the force signal data collected during the cutting process into an acquisition card and transmits it to the bone auxiliary identification subsystem.

[0103] In this embodiment, the force sensor model selected is the six-dimensional force sensor KWR75B of Kunwei Company, which can collect force and torque signals in three directions of Fx, Fy, and Fz, and the information collection accuracy can reach 0.1% FS.

[0104] Specifically, in the compensation module, gravity compensation for the six-dimensional force sensor includes:

[0105] S1. In order to ensure the accuracy of the sensor measurement results, the sensor is placed under a known weight for testing, and the displayed value is calibrated using the tested reference value to eliminate the sensor error;

[0106] S2. Calculate the center of gravity of the bone drill and the magnitude of the gravity it is subjected to through coordinate detection and the measurement value of the three-axis accelerometer;

[0107] During the bone cutting process, the force F measured by the six-dimensional force sensor includes the bone cutting force f y , the gravity of the bone drill f g , sensor zero point f0, the expression is:

[0108] F=f y +f g +f0 (1)

[0109] In the formula, each force measured is in the sensor coordinate system, and the sensor zero point f0 represents the data displayed when the sensor is turned on and under no-load conditions;

[0110] The center of gravity coordinates of the bone drill in the six-dimensional force sensor coordinate system are (x, y, z), and the bone drill gravity is set at the same time. S The components of G in the directions of the three coordinate axes are G x , G y , G z The acting moments are M gx 、M gy 、M gz , combined with the force diagram, according to the relationship between force and moment, the following formula is obtained:

[0111]

[0112] The feedback data measured from the six-dimensional force sensor is set as Fx, Fy, Fz, and the three torque components are set as M x 、M y 、M z , when the tool is unloaded, only the tool gravity and zero value are affected. Combining equations (1) and (2), we get:

[0113]

[0114] Substituting formula (2) into formula (3), the relationship between the torque generated by the tool gravity and the torque data measured by the six-dimensional force sensor coordinate system is shown as follows:

[0115]

[0116] Among them: F x0 、F y0 、F z0 、M x0 、M y0 、M z0 All of them are the force data of the six-dimensional force sensor under no-load, which is a constant. No-load means that there is no cutting force when the ball end mill is not cutting.

[0117] To facilitate calculation, replace the constant term in equation (4) so ​​that:

[0118]

[0119] Combining equations (4) and (5), we can get the following equation in matrix form:

[0120]

[0121] To solve the equation, we take k different cutting postures to operate the robot, k ≥ 3, and at least three positions and postures must have the sensor end pointing vectors that are not coplanar, or as close to coplanar as possible, so as to obtain the following k groups of six-dimensional sensor data, recorded as M xi 、M yi 、M zi , i=1,...,k, the expression is:

[0122]

[0123] Solve equation (7) to calculate the coordinates (x, y, z) of the bone drill's center of gravity in the six-dimensional force sensor coordinate system and the constant terms c1, c2, c3.

[0124] like Figure 2 As shown in the figure, before gravity compensation, the force data measured by the sensor is roughly in the 5N range. Since the connecting flange, cutting fixture, cutting tool and other components are installed below the sensor, the gravity of the above devices is measured. After the gravity coordinate position is confirmed and converted, the calibrated sensor no-load value is restored to near 0. The lateral feed movement involved in the cutting process cannot completely eliminate the fluctuation of the data, but the influence of the cutting experiment conducted under the unified standard on the signal acquisition is almost negligible, ensuring the reference value of the experimental data.

[0125] S3. According to the measured gravity value, accurate force measurement results are obtained through coordinate conversion;

[0126] In the robot base coordinate system, the negative direction of the z-axis is set as the direction of gravity, and the bone drill gravity is expressed as B G. Since the robot base is fixed, B G is a constant, expressed as:

[0127]

[0128] Wherein, g is the gravity value of the bone drill;

[0129] like Figure 3 As shown in the figure, the six-dimensional force sensor coordinate system is recorded as S-X0Y0Z0, and the user coordinate system is recorded as U-X1Y1Z1. The user coordinate system can be obtained by rotating the sensor coordinate system S-X0Y0Z0 around the X-axis; according to the conversion relationship, the bone drill gravity in the user coordinate system is U G is converted into the six-dimensional force sensor coordinate system to obtain S G, the expression is:

[0130]

[0131] in, is the rotation matrix between the sensor coordinate system and the user coordinate system, is the rotation matrix between the robot base coordinate system and the user coordinate system; and It only involves rotation around the X-axis and Y-axis. Based on the definition of Euler angles, let the rotation angle around the X-axis be U and the rotation angle around the Y-axis be V, then the rotation matrix is The expression is:

[0132]

[0133] Substituting equations (8) and (10) into equation (9), we obtain:

[0134]

[0135] Combining equation (3), substituting equation (11) into equation (11) yields:

[0136]

[0137] make:

[0138]

[0139] Combining equations (12) and (13), we can change the equation to:

[0140]

[0141] Where I is a 3×3 identity matrix;

[0142] Take k different postures in equation (14), and record the feedback data measured by the six-dimensional force sensor under different postures as F xj 、F yj 、F zj , j=1,...,k, the constructed equation group is:

[0143]

[0144] Where: represents the rotation matrix between the six-dimensional force sensor coordinate system and the user coordinate system in the jth posture, j = 1,...,k;

[0145] Solving equation (15), we get b x 、b y 、b z and F x0 、F y0 、F z0 ;

[0146] Combining formula (5), we get the zero point of the sensor torque as:

[0147]

[0148] According to formula (13), the tool gravity g is obtained as:

[0149]

[0150] According to formula (13), the rotation angles U and V between the coordinate systems can be obtained:

[0151]

[0152] Specifically, the acquisition module includes:

[0153] Pig bones have certain similarities with human bones in terms of bone density and mechanical properties. We selected several pieces of vertebrae from different pigs as experimental objects and split them into different lamina bones. The lamina bones are divided into three parts: outer cortical bone, cancellous bone, and inner cortical bone.

[0154] In the bone cutting test, an orthogonal test variation level table with feed speed, motor spindle speed, bone layer cutting depth, and bone density parameters as factors is set, as shown in Table 1:

[0155] Table 1 Bone cutting parameter change level table

[0156]

[0157] Cutting force signals under different cutting parameters and different bone density layers are collected to form a data set.

[0158] For the sensitivity analysis of various cutting parameters including bone density ρ, cutting depth d, feed speed v, and spindle speed n, a lateral cutting force prediction model based on different cutting parameters was established by response surface methodology, and the lateral cutting force value was used as the objective function of global sensitivity analysis, thereby calculating the global sensitivity S of each cutting parameter.Ti , and then the contribution rate of each cutting parameter to the output model is obtained. Through the sensitivity analysis of cutting parameters, it is found that the parameter with the greatest impact on cutting force is bone density, which will be used as the output judgment item of the bone recognition model, followed by the impact of feed speed on cutting force, followed by the impact of tool speed and cutting depth on cutting force.

[0159] According to the contribution of different parameters to the cutting force signal, the lateral cutting force signal F after noise reduction is y , feed speed v, cortical bone cutting depth d1, and cancellous bone cutting depth d2 are used as parameters to establish a unit energy consumption function that reflects the bone characteristics of the bone layer.

[0160] In the feature vector extraction module, the energy consumption function in the cutting process is established as:

[0161] E c =E y -Q fy (19)

[0162] Among them, E c is the total energy consumed in transverse cutting, E y is the lateral feed energy of the milling cutter, Q fy The energy consumed by the friction of the transverse feed;

[0163] In order to improve the feedback efficiency during the cutting process, the cutting process is subdivided into quantitative time intervals, and the volume of the bone layer cut during the interval is V y , energy consumption function per unit volume W y for:

[0164]

[0165] (1) Transverse feed energy E y :

[0166]

[0167] Where: F yi It represents the instantaneous cutting force during the cutting process, v represents the tool lateral feed speed set in the cutting process, and T is the interval cycle time;

[0168] In order to facilitate calculation, according to the sampling period of the sensor, the cutting force signal F within the period T is yi Take it as the average value F in period T ya , we get the following formula (22):

[0169] but:

[0170] E y =T·F ya·v (22)

[0171] (2) Energy consumed by transverse feed friction Q fy :

[0172]

[0173] Where: F zi Indicates the instantaneous longitudinal cutting force during the cutting process.

[0174] The longitudinal cutting force F within the cutting cycle T is obtained by averaging method. za , and then get the following formula:

[0175] Q fy =μ·T·F za ·v (24)

[0176] Where: μ is the cutting coefficient between the tool and the bone during the cutting process;

[0177] (3) In order to obtain the cutting coefficient between the tool and the bone, a friction experiment is used to solve it. The tool speed is set to 0, and the cutting force data is obtained through the relative sliding between the tool and the bone, and then the friction coefficient μ0 is calculated. In subsequent calculations, μ0 is used to represent it. The calculation formula of the friction coefficient μ0 is shown in equations (25) and (26):

[0178] F y0 =μ0·F z0 (25)

[0179]

[0180] Where: F y0 is the transverse feed force measured along the y direction when the tool speed is 0, F z0 is the pressure applied by the cutter to the bone block along the z direction, which is collected by the sensor. The value of the friction coefficient μ0 obtained through experiments in this embodiment is 0.1206.

[0181] Substituting μ0 obtained in equation (26) into equation (24) yields:

[0182] Q fy =μ0·T·F za ·v (27)

[0183] (4) Volume of bone layer cut during the cycle V y :

[0184] like Figure 4 As shown, during the cutting process, the tool is mainly cut by the cutting edge at the bottom, and the main area of ​​its cross section is the bottom of the fan-shaped area. Its main expression is:

[0185] S y =S0-S1 (28)

[0186] Where: S y represents the bone layer cutting surface area, S0 is the fan-shaped area in the figure, and S1 is the triangular area of ​​the upper part of the bone layer cutting surface area;

[0187] The calculation formula of S1 is:

[0188]

[0189] Where: d0 represents the diameter of the tool, d a Indicates the cutting depth when cutting bone layers;

[0190] The calculation method of the angle α is expressed by the relationship between the tool diameter and the cutting depth, that is:

[0191]

[0192] The formula for the sector area in the figure is:

[0193]

[0194] The cutting cross-sectional area formula obtained from the above formula is:

[0195]

[0196] The volume of bone layer cut in a cycle V y :

[0197]

[0198] Based on the above formula, the calculation formula of the unit energy consumption function within the cutting cycle T is as follows:

[0199]

[0200] like Figure 5 As shown, two independent eigenvalues ​​of the unit energy consumption function are extracted to form a new eigenvector E T =(E T1 , E T2 ) T , as shown below:

[0201]

[0202] In the data dimension reduction module, the lateral cutting force F is introduced yThe cutting parameters including feed speed v, cutting depth d and the two-dimensional feature vector constructed based on the unit energy consumption function form a five-dimensional vector for principal component analysis. By controlling the size of the cumulative contribution rate and determining the value of the number of principal components m, the features are combined and extracted.

[0203] As shown in Table 2, the results of principal component analysis show that the cumulative contribution rate of the first four principal components has reached 99.87%. Therefore, the value of m is determined to be 4, and the first four principal components are selected to form new feature parameters as new feature vector parameters for subsequent support vector machine classification model training.

[0204] Table 2 Contribution rate of each principal component

[0205]

[0206]

[0207] In the bone identification prediction module, the data after dimensionality reduction is normalized, and the expression is:

[0208]

[0209] Where: x l ' is the normalized value, x l is the lth value in the data sequence, x min is the minimum value of the data sequence, x max is the maximum value of the data series.

[0210] The normalized feature vector is used as the training data of the bone recognition model. The libsvm open source system is used to load the libsvm toolbox into the Matlab software through the toolbox function in Matlab. The parameters are adjusted based on the Matlab software to train the bone recognition model based on the support vector machine. The kernel function selects the RBF kernel function.

[0211] In order to further optimize the classification effect of the support vector machine, the particle swarm optimization algorithm is used to optimize the penalty factor C1 and the kernel function γ so that the classifier can achieve the best prediction effect under cross validation.

[0212] In order to compare the effects of feature extraction and particle swarm optimization, the data was further verified and analyzed. In order to facilitate intuitive observation, the bone recognition result display interface displayed the recognition results according to the set cutting trajectory. Figure 6 , Figure 7The recognition effects of the characteristic signals before and after the optimization process are shown. From the comparison chart, when the first layer is cut, the result is cortical bone cutting. When using cortical bone signals for identification, the signal recognition of the two models is accurate and very stable; when the second layer is cut, the results of the two recognition modes have different degrees of deviation, but the recognition results of the optimized model are significantly improved compared with those before optimization; when cutting along the third layer of cutting trajectory, the recognition results of the model after the final optimization are stable, with a high accuracy rate, which is better than the recognition mode before optimization.

[0213] In order to avoid the contingency of data, experiments with multiple groups of samples were conducted and the overall experimental results were obtained, as shown in Tables 3 and 4. From the data analysis in the tables, it can be seen that for the optimized and improved bone recognition model, when cutting cortical bone, due to the hardness of cortical bone and obvious characteristics, the recognition accuracy of the cortical bone layer in the three groups of experiments reached 100%. In the recognition of cancellous bone and transition area, due to the mixed nature of the bone layer and the acquisition error of the experimental signal, the recognition accuracy of the transition area is the bone layer with the lowest recognition accuracy. In two of the experiments, the recognition results of some cancellous bones were biased. According to the recognition results, it was found that they were all misjudgments between cancellous bone and transition area. The comprehensive accuracy was measured to be above 90%, which was much higher than before optimization, and a good recognition and classification effect was achieved. Based on the monitoring results, as the cutting progressed, some signal recognition errors appeared in the recognition results of the transition layer and cancellous bone layer, and the recognition difficulty was relatively large. However, according to the structural distribution characteristics of the innermost distribution of the lamina as cortical bone, combined with the overall cutting trend, it will not affect the safety of cutting.

[0214] Table 3 Bone recognition accuracy before optimization and improvement

[0215]

[0216]

[0217] Table 4 Bone recognition accuracy after optimization and improvement

[0218]

[0219] Through multiple cutting trajectory planning experiments and analysis of the recognition accuracy during each cutting, it can be seen that in the bone recognition model based on particle swarm optimization support vector machine constructed in the present invention, by extracting the established energy consumption function features, the bone layer recognition signal in the robot lamina cutting process can be effectively improved, thereby greatly improving the dynamic perception ability of the orthopedic robot.

[0220] Of course, the above description is not a limitation of the present invention, and the present invention is not limited to the above examples. Changes, modifications, additions or substitutions made by technicians in this technical field within the essential scope of the present invention should also fall within the protection scope of the present invention.

Claims

1. A bone-assisted identification system based on a laminectomy robot, characterized in that: It includes a robot subsystem and a bone auxiliary identification subsystem, wherein the bone auxiliary identification subsystem integrates an acquisition module, a feature vector extraction module, a data dimension reduction module, and a bone identification prediction module; The robot subsystem includes a six-degree-of-freedom mechanical arm and a compensation module. The six-degree-of-freedom mechanical arm is provided with a force sensor. The compensation module is used to perform gravity compensation on the force sensor to eliminate the influence of errors. The acquisition module is used to design an orthogonal cutting test, collect cutting force signals under different bone qualities through a robot subsystem, and form a data set; The feature vector extraction module is used to establish a unit energy consumption function for lamina cutting and extract feature vectors from a data set; The data dimension reduction module is used to perform secondary extraction and noise reduction processing on the extracted feature vectors using the principal component analysis method to achieve data dimension reduction; The bone identification prediction module is used to construct a bone identification prediction model using a support vector machine, train the model using dimension-reduced data, and optimize the model parameters using a particle swarm optimization algorithm.

2. The bone-assisted identification system based on a laminectomy robot according to claim 1 is characterized in that: In the robot subsystem, an end cutting device is provided at the end of the robot arm, and the end cutting device includes a bone drill and a six-dimensional force sensor. The bone drill is composed of a DC brushless motor, a connector and a ball-end milling cutter. The connector is installed on the measuring end of the six-dimensional force sensor. The six-dimensional force sensor is fixed on the flange at the end of the robot. The speed of the ball-end milling cutter is controlled by setting the motor speed. The six-dimensional force sensor stores the force signal data collected during the cutting process in an acquisition card and transmits it to the bone auxiliary identification subsystem.

3. The bone auxiliary identification system based on the laminectomy robot according to claim 2 is characterized in that: In the compensation module, performing gravity compensation on the six-dimensional force sensor includes: Place the sensor under a known weight for testing, and use the tested reference value to calibrate the displayed value to eliminate the sensor error; The center of gravity of the bone drill and the magnitude of the gravity it is subjected to are calculated through coordinate detection and the measurement values ​​of the three-axis accelerometer; According to the measured gravity value, accurate force measurement results can be obtained through coordinate conversion.

4. The bone auxiliary identification system based on the laminectomy robot according to claim 3 is characterized in that: Calculating the center of gravity of the bone drill and the gravity it is subjected to specifically includes: During the bone cutting process, the force F measured by the six-dimensional force sensor includes the bone cutting force f y , the gravity of the bone drill f g , sensor zero point f0, the expression is: F=f y +f g +f0 (1) The center of gravity coordinates of the bone drill in the six-dimensional force sensor coordinate system are (x, y, z), and the bone drill gravity is set at the same time. S The components of G in the directions of the three coordinate axes are G x , G y , G z The acting moments are M gx 、M gy 、M gz , we get the following formula: The feedback data measured from the six-dimensional force sensor is set as Fx, Fy, Fz, and the three torque components are set as M x 、M y 、M z , when the tool is unloaded, only the tool gravity and zero value are affected. Combining equations (1) and (2), we get: Substituting formula (2) into formula (3), we can obtain the relationship between the torque generated by the tool gravity and the torque data measured by the six-dimensional force sensor coordinate system: Among them: F x0 、F y0 、F z0 、M x0 、M y0 、M z0 All are the force data of the six-dimensional force sensor under no-load condition, which are constants; Substitute the constant term in equation (4) to obtain: Combining equations (4) and (5), we can get the following equation in matrix form: Take k different cutting postures to operate the robot, k ≥ 3, and at least 3 positions and postures must have sensor end pointing vectors that are not coplanar, so as to obtain the following k groups of six-dimensional sensor data, denoted as M xi 、M yi 、M zi , i=1,...,k, the expression is: Solve equation (7) to calculate the coordinates (x, y, z) of the bone drill's center of gravity in the six-dimensional force sensor coordinate system and the constant terms c1, c2, c3.

5. The bone auxiliary identification system based on the laminectomy robot according to claim 4 is characterized in that: According to the measured gravity value, through coordinate conversion, accurate force measurement results are obtained, including: In the robot base coordinate system, the negative direction of the z-axis is set as the direction of gravity, and the bone drill gravity is expressed as B G. Since the robot base is fixed, B G is a constant, expressed as: Wherein, g is the gravity value of the bone drill; According to the conversion relationship, the bone drill gravity in the user coordinate system is U G is converted into the six-dimensional force sensor coordinate system and obtained S G, the expression is: in, is the rotation matrix between the six-dimensional force sensor coordinate system and the user coordinate system, is the rotation matrix between the robot base coordinate system and the user coordinate system; and It only involves rotation around the X-axis and Y-axis. Based on the definition of Euler angles, let the rotation angle around the X-axis be U and the rotation angle around the Y-axis be V, then the rotation matrix is The expression is: Substituting equations (8) and (10) into equation (9), we obtain: Combining equation (3), substituting equation (11) into equation (11) yields: make: Combining equations (12) and (13), we can change the equation to: Where I is a 3×3 identity matrix; Take k different postures in equation (14), and record the feedback data measured by the six-dimensional force sensor under different postures as F xj 、F yj 、F zj , j=1,...,k, the constructed equation group is: Where: represents the rotation matrix between the six-dimensional force sensor coordinate system and the user coordinate system in the jth posture, j = 1,...,k; Solving equation (15), we get b x 、b y 、b z and F x0 、F y0 、F z0 ; Combining formula (5), we get the zero point of the sensor torque as: According to formula (13), the tool gravity g is obtained as: According to formula (13), the rotation angles U and V between the coordinate systems are:

6. The bone auxiliary identification system based on the laminectomy robot according to claim 5 is characterized in that: The acquisition module comprises: The vertebrae of several different pigs were selected as experimental objects and separated into different lamina bones; the lamina bones are divided into three parts: outer cortical bone, cancellous bone, and inner cortical bone; In the bone cutting test, an orthogonal test variation level table with feed speed, motor spindle speed, bone layer cutting depth, and bone density parameters as factors was set; Cutting force signals under different cutting parameters and different bone density layers are collected to form a data set.

7. The bone auxiliary identification system based on the laminectomy robot according to claim 6 is characterized in that: In the feature vector extraction module, the energy consumption function in the cutting process is established as: AND c =And y -Q fy (19) Among them, E c is the total energy consumed in transverse cutting, E y is the lateral feed energy of the milling cutter, Q fy The energy consumed by the friction of transverse feed; The volume of bone layer cut during the interval is V y , energy consumption function per unit volume W y for: but: E y =T·F ya ·v (22) Q fy =μ0·T·F za ·v (24) Where: F ya is the average value of the lateral cutting force in the cutting cycle T, v is the lateral feed speed of the tool, F za is the average longitudinal cutting force within the cutting cycle T, μ0 is the friction coefficient; The volume of bone layer cut in a cycle V y : Where d0 is the diameter of the tool, d a is the cutting depth when cutting the bone layer, α is the included angle, Based on the above formula, the calculation formula of the unit energy consumption function within the cutting cycle T is as follows: Extract two independent eigenvalues ​​of the unit energy consumption function to form a new eigenvector E T =(E T1 , E T2 ) T , as shown below:

8. The bone auxiliary identification system based on the laminectomy robot according to claim 7 is characterized in that: In the data dimension reduction module, the lateral cutting force F is introduced. y The cutting parameters including feed speed v, cutting depth d and the two-dimensional feature vector constructed based on the unit energy consumption function form a five-dimensional vector for principal component analysis. By controlling the size of the cumulative contribution rate and determining the value of the number of principal components m, the features are combined and extracted.

9. The bone auxiliary identification system based on the laminectomy robot according to claim 8, characterized in that: In the bone identification prediction module, the data after dimension reduction is normalized, and the expression is: Where: x l ' is the normalized value, x l is the lth value in the data sequence, x min is the minimum value of the data sequence, x max is the maximum value of the data sequence; The normalized feature vector is used as the training data of the bone recognition model, and the bone recognition model training based on the support vector machine is carried out, and the kernel function is selected as the RBF kernel function; The particle swarm optimization algorithm is used to optimize the penalty factor C1 and the kernel function γ.

Images