Bone grinding file component capable of adaptively optimizing cutting temperature, control method and application

Abstract

The present invention discloses a bone grinding file component, a control method and an application capable of adaptively optimizing cutting temperature; the present invention relates to the technical field of surgical instruments for artificial joint replacement; the invention comprises a housing and a transmission component built therein and driven to rotate based on an external rotating member, wherein the transmission component converts a rotational action into a feeding action and drives a bone file for grinding bones to perform swing feeding; an input variable Input is input into a single-output PID neural network SPIDNN; a current population is subjected to operations such as crossover and mutation based on a degree value DV through a differential evolution algorithm to generate a new population; torsional vibration force is generated in the form of reverse same frequency during the feeding and retraction of the bone file, and the magnitude of the cutting force is effectively reduced through the superposition effect of vibration, thereby achieving the cooling of the cutting temperature; the artificial intelligence technology (single-output PID neural network SPIDNN) adopted realizes a mechanism for adaptively adjusting the frequency, which can ensure that the optimal cutting state is always maintained during the cutting process.

Description

Technical Field

[0001] The present invention relates to the technical field of surgical instruments for artificial joint replacement, and in particular to grinding the surfaces of soft tissue and bone tissue and optimizing their cutting rasp and surface temperature during arthroscopic surgery or open surgery, and in particular to a bone grinding rasp assembly, control method and application that can adaptively optimize cutting temperature. Background Art

[0002] Bone rasps used in orthopedic surgery include oscillating saws (oscillating rasps), osteotome and acetabular rasps, etc., which play an indispensable role in orthopedic surgery. Through their specific design and materials, they can effectively remove irregular parts of the bone surface, making the bone structure smoother, which is conducive to subsequent surgical operations or implant fixation. The general functions of bone rasps include but are not limited to removing sharp edges of fracture ends, reducing postoperative pain and soft tissue damage. Or in operations such as joint replacement, by trimming the joint surface, a suitable interface is provided for the installation of artificial joints. And it can be used to remove diseased bone tissue in operations such as bone tumor resection.

[0003] The mechanical properties of bone tissue vary in different directions. For example, cortical bone has greater stiffness and stress resistance than cancellous bone, but its strain is smaller. This anisotropy causes bone tissue to exhibit different mechanical responses when subjected to loads in different directions. At high loading rates, the strength and stiffness of bone tissue will be correspondingly improved. This phenomenon is due to the fact that the microstructure inside the bone tissue is more closely arranged at high strain rates, thereby improving its ability to resist damage. However, during bone processing, this strain rate sensitivity will lead to increased energy consumption, which will have an adverse effect on the bone cutting process. At the same time, bone tissue is more sensitive to temperature, but its heat dissipation performance is poor. When bone tissue is exposed to a temperature environment above 47°C for more than one minute, its regeneration ability will be affected; temperatures exceeding 56°C may directly cause osteocyte necrosis.

[0004] That is to say, excessive cutting frequency and the frictional heat generated by it will have an adverse effect on surgical operations. To this end, many existing technologies attempt to optimize this phenomenon, such as:

[0005] (1) The Chinese utility model patent with publication number: CN209360804U discloses a low cutting temperature acetabular rasp (publication date September 10, 2019), which evenly distributes the cutting teeth of the acetabular rasp on the spherical spiral line at equal spiral angles, and at the same time opens a cooling through hole on each cutting tooth to further reduce the contact area between the cutting tooth and the bone tissue, thereby reducing the cutting temperature. However, this cooling method is passive and highly dependent on the use environment. If the feed speed or rotation speed is too high, high temperature will still occur, and its cooling effect is limited, and it cannot completely solve the high temperature problem.

[0006] (2) The Chinese invention patent with publication number: CN107753086B discloses a mortar and mortar assembly for joint replacement surgery (publication date: April 20, 2021), which achieves a function similar to water cooling through liquid circulation, thereby reducing the cutting temperature. However, this method will lead to a high level of structural redundancy in the equipment, increase the volume, and have an adverse effect on surgical operations. Therefore, the practicality of this technology has certain limitations, especially when compact and lightweight surgical tools are required.

[0007] (3) The document "Wang Da. Research on vibration minimally invasive bone processing technology for joint replacement [D]. Hebei University of Technology. 2019" confirmed through experimental analysis that within the parameter range of the bone grinder, an increase in the feed speed will lead to an increase in cutting temperature and cutting force. Similarly, an increase in the spindle speed will also increase the cutting force and temperature; at the same time, the document found through research and experiments that the built-in parallel vibration source in the bone grinder will reduce the cutting temperature and cutting force based on the superposition of vibration, and gave a set of parameter combinations with lower osteotomy temperature - "feed speed 4mm / min, spindle speed 40rpm, vibration frequency 49Hz"; but although the built-in vibration source and the superposition effect can not only optimize the cutting effect, but also optimize the cutting temperature under the premise of low equipment redundancy, its fixed parameter range may not necessarily adapt to different bone tissues, and the phenomenon of poor cutting temperature may still exist.

[0008] For example, for acetabulum cutting, experimental analysis shows that the operating mode of "feed speed 4mm / min, spindle speed 40rpm, vibration frequency 49Hz" is the optimal parameter. However, when the acetabulum is cut in practice, the feed speed and vibration frequency on the acetabulum cannot truly adapt to the predetermined values ​​of the parameters.

[0009] To this end, the present invention proposes a bone grinding rasp assembly, a control method and an application that can adaptively optimize the cutting temperature. Summary of the invention

[0010] In view of this, the present invention hopes to provide a bone grinding rasp assembly, a control method and an application that can adaptively optimize the cutting temperature, so as to solve or alleviate the technical problems existing in the prior art, namely: how to further allow the vibration source to adaptively adjust the frequency on the basis of optimizing the cutting temperature based on the built-in vibration source and the superposition effect, so as to adapt to the limitations and interference of the actual surgical environment of the bone tissue, and at least provide a beneficial option for this; the technical solution of the present invention is implemented as follows:

[0011] First, a bone grinding rasp assembly that can adaptively optimize cutting temperature:

[0012] The bone grinding file assembly includes a housing, a transmission assembly located in the housing, and a bone file. The transmission assembly includes a transmission shaft rotatably matched in the housing, and a bevel rotor is rotatably matched on the transmission shaft; the slide rail is divided into two sections, one of which is fixed with a bayonet pin, and the bevel rotor is clamped by the bayonet pin, and the slide rail is slidably matched with a first slide groove in the housing;

[0013] The transmission assembly further includes a forced vibration system; the forced vibration system includes two excitation units; the excitation unit is located below the transmission shaft; the excitation unit includes a servo motor and an eccentric shaft driven by its output shaft; the body of the servo motor is fixed on a section of the slide rail;

[0014] One end surface and another end surface of the other section of the slide rail are fixedly connected to the sleeve and the bone rasp respectively;

[0015] The output shaft of the servo motor close to the bone rasp is also fixed to the inner bush of the needle roller bearing, and the outer bush of the needle roller bearing is fixed in the sleeve;

[0016] A second sliding groove is provided in the housing, and the excitation unit is accommodated in the second sliding groove.

[0017] In one implementation manner, a counterweight is installed on the transmission shaft directly above the vibration excitation unit.

[0018] In one embodiment, a water inlet and a water outlet are provided on the bone rasp.

[0019] In the second aspect, a control method for a bone grinding rasp assembly capable of adaptively optimizing cutting temperature:

[0020] 1. Overview:

[0021] This scheme aims to combine the above-mentioned bone rasp and use advanced control algorithms and optimization techniques to achieve adaptive optimization of cutting temperature control of bone rasp components in orthopedic surgery. By constructing a single output PID neural network (SPIDNN) containing proportional, integral, and differential neurons, and combining the differential evolution algorithm for weight and parameter optimization, independent control of the two excitation units can be achieved. During the operation, the system collects the parameter data of the bone rasp in real time, and adaptively adjusts the control parameters of the PIDNN based on these data, thereby optimizing the cutting temperature and cutting force, and improving the accuracy and safety of the operation. In addition, through continuous iterative optimization and real-time adjustment, the system can gradually adapt to different surgical needs and bone tissue characteristics. In summary, through advanced control algorithms and optimization techniques, precise control of bone rasp components in orthopedic surgery can be achieved.

[0022] (II) Technical solution:

[0023] In order to achieve the above technical objectives, the input variable Input is first established, including the set vibration frequency F when the bone rasp 3 is extended. 1S and the actual vibration frequency F 2S , and the set vibration frequency F of the bone rasp 3 in the retracted state 1R and the actual vibration frequency F 2R ; Then the present invention chooses to perform the following steps:

[0024] 2.1 Step S1, initialize the PID neural network SPIDNN:

[0025] When the device is started, a single output PID neural network SPIDNN of each vibration source (excitation unit) in the forced vibration system is activated, and the PID neural network SPIDNN includes a neuron of proportion P, an neuron of integral I, and a neuron of differential D;

[0026] 2.1.1 Step S100, initializing neurons:

[0027] (1) Proportional neuron P: P(t) = Kp·e(t);

[0028] Among them, K p is the proportionality coefficient, e(t) = F1-F2 is the error between the set vibration frequency and the actual vibration frequency (for the extended state, F1 = F 1S , F2=F 2S ; For the retracted state, F1 = F 1R , F2=F 2R ).

[0029] (2) Integrating neuron I:

[0030] Among them, K i is the integration coefficient, b I is the bias of the integrating neuron; y I (t) is the output of the integrating neuron at time t; It is the integral of the error signal e(t) from 0 to t, reflecting the cumulative effect of the error; t is the time unit variable;

[0031] (3) Differentiable neuron D:

[0032] Among them, K d is the differential coefficient, b D is the bias of the differentiating neuron; y D (t) is the output of the differentiable neuron at time t; It is the derivative of the error signal e(t), which reflects the rate of change of the error;

[0033] 2.1.2 Step S101, initialize network output:

[0034] y(t)=w P ·y P (t)+w I ·y I (t)+w D ·y D (t)+b o ;

[0035] Where y(t) is the total output of the PID neural network at time t; w P 、w I and w D are the weights of proportional, integral and differential neurons, respectively, which determine the contribution of each neuron output to the total output of the network; b o is the bias of the output layer, which is used to adjust the benchmark of the total output of the network. It can be set to a small random number to ensure that the initial state of the network has diversity. For example, use uniform distribution or normal distribution to generate these initial values:

[0036] w P ,w I ,w D ~U(-∈,∈) or w P ,w I ,w D ~N(0,σ 2 );

[0037] b I ,b D ,b o ~U(-δ,δ) or b I ,b D ,b o ~N(0,μ2);

[0038] Among them, ∈ and δ are the control initial weights w P 、w I and w D and bias b o A small positive number in the range of 2 and μ are the standard deviations of the normal distribution, respectively.

[0039] 2.2 Step S2, control signal generation:

[0040] Input variable Input is input into the single output PID neural network SPIDNN, according to the internal parameters (P, I and D), and the set vibration frequency F 1S and the actual vibration frequency F 2S The error between E1, or according to the set vibration frequency F 1R and the actual vibration frequency F 2RThe error E2 between them is used to adaptively calculate the corresponding control electrical signals e1 and e2 to control the vibration source of the corresponding forced vibration system and adjust its vibration parameters.

[0041] 2.2.1 Step S200, activate the input layer:

[0042] Receive input variable Input={F 1S ,F 2S ,F 1R ,F 2R};

[0043] At the same time, for the extended state, the calculated error E1 = F 1S -F 2S ; For the retracted state, calculate the error E2 = F 1R -F 2R ;

[0044] 2.2.2 Step S201, activate the hidden layer:

[0045] Contains proportional (P), integral (I) and differential (D) neurons. Each neuron calculates the error e(t) and output y according to the input and internal parameters. I (t) and output y D (t).

[0046] 2.2.3 Step S202, the output layer adaptively calculates the corresponding control electrical signals e1 and e2:

[0047] (1) For the control electrical signal e1 of the extended state:

[0048]

[0049] in, and They respectively represent the outputs of the proportional P neuron, the integral I neuron, and the differential D neuron in the extended state.

[0050] (2) For the control electrical signal e2 in the retracted state:

[0051]

[0052] in, and They represent the outputs of the proportional P neuron, the integral I neuron, and the differential D neuron in the retracted state respectively.

[0053] 2.3 Step S3, autonomous learning and updating (implementing differential evolution algorithm DE optimization):

[0054] Calculate the degree value DV of the control effect of the current single output PID neural network SPIDNN.

[0055] The differential evolution algorithm performs crossover and mutation operations on the current population based on the degree value DV to generate a new population. The best performing individual is selected from the new population to update the weights of the PID neural network SPIDNN, including the weight w P , weight w I and weight w D .

[0056] 2.3.1 Step S300, calculate the degree value DV:

[0057] DV = |TT′|;

[0058] Where T is the actual detected cutting temperature and T′ is the set theoretical temperature. The smaller the DV value, the better the control effect.

[0059] 2.3.2 Step S301, population initialization:

[0060] P0={x1,x2,…,x NP};

[0061] Among them, P0 represents the initial population, NP represents the population size; x i Represents the i-th individual (vector), containing the weight w of the PID neural network SPIDNN P , weight w I and weight w D .

[0062] 2.3.3 Step S302, crossover operation:

[0063] For each individual x i , generate the test vector v i :v i =x r1 +F·(x r2 -x r3 );

[0064] Where r1, r2 and r3 are different random indices and different from i, and F is a scaling factor.

[0065] At the same time, for the test vector v i Perform a bounds check to ensure it is within the valid range.

[0066] 2.3.4 Step S303, select operation:

[0067] Evaluate the trial vector v according to the degree value DV i and the current individual x i performance, and select those with better performance to enter the next generation population

[0068] Among them, DV(x i ) means using the current individual x i (i.e. a specific set of weights w P , weight w I and weight w D ), the degree value between the cutting temperature and the set temperature obtained under the control of the single output PID neural network SPIDNN. DV(v i ) means using the test vector v i (i.e., a set of new weights generated by the crossover and mutation operations of the differential evolution algorithm DE), the degree value between the cutting temperature and the set temperature obtained under the control of the single-output PID neural network SPIDNN.

[0069] Then, we get a new population P′={x1′,x2′,…,x NP '};

[0070] 2.3.5 Step S304, iterative update:

[0071] Repeat steps S302 to S303 until a predetermined number of iterations is reached or a convergence condition is met.

[0072] 2.3.6 Step S305, update weight w P , weight w I and weight w D :

[0073] After optimization by differential evolution algorithm DE, the individual x with the best performance is selected. best , which contains the weight w P , weight w I and weight w D These weight values ​​are assigned to the single output PID neural network SPIDNN to improve its control effect.

[0074] (III) Mechanism for solving technical problems:

[0075] 3.1 Adaptive control strategy:

[0076] The PID neural network (SPIDNN) constructs a single-output PID neural network for each excitation unit. The network can automatically adjust the proportional (P), integral (I), and differential (D) parameters according to the input error (the difference between the set vibration frequency and the actual vibration frequency) and output an accurate control signal.

[0077] At the same time, the differential evolution algorithm is used to continuously optimize the weights and parameters of PIDNN to ensure that the control strategy can be automatically adjusted as the surgical environment changes to achieve the optimal control effect.

[0078] 3.2 Adaptive adjustment of vibration source frequency:

[0079] The present invention adjusts the vibration frequency of the excitation unit in real time according to the control signal output by PIDNN. When encountering situations such as increased hardness of bone tissue and increased cutting force, the system can automatically increase the vibration frequency to maintain cutting efficiency; conversely, it reduces the vibration frequency to reduce cutting heat and wear. Through the real-time feedback mechanism, the system can continuously monitor and adjust the vibration frequency to ensure that the optimal cutting state can be maintained throughout the entire surgical process.

[0080] Thirdly, the application of bone grinding files that can adaptively optimize cutting temperature:

[0081] The bone rasp assembly can be widely used in artificial joint replacement surgical instruments, for example, in oscillating handheld bone rasps (such as oscillating saws, osteotomes or acetabular rasps), or in bone rasp surgical robots in the form of mechanical arms.

[0082] Compared with the prior art, the present invention has the following beneficial effects:

[0083] 1. Optimize cutting temperature: By integrating a forced vibration system and setting an excitation unit on the transmission shaft, the present invention can generate precisely controllable torsional vibration forces. These torsional vibration forces are generated in the form of reverse frequency during the feeding and retraction of the bone rasp. Through the superposition effect of vibration, the magnitude of the cutting force is effectively reduced, thereby optimizing the friction force and achieving a reduction in the cutting temperature. It helps to reduce the risk of thermal damage and improve the safety and effectiveness of the operation. At the same time, the addition of the counterweight makes the torsional vibration force generated by the transmission shaft during the rotation process more uniform and stable. This not only optimizes the control effect of the cutting temperature, but also improves the cutting efficiency of the bone grinding rasp assembly.

[0084] 2. Improve surgical accuracy and safety: The artificial intelligence technology (single-output PID neural network SPIDNN) adopted in the present invention realizes a mechanism of adaptive frequency adjustment, which can ensure that the optimal cutting state is always maintained during the cutting process, avoiding unnecessary damage to bone tissue due to excessive cutting force or high cutting temperature, thereby reducing surgical risks.

[0085] 3. Improve surgical efficiency: The artificial intelligence technology used in this invention can automatically adjust cutting parameters such as vibration frequency according to changes in the surgical environment, ensuring that the cutting process always remains efficient. This helps shorten surgical time and improve surgical efficiency.

[0086] 4. Enhanced user experience: The present invention provides doctors with intelligent auxiliary tools, enabling them to cope with complex surgical environments more easily and reduce their workload. BRIEF DESCRIPTION OF THE DRAWINGS

[0087] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or technical descriptions will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0088] Figure 1 It is a three-dimensional schematic diagram of the present invention;

[0089] Figure 2 It is a three-dimensional schematic diagram of the present invention with part of the casing structure hidden;

[0090] Figure 3 It is a three-dimensional schematic diagram of the forced vibration system of the present invention;

[0091] Figure 4 It is a three-dimensional schematic diagram of the cooperation between an excitation unit and a slide rail of the present invention (wherein the sleeve 2054 is half-sectioned);

[0092] Figure 5 It is a schematic diagram of a swing-type handheld bone rasp disclosed in the second embodiment of the present invention;

[0093] Figure 6 It is a schematic diagram of a bone rasp surgical robot disclosed in Embodiment 3 of the present invention;

[0094] Figure 7 It is a schematic flow chart of the control method of the present invention;

[0095] Figure 8 It is a schematic diagram of the architecture of a single-output PID neural network SPIDNN of the control method of the present invention;

[0096] Fig. 9 It is a schematic diagram of finite element simulation of a test example of the present invention (wherein part A is a control group and part B is an experimental group);

[0097] Fig.10 A control error comparison schematic diagram of a test example of the present invention;

[0098] Fig.11 It is a schematic diagram for comparing the coupling degrees of the test examples of the present invention.

[0099] Description of the drawings: 1. Casing; 101. First slide slot; 102. Second slide slot; 2. Transmission assembly; 201. Transmission shaft; 202. Slide rail; 2021. Pin; 203. Oblique rotor; 204. Counterweight; 205. Excitation unit; 2051. Servo motor; 2052. Eccentric shaft; 2053. Needle bearing; 2054. Sleeve; 3. Bone rasp; 301. Water inlet; 302. Water outlet; 4. Swinging handheld bone rasp; 5. Bone rasp surgical robot. DETAILED DESCRIPTION

[0100] In order to make the above-mentioned purposes, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the accompanying drawings. In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention can be implemented in many other ways different from those described herein, and those skilled in the art can make similar improvements without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below;

[0101] It should be noted that the various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments, and the same or similar parts between the various embodiments can be referred to each other. For the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part description.

[0102] Example 1: Please refer to Figures 1 to 4 The present embodiment discloses a bone file assembly that can adaptively optimize the cutting temperature: it is mainly composed of a housing 1, a transmission assembly 2 and a bone file 3. The housing 1 serves as the outer shell of the entire assembly, protecting the internal structure and providing support. The transmission assembly 2 is built into the housing 1, and realizes a rotational motion through the drive of an external rotating part, and converts this rotational motion into a feeding motion of the bone file 3. Driven by the transmission assembly 2, the bone file 3 performs swing feeding, thereby achieving grinding of the surface of soft tissue and bone tissue. In particular, a forced vibration system is also integrated in the transmission assembly 2, which can generate at least two relative and axial torsional vibration forces, and these torsional vibration forces are based on the direction of the feeding motion of the transmission assembly 2 and the bone file 3.

[0103] Specifically: The design principle of the transmission assembly 2 is to use the mechanical transmission principle to convert the rotational motion of the external rotating part into the linear or swinging feed motion of the bone rasp 3. The forced vibration system is based on the principle of vibration, and through a specific mechanical structure or electromagnetic device, two relative and opposite torsional vibration forces are generated in the direction of the transmission assembly 2. When these torsional vibration forces act on the bone rasp 3, a vibration superposition effect will be generated, which helps to reduce the friction heat generated during the cutting process, thereby achieving the purpose of optimizing the cutting temperature.

[0104] Furthermore, when the bone rasp 3 moves based on the feeding action, except for the torsional vibration force closest to the bone rasp 3, the remaining torsional vibration forces are generated in the form of reverse and same frequency; when the bone rasp 3 retracts based on the feeding action, except for the torsional vibration force farthest from the bone rasp 3, the remaining torsional vibration forces are generated in the form of reverse and same frequency. In the process of vibration superposition, if the directions of two vibrations are opposite and the amplitudes are similar, they will cancel each other out. This phenomenon is called destructive interference. In this scheme, except for the torsional vibration force closest to or farthest from the bone rasp 3, the remaining torsional vibration forces are all generated in the form of reverse and same frequency, which is equivalent to that at certain positions of the transmission shaft 201, adjacent torsional vibration forces will produce destructive interference, thereby reducing the overall vibration amplitude. Due to the effects of vibration superposition and destructive interference, the vibration on the transmission shaft 201 becomes more uniform and stable. This stable vibration state helps to reduce the reaction force generated during the cutting process, thereby reducing the size of the cutting force. The reduction of cutting force is equivalent to the reduction of heat generated during the cutting process (because the heat is mainly generated by the work done by the cutting force). The existence of torsional vibration force also optimizes the friction during the cutting process. Vibration makes the contact between the cutting edge and the bone tissue more uniform and stable, reducing the fluctuation of friction. Stable friction helps to improve the accuracy and stability of cutting and reduce the heat generated by excessive friction.

[0105] It can be understood that, in order to control the magnitude of the above-mentioned superposition effect, the cutting temperature can be adaptively controlled.

[0106] Furthermore, a vibration sensor may be built into the bone rasp 3 to detect the actual vibration magnitude; meanwhile, a temperature sensor or an infrared sensor may be installed outside the housing 1 to detect the cutting temperature.

[0107] It can be understood that in the above scheme: the bone grinding rasp assembly of the present technology realizes adaptive optimization of cutting temperature by integrating a forced vibration system. During the bone grinding process, the torsional vibration force generated by the forced vibration system can reduce the cutting force and cutting temperature, reduce the risk of thermal damage, and improve the safety and effect of the operation.

[0108] It should be particularly pointed out that when multiple torsional vibration forces act in the form of reverse and same frequency, they will produce a vibration superposition effect on the transmission shaft 201. This superposition effect helps to offset part of the reaction force generated during the cutting process, thereby reducing the size of the cutting force. Since cutting force is one of the main factors that generate cutting heat, reducing the cutting force helps to reduce the heat generated during the cutting process, thereby optimizing the cutting temperature. At the same time, although the cutting force is reduced, due to the presence of torsional vibration force, the cutting edge of the bone rasp 3 can still contact the bone tissue in the form of high-frequency vibration. This high-frequency vibration helps to crush and separate the bone tissue, making the cutting process smoother and more efficient. Therefore, although the cutting force is reduced, the cutting effect is not significantly affected. And the presence of torsional vibration force also optimizes the friction during the cutting process. Due to the effect of vibration, the contact between the bone rasp 3 and the bone tissue becomes more uniform and stable, reducing the fluctuation and uncertainty of the friction force. This helps to further improve the accuracy and stability of cutting, while reducing the heat generated by excessive friction.

[0109] In the technical solution provided in this embodiment, please refer to Figure 2-4 : In the specific implementation of the present technology, the core component of the transmission assembly 2 is the transmission shaft 201, which is rotatably fitted in the housing 1 and defines the axial direction of the entire system. The excitation unit 205 of the forced vibration system is directly arranged below or above the transmission shaft 201 (only the below is shown in the figure) to generate the required torsional vibration force. In addition, the transmission shaft 201 is also tilted and rotatably fitted with an oblique rotor 203 (such as a special bearing arranged obliquely), which is connected to the slide rail 202 in the housing 1 through a bayonet 2021 (stuck as shown in the figure). The slide rail 202 is slidably fitted in the first slide groove 101 of the housing 1, and the other end is connected to the bone rasp 3. When the transmission shaft 201 is driven to rotate by an external rotating member (such as a motor), the oblique rotor 203 will rotate with it, and according to different rotation angles, it will be tilted at different angles relative to the bayonet 2021. This inclination will push the latch 2021 and the slide rail 202 connected thereto to perform feeding movement, thereby driving the bone rasp 3 to perform swing feeding, and then based on the rotation angle and direction of the external rotating part, the swing feeding distance and direction of the bone rasp 3 can be controlled.

[0110] Specifically: the torsional vibration force generated by the excitation unit 205 on the transmission shaft 201 will be generated in the form of reverse frequency during the feeding and retraction of the bone file 3. This design makes it possible for the torsional vibration force, except for the torsional vibration force closest to the bone file 3, to exist in the form of reverse frequency when the bone file 3 is fed, thereby reducing the size of the cutting force under the effect of the vibration superposition effect, and then achieving the cooling of the cutting temperature by optimizing the form of friction. It should also be pointed out that the so-called "reverse frequency" means that in the process of feeding or retracting the bone file 3, except for the torsional vibration force at a specific position (closest or farthest), the directions of the remaining torsional vibration forces are opposite, but the frequency is the same. That is, when one torsional vibration force acts in a certain direction, another torsional vibration force will act simultaneously in the opposite direction. This design allows the torsional vibration force to form a dynamic balance on the transmission shaft 201, which helps to reduce unnecessary vibration and noise, while improving the stability and accuracy of cutting.

[0111] It can be understood that in the above scheme: through the implementation of this technology, the bone rasp assembly not only realizes the function of converting rotational motion into swinging feeding motion, but also realizes adaptive optimization of cutting temperature through the torsional vibration force generated by the forced vibration system. During the feeding and retraction process of the bone rasp 3, the reverse frequency generation of the torsional vibration force and the vibration superposition effect effectively reduce the cutting force and cutting temperature, and reduce the risk of thermal damage. This function of adaptively optimizing the cutting temperature enables the bone rasp assembly of this technology to be operated more safely and efficiently in orthopedic surgery, improving the success rate of the operation and the recovery speed of the patient.

[0112] In the technical solution provided in this embodiment, please refer to Figure 2-4 The excitation unit 205 is composed of a servo motor 2051 and an eccentric shaft 2052. The servo motor 2051 is fixedly arranged on the slide rail 202, and its output shaft is connected to the eccentric shaft 2052. When the output shaft of the servo motor 2051 rotates, the eccentric shaft 2052 rotates accordingly, and generates a torsional vibration force due to its eccentric design. This design enables the excitation unit 205 to directly act on the slide rail 202, thereby affecting the cutting action of the bone rasp 3 through the slide rail 202.

[0113] Meanwhile, a second slide slot 102 is provided in the housing 1 for accommodating the excitation unit 205. When the slide rail 202 is pushed or retracted, the servo motor 2051 moves in the second slide slot 102 to avoid being affected by mechanical interference.

[0114] Specifically: The principle of this technology is to utilize the precise control capability of the servo motor 2051 and the eccentric design of the eccentric shaft 2052 to generate the required torsional vibration force. The servo motor 2051 is a motor that can precisely control the speed and direction. The rotation of its output shaft will drive the eccentric shaft 2052 to rotate. Since the center of mass of the eccentric shaft 2052 deviates from the rotation axis, a periodic torsional vibration force is generated during the rotation process. This torsional vibration force is transmitted to the bone rasp 3 through the slide rail 202 to optimize the cutting temperature.

[0115] It can be understood that in the above scheme: through the implementation of this technology, the excitation unit 205 can generate a precisely controllable torsional vibration force, further optimizing the control effect of the cutting temperature. The precise control capability of the servo motor 2051 makes the generation of torsional vibration force more stable and reliable. This design not only improves the cutting efficiency of the bone grinding rasp assembly, but also reduces the heat generated during the cutting process, reducing the risk of thermal damage.

[0116] On the basis of the above-mentioned excitation unit 205, the excitation unit 205 close to the bone rasp 3 is located in the middle of the slide rail 202, and when the servo motor 2051 of the excitation unit 205 rotates, mechanical interference will inevitably occur. Figure 4 The main body of the servo motor 2051 is normally fixedly connected to the slide rail 202, but its output shaft is fixedly provided with a needle bearing 2053 (fixed on the inner bushing), the outer bushing of the needle bearing 2053 is fixed in the sleeve 2054, the end surface of the sleeve 2054 is fixed on the surface of the slide rail 202, and the front end of the slide rail 202 is the connection position with the bone rasp 3. Furthermore, when the output shaft of the servo motor 2051 rotates, the rotation force is offset by the needle bearing 2053, and the normal feed force of the slide rail 202 is borne by the sleeve 2054, so that no mechanical interference will occur.

[0117] It can be understood that the slide rail 202 in this solution is actually divided into two parts; one of the parts is fixed with a bayonet 2021, and one end face and the other end face of the other part are fixedly connected to the sleeve 2054 and the bone rasp 3 respectively; Figures 3-4 It can also be known from the corresponding structure in that the bodies of the two servo motors 2051 are both fixed on a section of the slide rail 202.

[0118] It is understandable that, because the slide rail 202 is divided into two parts, the first slide groove 101 should also be divided into two parts, one end of which needs to extend to the front end of the second slide groove 102 .

[0119] In the technical solution provided in this embodiment, please refer to Figure 2: The transmission shaft 201 is provided with a counterweight 204 at a position corresponding to the vibration unit 205 (directly above the vibration unit 205). The counterweight 204 is used to coordinate with the torsional vibration force generated by the vibration unit 205 to adjust the vibration characteristics of the entire transmission system. By properly designing the mass and position of the counterweight 204, the transmission shaft 201 can generate a more uniform and stable torsional vibration force during rotation, thereby optimizing the control effect of the cutting temperature.

[0120] Specifically, when the transmission shaft 201 rotates, the counterweight 204 generates a certain inertial force due to its mass, and this inertial force interacts with the torsional vibration force generated by the excitation unit 205 to jointly affect the vibration state of the transmission shaft 201. By reasonably designing the mass and position of the counterweight 204, the transmission shaft 201 can generate a more ideal torsional vibration force distribution during the rotation process, thereby achieving optimal control of the cutting temperature.

[0121] It can be understood that in the above scheme: the addition of the counterweight block 204 further improves the performance of the bone grinding rasp assembly of the present technology. Through the synergistic effect with the excitation unit 205, the counterweight block 204 makes the torsional vibration force generated by the transmission shaft 201 during the rotation process more uniform and stable, thereby optimizing the control effect of the cutting temperature. This design not only improves the cutting efficiency of the bone grinding rasp assembly, but also reduces the heat generated during the cutting process and reduces the risk of thermal damage.

[0122] In the technical solution provided in this embodiment, please refer to Figures 1 to 4 The bone rasp 3 is also provided with a water inlet 301 and a water outlet 302. The water inlet 301 can be connected to an external water body, and the bone debris generated by cutting can be flushed through the water outlet 302.

[0123] The technical solution provided in this embodiment also includes a controller for controlling all the electrical components in this embodiment. The controller is also used to receive the control electrical signals e1 and e2 described in the fourth embodiment to control the corresponding excitation unit 205 (ie, its servo motor 2051).

[0124] Embodiment 2: Figure 5 As shown, this embodiment further discloses an oscillating handheld bone rasp, which adopts the bone rasp assembly as described in the first embodiment, and the shape of the bone rasp 3 can be designed to be flat, hooked or spherical to achieve the functions of a traditional oscillating saw (oscillating rasp), osteotome and acetabular rasp:

[0125] (1) Oscillating saws and osteotomes: The forced vibration system is directly integrated into the handle or transmission part of the oscillating saw, so that the oscillating saw can generate stable torsional vibration force during the cutting process. Through the torsional vibration force of the opposite frequency, the oscillating saw can significantly reduce the cutting force when cutting bones, reduce thermal damage, and improve the accuracy and smoothness of the cutting. The addition of the vibration system also helps to reduce the reaction force felt by the doctor when operating the oscillating saw, improving the comfort and efficiency of the operation.

[0126] (2) Acetabular reamer: By integrating a forced vibration system, the acetabular reamer can generate stable vibration during the trimming process, optimize the cutting effect, and reduce thermal damage and unevenness. Stable vibration helps doctors control the trimming process more accurately, improving the quality of surgery and the patient's recovery effect.

[0127] Embodiment 3: Figure 6 As shown, this embodiment further discloses a bone rasp surgical robot. The bone rasp surgical robot 5 adopts the bone rasp assembly as described in the first embodiment, and realizes optimal control of the cutting force by precisely controlling vibration parameters (such as frequency, amplitude, etc.).

[0128] Specifically, the bone rasp assembly is placed on the end effector in the form of a mechanical arm of the bone rasp surgical robot 5 .

[0129] Example 4: In bone grinding surgery, precise control of the bone grinding rasp assembly is crucial to ensure surgical results and patient safety. However, there are two problems in practical applications:

[0130] First, under the vibration superposition force generated by the excitation unit 205 on the bone rasp 3, the actual feed speed and vibration frequency on the bone often cannot accurately match the preset parameters, resulting in the impact on the surgical accuracy and effect; in the solution provided in the first embodiment, the method of "adaptively controlling the cutting temperature according to the size of the above superposition effect" can solve this problem to a certain extent, but it is more dependent on the user's subjective inference and control, and lacks intelligence;

[0131] Secondly, when two excitation units 205 are used, they will interfere with each other during the process of switching between start and stop, resulting in a coupling effect, which further affects the control accuracy.

[0132] In order to solve the above two problems, Figure 7 As shown, this embodiment further discloses a control method for a bone grinding rasp assembly that can adaptively optimize cutting temperature. That is, a multi-output PID neural network (MPIDNN) system composed of a single output PID neural network (SPIDNN) is introduced. The system achieves independent and precise control of the two excitation units 205 through neural network modeling and dynamic adjustment, thereby optimizing the cutting temperature and improving surgical efficiency and safety.

[0133] The multi-output PID neural network (MPIDNN) system is a 4-input 2-output multivariable coupling system, wherein the 4 inputs correspond to the set vibration frequencies (F 1S ,F 1R ) and the actual vibration frequency (F 2S ,F 2R ). Two single output PID neural networks (SPIDNNs) are constructed, each SPIDNN is responsible for controlling one excitation unit 205. Each SPIDNN includes:

[0134] 1) Input layer: It consists of two neurons and receives the setting and actual vibration frequency in the corresponding state;

[0135] 2) Hidden layer: It consists of three neurons: proportional (P), integral (I) and differential (D), which are responsible for the three core functions of the PID control algorithm respectively;

[0136] 3) Output layer: Combined with the differential evolution (DE) algorithm, a single control signal is output to the corresponding excitation unit 205;

[0137] Furthermore, the two SPIDNNs work in parallel to control the two excitation units 205 respectively, thus avoiding the coupling effect between them. At the same time, by sharing some input information (such as the state of the bone rasp), the coordinated control at the system level is realized.

[0138] In practical application of this embodiment, it is necessary to first establish an input variable Input, including the set vibration frequency F when the bone rasp 3 is in the extended state. 1S and the actual vibration frequency F 2S , and the set vibration frequency F of the bone rasp 3 in the retracted state 1R and the actual vibration frequency F 2R ; Then execute the following steps S1 to S3.

[0139] In this embodiment, regarding step S1, the PID neural network SPIDNN is initialized: when the bone grinding file assembly is powered on, it is necessary to activate the single output PID neural network SPIDNN of each vibration source (i.e., the excitation unit 205) in the forced vibration system. Each SPIDNN is composed of three neurons: proportional (P), integral (I) and differential (D), which are used to implement the PID control algorithm.

[0140] Among them, the two input neurons in a single-output PID neural network SPIDNN are connected to the neurons of proportion P, the neurons of integral I and the neurons of differential D, and the two input neurons in another single-output PID neural network SPIDNN are also connected to the neurons of proportion P, the neurons of integral I and the neurons of differential D in the previous single-output PID neural network SPIDNN, and vice versa. The structure is as follows Figure 8 shown.

[0141] Specifically, step S100, initializing neurons of each single-output PID neural network SPIDNN, includes:

[0142] (1) A proportional neuron P for adjusting the amplitude of the error signal: P(t) = Kp·e(t);

[0143] Among them, K p is the proportional coefficient, which is used to adjust the amplitude of the error signal; e(t) = F1-F2 is the error between the set vibration frequency and the actual vibration frequency (for the extended state, F1 = F 1S , F2=F 2S ; For the retracted state, F1 = F 1R , F2=F 2R ).

[0144] (2) Integral neuron I that eliminates the static error of the system and improves control accuracy by accumulating error signals:

[0145] Among them, K i is the integral coefficient, which determines the speed of error accumulation; b I is the bias of the integral neuron, which is used to adjust the benchmark of the integral output; I (t) is the output of the integrating neuron at time t; It is the integral of the error signal e(t) from 0 to t, reflecting the cumulative effect of the error; t is the time unit variable;

[0146] (3) Differential neuron D that adjusts the output in advance and enhances the dynamic performance of the system by predicting the changing trend of the error:

[0147] Among them, K d is the differential coefficient, which determines the sensitivity to the rate of change of the error; b D is the bias of the differential neuron, which is used to adjust the benchmark of the differential output; D (t) is the output of the differentiable neuron at time t; It is the derivative of the error signal e(t), which reflects the rate of change of the error;

[0148] Specifically, step S101, initialize the network output of each single-output PID neural network SPIDNN: y(t)=w P ·y P (t)+w I ·y I (t)+w D ·y D (t)+b o ;

[0149] Where y(t) is the total output of the single-output PID neural network at time t; w P 、w I and w D are the weights of proportional, integral and differential neurons, respectively, which determine the contribution of each neuron output to the total output of the network; b o is the bias of the output layer, which is used to adjust the benchmark of the total output of the network. It can be set to a small random number to ensure that the initial state of the network has diversity. For example, use uniform distribution or normal distribution to generate these initial values:

[0150] w P ,w I ,w D ~U(-∈,∈) or w P ,w I ,w D ~N(0,σ 2 );

[0151] b I ,b D ,b o ~U(-δ,δ) or b I ,b D ,b o ~N(0,μ2);

[0152] Among them, ∈ and δ are the control initial weights w P 、w I and w D and bias b o A small positive number in the range of 2 and μ are the standard deviations of the normal distribution, respectively.

[0153] Further, such as Figure 8As shown, in the multi-output PID neural network MPIDNN, the input layer to the hidden layer of each single-output PID neural network SPIDNN is independently operated, but the hidden layer to the output layer are cross-correlated. This structure not only ensures the independent control of each excitation unit 205, but also realizes the coordination at the system level. By defining different activation functions for the proportion, integral and differential neurons of the hidden layer, there is a corresponding relationship between the proportion, integral and differential. It is helpful to continuously train and learn according to the error function through forward calculation and reverse calculation, modify and improve the connection weights, and then solve the coupling effect.

[0154] Specific:

[0155] (1) Activation function of proportional neuron P: The output of a proportional neuron can be directly determined by its input and the proportional coefficient, so its activation function can be expressed as a linear function:

[0156] y P (t) = f P (P(t))=K p ·e(t);

[0157] Among them, f P represents the activation function of the proportional neuron, y P (t) is the output of the ratio neuron.

[0158] (2) Activation function of the integrating neuron I: Since the output of the integrating neuron is the accumulation of its input, its activation function is a ReLU accumulation function:

[0159]

[0160] Among them, f I represents the activation function of the integrating neuron, y I (t) is the output of the integrating neuron (note that y I (t)' is the output of the integrating neuron defined previously, and y I (t) is the output after the activation function processing; however, for the sake of uniformity, it will be referred to as y in the following text. I (t)).

[0161] (3) Activation function of the differential neuron D: The output of a differential neuron is the rate of change of its input, so its activation function can be a function that takes the derivative of the input:

[0162]

[0163] Among them, f D represents the activation function of the differentiable neuron, y D (t) is the output of the differentiating neuron (note also that yD (t)' is the output of the differentiable neuron defined previously, and y D (t) is the output after the activation function is processed. For the sake of uniformity, it will be referred to as y in the following text. D (t)).

[0164] It is understandable that under the above-mentioned cross-correlated single-output PID neural network SPIDNN architecture, each excitation unit 205 is controlled by an independent SPIDNN, ensuring that each unit can be accurately controlled according to its own characteristics and needs. Each SPIDNN has its own independent input layer, hidden layer and output layer, which are locally optimized according to their own error functions. This independent control strategy helps to reduce direct interference between different excitation units. In practice (especially during training), the effect is: when the coupling effect described above occurs, SPIDNN essentially regards it as part of the error e(t), and the input layer to the hidden layer of each SPIDNN is independently calculated. Because the hidden layer to the output layer is designed to be cross-correlated, it allows the hidden layer neurons of different SPIDNNs to exchange information, thereby achieving system-level coordination. This coordination mechanism helps to consider the interaction between them while maintaining the independent control of each excitation unit 205. And in the process of transmission from the hidden layer to the output layer, the output is dynamically adjusted according to the size and rate of change of the error signal through the activation function, thereby achieving precise regulation of the control object. Since each SPIDNN has its own PID neuron, they can perform independent PID control based on their own error signals, thus avoiding the influence of coupling.

[0165] In layman's terms, since each SPIDNN in MPIDNN learns independently (such as the solution disclosed in Example 5), and there are cross-correlations between hidden layers, the network can gradually learn how to coordinate the control actions between different excitation units to reduce the impact of coupling effects. During the training process, the network will continuously adjust the parameters according to the actual error signal to achieve dynamic compensation for the coupling effect. This adaptability enables MPIDNN to maintain high control accuracy and stability in the presence of coupling.

[0166] It can be understood that through the adaptive adjustment and modeling of the PID neural network, the actual feed speed and vibration frequency on the bone can be matched more accurately, the error can be reduced, and the control accuracy can be improved. The integral neuron can eliminate the static error of the system, and the differential neuron can predict the changing trend of the error and adjust the output in advance, thereby enhancing the stability and dynamic performance of the system. The two excitation units 205 are controlled by independent SPIDNNs to avoid mutual influence and coupling effects, and improve the reliability and independence of the system. Precise control helps to maintain temperature stability during cutting, reduce thermal damage, and improve surgical results and patient safety.

[0167] Furthermore, the MATLAB architecture program of the multi-output PID neural network MPIDNN in step S1 is as follows:

[0168] % Initialization parameters

[0169] numInputs=2;% Input layer neurons

[0170] numHidden=10;% Hidden layer neurons

[0171] numOutputs = 2; % Output layer neurons

[0172] learningRate=0.01;% learning rate

[0173] numEpochs = 1000; % Number of training rounds

[0174] % Activation function definition

[0175] sigmoid=@(x)1. / (1+exp(-x));

[0176] sigmoidDerivative=@(x)x.*(1-x);

[0177] % Initialize weights and biases

[0178] W_input_hidden=rand(numInputs,numHidden)-0.5;

[0179] b_hidden=rand(1,numHidden)-0.5;

[0180] W_hidden_output=rand(numHidden,numOutputs)-0.5; b_output=rand(1,numOutputs)-0.5;

[0181] %Define the activation function of the PID neuron (only the proportional neuron is shown as an example) Kp=1; %Proportional coefficient

[0182] Ki=0;%Integral coefficient

[0183] Kd=0;% differential coefficient

[0184] % Input data is a matrix of numInputs x numSamples

[0185] % Output data is the matrix X_train = [] of numOutputs x numSamples; % Input data

[0186] Y_train=[];% output data

[0187] % Training process

[0188] for epoch=1:numEpochs

[0189] % Forward calculation

[0190] net_hidden=X_train*W_input_hidden+b_hidden;

[0191] y_hidden=sigmoid(net_hidden);

[0192] Calculation of %PID neuron (only the proportional part is used as an example)

[0193] net_output=y_hidden*W_hidden_output+b_output;

[0194] y_output = Kp*net_output;

[0195] % Calculation error

[0196] E=0.5*sum((Y_train-y_output).^2,'all');

[0197] % Reverse calculation

[0198] delta_output = Y_train-y_output; % output layer error

[0199] delta_hidden=(delta_output*W_hidden_output').*

[0200] sigmoidDerivative(y_hidden);% Hidden layer error

[0201] % Update weights and biases

[0202] W_hidden_output=W_hidden_output+learningRate*(y_hidden′*delta_output);

[0203] b_output=b_output+learningRate*sum(delta_output,1);

[0204] W_input_hidden=W_input_hidden+learningRate*(X_train'*

[0205] delta_hidden);

[0206] b_hidden=b_hidden+learningRate*sum(delta_hidden,1);

[0207] % Printing Error

[0208] ifmod(epoch,100) == 0

[0209] fprintf('Epoch%d,Error:%.4f\n',epoch,E);

[0210] end

[0211] end

[0212] %Testing process (using the trained network for control)

[0213] X_test=[];%Test input data

[0214] y_output_test = []; % Test output data (used to store network control output)

[0215] % Use the trained network to calculate the test output

[0216] net_hidden_test=X_test*W_input_hidden+b_hidden;

[0217] y_hidden_test=sigmoid(net_hidden_test);

[0218] net_output_test=y_hidden_test*W_hidden_output+b_output;

[0219] y_output_test=Kp*net_output_test; %Only consider the proportional part

[0220] In this embodiment, if Figure 7 and Figure 8 As shown, regarding step S2, control signal generation: input variable Input is input into the single output PID neural network SPIDNN, according to the internal parameters (P, I and D), and the vibration frequency F is set 1S and the actual vibration frequency F 2S The error between E1, or according to the set vibration frequency F 1R and the actual vibration frequency F 2R The error E2 between the two excitation units is calculated adaptively to calculate the corresponding control electrical signals e1 and e2 to control the forced vibration system and adjust its vibration parameters, so as to solve the problem that the feeding speed and vibration frequency on the bone cannot really adapt to the predetermined value when the excitation unit 205 generates the vibration superposition force on the bone file, and reduce the coupling effect between the two excitation units. It includes the following steps S200-S202.

[0221] Specifically, step S200, activate the input layer: receive input variable Input = {F 1S ,F 2S ,F 1R ,F 2R}; At the same time, for the extended state, the calculated error E1 = F 1S -F 2S ; For the retracted state, calculate the error E2 = F 1R -F 2R ;

[0222] Specifically, step S201 activates the hidden layer: including proportional (P), integral (I) and differential (D) neurons, each neuron calculates the error e(t) and outputs y according to the input and internal parameters I (t) and output y D (t);

[0223] Among them, in the proportional neuron P(t)=Kp·e(t), E(t) is E1 or E2, depending on the current state. The forms of the integral neuron and the differential neuron are as described in the previous step S1, which will not be repeated here.

[0224] Specifically, in step S202, the output layer adaptively calculates the corresponding control electrical signals e1 and e2:

[0225] (1) For the control electrical signal e1 of the extended state:

[0226]

[0227] in, and They respectively represent the outputs of the proportional P neuron, the integral I neuron, and the differential D neuron in the extended state.

[0228] (2) For the control electrical signal e2 in the retracted state:

[0229]

[0230] in, and They represent the outputs of the proportional P neuron, the integral I neuron, and the differential D neuron in the retracted state respectively.

[0231] It is understandable that through the adaptive calculation of SPIDNN, the control signal can be dynamically adjusted according to the error between the actual vibration frequency and the set vibration frequency, so that the vibration parameters are more in line with the predetermined value, thereby improving the control accuracy of the cutting temperature. Since SPIDNN can generate control signals for the extended and retracted states respectively, it can more effectively deal with the mutual influence between the two excitation units and reduce the interference caused by the coupling effect. Through the combined action of proportional, integral and differential neurons, the immediate response, accumulation and change rate of the error can be more comprehensively considered, thereby improving the stability and response speed of the entire control system.

[0232] In this embodiment, regarding step S3, autonomous learning and updating (differential evolution algorithm DE optimization): within one execution cycle of each single-output PID neural network SPIDNN, the degree value DV of the control effect of the current single-output PID neural network SPIDNN is calculated.

[0233] The differential evolution algorithm performs crossover and mutation operations on the current population based on the degree value DV to generate a new population. The best performing individual is selected from the new population to update the weights of the PID neural network SPIDNN, including the weight w P , weight w I and weight w D The method specifically includes the following steps S300 to S305.

[0234] Specifically, step S300, calculating the degree value DV: DV = |TT′|;

[0235] Where T is the actual detected cutting temperature and T′ is the set theoretical temperature. The smaller the DV value, the better the control effect.

[0236] Specifically, step S301, population initialization: P0 = {x1, x2, ..., x NP};

[0237] Among them, P0 represents the initial population, NP represents the population size; x i Represents the i-th individual (vector), containing the weight w of the PID neural network SPIDNN P , weight w I and weight w D Population initialization is the basis of the differential evolution algorithm. By randomly generating a group of individuals with different weight combinations, it provides an initial solution space for subsequent crossover and mutation operations. A diverse initial population helps increase the possibility of finding the global optimal solution and improves the search efficiency of the algorithm.

[0238] Specifically, step S302, crossover operation: for each individual x i , generate the test vector v i :

[0239] v i =x r1 +F·(x r2 -x r3 );

[0240] Among them, r1, r2 and r3 are different random indexes and different from i; F is the scaling factor, whose main function is to control the degree of scaling of the differential vector, thereby affecting the search step and exploration range of the newly generated test vector (solution vector). The simplest method is to use a fixed constant as the value of the scaling factor F. This method is simple to implement, but may lack flexibility when facing different problems.

[0241] It is understandable that the crossover operation is one of the core of the differential evolution algorithm. By introducing the difference information of other individuals, a new solution vector is generated to explore new areas in the solution space. It helps to increase the diversity of the population and improve the global search ability of the algorithm, making it more likely to find a better weight combination. At the same time, for the test vector v i Perform a bounds check to ensure it is within the valid range.

[0242] Specifically, step S303, selection operation: The selection operation is based on the principle of "survival of the fittest", retaining individuals with better performance and eliminating individuals with poorer performance to gradually approach the global optimal solution. That is, the test vector v is evaluated according to the degree value DV i and the current individual x i performance, and select those with better performance to enter the next generation population

[0243] Among them, DV(x i ) means using the current individual xi (i.e. a specific set of weights w P , weight w I and weight w D ), the degree value between the cutting temperature and the set temperature obtained under the control of the single output PID neural network SPIDNN. DV(v i ) means using the test vector v i (i.e., a set of new weights generated by the crossover and mutation operations of the differential evolution algorithm DE), the degree value between the cutting temperature and the set temperature obtained under the control of the PID neural network SPIDNN.

[0244] Then, we get a new population P′={x1′,x2′,…,x NP ′}; The selection operation ensures that the population evolves in a better direction, which helps to improve the optimization efficiency and accuracy of the algorithm.

[0245] Specifically, step S304, iterative update: in one execution cycle of each single-output PID neural network SPIDNN, steps S302 to S303 are executed cyclically until the equipment stops. At the same time, in each execution cycle of each single-output PID neural network SPIDNN, the new population P′ based on the previous execution cycle is used as the initial population P0. Iterative update ensures that the algorithm can continuously adapt to changing cutting conditions and maintain the stability and accuracy of the control effect.

[0246] Specifically, step S305, update the weight w P , weight w I and weight w D :After optimization by differential evolution algorithm DE, select the individual x with the best performance best , which contains the weight w P , weight w I and weight w D These are the updated weight values. These weight values ​​are assigned to each single-output PID neural network SPIDNN to improve its control effect. After the weights are updated, the control effect of SPIDNN is optimized, which can better adapt to the changes in the cutting process and improve the control accuracy and stability of the cutting temperature.

[0247] It can be understood that step S304 provides a continuous learning and optimization capability enhancement mechanism for each single-output PID neural network SPIDNN, rather than a simple weight assignment effect. This mechanism achieves a gradual improvement in the control performance of the device by continuously executing the key steps of the differential evolution algorithm, namely the crossover operation (S302) and the selection operation (S303), and using the optimization result of the previous execution cycle as the starting point of the next cycle. Because in the execution cycle of each single-output PID neural network (SPIDNN), steps S302 to S303 are executed cyclically. This is equivalent to the device constantly generating a new solution vector (trial vector v) according to the current control effect (evaluated by calculating the degree value DV) during operation. i ) and select, retaining individuals with better performance. This cyclic iterative process is actually constantly exploring and optimizing the weight space to find a weight combination that can better adapt to the current cutting conditions. Moreover, at the beginning of each execution cycle, the new population P' optimized in the previous execution cycle will be used as the initial population P0 of the current cycle. This approach makes full use of historical optimization information and avoids the inefficiency of starting from scratch every time optimization is performed.

[0248] Because the differential evolution algorithm gradually approaches the global optimal solution through continuous crossover, mutation and selection operations, as the number of iterations increases, the evaluation of the equipment control performance (i.e., the degree value DV) will tend to stabilize and may reach a better level. In this process, the control strategy of the equipment (i.e., the weight combination of SPIDNN) is continuously optimized, so that the equipment can make adjustments faster and maintain a stable control effect when facing changing cutting conditions. Therefore, in the long run, the equipment seems to have the ability to "learn" and "remember", and can make smarter decisions based on historical experience and current conditions, that is, it achieves the effect of "getting smarter with use".

[0249] Embodiment 5: This embodiment further discloses the training method of the multi-output PID neural network MPIDNN as described in Embodiment 4, and the following training relationship needs to be established:

[0250] P1, Error function: It is the sum of the squares of the differences between the actual output and the target output of each output unit. For the i-th output unit, its error function E i It is expressed as:

[0251] Among them, y i (t) is the actual output of the i-th output unit at time t, is the corresponding target output, and Time is the total number of time steps.

[0252] For the entire network, the total error function E is expressed as the sum E of the error functions of each output unit:

[0253] Where N is the number of output units.

[0254] P2, connection weight is set to w ijk Represents the connection weight from the i-th neuron to the j-th neuron in the k-th layer. During the training process, these weights need to be continuously adjusted to minimize the error function.

[0255] P3. Forward calculation: Starting from the input layer, the output of each neuron is calculated layer by layer until the output of the output layer is obtained. For each single-output PID neural network (SPIDNN) in the multi-output PID neural network, its forward calculation process is expressed as:

[0256] P3.1 Input layer to hidden layer: For the i-th input neuron and the j-th hidden layer neuron, the input net of the hidden layer neuron j and output It is expressed as:

[0257]

[0258] Among them, x i is the input of the ith neuron in the input layer, input→hidden is the connection weight from the input layer input (note that the input here is not the input variable described in Example 4) to the hidden layer hidden, is the bias of the jth neuron in the hidden layer, f hidden (·) is the activation function of the hidden layer neurons (sigmoid, tanh, etc.).

[0259] P3.2, hidden layer to output layer (proportional, integral, differential neurons): their outputs can be calculated according to the activation functions defined previously. Let y P ,y I and D Represent the outputs of proportional, integral and differential neurons respectively, then:

[0260]

[0261]

[0262]

[0263] in, and are the connection weights from the hidden layer to the proportional, integral and differential neurons, bI and bD are the biases of the integral and differential neurons, respectively; t is the time unit variable;

[0264] The output y of the final output layer is the weighted sum of the outputs of the proportional, integral, and derivative neurons:

[0265] y=w P ·y P +w I ·y I +w D ·y D +b output ;

[0266] Among them, w P 、w I and w D are the connection weights from the proportional, integral and differential neurons to the final output, and b output is the bias of the output layer.

[0267] P4, reverse calculation: refers to calculating the error gradient of each neuron layer by layer starting from the output layer according to the error function, and updating the connection weights according to the gradient descent method. For a multi-output PID neural network, its reverse calculation process is expressed as:

[0268] P4.1, output layer error gradient: For the i-th output unit, its error gradient δ ioutput It is expressed as:

[0269]

[0270] P4.2, Hidden layer error gradient: For the jth neuron in the hidden layer, its error gradient δ jhidden The error gradient of the output layer is back-propagated to obtain:

[0271]

[0272] Among them, f′ hidden is the derivative of the activation function of the hidden layer neurons.

[0273] P4.3, Update connection weights: According to the error gradient, use the gradient descent method to update the connection weights. For the connection weights w between any two adjacent layers ij , and its update formula is:

[0274] w ij (new) = w ij (old)-η·δ j ·y i ;

[0275] Where η is the learning rate, δ j is the error gradient of the next layer of neurons, y i is the output of the neurons in the previous layer.

[0276] In summary, in practice, by collecting training sets and verification sets in the form of input variables such as those in Example 4, and continuously repeating the above-mentioned forward calculation and reverse calculation processes, the connection weights can be gradually adjusted so that the output of the network is closer and closer to the target output, thereby realizing PID control of the multi-output system and training of the neural network.

[0277] Embodiment 6: In step S302 of embodiment 4, the main function of the scaling factor F is to control the degree of scaling of the differential vector, thereby affecting the search step and exploration range of the newly generated test vector (solution vector). Specifically, the scaling factor F determines the degree of amplification of the difference information between the current individual and two randomly selected individuals, thereby affecting the distance between the newly generated test vector and the current individual. The solution adopted in embodiment 4 is to use a fixed constant as the value of the scaling factor F. This method is simple to implement, but lacks flexibility and objectivity when facing different problems. To this end, this embodiment further provides a preferred solution, which is to draw on the concept of the entropy weight method.

[0278] The entropy weight method is a multi-index evaluation method based on information entropy. Although the entropy weight method itself is not directly used to assign the scaling factor F, it can dynamically adjust the F value according to the fitness differences of individuals in the population, making the assignment of F more reasonable and effective. It includes the following steps:

[0279] S3020, calculate the fitness entropy: First, calculate the fitness entropy of each individual x in the population P i The fitness value of the population is obtained, and the fitness entropy is calculated based on the fitness value. The fitness entropy reflects the discrete degree of individual fitness in the population. If there are N individuals in the population, the fitness value of the i-th individual is fit i (i=1,2,...,N), then the fitness entropy E is expressed as:

[0280] in, is the ratio of the fitness value of individual i to the total fitness value of the population, k is a constant (k = 1 / N); fit j is the fitness value of the jth individual.

[0281] S3021, determine the value range of F: dynamically adjust the value range of the scaling factor F according to the size of the fitness entropy E. When the fitness entropy E is large, it means that the fitness differences of individuals in the population are large. At this time, the value range of F can be increased to expand the search space; when the fitness entropy E is small, it means that the fitness differences of individuals in the population are small. At this time, the value range of F can be reduced to perform a more detailed search. Then, the value range of F can be dynamically adjusted according to the size of the fitness entropy E. For example, a basic value F is set. base and an adjustment coefficient α, so that: F = F base +α(EEmin )

[0282] Among them, E min is the minimum possible value of fitness entropy (obtained when all individuals have the same fitness), and α is a positive number used to control the sensitivity of F to changes in E.

[0283] It can be understood that the entropy weight method can dynamically adjust the value of the scaling factor F according to the fitness differences of individuals in the population. This is equivalent to that at different stages of the algorithm, F can automatically adapt to the search status of the current population, thereby guiding the search process more effectively. By adjusting F through the entropy weight method, a better balance can be achieved between global search and local search. When the population diversity is high, increasing the F value can expand the search range and enhance the global search capability; when the population diversity is low, reducing the F value can perform more detailed local searches and improve search accuracy. Moreover, the entropy weight method dynamically adjusts F according to the fitness entropy, which can make the algorithm more efficient in the search process. Fitness entropy reflects the discrete degree of individual fitness in the population. By adjusting F through this indicator, the search can be more targeted and invalid searches and repeated searches can be reduced. At the same time, using the entropy weight method to assign F can enhance the adaptability and robustness of the algorithm to different problems. Because the entropy weight method is an evaluation method based on information entropy, it can dynamically adjust F according to the specific characteristics of the problem, so that the algorithm can perform better performance when facing different optimization problems.

[0284] Experimental example: This example aims to evaluate the effectiveness of a new bone grinding rasp component control method that can adaptively optimize cutting temperature by comparing it with the traditional PID controller algorithm. The experiment uses commercially available bovine vertebral segments as the simulation object, focusing on indicators such as temperature divergence, control error, and coupling degree during the cutting process to verify the superiority of the proposed control method.

[0285] 1. Experimental design

[0286] 1.1 Experimental and control group settings:

[0287] Experimental group: The bone grinding rasp assembly provided in Example 1 was used in hardware; the control method capable of adaptively optimizing cutting temperature in Examples 4 to 6 was used in control, which is based on a differential evolution algorithm and has continuous learning and optimization capabilities.

[0288] Control group: The hardware is the same as that of the experimental group; the traditional PID controller algorithm is used for control.

[0289] 1.2 Experimental scenario

[0290] Simulation object: commercially available bovine vertebral segment.

[0291] Cutting operation: The bone grinding rasp assembly applies a tangential force along the positive X direction of the spine for 0.2 minutes.

[0292] Control cycle: one cycle is 0.02 minutes, during which the feeding direction of the bone rasp 3 is switched.

[0293] 1.3 Measuring tools:

[0294] A force sensing unit (pressure sensor) of a multi-component dynamometer Kistler9257B is placed inside the spine to detect changes in the cutting pressure of the spine in each control cycle, which is converted into a control error and then into a degree value (%). A force sensing unit (vibration sensor) of an LMS vibration test system is placed inside the bone file 3 of the bone grinding file assembly of the experimental group and the control group to detect the degree of coupling when switching the feed direction of the bone file 3 in each control cycle, and convert it into a degree value (%).

[0295] ANSYS finite element software was used to model the bone grinding rasp components and spine to simulate the cutting process and temperature divergence effect.

[0296] (II) Experimental results:

[0297] 2.1 Finite element analysis results:

[0298] like Fig. 9 As shown in the figure, part (A) is the control group, and part (B) is the experimental group. Comparative analysis shows that the temperature divergence and temperature magnitude on the spine of the experimental group are significantly lower than those of the control group, indicating that the proposed control method can effectively control the cutting temperature and reduce thermal damage.

[0299] 2.2 Control error analysis:

[0300] like Fig.10 As shown in the figure, the control error of the experimental group is always lower than the preset threshold, and the fluctuation range is significantly smaller than that of the control group. The error curve of the experimental group shows a fluctuating regularity, which is attributed to the continuous learning and optimization ability of the differential evolution algorithm, indicating that this method can better adapt to the dynamic changes in the cutting process.

[0301] 2.3 Coupling degree analysis:

[0302] like Fig.11 As shown in the figure, the coupling degree fluctuation of the experimental group is significantly better than that of the control group, and its curve also shows a regular decline in the form of fluctuation, which further verifies the effectiveness of the differential evolution algorithm in optimizing the cutting process. The control group shows irregular fluctuations, indicating that the traditional PID controller has limitations in dealing with complex cutting conditions.

[0303] (III) Conclusion:

[0304] The experimental results clearly show that the bone rasp component control method that can adaptively optimize the cutting temperature not only performs well in controlling the cutting temperature and reducing thermal damage, but also significantly outperforms the traditional PID controller in optimizing the control error and coupling degree. This is mainly due to the continuous learning and optimization capabilities of the differential evolution algorithm, which enables it to better adapt to various changes in the cutting process.

[0305] This case verified the superiority of the bone rasp component control method that can adaptively optimize cutting temperature through comparative experiments. This method has significant advantages in controlling cutting temperature, reducing control errors, and optimizing coupling, providing a more accurate and safe solution for orthopedic surgery.

[0306] All the above embodiments only express the implementation methods of the relevant practical applications of the present invention, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the invention patent. It should be pointed out that for ordinary technicians in this field, several modifications and improvements can be made without departing from the concept of the present invention, which all belong to the protection scope of the present invention. Therefore, the protection scope of the patent of the present invention shall be based on the attached claims.

[0307] For those skilled in the art, it can be further appreciated that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the composition and steps of each example have been generally described in the above description according to function. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.

[0308] At the same time, those skilled in the art can understand that all or part of the processes in all the above-mentioned embodiments can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media provided in this application and used in the embodiments may include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (SSRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

Claims

1. A bone rasp assembly capable of adaptively optimizing cutting temperature, comprising a housing (1), a transmission assembly (2) located in the housing (1), and a bone rasp (3); the transmission assembly (2) comprises a transmission shaft (201) rotatably engaged in the housing (1), and a slanted rotor (203) rotatably engaged on the transmission shaft (201); a slide rail (202) and a first slide groove (101) are further arranged in the housing (1), the slide rail (202) and the first slide groove (101) are slidably engaged, the slide rail (202) is divided into two sections, wherein a latch (2021) is fixedly arranged on the first section of the slide rail, and the slanted rotor (203) is latched by the latch (2021); The transmission assembly (2) further comprises a forced vibration system; the forced vibration system comprises two excitation units (205); the two excitation units (205) are located below the transmission shaft (201); each excitation unit (205) comprises a servo motor (2051) and an eccentric shaft (2052) driven by its output shaft, the body of the servo motor (2051) is fixedly mounted on the first section slide rail, one end surface of the second section slide rail is fixedly connected to a sleeve (2054), and the other end surface of the second section slide rail is fixedly connected to the bone rasp (3); The output shaft of the servo motor (2051) close to the bone rasp (3) is fixedly provided with a needle bearing (2053), and the output shaft is fixed to the inner bushing of the needle bearing (2053), and the outer bushing of the needle bearing (2053) is fixed in the sleeve (2054); A second sliding groove is provided in the housing (1), and the excitation unit (205) is accommodated in the second sliding groove (102); The torsional vibration force generated by the excitation unit (205) on the transmission shaft (201) is generated in the form of reverse and same frequency during the feeding and retraction process of the bone rasp (3).

2. The bone grinding rasp assembly according to claim 1, characterized in that: A counterweight block (204) is installed on the transmission shaft (201) directly above the vibration excitation unit (205).

3. The bone rasp assembly according to claim 2, characterized in that: The bone rasp (3) is provided with a water inlet (301) and a water outlet (302).

4. A control method for a bone rasp assembly according to any one of claims 1 to 3, establishing an input variable Input, including a set vibration frequency F of the bone rasp (3) in an extended state 1S and the actual vibration frequency F 2S , and the set vibration frequency F of the bone rasp (3) in the retracted state 1R and the actual vibration frequency F 2R ; It is characterized by: S1, activating a single-output PID neural network SPIDNN of the forced vibration system, wherein the single-output PID neural network SPIDNN includes a proportional P neuron, an integral I neuron, and a differential D neuron; S2, inputting the input variable Input into the single output PID neural network SPIDNN, based on the set vibration frequency F 1S and the actual vibration frequency F 2S The error between E1, or according to the set vibration frequency F 1R and the actual vibration frequency F 2R The error E2 between them is used to adaptively calculate the corresponding control electrical signals e1 and e2 to control the forced vibration system; S3, calculate the degree value DV of the control effect of the current single output PID neural network SPIDNN, and use the differential evolution algorithm to update the weights w corresponding to the neurons of the proportion P, the integral I and the differential D P , weight w I and weight w D .

5. The control method according to claim 4, characterized in that: In S1, the single output PID neural network SPIDNN includes: Neurons with ratio P: P(t) = K p ·e(t); Neuron of integral I: Differential D's Nerve: Where P(t) is the output of the neuron of proportion P at time t; K p is the proportionality coefficient; e(t) is the error between the set vibration frequency and the actual vibration frequency; K i is the integration coefficient; b I is the bias of the integrating neuron; y I (t) is the output of the neuron integrating I at time t; is the integral of the error signal e(t) from 0 to t; t is the time unit variable; K d is the differential coefficient; b D is the bias of the neuron with the differential D; y D (t) is the output of the neuron of the differential D at time t; is the derivative of the error signal e(t); And the network output y(t) = w P ·y P (t)+w I ·y I (t)+w D ·y D (t)+b o ; Among them, w P 、w I and w D are the weights of proportional, integral and differential neurons, respectively, and b o is the bias of the output layer.

6. The control method according to claim 5, characterized in that: In S2, the input layer of the single output PID neural network SPIDNN includes: Receive input variable Input={F 1S ,F 2S ,F 1R ,F 2R }; At the same time, the calculated error E1 = F 1S -F 2S Or / and error E2 = F 1R -F 2R ; In the hidden layer of the single output PID neural network SPIDNN, each neuron calculates the error e(t) and output y according to the input and internal parameters. I (t) and output y D (t).

7. The control method according to claim 6, characterized in that: In S2, the output layer of the single output PID neural network SPIDNN includes: Output the control electrical signal e1 of the extended state: in, and They represent the outputs of the proportional P neuron, the integral I neuron, and the differential D neuron in the extended state respectively; Output retracted state control electrical signal e2: in, and They respectively represent the outputs of the neuron of example P, the neuron of integration I and the neuron of differentiation D in the retracted state.

8. The control method according to claim 4, characterized in that: In S3, the degree value DV=|TT′|; wherein T is the actually detected cutting temperature, and T′ is the set theoretical temperature; The execution process of the differential evolution algorithm is as follows: Population initialization: P0 = {x1, x2, ..., x NP }; For each individual x i , generate the test vector v i :v i =x r1 +F·(x r2 -x r3 ); Evaluate the trial vector v according to the degree value DV i and the current individual x i performance, select the best performance to enter the next generation population x i ′: Among them, P0 represents the initial population; NP represents the population size; x i represents the i-th individual, including the weight w of the single output PID neural network SPIDNN P , weight w I and weight w D ; r1, r2 and r3 are different random indices and different from i; F is the scaling factor; DV(x i ) means using the current individual x i When the single output PID neural network SPIDNN is used to control the cutting temperature, the value between the cutting temperature and the set temperature is DV(v i ) means using the test vector v i When the single output PID neural network SPIDNN is used to control the cutting temperature, the degree value between the cutting temperature and the set temperature is obtained.

9. The control method according to claim 8, characterized in that: In S3, the updating method is: repeatedly generating the test vector v i and select the best performers to enter the next generation population x i ′ steps until the predetermined number of iterations is reached, and then the individual x with the best performance is selected best , which contains the weight w P , the weight w I and the weight w D This is the updated weight value.

10. An artificial joint replacement surgical instrument, characterized in that: The artificial joint replacement surgical instrument includes the bone grinding rasp assembly according to any one of claims 1 to 3.

Images