Dynamic modeling of six-axis force fiber optic bone drill handpiece and its online error compensation method

Abstract

The application provides a kind of six-dimensional force fiber optical bone drill operator dynamic modeling and its online error compensation method, it is related to optical fiber sensing technical field, the method steps include the six-dimensional force sensor with eight optical fibers is configured, four optical fibers are vertically suspended, and the other four optical fibers are inclined to be suspended, a segment of fiber bragg grating is arranged on each optical fiber;The six-dimensional force sensor is subjected to step calibration test, the center wavelength drift of eight gratings is recorded, and the actual applied force value is used as reference value to construct measurement dataset;Based on the measurement dataset, the improved great white shark optimization algorithm is used to optimize the least squares support vector machine model, the six-dimensional force sensor is dynamically modeled, and the best decoupling model representing the nonlinear dynamic relationship between the six-dimensional force sensor wavelength drift and the six-dimensional force value is obtained, and the best decoupling model is used to decouple the center wavelength drift of fiber bragg grating into six-dimensional force output.

Description

Technical Field

[0001] This invention relates to the field of fiber optic sensing technology, and in particular to a dynamic modeling method for a six-dimensional force fiber optic bone drill manipulator and its online error compensation method. Background Technology

[0002] During high-speed bone drilling, the drill bit is subjected to complex dynamic forces, including axial, radial, and torque forces. The ability to measure and provide real-time, accurate feedback of these six forces on the end effector directly impacts the safety, precision, and intelligence of the bone drill operation. However, the confined operating space and the high-speed rotation of the bone drill place extremely high demands on the size, dynamic response speed, electromagnetic interference resistance, and multi-dimensional force decoupling accuracy of the force sensors.

[0003] Currently, force measurement technologies used in bone drill manipulators mainly include resistance strain gauge, piezoelectric, and capacitive sensors. While resistance strain gauge sensors are relatively inexpensive, their limited dynamic response frequency makes it difficult to capture instantaneous force fluctuations during high-speed drilling, and the complex wiring and signal transmission challenges associated with multiple bridge circuits in a rotating environment further complicate the process. Piezoelectric sensors, while offering good dynamic performance, cannot measure static forces and suffer from charge leakage issues, requiring complex rotating wireless power supply modules, increasing system size and complexity. These existing technologies generally suffer from electromagnetic compatibility problems, and their dynamic measurement accuracy is severely affected by nonlinear coupling and dynamic errors, making it difficult to achieve high-precision dynamic force decoupling and modeling under complex conditions of high-speed rotation and multidimensional coupling. Therefore, they cannot meet the clinical needs of minimally invasive rotating bone drill manipulators for in-situ integration and dynamic measurement. Summary of the Invention

[0004] The purpose of this invention is to provide a dynamic modeling method for a six-dimensional force fiber optic bone drill manipulator and its online error compensation method, so as to solve the problem mentioned in the background art of the difficulty in achieving high-precision dynamic force decoupling and modeling under complex working conditions of high-speed rotation and multi-dimensional coupling in the force measurement technology of existing bone drill manipulators.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a dynamic modeling method for a six-dimensional force fiber optic bone drill manipulator, comprising the following steps: configuring a six-dimensional force sensor with eight optical fibers, wherein four optical fibers are vertically suspended and the other four optical fibers are tilted and suspended, and each optical fiber is provided with a fiber Bragg grating; performing a step calibration test on the six-dimensional force sensor, recording the center wavelength drift of the eight gratings, and using the actual applied force value as a reference value to construct a measurement dataset; based on the measurement dataset, using an improved Great White Shark optimization algorithm to optimize the least squares support vector machine model, dynamically modeling the six-dimensional force sensor, and obtaining the optimal decoupling model characterizing the nonlinear dynamic relationship between the wavelength drift of the six-dimensional force sensor and the six-dimensional force value.

[0006] Optionally, the improved great white shark optimization algorithm specifically includes: initializing the training data of the least squares support vector machine using chaotic mapping to generate an initial great white shark population; using the mean square error between the predicted output of the least squares support vector machine model and the reference force value as the fitness function to evaluate the fitness of each great white shark individual; dynamically dividing the initial great white shark population into an elite subgroup and a normal subgroup according to the individual fitness, and adaptively adjusting the ratio of the two subgroups according to the population diversity; updating the movement speed and position of the elite subgroup and the normal subgroup, and performing crossover mutation and selective back-generation on the elite subgroup to replace inferior individuals in the normal subgroup with superior individuals in the elite subgroup; after iterative optimization, passing the inertia weight and shrinkage factor parameters of the great white shark individual in the best position to the least squares support vector machine model for six-dimensional force decoupling.

[0007] Optionally, the step of initializing the training data of the least squares support vector machine using chaotic mapping specifically includes: initializing the measurement data using chaotic mapping through polar coordinate transformation; finding the optimal solution in the data space, and generating an initial great white shark population centered on the optimal solution through small-range random perturbation, so that the initial individuals are evenly distributed in the search space.

[0008] Optionally, the step of adaptively adjusting the proportion of the two subgroups based on population diversity specifically includes: the proportion of the elite subgroup changes linearly and continuously with the standard deviation of the overall fitness of the population; when population diversity decreases, the proportion of the elite subgroup is automatically increased to enhance local development; when population diversity increases, the proportion of the elite subgroup is reduced to maintain global exploration capability.

[0009] Optionally, the step of updating the movement speed and position of the elite subgroup and the ordinary subgroup specifically includes: the elite subgroup moves towards its own historical optimal direction, and the speed update of the elite subgroup introduces an adaptive shrinkage factor and a nonlinear inertia weight. The adaptive shrinkage factor is dynamically adjusted according to the six-dimensional force coupling degree and the individual fitness, and the nonlinear inertia weight changes adaptively with the iteration stage; the ordinary subgroup selects to move towards the global optimal direction or towards the random individual optimal direction according to the relationship between random numbers and dynamic thresholds, and introduces a chaotic perturbation term in the speed update to avoid getting trapped in local optima.

[0010] Optionally, the construction steps of the least squares support vector machine model specifically include: constructing feature vectors based on the measurement dataset; introducing a radial basis function kernel to map the feature vectors to a high-dimensional feature space, performing linear fitting using a least squares support vector regression algorithm in the high-dimensional feature space, and constructing a nonlinear mapping model that can decouple the wavelength signal into six-dimensional force components; evaluating the modeling accuracy through normalized root mean square error, and optimizing and adjusting the kernel parameters and regularization parameters of the radial basis function kernel.

[0011] On the other hand, the present invention also provides an online error compensation method for a six-dimensional force fiber optic bone drill manipulator based on the above-mentioned dynamic modeling method for the six-dimensional force fiber optic bone drill manipulator. The steps include: real-time acquisition of spectral data from the six-dimensional force sensor, obtaining center wavelength drift signals of multiple fiber Bragg gratings and performing preprocessing; inputting the preprocessed wavelength signals into a pre-trained error compensation inverse mapping model for online real-time decoupling and dynamic error compensation, and outputting high-precision six-dimensional force values. The error compensation inverse mapping model is obtained by constructing the inverse mapping of a nonlinear dynamic system based on the optimal decoupling model; and based on the compensated six-dimensional force data, calling a hybrid deep learning model based on multi-instance learning to identify abnormal patterns in the drilling process.

[0012] Optionally, the step of calling a hybrid deep learning model based on multi-instance learning to identify abnormal patterns in the drilling process specifically includes: dividing the compensated six-dimensional force time-series signal into continuous time-series segments of equal length; packaging multiple continuous segments belonging to the same drilling process into a package; and marking the package as positive or negative based on whether it contains abnormal segments; using a gated recurrent unit network to extract temporal dynamic features for each time-series segment, and converting the temporal features into a spatial feature matrix through a difference matrix; using a convolutional neural network to extract high-order spatial features; introducing an attention mechanism to calculate the attention weight of each segment within its package; weighting and fusing the features of all segments according to their attention weights to generate a package-level feature representation; inputting the package-level features into a classifier for binary classification to determine whether an abnormality has occurred in the current drilling stage; if an abnormality is determined, an abnormality alarm is output and the abnormal time period is located.

[0013] On the other hand, the present invention also provides a dynamic modeling system for a six-dimensional force fiber optic bone drill manipulator, comprising: a configuration module for configuring a six-dimensional force sensor with eight optical fibers, wherein four optical fibers are vertically suspended and the other four optical fibers are tilted and suspended, and each optical fiber is provided with a fiber Bragg grating; a data set module for performing a step calibration test on the six-dimensional force sensor, recording the center wavelength drift of the eight gratings, and constructing a measurement data set using the actual applied force value as a reference value; and a modeling module for dynamically modeling the six-dimensional force sensor based on the measurement data set, using an improved Great White Shark optimization algorithm to optimize the least squares support vector machine model, and obtaining the optimal decoupling model characterizing the nonlinear dynamic relationship between the wavelength drift of the six-dimensional force sensor and the six-dimensional force value.

[0014] On the other hand, the present invention also provides an online error compensation system for a six-dimensional force fiber optic bone drill manipulator based on the above-mentioned dynamic modeling system for a six-dimensional force fiber optic bone drill manipulator, comprising: an acquisition module for real-time acquisition of spectral data from the six-dimensional force sensor, obtaining center wavelength drift signals of multiple fiber Bragg gratings and performing preprocessing; a compensation module for inputting the preprocessed wavelength signals into a pre-trained error compensation inverse mapping model for online real-time decoupling and dynamic error compensation, outputting high-precision six-dimensional force values, wherein the error compensation inverse mapping model is obtained by constructing an inverse mapping of a nonlinear dynamic system based on the optimal decoupling model; and an anomaly identification module for identifying abnormal patterns in the drilling process based on the compensated six-dimensional force data and calling a hybrid deep learning model based on multi-instance learning.

[0015] Compared with the prior art, the beneficial effects of the present invention are: This invention provides a unique hardware foundation for the differentiated sensing and decoupling of multidimensional force signals by configuring a six-dimensional force sensor with a vertical and tilted eight-fiber suspension structure. Furthermore, by combining a high-quality measurement dataset constructed from step calibration experiments, and employing an improved Great White Shark optimization algorithm to optimize the parameters of the least squares support vector machine model, this invention accurately captures and characterizes the nonlinear dynamic relationship between wavelength drift and six-dimensional force values ​​under high-speed rotation and multidimensional coupling conditions. This yields a high-precision optimal decoupling model, significantly improving the accuracy of sensor dynamic decoupling and the robustness of modeling. This provides a reliable theoretical model foundation for the real-time and accurate measurement of cutting forces during subsequent bone drilling operations, effectively ensuring the reliability of force feedback control under complex operating conditions. Attached Figure Description

[0016] Figure 1 This is a schematic diagram of the steps in the dynamic modeling method of the six-dimensional force fiber optic bone drill manipulator of the present invention.

[0017] Figure 2 This is a schematic diagram of the steps of the online error compensation method for the six-dimensional force fiber optic bone drill manipulator of the present invention.

[0018] Figure 3 This is a schematic diagram of the sensor structure of the six-dimensional force fiber optic bone drill manipulator of the present invention.

[0019] Figure 4 This is a flowchart of the optimization algorithm for the six-dimensional force fiber optic bone drill manipulator of this invention.

[0020] Figure 5 This is a flowchart of the online compensation process for the six-dimensional force fiber optic bone drill manipulator of the present invention.

[0021] Figure 6 This is a flowchart of the abnormal mode recognition process for the six-dimensional force fiber optic bone drill manipulator of the present invention.

[0022] Figure 7 This is a schematic diagram of the dynamic modeling system for the six-dimensional force fiber optic bone drill manipulator of the present invention.

[0023] Figure 8 This is a schematic diagram of the online error compensation system for the six-dimensional force fiber optic bone drill manipulator of the present invention.

[0024] In the diagram: 1-elastic body, 2-FBG fiber, 3-connector, 10-configuration module, 20-dataset module, 30-modeling module, 40-acquisition module, 50-compensation module, 60-anomaly identification module. Detailed Implementation

[0025] The present invention will now be clearly and completely described in conjunction with the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0026] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be used interchangeably where appropriate for the embodiments of this application described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0027] Those skilled in the art will understand that, unless explicitly stated otherwise, the singular forms “a,” “an,” “the,” and “the” used herein may also include the plural forms. It should be further understood that the term “comprising” as used in the specification of this application means the presence of features, integers, steps, operations, elements, and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups thereof. It should be understood that when we say an element is “connected” or “coupled” to another element, it can be directly connected or coupled to the other element, or there may be intermediate elements. Furthermore, “connected” or “coupled” as used herein can include wireless connections or wireless coupling. The term “and / or” as used herein includes all or any units and all combinations of one or more associated listed items.

[0028] It will be understood by those skilled in the art that, unless otherwise defined, all terms used herein (including technical and scientific terms) have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains. It should also be understood that terms such as those defined in general dictionaries should be understood to have the same meaning as in the context of the prior art, and should not be interpreted in an idealized or overly formal sense unless specifically defined as herein.

[0029] It should be understood that the sequence number and size of each step in this embodiment do not imply the order of execution. The execution order of each process is determined by its function and internal logic, and should not constitute any limitation on the implementation process of this application embodiment.

[0030] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. This application will now be described in detail with reference to the accompanying drawings and embodiments.

[0031] Please refer to Figures 1-6 The present invention discloses a dynamic modeling method for a six-dimensional force fiber optic bone drill manipulator, comprising the following steps: configuring a six-dimensional force sensor with eight optical fibers, wherein four optical fibers are vertically suspended and the other four optical fibers are tilted and suspended, and each optical fiber is provided with a fiber Bragg grating; performing a step calibration test on the six-dimensional force sensor, recording the center wavelength drift of the eight gratings, and using the actual applied force value as a reference value to construct a measurement dataset; based on the measurement dataset, using an improved Great White Shark optimization algorithm to optimize the least squares support vector machine model, dynamically modeling the six-dimensional force sensor, and obtaining the optimal decoupling model characterizing the nonlinear dynamic relationship between the wavelength drift of the six-dimensional force sensor and the six-dimensional force value.

[0032] Specifically, a six-dimensional force sensor with eight optical fibers is configured. The sensor structure includes an elastic body 1, eight FBG optical fibers 2, and connectors 3. Four of the optical fibers are in a vertically tensioned suspension state, while the other four are in an inclined tensioned suspension state. Each optical fiber has a fiber Bragg grating section. The eight FBG optical fibers 2 extend from the connectors 3 and are embedded in the elastic body 1, which is made of an elastic material.

[0033] A step calibration test was performed on the six-dimensional force sensor. The six-dimensional force sensor was securely mounted on the calibration platform and tightened with screws. The calibration cap was also tightened above the six-dimensional force sensor with screws. During the test, a standard weight was suspended from one end of a fishing line, and the other end was securely attached to the calibration cap via a fixed pulley on the step force loading bracket. It was strictly necessary to ensure that the tension direction of the fishing line was aligned with the single-dimensional force direction of the six-dimensional force sensor. At a certain instant, a negative step excitation was applied to the sensor by burning the fishing line, independently exciting each of the six degrees of freedom (Fx, Fy, Fz, Mx, My, Mz). The force readings of the six-dimensional force sensor were acquired by an external high-performance data acquisition card and used as reference force values. Simultaneously, the center wavelength drift of the eight gratings was recorded by an external demodulator. Based on the recorded center wavelength drift of the eight gratings and the corresponding reference force values, a measurement dataset of the six-dimensional force sensor was constructed.

[0034] Based on the aforementioned measurement dataset, an improved Great White Shark optimization algorithm is used to optimize the least squares support vector machine model for dynamic modeling of the six-dimensional force sensor. Specifically, samples are constructed using raw data obtained from step calibration experiments, divided into training and test sets. The Great White Shark algorithm is applied for global parameter optimization. The optimized least squares support vector machine is configured using the best parameters and tested for verification, thereby obtaining the optimal decoupling model that can characterize the nonlinear dynamic relationship between the wavelength drift of the six-dimensional force sensor and the six-dimensional force value. This optimal decoupling model can be used to subsequently decouple the center wavelength drift of the fiber Bragg grating into a six-dimensional force output, achieving high-precision measurement of dynamic interactive forces during bone drill cutting.

[0035] This application provides a unique hardware foundation for the differentiated sensing and decoupling of multidimensional force signals by configuring a six-dimensional force sensor with a vertical and tilted eight-fiber suspension structure. Furthermore, by combining the high-quality measurement dataset constructed from step calibration experiments, an improved Great White Shark optimization algorithm is used to optimize the parameters of the least squares support vector machine model. This accurately captures and characterizes the nonlinear dynamic relationship between the wavelength drift and the six-dimensional force value of the six-dimensional force sensor under high-speed rotation and multidimensional coupling conditions, thereby obtaining a high-precision optimal decoupling model. This significantly improves the accuracy of sensor dynamic decoupling and the robustness of modeling, providing a reliable theoretical model foundation for the real-time and accurate measurement of cutting forces during subsequent bone drilling operations, and effectively ensuring the reliability of force feedback control under complex operating environments. In some embodiments, the improved great white shark optimization algorithm specifically includes: initializing the training data of the least squares support vector machine using chaotic mapping to generate an initial great white shark population; using the mean square error between the predicted output of the least squares support vector machine model and the reference force value as a fitness function to evaluate the fitness of each great white shark individual; dynamically dividing the initial great white shark population into an elite subpopulation and a normal subpopulation according to the individual fitness, and adaptively adjusting the ratio of the two subpopulations according to the population diversity; updating the movement speed and position of the elite subpopulation and the normal subpopulation, and performing crossover mutation and selective back-generation on the elite subpopulation to replace inferior individuals in the normal subpopulation with superior individuals in the elite subpopulation; after iterative optimization, passing the inertia weight and shrinkage factor parameters of the great white shark individual in the best position to the least squares support vector machine model for six-dimensional force decoupling.

[0036] Specifically, the step of initializing the training data of the least squares support vector machine using chaotic mapping includes: initializing the measurement data using chaotic mapping through polar coordinate transformation; finding the optimal solution in the data space, and generating an initial great white shark population centered on the optimal solution through small-scale random perturbation, so that the initial individuals are evenly distributed in the search space. This application uses Circle chaotic mapping to initialize the measurement data through polar coordinate transformation, which can find a better initial solution in the search space; generating an initial population centered on this optimal solution through small-scale random perturbation, so that the individuals are evenly distributed throughout the search space, effectively avoiding the premature convergence problem that may be caused by traditional initialization methods, providing a high-quality starting point for subsequent global optimization search, thereby improving the overall optimization efficiency of the algorithm and the stability of the decoupled model.

[0037] The steps of adaptively adjusting the proportion of the two subgroups based on population diversity specifically include: the proportion of the elite subgroup changes linearly and continuously with the standard deviation of the overall fitness of the population; when population diversity decreases, the proportion of the elite subgroup is automatically increased to enhance local development; when population diversity increases, the proportion of the elite subgroup is reduced to maintain global exploration capability.

[0038] The steps of updating the movement speed and position of the elite subgroup and the ordinary subgroup specifically include: the elite subgroup moves towards its own historical optimal direction, and the speed update of the elite subgroup introduces an adaptive shrinkage factor and a nonlinear inertial weight. The adaptive shrinkage factor is dynamically adjusted according to the six-dimensional force coupling degree and the individual fitness, and the nonlinear inertial weight changes adaptively with the iteration stage; the ordinary subgroup selects to move towards the global optimal direction or towards the random individual optimal direction according to the relationship between random numbers and dynamic thresholds, and introduces a chaotic perturbation term in the speed update to avoid getting trapped in local optima.

[0039] For example, the Circle chaotic mapping is used to initialize the data during the training of the least squares support vector machine through polar coordinate transformation to obtain the optimal solution. Then, the initial great white shark population is obtained by using the optimal solution as the center and completing the initialization through small-range random perturbation. This can enhance the population diversity and avoid getting trapped in local optima.

[0040] The fitness function F is used to evaluate the current inertia weight and shrinkage factor. The fitness function F is set as the mean square error between the six-dimensional force prediction output of the least squares support vector machine and the reference force value, expressed as: Where n is the number of training data; and Let represent the predicted value and reference value of the i-th training data, respectively. The mean squared error is used as the fitness function to ensure that the optimization direction is consistent with the modeling accuracy target.

[0041] Individual fitness was sorted in ascending order, and at different stages, great white sharks were divided into elite and ordinary subgroups based on their fitness at a certain ratio. ;in For each individual, the fitness value This represents the average fitness value of all individuals in the current population. The difference in fitness values ​​between two consecutive iterations for each individual. The difference between the average fitness values ​​of all individuals in the current population over two consecutive iterations. and , where is the weighting coefficient. By dynamically dividing the elite subgroup into elite and ordinary subgroups and adaptively adjusting their ratio, a balance between the algorithm's global exploration and local exploitation capabilities is achieved. When population diversity decreases, the proportion of the elite subgroup is automatically increased to enhance local exploitation capabilities; when population diversity increases, the proportion of the elite subgroup is automatically decreased to maintain global exploration capabilities. This dynamic adaptive adjustment mechanism effectively avoids the algorithm getting trapped in local optima or experiencing low search efficiency, ensuring that the optimization process maintains ideal search performance at different stages.

[0042] Calculate the overall fitness standard deviation of the population , ,Will Sort in ascending order; the proportion of elite subgroups varies with standard deviation. Linear continuous change, The top ρ% are taken as the elite subgroup. Population diversity is quantified by calculating the standard deviation of the overall fitness of the population, and the proportion of the elite subgroup changes linearly and continuously with the standard deviation.

[0043] The elite subgroup moves towards the direction optimal for its own individuals, with its speed updated and defined as follows: Among them, adaptive shrinkage factor Nonlinear inertial weights , The random numbers generated by individuals in a normal subgroup are greater than the parameter. When moving towards the globally optimal direction, the speed is updated and defined as follows: ,in The random numbers generated by individuals in a normal subgroup are less than or equal to the parameter. At that time, it randomly moves in the direction optimal for other individuals, and its speed is updated and defined as follows: ,in, An adaptive shrinkage factor and nonlinear inertia weight are introduced into the velocity update process. The shrinkage factor dynamically adjusts based on the six-dimensional force coupling degree and individual fitness, while the inertia weight adaptively changes with each iteration, enhancing the algorithm's adaptability to complex optimization problems. The ordinary subgroup selects different movement directions based on the relationship between random numbers and dynamic thresholds, and a chaotic perturbation term is introduced into the velocity update, effectively avoiding getting trapped in local optima. These mechanisms enable the algorithm to better adapt to the nonlinear characteristics of the multidimensional coupling of the six-dimensional force sensor, improving the accuracy and robustness of parameter optimization. After updating the velocity of the great white shark population, the individual positions are corrected by position projection, adhering to the principles of position update. P[·] is the projection mapping operator used for selective back-introduction of the elite subgroup. Crossover mutation and selective back-introduction of the elite subgroup effectively preserve high-quality genes and improve the overall quality of the population.

[0044] Repeat the above steps to pass the inertia weight and shrinkage factor of the great white shark with the best position to the least squares support vector machine.

[0045] In some embodiments, the construction steps of the least squares support vector machine model specifically include: constructing feature vectors based on the measurement dataset; introducing a radial basis function kernel to map the feature vectors to a high-dimensional feature space, performing linear fitting using a least squares support vector regression algorithm in the high-dimensional feature space, and constructing a nonlinear mapping model that can decouple the wavelength signal into six-dimensional force components; evaluating the modeling accuracy by normalized root mean square error, and optimizing and adjusting the kernel parameters and regularization parameters of the radial basis function kernel.

[0046] Specifically, the model input feature vector is constructed based on the output force of the six-dimensional force sensor. Let the output of the six-dimensional force sensor be... The input is Then the model input feature vector .

[0047] Introducing the RBF radial basis function maps the input space to a high-dimensional feature space, where the RBF radial basis function is: .

[0048] Linear regression fitting is performed within this space, using the least squares support vector regression algorithm. The model training objective is... The constraints are ,in, For regularization parameters, Using the kernel mapping function, a dual problem is constructed through the Lagrange multiplier method. The final solution is represented by the kernel matrix composed of training samples, thus realizing the system modeling of the nonlinear dynamic relationship of the six-dimensional force sensor.

[0049] The modeling error is represented by the normalized root mean square error between the actual output yd and the model output ym after preprocessing by the six-dimensional force sensor. The normalized root mean square error is calculated based on the data to verify and optimize the hyperparameters. and .

[0050] This application fully explores the temporal correlation of the dynamic response of a six-dimensional force sensor by constructing feature vectors containing historical outputs and inputs. It introduces a radial basis function kernel to map the input to a high-dimensional feature space, performing linear regression fitting in this high-dimensional space, which effectively handles the nonlinear mapping relationship between wavelength signals and six-dimensional force values. By evaluating the modeling accuracy through normalized root mean square error and optimizing the kernel and regularization parameters, the application ensures that the constructed model accurately describes the nonlinear dynamic characteristics of the sensor, providing a high-precision mathematical mapping model for subsequent force value decoupling.

[0051] On the other hand, the present invention also provides an online error compensation method for a six-dimensional force fiber optic bone drill manipulator based on the above-mentioned dynamic modeling method for the six-dimensional force fiber optic bone drill manipulator. The steps include: real-time acquisition of spectral data from the six-dimensional force sensor, obtaining center wavelength drift signals of multiple fiber Bragg gratings and performing preprocessing; inputting the preprocessed wavelength signals into a pre-trained error compensation inverse mapping model for online real-time decoupling and dynamic error compensation, and outputting high-precision six-dimensional force values. The error compensation inverse mapping model is obtained by constructing the inverse mapping of a nonlinear dynamic system based on the optimal decoupling model; and based on the compensated six-dimensional force data, calling a hybrid deep learning model based on multi-instance learning to identify abnormal patterns in the drilling process.

[0052] Specifically, the processed wavelength signal data is input into a pre-trained optimization algorithm-optimized least squares support vector machine error compensation model. This optimization model is used to perform online real-time decoupling and dynamic error compensation calculations on the wavelength signal, and outputs the calculated high-precision six-dimensional force value after compensation in real time. The input of the processed wavelength signal data into the pre-trained optimization algorithm-optimized least squares support vector machine error compensation model is accomplished through offline calibration data from a six-dimensional force sensor.

[0053] The center wavelength drift signal data of the fiber Bragg grating is preprocessed. After dividing the data into training and test sets, the center wavelength is input into the decoupling model for training. After completing the dynamic modeling of the six-dimensional force, it is imported into the error compensation model for training. The trained error compensation model is saved, resulting in a pre-trained optimized least squares support vector machine error compensation model. The online real-time decoupling and dynamic error compensation calculation of the wavelength signal using this optimized model is achieved by adopting a data-driven machine learning method. Error compensation is realized by performing online real-time decoupling and dynamic error compensation calculation of the wavelength signal.

[0054] The Great White Shark optimization algorithm is selected to optimize the least squares support vector machine, and an inverse mapping system model of the six-dimensional force sensor nonlinear dynamic system is constructed. The true strength estimate after compensation is obtained. The system acquires the center wavelength signal data of the six-dimensional force sensor in real time online. After preprocessing the data, the trained error compensation inverse mapping system model is embedded into the established Django framework to perform dynamic real-time error compensation on the acquired center wavelength data. The six-dimensional force value is then visualized and output, thus completing the online dynamic error compensation for the six-dimensional force fiber optic bone drill manipulator.

[0055] The calculated six-dimensional force sensor time-series signal is denoised and normalized before being divided into continuous time-series segments of equal length. Multiple continuous segments belonging to the same drilling process or drilling stage are packaged into a single package, which is then labeled as a positive or negative package.

[0056] The GRU is used to extract temporal dynamic features for each time segment and convert them into a spatial feature matrix. The CNN is then used to perform deep convolution and pooling operations on the spatial matrix to extract high-order spatial features, thus obtaining the feature representation of each segment.

[0057] Based on the features of the segments, an attention mechanism is introduced. According to the features of each segment obtained in S62, the attention weight of that segment within its respective bag is calculated. The features of all segments are then weighted and summed according to their attention weights to generate a global bag-level feature representation. The selection of DIP is based on an evaluation function. In the formula ,in, The attention weights corresponding to the instances. , and These are the training parameters for the feature extraction model.

[0058] Packet-level features are input into a classifier for binary classification. The classifier consists of a fully connected layer and a Softmax activation function. The classifier outputs the probability that the packet is a positive or negative packet. Based on the probability threshold, it is determined whether an anomaly has occurred in the current drilling stage. The above hybrid deep learning model is trained end-to-end using the constructed packet-label dataset. The model parameters are updated iteratively multiple times through the backpropagation algorithm, highlighting packets with higher attention weights calculated by the attention mechanism. The trained model parameters are then solidified, and the model is imported into the control system of the six-dimensional force fiber optic bone drill manipulator to complete the identification of drilling anomalies by the six-dimensional force fiber optic bone drill manipulator.

[0059] In this application, the output of the FBG sensor in the six-dimensional force fiber optic bone drill manipulator is affected by dynamic errors under high-speed rotation, instantaneous impact, and multi-dimensional coupling environments, directly weakening the accuracy of multi-dimensional force decoupling and feedback control. This solution introduces system identification and real-time compensation methods to model and correct the dynamic errors of the six-dimensional force sensor online, and identifies abnormal modes of the six-dimensional force fiber optic bone drill manipulator during the drilling process, improving the force measurement accuracy and system reliability in complex environments.

[0060] This application acquires and preprocesses sensor spectral data in real time, then inputs the wavelength signal into a pre-trained error compensation inverse mapping model for online real-time compensation of dynamic errors. This inverse mapping model, based on an optimal decoupling model, constructs an inverse mapping of the nonlinear dynamic system, effectively eliminating dynamic errors in sensor output under high-speed rotation and multi-dimensional coupling conditions. A training dataset suitable for multi-instance learning is constructed by dividing the six-dimensional force time-series signal into equal-length segments and packaging them into packets. Gated recurrent units are used to extract temporal dynamic features, which are then converted into spatial feature matrices using a difference matrix. High-order spatial features are extracted using a convolutional neural network, fully exploiting the spatiotemporal correlation information in the force signal. An attention mechanism is introduced to calculate the weights of each segment within the packet and performs weighted fusion, generating packet-level features with global representation capabilities. These packet-level features are input into a classifier for binary classification, accurately determining whether anomalies occur during the drilling stage and locating abnormal time periods, providing real-time safety warnings and decision support for bone drill operations.

[0061] Please refer to Figure 7On the other hand, the present invention also provides a dynamic modeling system for a six-dimensional force fiber optic bone drill manipulator, comprising: a configuration module 10 for configuring a six-dimensional force sensor with eight optical fibers, wherein four optical fibers are vertically suspended and the other four optical fibers are tilted and suspended, and each optical fiber is provided with a fiber Bragg grating; a dataset module 20 for performing a step calibration test on the six-dimensional force sensor, recording the center wavelength drift of the eight gratings, and constructing a measurement dataset using the actual applied force value as a reference value; and a modeling module 30 for dynamically modeling the six-dimensional force sensor based on the measurement dataset, using an improved Great White Shark optimization algorithm to optimize the least squares support vector machine model, and obtaining the optimal decoupling model characterizing the nonlinear dynamic relationship between the wavelength drift of the six-dimensional force sensor and the six-dimensional force value.

[0062] Please refer to Figure 8 On the other hand, the present invention also provides an online error compensation system for a six-dimensional force fiber optic bone drill manipulator based on the above-mentioned dynamic modeling system for a six-dimensional force fiber optic bone drill manipulator, comprising: an acquisition module 40, used to acquire the spectral data of the six-dimensional force sensor in real time, obtain the center wavelength drift signals of multiple fiber Bragg gratings and perform preprocessing; a compensation module 50, used to input the preprocessed wavelength signals into a pre-trained error compensation inverse mapping model, perform online real-time decoupling and dynamic error compensation, and output high-precision six-dimensional force values, wherein the error compensation inverse mapping model is obtained by constructing the inverse mapping of a nonlinear dynamic system based on the optimal decoupling model; and an anomaly identification module 60, used to identify abnormal patterns in the drilling process by calling a hybrid deep learning model based on multi-instance learning based on the compensated six-dimensional force data.

[0063] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause an electronic device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0064] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. 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 methods. Any references to memory, storage, database, or other media used in the embodiments provided by this invention can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and RAMbus dynamic RAM (RDRAM), etc.

[0065] The above are merely embodiments of the present invention and do not limit the patent scope of the present invention. Any equivalent modifications made based on the content of the present invention's specification and drawings, or direct or indirect applications in related technical fields, are similarly included within the patent protection scope of the present invention.

Claims

1. A dynamic modeling method for a six-dimensional force fiber optic bone drill manipulator, characterized by the following steps: include: A six-dimensional force sensor with eight optical fibers is configured, of which four fibers are suspended vertically and the other four fibers are suspended at an angle. Each fiber is equipped with a fiber Bragg grating. A step calibration test was performed on the six-dimensional force sensor, the center wavelength drift of the eight gratings was recorded, and the actual applied force value was used as a reference value to construct a measurement dataset. Based on the measurement dataset, the least squares support vector machine model is optimized using the improved Great White Shark optimization algorithm to dynamically model the six-dimensional force sensor, thereby obtaining the optimal decoupling model that characterizes the nonlinear dynamic relationship between the wavelength drift of the six-dimensional force sensor and the six-dimensional force value.

2. The dynamic modeling method for a six-dimensional force fiber optic bone drill manipulator according to claim 1, characterized in that, The improved Great White Shark optimization algorithm specifically includes: The training data for the least squares support vector machine was initialized using chaotic mapping to generate the initial population of great white sharks. The fitness of each great white shark individual was evaluated by using the mean square error between the predicted output of the least squares support vector machine model and the reference force value as the fitness function. The initial population of great white sharks was dynamically divided into elite and ordinary subpopulations based on individual fitness, and the ratio of the two subpopulations was adaptively adjusted according to population diversity. Update the movement speed and position of the elite subgroup and the ordinary subgroup, and perform crossover mutation and selective back-generation on the elite subgroup, replacing inferior individuals in the ordinary subgroup with superior individuals in the elite subgroup. After iterative optimization, the inertial weight and contraction factor parameters of the great white shark individual in the optimal position are passed to the least squares support vector machine model for six-dimensional force decoupling.

3. The dynamic modeling method for a six-dimensional force fiber optic bone drill manipulator according to claim 2, characterized in that, The steps for initializing the training data of the least squares support vector machine using chaotic mapping specifically include: The measurement data is initialized using chaotic mapping through polar coordinate transformation; Find the optimal solution in the data space, and generate an initial great white shark population centered on the optimal solution through small-scale random perturbation, so that the initial individuals are evenly distributed in the search space.

4. The dynamic modeling method for a six-dimensional force fiber optic bone drill manipulator according to claim 3, characterized in that, The step of adaptively adjusting the ratio of the two subgroups based on population diversity specifically includes: The proportion of the elite subpopulation changes linearly and continuously with the standard deviation of the overall fitness of the population. When population diversity decreases, the proportion of elite subpopulations is automatically increased to enhance local exploitation; When population diversity increases, the proportion of elite subpopulations is reduced to maintain global exploration capabilities.

5. The dynamic modeling method for a six-dimensional force fiber optic bone drill manipulator according to claim 4, characterized in that, The steps of updating the movement speed and position of the elite subgroup and the ordinary subgroup specifically include: The elite subgroup moves toward its historical best direction. The speed update of the elite subgroup introduces an adaptive shrinkage factor and a nonlinear inertia weight. The adaptive shrinkage factor is dynamically adjusted according to the six-dimensional force coupling degree and the individual fitness level. The nonlinear inertia weight changes adaptively with the iteration stage. The ordinary subgroup selects to move towards the global optimum or towards the random individual optimum based on the relationship between the random number and the dynamic threshold, and introduces a chaotic perturbation term in the velocity update to avoid getting trapped in local optima.

6. The dynamic modeling method for a six-dimensional force fiber optic bone drill manipulator according to claim 1, characterized in that, The specific steps for constructing the least squares support vector machine model include: Construct feature vectors based on the measurement dataset; A radial basis function is introduced to map the feature vector to a high-dimensional feature space. In the high-dimensional feature space, the least squares support vector regression algorithm is used for linear fitting to construct a nonlinear mapping model that can decouple the wavelength signal into six-dimensional force components. The modeling accuracy is evaluated by normalized root mean square error, and the kernel parameters and regularization parameters of the radial basis kernel function are optimized and adjusted.

7. An online error compensation method for a six-dimensional force fiber optic bone drill manipulator based on the dynamic modeling method of the six-dimensional force fiber optic bone drill manipulator according to claim 1, characterized in that the steps are as follows: include: The spectral data of the six-dimensional force sensor is acquired in real time, and the center wavelength drift signals of multiple fiber Bragg gratings are obtained and preprocessed. The preprocessed wavelength signal is input into a pre-trained error compensation inverse mapping model for online real-time decoupling and dynamic error compensation, and outputs a high-precision six-dimensional force value. The error compensation inverse mapping model is based on the optimal decoupling model and is obtained by constructing the inverse mapping of a nonlinear dynamic system. Based on the compensated six-dimensional force data, a hybrid deep learning model based on multi-instance learning is invoked to identify abnormal patterns in the drilling process.

8. The online error compensation method for the six-dimensional force fiber optic bone drill manipulator according to claim 7, characterized in that, The steps of calling the hybrid deep learning model based on multi-instance learning to identify abnormal patterns in the drilling process specifically include: The compensated six-dimensional force time-series signal is divided into continuous time-series segments of equal length. Multiple continuous segments belonging to the same drilling process are packaged into a single package, and the package is marked as a positive or negative package based on whether it contains abnormal segments. Gated recurrent unit networks are used to extract temporal dynamic features for each time segment, and the temporal features are converted into spatial feature matrices through difference matrices. Convolutional neural networks are then used to extract high-order spatial features. An attention mechanism is introduced to calculate the attention weight of each segment within its respective bag. The features of all segments are weighted and fused according to their attention weights to generate a bag-level feature representation. The package-level features are input into the classifier for binary classification to determine whether an anomaly has occurred in the current drilling stage. If an anomaly is detected, an anomaly alarm is output and the abnormal time period is located.

9. A dynamic modeling system for a six-dimensional force fiber optic bone drill manipulator, characterized in that, include: The configuration module is used to configure a six-dimensional force sensor with eight optical fibers, four of which are suspended vertically and the other four are suspended at an angle. Each fiber is equipped with a fiber Bragg grating. The dataset module is used to perform a step calibration test on the six-dimensional force sensor, record the center wavelength drift of the eight gratings, and construct a measurement dataset using the actual applied force value as a reference value. The modeling module is used to optimize the least squares support vector machine model based on the measurement dataset using the improved Great White Shark optimization algorithm, and to dynamically model the six-dimensional force sensor to obtain the optimal decoupling model characterizing the nonlinear dynamic relationship between the wavelength drift of the six-dimensional force sensor and the six-dimensional force value.

10. An online error compensation system for a six-dimensional force fiber optic bone drill manipulator based on the dynamic modeling system of the six-dimensional force fiber optic bone drill manipulator according to claim 9, characterized in that, include: The acquisition module is used to acquire the spectral data of the six-dimensional force sensor in real time, obtain the center wavelength drift signals of multiple fiber Bragg gratings and perform preprocessing. The compensation module is used to input the preprocessed wavelength signal into the pre-trained error compensation inverse mapping model to perform online real-time decoupling and dynamic error compensation, and output a high-precision six-dimensional force value. The error compensation inverse mapping model is based on the optimal decoupling model and is obtained by constructing the inverse mapping of the nonlinear dynamic system. The anomaly identification module is used to identify abnormal patterns in the drilling process by calling a hybrid deep learning model based on multi-instance learning, based on the compensated six-dimensional force data.

Images