Mechanical arm assisted orthodontic arch wire bending system and method for machine learning optimization

Abstract

The invention relates to the technical field of orthodontic treatment, and discloses a mechanical arm assisted orthodontic arch wire bending system and method for machine learning optimization. The system comprises a patient data acquisition and three-dimensional modeling module, a personalized sequence bending design module, a mechanical arm bending module, a real-time feedback and closed-loop control module and a user interaction and operation interface module. And designing an arch wire bending path through machine learning based on the model, then driving a mechanical arm to bend the arch wire according to the path, and performing real-time monitoring and closed-loop control to correct the deviation so as to ensure the precision. Through personalized design driven by machine learning, precise bending of the mechanical arm and real-time closed-loop control, the precision, efficiency and consistency of orthodontic arch wire manufacturing are remarkably improved, the treatment risk is reduced through biomechanical simulation and intelligent interaction, operation and management are optimized, and many defects of the traditional technology are overcome.

Description

Technical Field

[0001] This invention relates to the field of orthodontic treatment technology, specifically to a machine learning-optimized robotic arm-assisted orthodontic archwire bending system and method. Background Technology

[0002] In the field of orthodontic treatment, fixed braces remain the mainstream treatment mode. The core of fixed braces lies in guiding tooth movement by applying appropriate biomechanical forces through archwires. As the primary force-applying device, the degree of personalization in the design and the precision in the fabrication of the orthodontic archwire play a decisive role in the final treatment outcome. For a long time, clinicians and related technology researchers have been committed to improving the quality and efficiency of archwire fabrication to meet increasingly sophisticated clinical needs. However, in existing and previous technical practices, the design and manufacturing process of orthodontic archwires still faces some inherent challenges and areas requiring optimization.

[0003] In classic edgewise orthodontic techniques, archwire bending begins from the very first step. In the currently widely used straight-wire orthodontic techniques, the alignment and leveling stages rely on pre-fabricated archwires. Personalized archwire bending primarily focuses on the fine-tuning stage to compensate for the initial precise fit between the pre-fabricated archwire and the patient's dentition. This fine-tuning stage relies heavily on the clinician's professional experience, understanding of standard arch morphology, and assessment of the patient's specific situation. This method is effective for routine cases. However, when dealing with cases of complex tooth alignment and significant individual differences, relying solely on experience and two-dimensional analysis makes it difficult to accurately predict and plan the ideal archwire shape in three-dimensional space. This makes achieving a high degree of individualization and biomechanical optimization in the archwire design process challenging.

[0004] Furthermore, in routine treatment plans, the assessment and adjustment of orthodontic forces largely rely on indirect inferences and empirical adjustments by the physician. For complex biomechanical factors such as the periodontal tissue condition and the resistance center of tooth movement in a specific patient, there is a lack of direct, quantitative analytical methods to accurately predict and optimize the biomechanical system. Therefore, clinical decision-making faces considerable complexity in ensuring efficient and safe tooth movement.

[0005] In the physical forming and manufacturing of bowwires, traditional manual bending techniques have long been widely used. While flexible, this method is relatively inefficient and requires a high level of skill and experience from the operator. The accuracy and consistency of bowwires produced by different operators, or even by the same operator at different times, are difficult to guarantee consistently. Subsequent mechanical bending tools have improved the convenience of bending to some extent, but precise control of complex three-dimensional bending and effective compensation for the springback effect of bowwire materials (especially high-elasticity alloys) during bending remain technical challenges. These factors can all lead to deviations between the final bowwire and the intended design.

[0006] Meanwhile, with the gradual penetration of digital technology into the field of oral medicine, related software and hardware tools have begun to be applied in orthodontic clinical practice. However, how to effectively integrate these scattered technical points to form a smooth, efficient, and complete digital workflow covering data acquisition, treatment planning, archwire fabrication, treatment monitoring, and data management remains a direction for continuous exploration in the industry. In the existing process, there is still room for further improvement and refinement in terms of data conversion between different stages, the convenience of human-computer interaction, and the systematic nature of information management. Summary of the Invention

[0007] To address the shortcomings of existing technologies, this invention provides a machine learning-optimized robotic arm-assisted orthodontic archwire bending system and method, which solves the problems of existing technologies in efficiently and accurately achieving personalized orthodontic archwire design, manufacturing, and biomechanical optimization, as well as the lack of intelligent and integrated operation and management processes.

[0008] To achieve the above objectives, the present invention provides the following technical solution: a machine learning-optimized robotic arm-assisted orthodontic archwire bending system, comprising: The patient data acquisition and 3D modeling module is used to acquire patient oral data and generate personalized 3D oral models of patients. A personalized sequence bending design module is connected to the patient data acquisition and 3D modeling module, and is used to design a personalized sequence archwire bending path based on the personalized 3D oral model using artificial intelligence algorithms. The robotic arm bending module is connected to the personalized sequence bending design module and is used to drive the robotic arm to perform bowwire bending according to the personalized sequence bowwire bending path. A real-time feedback and closed-loop control module is connected to the robotic arm bending module to monitor the bending status in real time during the bowwire bending process and perform closed-loop control to correct execution deviations. The user interaction and operation interface module is connected to the personalized sequence bending design module and the real-time feedback and closed-loop control module, and is used to visually monitor the bending process, adjust the design parameters, and provide intelligent assistance.

[0009] Preferably, the patient data acquisition and 3D modeling module includes: The data acquisition unit is used to acquire three-dimensional data of the crown surface through an intraoral scanner and to acquire three-dimensional image data including the tooth root and jawbone structure through a cone-beam computed tomography (CBCT) device. The data processing and fusion unit is used to preprocess, register, and fuse the collected three-dimensional data and three-dimensional image data of the crown surface to generate a personalized three-dimensional oral model, and to optimize the personalized three-dimensional oral model.

[0010] Preferably, the personalized sequence bending design module includes: A deep learning analysis unit is used to perform feature analysis and initial path planning on the personalized 3D oral model. The biomechanical modeling and simulation unit is used for biomechanical verification and optimization of the initial path planning; The reinforcement learning optimization unit is used to dynamically optimize the archwire bending path sequence to form bending strategies for different treatment stages.

[0011] Preferably, the deep learning analysis unit uses convolutional neural networks (CNN) and graph neural networks (GNN) to process personalized three-dimensional oral cavity model data. To extract the morphological features of teeth and the topological relationships between teeth.

[0012] Preferably, the biomechanical modeling and simulation unit employs the finite element analysis (FEA) method, based on the personalized three-dimensional oral cavity model and the preliminary archwire bending path. Simulates the stress on periodontal tissues after archwire loading. And according to the preset physiological safety threshold The evaluation and optimization of the path can be expressed as follows: .

[0013] Preferably, the reinforcement learning optimization unit achieves dynamic optimization of the archwire bending path sequence and forms the bending strategy for different treatment stages in the following manner: This unit models the archwire bending process associated with the archwire bending path sequence as a Markov decision process and optimizes a strategy. To maximize expected cumulative reward The cumulative reward is defined based on the healing effect and efficiency after the bowwire is bent, whereby... This is the current treatment status. For adjusting the bowwire bending parameters, ,in, For instant rewards, This is the discount factor.

[0014] Preferably, the robotic arm bending module includes a multi-degree-of-freedom robotic arm and a dedicated end effector for bending wire, and has a dynamic bending force and angle compensation function based on the characteristics of the selected bowwire material. This angle compensation amount... Based on the springback model of bowwire materials Target bending angle and bending radius Calculations are performed to make the actual forming angle close to the target bending angle.

[0015] Preferably, the real-time feedback and closed-loop control module includes: The multimodal sensor data real-time acquisition unit is used to acquire the joint position, end-effector posture, actual bending angle of the archwire, force on the archwire, and real-time three-dimensional morphological data of the archwire of the robotic arm. The real-time deviation calculation and error assessment unit is used to compare the real-time acquired bending state parameters with the personalized sequence bowwire bending path to calculate the execution deviation. ; An AI-based closed-loop control command generation unit is used to generate commands based on the execution deviation. and current system status Generate correction control instructions This unit employs the Model Predictive Control (MPC) algorithm, which optimizes a series of control inputs in the prediction time domain to minimize the deviation of the predicted system output from the personalized sequence of archwire bending paths. Simultaneously considering the cost of the control inputs, it determines the optimal corrective control command. This is to drive the robotic arm to adjust its movements.

[0016] Preferably, the user interaction and operation interface module includes: The 3D visualization and dynamic rendering unit is used to render and dynamically display the personalized 3D oral cavity model, the personalized sequence archwire bending path, the bending action of the robotic arm, and the archwire forming process in real time in 3D form. The intelligent decision support and suggestion unit is used to push intelligent decision support information and clinical suggestions to users based on real-time monitoring data and treatment rule engine; The treatment report and case management unit is used to automatically integrate bending process data to generate a structured treatment report and support case archiving after treatment is completed. The bending process data includes parameters of the personalized sequential archwire bending path and actual execution parameters monitored by the real-time feedback and closed-loop control module.

[0017] This invention also provides a machine learning-optimized robotic arm-assisted orthodontic archwire bending method, comprising the following steps: Collect and model patient oral data: Acquire patient oral data and generate a personalized three-dimensional oral model based on the patient's oral data, thereby obtaining an accurate digital oral representation; Personalized sequential bending design is carried out based on the obtained accurate digital oral cavity representation: according to the personalized three-dimensional oral cavity model, a personalized sequential archwire bending path is designed through artificial intelligence algorithm, thereby obtaining digital path parameters to guide subsequent bending. Based on the obtained digital path parameters, the robotic arm performs bending: it drives the robotic arm and performs physical bending of the bowwire according to the personalized bowwire bending path sequence; During the physical bending process of the bowwire, real-time feedback and closed-loop control are implemented: the bending status is monitored in real time, and the monitored bending status is compared with the digital path parameters. Based on the deviation generated by the comparison, closed-loop control is performed to correct the execution action of the robotic arm, thereby ensuring the final bending accuracy.

[0018] This invention provides a machine learning-optimized robotic arm-assisted orthodontic archwire bending system and method. It offers the following advantages: 1. This invention integrates a complete technical process of patient oral data acquisition, AI-based personalized path design, and high-precision execution by robotic arms. This solution achieves individualized and precise archwire design. Compared with existing technologies that rely on experience or standardized templates, this invention solves the defects of insufficient personalization and difficulty in unifying precision.

[0019] 2. This invention uses a robotic arm to bend the archwire and combines real-time feedback with artificial intelligence closed-loop control technology for process monitoring and adjustment. This method improves the efficiency, accuracy and consistency of archwire manufacturing. Compared with traditional manual bending or simple mechanical-assisted bending, this invention overcomes the problems of low production efficiency and large fluctuations in finished product accuracy, thereby enhancing the predictability of treatment.

[0020] 3. This invention introduces a biomechanical modeling and simulation unit into the personalized sequential bending design module. This unit pre-evaluates and optimizes the mechanical effects applied to periodontal tissues before the archwire is physically formed, ensuring that the orthodontic force is within the physiologically safe range. Compared with the existing technology that mainly relies on the clinical doctor's experience to judge the orthodontic force, this invention reduces the risk of periodontal tissue damage caused by improper force values.

[0021] 4. This invention is equipped with a user interaction and operation interface module, providing three-dimensional visualization, intelligent decision support and treatment data management functions. Users can intuitively monitor the entire design and bending process and obtain auxiliary information provided by the system based on real-time data and rule engine. After treatment, a structured report can be automatically generated. Compared with the problems of insufficient intuitive human-computer interaction and low information integration in the prior art, this invention improves the ease of use and clinical data management efficiency of the system. Attached Figure Description

[0022] Figure 1 This is a flowchart illustrating the framework of the system modules of the present invention. Figure 2 This is a flowchart illustrating the framework of the personalized three-dimensional oral cavity module of this invention. Figure 3 This is a flowchart illustrating the framework of the robotic arm bending module of the present invention. Figure 4 This is a flowchart illustrating the framework of the real-time feedback and closed-loop control module of the present invention. Figure 5 This is a flowchart illustrating the framework of the user interaction and operation interface module of the present invention. Figure 6 This is a flowchart illustrating the framework of the treatment report and pathology management unit of the present invention. Figure 7 This is a flowchart of the method of the present invention. Detailed Implementation

[0023] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0024] Please see the appendix Figure 1 This invention provides a machine learning-optimized robotic arm-assisted orthodontic archwire bending system, comprising: The patient data acquisition and 3D modeling module is used to acquire patient oral data and generate personalized 3D oral models of patients. The core function of the patient data acquisition and 3D modeling module is to acquire the patient's oral cavity data and generate a personalized 3D oral cavity model based on this data. This provides a precise digital foundation for subsequent machine learning-driven personalized archwire bending path design.

[0025] The patient data acquisition and 3D modeling module, exemplarily, includes a data acquisition unit and a data processing and fusion unit. The data acquisition unit and the data processing and fusion unit are electrically or digitally connected to achieve data transmission and processing.

[0026] The data acquisition unit is responsible for obtaining raw oral anatomy information from the patient. For example, the operator first uses an intraoral scanner to optically scan the surface of the tooth crowns and adjacent soft tissues within the patient's oral cavity. This process acquires high-precision three-dimensional point cloud data of the tooth crown surface, denoted as... The three-dimensional point cloud data of the crown surface It accurately reflects the external shape and arrangement of the teeth.

[0027] Subsequently, or in parallel, the operator uses cone-beam computed tomography (CBCT) to scan the patient's craniofacial region. This CBCT scan acquires three-dimensional image data including tooth and root morphology, alveolar bone structure, and internal anatomical structures of the jawbone, denoted as... The three-dimensional image data Typically in DICOM format, it provides detailed information about the interior of the tooth and periodontal tissues. (3D point cloud data of the tooth crown surface) With 3D image data Together, these constitute the patient's oral cavity data required for subsequent treatment.

[0028] The data processing and fusion unit receives three-dimensional point cloud data of the crown surface from the data acquisition unit. and 3D image data First, this unit preprocesses both types of data separately. For the 3D point cloud data of the tooth crown surface... Preprocessing may include denoising algorithms, such as statistical outlier removal or bilateral filtering, as well as the removal of unwanted non-dental structures.

[0029] For 3D image data Preprocessing may include image denoising, such as using median filtering or anisotropic diffusion filtering, and possible image enhancement operations. Furthermore, for subsequent registration and fusion, it may be necessary to extract data from 3D image data. The area of ​​interest, such as teeth and alveolar bone, is divided into sections.

[0030] After preprocessing, the data processing and fusion unit processes the three-dimensional point cloud data of the crown surface. and 3D image data (or its derived data, such as tooth surface data extracted from CBCT) undergo precise 3D spatial registration and fusion. The purpose of registration is to unify data from two different sources and modalities into the same coordinate system.

[0031] For example, a registration method based on the Iterative Closest Point (ICP) algorithm or a variant thereof can be used. This method iteratively optimizes a transformation matrix. This enables the three-dimensional point cloud data of the crown surface. Transformed and compared with 3D image data The goal is to achieve optimal alignment between the corresponding crown portions of the teeth. This involves minimizing the distance between corresponding points.

[0032] After registration, data fusion is performed to combine the three-dimensional point cloud data of the crown surface. High-precision surface information and 3D image data The internal structural information (especially information about the tooth root and alveolar bone) is effectively combined. The fused data... This provides a complete and accurate data foundation for subsequent 3D model reconstruction.

[0033] Based on the fused data The data processing and fusion unit uses 3D reconstruction algorithms to generate personalized 3D oral models for patients. These algorithms can be surface reconstruction-based, such as Poisson surface reconstruction, or they can directly generate mesh models from fused point clouds or segmented volume data. The generated initial personalized 3D oral model is denoted as... .

[0034] Finally, the data processing and fusion unit analyzes the initial personalized 3D oral model. Optimization processes may be performed. These processes may include mesh smoothing to eliminate minor irregularities or noise that may occur during the reconstruction process. Exemplary smoothing algorithms include Laplacian smoothing or Taubin smoothing.

[0035] Optimization may also include mesh hole filling to ensure model integrity; and mesh simplification or remeshing to reduce the number of facets while maintaining key anatomical features, thereby improving subsequent computational efficiency. The final output is an optimized personalized 3D oral cavity model, whose personalized 3D oral cavity model data is denoted as [data missing]. This personalized 3D oral model data They are typically stored in standard 3D model file formats (such as STL and PLY) and used as input for subsequent personalized sequential bending design modules.

[0036] Through the above implementation methods, the patient data acquisition and 3D modeling module can accurately and efficiently acquire comprehensive oral information of patients and generate high-quality personalized 3D digital oral models, providing a solid data foundation for the entire machine learning-optimized robotic arm-assisted orthodontic archwire bending process, and enabling precise representation of the patient's oral structure.

[0037] Please see the appendix Figure 2 The personalized sequence bending design module is connected to the patient data acquisition and 3D modeling module. It is used to design personalized sequence archwire bending paths based on personalized 3D oral models and through artificial intelligence algorithms. The personalized sequence curvature design module connects to the data output of the aforementioned patient data acquisition and 3D modeling module to receive the patient's personalized 3D oral model data. Its core function is to automatically design personalized, phased archwire bending paths based on this model and through a series of artificial intelligence algorithms.

[0038] The personalized sequence bending design module, exemplarily, includes a deep learning analysis unit, a biomechanical modeling and simulation unit, and a reinforcement learning optimization unit. These three units work in series or collaboratively in a feedback iterative manner to achieve a complete design flow from initial planning to final optimization.

[0039] First, the deep learning analysis unit receives personalized 3D oral model data. This unit is used to perform in-depth feature analysis on personalized 3D oral models and to plan the initial archwire bending path based on the analysis results. For example, this unit employs a combined architecture of convolutional neural networks (CNN) and graph neural networks (GNN).

[0040] Convolutional Neural Networks (CNNs) first process personalized 3D oral cavity model data. The process involves extracting local geometric features of the teeth, such as the cusps, ridges, curvature of the tooth surface, and normal vectors. These local features, along with the spatial relationships between teeth, are then used to construct graph-structured data.

[0041] Graph Neural Networks (GNNs) then process this graph-structured data to learn and analyze the complex topological relationships, adjacency relationships, and potential interactions between teeth. In this way, the deep learning analysis unit can comprehensively understand the patient's oral structural features and any abnormalities. Its feature extraction process can be represented as follows: ; in: This represents the input of personalized 3D oral model data; This represents the operation of a convolutional neural network to extract local features from model data. This represents the operation of a graph neural network to perform global relational analysis on the extracted features and the constructed graph structure; This represents the final extracted comprehensive features, which include information on tooth morphology and the topological relationships between teeth.

[0042] Based on comprehensive features Combining knowledge learned from a large number of historical orthodontic cases (e.g., pre-trained models through supervised or unsupervised learning), the deep learning analysis unit plans one or more preliminary archwire bending paths, denoted as... This initial path serves as the starting point for subsequent optimizations.

[0043] Subsequently, the biomechanical modeling and simulation unit receives the preliminary archwire bending path output by the deep learning analysis unit. and personalized 3D oral model data The core function of this unit is to verify and optimize the initial archwire bending path from a biomechanical perspective, ensuring that the applied corrective force is within physiologically permissible limits.

[0044] For example, this unit employs the finite element analysis (FEA) method. First, based on personalized three-dimensional oral cavity model data... This assigns corresponding material properties (such as elastic modulus and Poisson's ratio) to different tissue structures such as teeth, periodontal ligament, and alveolar bone. Simultaneously, it establishes the initial archwire bending path. The bowwire it represents gives it material properties (such as the mechanical parameters of nickel-titanium alloys or stainless steel).

[0045] Then, the initial bowwire bending path is simulated in a virtual environment. The archwire is fitted onto the teeth of a personalized 3D oral model. FEA calculations are used to simulate the stress distribution in the periodontal tissues (especially the periodontal ligament) after the archwire is loaded, thus obtaining the stress experienced by the periodontal tissues. .

[0046] The calculated stress on the periodontal tissue With the preset physiological safety threshold Comparison. Preset physiological safety thresholds. This is the upper limit of stress that will not cause irreversible damage to periodontal tissues, determined based on clinical experience and biomechanical studies. The assessment criteria are as follows: ; in: This is the initial bowwire bending path input; For personalized 3D oral cavity model data; This represents the finite element analysis process, and its output is the stress on the periodontal tissue. The stress on periodontal tissues as calculated by finite element analysis; This is a preset physiological safety threshold.

[0047] If the periodontal tissue is subjected to stress Exceeding the preset physiological safety threshold This indicates the current preliminary archwire bending path. This could result in excessive corrective force. In this case, the unit will provide feedback adjustment information to the deep learning analysis unit based on the area and degree of stress exceeding the limit, or use built-in optimization algorithms (such as topology optimization or shape optimization methods) to adjust the initial archwire bending path. The bending point position, angle, and other parameters are iteratively modified until biomechanical safety requirements are met. The archwire bending path verified and optimized through this step is denoted as... .

[0048] Finally, the reinforcement learning optimization unit receives the validated bowwire bending path output by the biomechanical modeling and simulation unit. (Or define the initial state based on this). This unit is used to dynamically and multi-stage optimize the archwire bending path sequence to form a refined bending strategy for different treatment stages (such as the alignment and leveling stage, the gap closing stage, and the torque control stage), and finally output a personalized sequence of archwire bending paths.

[0049] For example, this unit models the archwire bending process associated with the archwire bending path sequence as a Markov decision process (MDP). In this MDP: current treatment state It can be characterized by factors such as the current position, posture, and alignment of the teeth, as well as the current shape of the archwire. This information can be obtained from personalized 3D oral model data. Obtained from the bowwire path determined in the previous stage. Bowwire bending parameter adjustment action. This refers to the adjustment amount or selection of the next bending operation for parameters such as the three-dimensional coordinates, bending angle, and segment length of a specific bending point on the bowwire.

[0050] The goal of this unit is to optimize a strategy. This strategy determines the treatment state given the current treatment status. What adjustment action should be taken to adjust the bowwire bending parameters? To maximize the expected cumulative reward Cumulative Rewards The treatment outcome (such as the accuracy of tooth movement to the target position, the degree of improvement in dental arch morphology, and optimization of occlusion) and treatment efficiency (such as the appropriateness of the orthodontic force and the potential reduction in treatment duration) after archwire bending are defined. Its mathematical expression is: ; in: In order to perform the action After that, from the state Transition to state Instant rewards obtained at that time; The discount factor is a constant between 0 and 1 used to balance the importance of current rewards and future rewards. This represents the expected value.

[0051] The reinforcement learning optimization unit can employ deep reinforcement learning algorithms, such as Deep Q-Network (DQN), Policy Gradient methods, or Actor-Critic methods, to learn the optimal policy through extensive trial and error with a simulated environment (based on a personalized 3D oral cavity model and biomechanical model). .

[0052] Through the aforementioned reinforcement learning optimization process, this unit can generate a phased, personalized sequential archwire bending path tailored to the individual patient's condition. This path not only considers the final treatment goal but also plans the archwire morphology and bending parameters to be used in each intermediate treatment stage (such as the first series of bends, the second series of bends, and the third series of bends) to achieve that goal. This final output of the personalized sequential archwire bending path will be passed as instructions to the subsequent robotic arm bending module.

[0053] Through the above implementation methods, the personalized sequence bending design module can combine the pattern recognition capabilities of deep learning, the safety verification of biomechanical simulation, and the sequence decision optimization capabilities of reinforcement learning to design a highly personalized, safe, effective, and phased archwire bending plan for each patient, which can significantly improve the accuracy and efficiency of orthodontic treatment.

[0054] Please see the appendix Figure 3 The robotic arm bending module, connected to the personalized sequence bending design module, is used to drive the robotic arm to perform bowwire bending according to the personalized sequence bowwire bending path. The input of the robotic arm bending module is connected to the output of the personalized sequence bending design module to receive personalized sequence bowwire bending paths. This path precisely defines the type, position, angle, and straight segment length between each bend on the bowwire.

[0055] The core function of the robotic arm bending module is to drive the robotic arm and its end effector with high precision according to the received personalized sequence archwire bending path, and automatically complete the physical bending and shaping of the orthodontic archwire.

[0056] For example, the robotic arm bending module includes a multi-degree-of-freedom robotic arm, such as a six- or seven-degree-of-freedom industrial robot. This multi-degree-of-freedom robotic arm has sufficient motion flexibility to precisely position the bowwire to be bent or its end effector, a dedicated bowwire bending actuator, to any target position and orientation in three-dimensional space.

[0057] An end effector for bending orthodontic archwires is mounted on the end flange of the robotic arm. This end effector is a device specifically designed for the precision bending of orthodontic archwires. Exemplarily, it may include an archwire clamping mechanism, an archwire feeding mechanism, and a bending forming mechanism.

[0058] The bowwire clamping mechanism securely holds the bowwire during bending, preventing slippage or undesirable deformation. The bowwire feed mechanism precisely feeds the bowwire to the next bending point according to the segment lengths defined in the personalized bowwire bending path sequence.

[0059] The bowwire bending mechanism directly applies force to the bowwire to create the desired bend. This mechanism may include one or more sets of bending pins, bending dies, or rollers whose position and angle can be precisely controlled. Through the relative movement and action of these components with the bowwire, a bend of the bowwire at a specific angle and radius is achieved at a specific location.

[0060] One of the key technical features of the robotic arm bending module is its ability to dynamically compensate for bending force and angle based on the characteristics of the selected bowwire material. Because bowwire materials (such as nickel-chromium alloys, titanium-molybdenum alloys, and stainless steel wire) are inherently elastic, they will spring back to a certain extent after the bending force is unloaded, resulting in the actual forming angle being smaller than the bending angle at the time of loading.

[0061] To address this issue, before performing each bending operation, the module considers the current archwire material type (pre-inputted or identified by sensors), diameter, and the target bending angle defined in the personalized archwire bending path sequence. and the bending radius determined by the wire bending forming mechanism Calculate the required angle compensation amount .

[0062] This angle compensation amount Using a pre-established springback model of the bowwire material The springback model of the bowwire material was calculated. It can be an empirical model calibrated from a large amount of experimental data, an analytical model derived from the theory of mechanics of materials, or a numerical model established through finite element simulation. This model establishes the mapping relationship between the bowwire material, the target bending angle, the bending radius, and the springback angle.

[0063] Therefore, when the robotic arm and the end effector are performing bending, the actual bending angle (i.e., the loading angle) is... (This should be the target bending angle) Add angle compensation This ensures that the final actual forming angle of the bowwire after springback can accurately reach the target bending angle. Their relationship can be expressed as: when ; Among them, loading angle It refers to the physical bending angle performed by the wire bending forming mechanism of the robotic arm bending module. This is the final forming angle after the bowwire springs back. The angle compensation is calculated as follows: ; in: A springback model representing the bowwire material; This represents the material type of the selected bowwire and its associated mechanical parameters (e.g., elastic modulus, yield strength, diameter). The target bending angle for the current bend, derived from the personalized sequence of bowwire bending paths; The bending radius formed by the bow wire during the bending process is usually determined by the geometry of the bending forming mechanism of the end-effector (such as the diameter of the bending pin and the curvature of the die).

[0064] Furthermore, the dynamic bending force function is manifested in the integration of a force sensor into the end effector or robotic arm wrist during the bending process. This force sensor monitors the force applied to the bowwire in real time during bending. The system can then compare the monitored actual bending force with a pre-set desired force range based on the bowwire material properties and bending parameters.

[0065] If the actual force deviates from the expected range, the system can dynamically adjust the movement speed of the multi-degree-of-freedom robotic arm or the stroke of the end effector to ensure the stability and consistency of the bending process and prevent damage to the bowwire.

[0066] In the entire bending process, the controller of the robotic arm bending module first analyzes the personalized sequence of bowwire bending paths. Then, for each bend in the path, the controller drives the bowwire feed mechanism to deliver the bowwire to the predetermined position, followed by the bowwire clamping mechanism securing the bowwire. Subsequently, the controller calculates the angle compensation amount according to the aforementioned formula. Determine the loading angle Finally, the wire bending forming mechanism is driven according to the loading angle. and preset bending radius Complete the current bend while monitoring the bending force. After completing a bend, release the clamp and feed to the next bend point. Repeat this process until the entire bowwire is bent according to the required path.

[0067] Through the above implementation methods, the robotic arm bending module can automatically and accurately bend orthodontic archwires according to the personalized path designed by machine learning, and ensure the forming accuracy of the archwires through dynamic angle and force compensation technology, enabling high-quality manufacturing of complex three-dimensional arch shapes.

[0068] Please see the appendix Figure 4 The real-time feedback and closed-loop control module is connected to the robotic arm bending module to monitor the bending status in real time during the bowwire bending process and perform closed-loop control to correct execution deviations. The real-time feedback and closed-loop control module is bidirectionally connected to the robotic arm bending module or forms a control loop. Its core function is to monitor the actual bending state in real time during the robotic arm's bowwire bending process and compare this state with the personalized bowwire bending path sequence. Once an execution deviation occurs, a correction command is generated through the closed-loop control strategy to dynamically adjust the robotic arm's movements, thereby ensuring the final accuracy and quality of the bowwire bending.

[0069] The real-time feedback and closed-loop control module, exemplarily, includes a multi-modal sensor data real-time acquisition unit, a real-time deviation calculation and error assessment unit, and an artificial intelligence-based closed-loop control command generation unit. These three units work together to form a complete real-time feedback control loop.

[0070] The multimodal sensor data real-time acquisition unit is responsible for continuously acquiring physical quantities related to the bending state from multiple sources during the bowwire bending process. For example, this unit integrates various sensors, including but not limited to: encoders for acquiring the angles of each joint of the robotic arm to calculate joint positions; inertial measurement units (IMUs) or vision sensors for acquiring the precise posture of the end effector for bowwire bending; dedicated angle sensors or machine vision systems for directly measuring the actual angle of the bend formed on the bowwire; force / torque sensors mounted on the end effector or the wrist of the robotic arm for measuring the forces acting on the bowwire; and possible optical scanning devices (such as line laser scanners or structured light cameras) for acquiring real-time three-dimensional morphological data of the bowwire during the bending process.

[0071] This unit performs necessary filtering and preprocessing on the raw sensor data collected, and outputs structured bending state parameters, such as joint position data of the robotic arm, end-effector posture data, actual bending angle data of the bowwire, force data of the bowwire, and real-time three-dimensional morphology data of the bowwire.

[0072] The real-time deviation calculation and error assessment unit receives bending state parameters from the multimodal sensor data real-time acquisition unit, and simultaneously receives personalized sequence bowwire bending paths (as target reference paths) from the personalized sequence bending design module. The core function of this unit is to accurately compare the real-time acquired bending state parameters with the target parameters corresponding to the current bending step in the personalized sequence bowwire bending path.

[0073] For example, the actual bending angle of the bowwire is compared with the target bending angle; the actual spatial position of the end effector is compared with the target position; or the instantaneous three-dimensional shape of the bowwire is compared with the target three-dimensional shape. Through such comparisons, the unit calculates the execution deviation in various dimensions. Execution deviation It can be a vector that contains the differences between the current bowwire formation and the expected value in terms of angle, position, and shape.

[0074] The AI-based closed-loop control command generation unit receives the execution deviation calculated by the real-time deviation calculation and error evaluation unit. And in combination with the current system status Generate control commands for correcting the robotic arm's movements. Current system status May include execution deviations Information such as the current joint angular velocity of the robotic arm, the speed of the end effector for bending wire, and the properties of the bow wire material and the geometry of the bent portion.

[0075] This unit exemplarily employs Model Predictive Control (MPC) algorithm to achieve intelligent closed-loop control. Model Predictive Control is an advanced control strategy that utilizes a dynamic model of the controlled system (in this case, the process of a robotic arm bending a bowwire) to predict the system's behavior over a future period by solving an online optimization problem in each control cycle, and to determine a series of optimal control inputs.

[0076] Specifically, the model predictive control algorithm optimizes a series of control input sequences in the prediction time domain. This ensures that the system's predicted output trajectory (i.e., the predicted bowwire bending state) under these control inputs deviates to the target value of the personalized bowwire bending path within the corresponding prediction time period, while also considering the magnitude or rate of change of the control inputs themselves to avoid overly drastic control actions. This optimization problem is typically expressed as a constrained quadratic programming problem, with its objective function... Examples are as follows: ; in, Represents the current moment For the future The predicted value of the bowwire bending state (such as angle, position, and morphological parameters) at any time is based on the system dynamic model and the control input sequence to be optimized. Representing personalized sequence archwire bending paths in the future The target state value corresponding to the given time; Representing the future The vector of corrective control commands applied at any given time is the decision variable for the optimization problem; The length of the prediction time domain, i.e., the time range within which the system behavior is predicted; To control the length of the time domain ( ≤ ), that is, the length of the future control input sequence; and The weight matrices are either positive definite or semi-positive definite, used to adjust the relative emphasis on tracking error and control energy consumption.

[0077] The system dynamic model can be built based on physical principles, or it can be modeled or adaptively adjusted online using historical data and artificial intelligence methods (such as machine learning and system identification technology) to improve the accuracy of the model and thus support "artificial intelligence-based" closed-loop control.

[0078] After solving the above optimization problem in each control cycle, an optimal control input sequence is obtained. However, according to the recedinghorizon principle of model predictive control, only the first corrected control command in this sequence is valid. The corrected control command was actually adopted and output as the current correction command. In the next control cycle, the state observation, deviation calculation, and optimization solution are repeated.

[0079] Correct control commands The data is then sent to the underlying controller of the robotic arm bending module to adjust the joint motion commands of the robotic arm or the operating parameters of the end effector wire bending actuator, thereby offsetting any execution deviations that have occurred. This allows the actual bowwire bending process to more closely track the personalized bowwire bending path.

[0080] Through the above implementation methods, the real-time feedback and closed-loop control module can continuously monitor the bowwire bending process, promptly detect and correct deviations, significantly improve the accuracy and consistency of bowwire forming, enhance the system's robustness to uncertainties such as changes in material properties and external disturbances, and ensure the precise manufacturing of complex and customized bowwires.

[0081] Please see the appendix Figure 5 - Appendix Figure 6The user interaction and operation interface module is connected to the personalized sequence bending design module and the real-time feedback and closed-loop control module, which are used to visually monitor the bending process, adjust the design parameters, and provide intelligent assistance.

[0082] The user interaction and operation interface module, personalized sequence bending design module, and real-time feedback and closed-loop control module are interconnected for data and control signal exchange. Its core function is to provide operators with an intuitive and efficient human-machine interaction platform for visual monitoring of the entire bowwire design and bending process, adjustment of key design parameters, and receiving intelligent auxiliary information provided by the system.

[0083] The user interaction and operation interface module, for example, includes a 3D visualization and dynamic rendering unit, an intelligent decision support and suggestion unit, and a treatment report and case management unit. These units together constitute the core of user-system interaction.

[0084] The 3D visualization and dynamic rendering unit is responsible for presenting abstract data and complex processes to users in an intuitive 3D graphical format. This unit receives personalized 3D oral model data from the patient data acquisition and 3D modeling module (or via the personalized sequence bending design module). Simultaneously, it receives personalized sequence archwire bending path data from the personalized sequence bending design module.

[0085] Furthermore, this unit acquires real-time motion data of the robotic arm during the bending process, as well as geometric data of the archwire gradually taking shape during the bending process, from the real-time feedback and closed-loop control module. Utilizing computer graphics technology, such as a 3D rendering engine, this unit can render and dynamically display a personalized 3D dental model in real time.

[0086] In this 3D scene, personalized sequenced bowwire bending paths can be overlaid as target trajectories. The bending motion of the robotic arm, including the movement of its joints and the operation of the end effector, can be dynamically simulated and displayed in real time. The bowwire forming process, that is, the process of gradually bending the bowwire from its initial straight state to its final designed shape, can also be visually tracked and displayed. This integrated visualization allows users to comprehensively monitor the entire bending process.

[0087] The intelligent decision support and suggestion unit aims to improve the intelligence level of operations and the quality of clinical decision-making. This unit receives real-time monitoring data from the real-time feedback and closed-loop control module, such as the force applied to the archwire, the deviation between the actual bending angle and the target value, and the motion stability of the robotic arm. An integrated treatment rule engine is also present within this unit.

[0088] The treatment rule engine pre-stores a rule base based on orthodontic clinical experience, biomechanical principles, materials science knowledge, and equipment safety operating procedures. When real-time monitoring data triggers specific rules, such as detecting that the bending force continuously exceeds the safety threshold or that the archwire forming deviation accumulates too much, the unit will proactively push intelligent decision-making assistance information to the user.

[0089] This information can manifest as warnings, risk alerts, parameter adjustment suggestions, or relevant clinical management recommendations. For example, the system might suggest that the user pause the operation to check the archwire clamping or fine-tune a bending parameter to optimize treatment outcomes. This helps users promptly identify and address potential problems, ensuring the safety and effectiveness of the treatment.

[0090] The Treatment Reporting and Case Management Unit is responsible for systematically organizing, recording, and archiving data from the entire process after the archwire bending task is completed, or after a specific stage of treatment. This unit obtains detailed parameters of the finalized personalized archwire bending path from the Personalized Sequence Bending Design Module, including the design angle, position, and radius of each bend.

[0091] Simultaneously, this unit obtains key execution parameters monitored during the actual bending process from the real-time feedback and closed-loop control module, such as the actual forming angle of each bend, the bending force used, the deviation value, and the record of correction control. The treatment report and case management unit automatically integrates these design parameters and actual execution parameters to generate a structured treatment report.

[0092] This report can include patient information, treatment date, archwire material used, a diagram of the designed archwire morphology, a diagram of the actual archwire morphology after bending (if the system has a scanning verification function), a comparison table of key parameters, and any warnings or suggestions generated during the process. This report not only helps physicians assess the quality of the archwire fabrication and the treatment process but also supports electronic case archiving, facilitating subsequent review, tracking of treatment progress, and clinical research.

[0093] Through the above implementation methods, the user interaction and operation interface module provides users with a comprehensive and convenient platform, enabling effective monitoring, intelligent assistance, and standardized management of complex archwire design and bending processes, thereby improving the system's usability and clinical application value.

[0094] The machine learning-optimized robotic arm-assisted orthodontic archwire bending method described below can be used as a reference to the machine learning-optimized robotic arm-assisted orthodontic archwire bending system described above.

[0095] Please see the appendix Figure 7 A machine learning-optimized robotic arm-assisted orthodontic archwire bending method includes the following steps: Collect and model patient oral data: Obtain patient oral data and generate personalized 3D oral model based on the patient's oral data to obtain accurate digital oral representation; Personalized sequential bending design based on the obtained accurate digital oral cavity representation: Based on the personalized three-dimensional oral cavity model, a personalized sequential archwire bending path is designed through artificial intelligence algorithms, thereby obtaining digital path parameters to guide subsequent bending. Based on the obtained digital path parameters, the robotic arm performs bending: it drives the robotic arm and performs physical bending of the bowwire according to the personalized bowwire bending path sequence; During the physical bending process of the bowwire, real-time feedback and closed-loop control are implemented: the bending status is monitored in real time, and the monitored bending status is compared with the digital path parameters. Based on the deviation generated by the comparison, closed-loop control is performed to correct the execution action of the robotic arm, thereby ensuring the final bending accuracy.

[0096] The method in this embodiment can be used to execute the above system embodiment, and its principle and technical effect are similar, so it will not be described again here.

[0097] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A machine learning-optimized robotic arm-assisted orthodontic archwire bending system, characterized in that, include: The patient data acquisition and 3D modeling module is used to acquire patient oral data and generate personalized 3D oral models of patients. A personalized sequence bending design module is connected to the patient data acquisition and 3D modeling module, and is used to design a personalized sequence archwire bending path based on the personalized 3D oral model using artificial intelligence algorithms. The robotic arm bending module is connected to the personalized sequence bending design module and is used to drive the robotic arm to perform bowwire bending according to the personalized sequence bowwire bending path. A real-time feedback and closed-loop control module is connected to the robotic arm bending module to monitor the bending status in real time during the bowwire bending process and perform closed-loop control to correct execution deviations. The user interaction and operation interface module is connected to the personalized sequence bending design module and the real-time feedback and closed-loop control module, and is used to visually monitor the bending process, adjust the design parameters, and provide intelligent assistance.

2. The machine learning-optimized robotic arm-assisted orthodontic archwire bending system according to claim 1, characterized in that, The patient data acquisition and 3D modeling module includes: The data acquisition unit is used to acquire three-dimensional data of the crown surface through an intraoral scanner and to acquire three-dimensional image data including the tooth root and jawbone structure through a cone-beam computed tomography (CBCT) device. The data processing and fusion unit is used to preprocess, register, and fuse the collected three-dimensional data and three-dimensional image data of the crown surface to generate a personalized three-dimensional oral model, and to optimize the personalized three-dimensional oral model.

3. The machine learning-optimized robotic arm-assisted orthodontic archwire bending system according to claim 1, characterized in that, The personalized sequence bending design module includes: A deep learning analysis unit is used to perform feature analysis and initial path planning on the personalized 3D oral model. The biomechanical modeling and simulation unit is used for biomechanical verification and optimization of the initial path planning; The reinforcement learning optimization unit is used to dynamically optimize the archwire bending path sequence to form bending strategies for different treatment stages.

4. The machine learning-optimized robotic arm-assisted orthodontic archwire bending system according to claim 3, characterized in that, The deep learning analysis unit uses Convolutional Neural Networks (CNN) and Graph Neural Networks (GNN) to process personalized 3D oral model data. To extract the morphological features of teeth and the topological relationships between teeth.

5. The machine learning-optimized robotic arm-assisted orthodontic archwire bending system according to claim 3, characterized in that, The biomechanical modeling and simulation unit employs the finite element analysis (FEA) method, based on the personalized three-dimensional oral cavity model and the preliminary archwire bending path. Simulates the stress on periodontal tissues after archwire loading. And according to the preset physiological safety threshold The evaluation and optimization of the path can be expressed as follows: .

6. The machine learning-optimized robotic arm-assisted orthodontic archwire bending system according to claim 3, characterized in that, The reinforcement learning optimization unit achieves dynamic optimization of the archwire bending path sequence and forms the bending strategy for different treatment stages in the following manner: This unit models the archwire bending process associated with the archwire bending path sequence as a Markov decision process and optimizes a strategy. To maximize expected cumulative reward The cumulative reward is defined based on the healing effect and efficiency after the bowwire is bent, whereby... This is the current treatment status. For adjusting the bowwire bending parameters, ,in, For instant rewards, This is the discount factor.

7. The machine learning-optimized robotic arm-assisted orthodontic archwire bending system according to claim 1, characterized in that, The robotic arm bending module includes a multi-degree-of-freedom robotic arm and a dedicated end effector for bending wire, and possesses dynamic bending force and angle compensation functions based on the characteristics of the selected bowwire material. This angle compensation amount... Based on the springback model of bowwire materials Target bending angle and bending radius Calculations are performed to make the actual forming angle close to the target bending angle.

8. The machine learning-optimized robotic arm-assisted orthodontic archwire bending system according to claim 1, characterized in that, The real-time feedback and closed-loop control module includes: The multimodal sensor data real-time acquisition unit is used to acquire the joint position, end-effector posture, actual bending angle of the archwire, force on the archwire, and real-time three-dimensional morphological data of the archwire of the robotic arm. The real-time deviation calculation and error assessment unit is used to compare the real-time acquired bending state parameters with the personalized sequence bowwire bending path to calculate the execution deviation. ; An AI-based closed-loop control command generation unit is used to generate commands based on the execution deviation. and current system status Generate correction control instructions This unit employs a Model Predictive Control (MPC) algorithm. By optimizing a series of control inputs in the prediction time domain, it minimizes the deviation of the predicted system output from the personalized sequence of archwire bending paths, while also considering the cost of the control inputs, thereby determining the optimal corrective control command. This is to drive the robotic arm to adjust its movements.

9. The machine learning-optimized robotic arm-assisted orthodontic archwire bending system according to claim 1, characterized in that, The user interaction and operation interface module includes: The 3D visualization and dynamic rendering unit is used to render and dynamically display the personalized 3D oral model, the personalized sequence archwire bending path, the bending action of the robotic arm, and the archwire forming process in real time in 3D form. The intelligent decision support and suggestion unit is used to push intelligent decision support information and clinical suggestions to users based on real-time monitoring data and treatment rule engine; The treatment report and case management unit is used to automatically integrate bending process data to generate a structured treatment report and support case archiving after treatment is completed. The bending process data includes parameters of the personalized sequential archwire bending path and actual execution parameters monitored by the real-time feedback and closed-loop control module.

10. A machine learning-optimized robotic arm-assisted orthodontic archwire bending method, characterized in that, Includes the following steps: Collect and model patient oral data: Acquire patient oral data and generate a personalized three-dimensional oral model based on the patient's oral data, thereby obtaining an accurate digital oral representation; Personalized sequential bending design is carried out based on the obtained accurate digital oral cavity representation: according to the personalized three-dimensional oral cavity model, a personalized sequential archwire bending path is designed through artificial intelligence algorithm, thereby obtaining digital path parameters to guide subsequent bending. Based on the obtained digital path parameters, the robotic arm performs bending: it drives the robotic arm and performs physical bending of the bowwire according to the personalized bowwire bending path sequence; During the physical bending process of the bowwire, real-time feedback and closed-loop control are implemented: the bending status is monitored in real time, and the monitored bending status is compared with the digital path parameters. Based on the deviation generated by the comparison, closed-loop control is performed to correct the execution action of the robotic arm, thereby ensuring the final bending accuracy.

Images