An intelligent physiotherapy control system of a physiotherapy device

Abstract

The application discloses a kind of intelligent physiotherapy control systems of physiotherapy equipment, belong to the technical field of physiotherapy instrument control system, the system includes integrated in physiotherapy comb body physiotherapy execution module, for real-time acquisition user physiological parameter physiological parameter sensing module, and with the communication connection of both intelligent control unit.State analysis module is included in intelligent control unit, for the current physiological state of user is analyzed according to the physiological parameter of acquisition;Strategy generation module is used to match or generate target physiotherapy strategy from preset physiotherapy mode library based on current physiological state;And drive control module is used to drive physiotherapy execution module to work according to target physiotherapy strategy.The application is fused by multi-source physiological signal, and the state of user is analyzed, and using the intelligent decision mechanism including reinforcement learning, generative algorithm, dynamically generates or optimizes physiotherapy strategy, realizes the real-time self-adaptation and high personalization of physiotherapy process, improves the safety and effectiveness of physiotherapy.

Description

Technical Field

[0001] This invention belongs to the technical field of physiotherapy equipment control systems, specifically relating to an intelligent physiotherapy control system for physiotherapy equipment. Background Technology

[0002] Traditional therapeutic combs are limited in function, typically offering only mechanical massage, single light therapy, or fixed-mode combinations. They cannot dynamically adjust based on the user's real-time physiological state, resulting in limited therapeutic effects. With the development of smart health devices, some products have begun to incorporate physiological parameter monitoring functions, but these are mostly limited to single signals such as heart rate and temperature, and the control strategies are often simple "threshold-response" modes, lacking in-depth analysis and comprehensive judgment of the user's physiological state.

[0003] In existing technologies, the control strategies of physiotherapy equipment are usually preset fixed programs or require manual selection by the user, failing to achieve true personalization and adaptive adjustment. Even if some systems can adjust parameters based on simple signals, their decision-making logic is often linear and rigid, unable to handle multi-source heterogeneous physiological signals, and lacking the ability to continuously optimize strategies based on long-term user feedback and group data. Furthermore, the coordinated control between different physiotherapy modules (such as mechanical, optical, and ultrasound) lacks intelligent scheduling and safety arbitration mechanisms, posing potential risks or offsetting effects.

[0004] Therefore, there is an urgent need in this field for a control system that can perceive the user's multidimensional physiological state in real time and accurately, and based on this, intelligently generate, optimize and safely execute personalized and collaborative physiotherapy strategies to improve the intelligence level of the physiotherapy comb and the user experience. Summary of the Invention

[0005] In view of this, the purpose of the present invention is to provide an intelligent physiotherapy control system for physiotherapy equipment, which analyzes the user's state through the fusion of multi-source physiological signals and uses intelligent decision-making mechanisms including reinforcement learning and generative algorithms to dynamically generate or optimize physiotherapy strategies, so as to achieve real-time adaptation and high personalization of the physiotherapy process.

[0006] To achieve the above objectives, the present invention provides the following technical solution: This invention provides an intelligent physiotherapy control system for a physiotherapy device, used for physiotherapy control of the physiotherapy comb body, comprising: The physiotherapy execution module, integrated into the physiotherapy comb body, includes a massage head for mechanical stimulation, a light-emitting element for phototherapy, a feeding unit for introducing substances, and an ultrasonic head for ultrasonic action. A physiological parameter sensing module is installed on the body of the physiotherapy comb to collect at least one physiological parameter of the user in real time. The intelligent control unit, which is communicatively connected to the physiological parameter sensing module and the physiotherapy execution module, includes: a state analysis module for analyzing the user's current physiological state based on the collected physiological parameters; a strategy generation module for matching or generating a target physiotherapy strategy from a preset physiotherapy mode library based on the current physiological state, wherein the target physiotherapy strategy includes at least the setting of working parameters for at least one execution element in the physiotherapy execution module; and a drive control module for generating control commands to drive the physiotherapy execution module to work according to the target physiotherapy strategy.

[0007] Furthermore, the physiological parameter sensing module includes: an optical sensing unit for acquiring photoplethysmography (PPG) signals from the scalp or neck region; an electrophysiological sensing unit for measuring skin conductance signals; and a motion sensing unit for monitoring head movement and posture signals. The state analysis module includes: a multi-source data fusion unit, used for time alignment, filtering, noise reduction, and normalization preprocessing of asynchronous heterogeneous data from optical sensing units, electrophysiological sensing units, and motion sensing units; a time-frequency domain feature extraction unit, used to extract features characterizing the user's physiological state from the preprocessed signal, including heart rate variability indicators calculated from photoplethysmography (PPG) signals, skin conductance response features extracted from skin conductance signals, and motion state features calculated from motion posture signals; and a state classifier, used to input the extracted feature vectors into a preset state classification model, which outputs the probability distribution of the user being in various preset physiological states, including deep relaxation, mild tension, cognitive fatigue, and physiological excitement.

[0008] Furthermore, the strategy generation module includes: a strategy decision engine, used to receive the user state probability distribution from the state classifier and perform the following operations: if the matching degree between the state corresponding to the highest probability in the probability distribution and the adapted state of a preset mode in the physiotherapy mode library is higher than a first threshold, then the preset mode is directly called as the target physiotherapy strategy; if the matching degree is lower than the first threshold but higher than the second threshold, then the parameter optimizer is started, taking the user's real-time physiological characteristics as input and historical physiotherapy effect data as a reference, and fine-tuning at least one running parameter of the preset mode online through a reinforcement learning model driven by a reward function; if the matching degree is lower than the second threshold, then the strategy synthesizer is triggered, based on the state probability distribution, the user's historical preference profile, and the group efficacy data obtained through the cloud platform, using a generative algorithm to construct a new strategy containing the synergistic temporal and intensity parameters of multiple physiotherapy modules as the target physiotherapy strategy.

[0009] Furthermore, the specific construction and operation methods of the reinforcement learning model include: Model definition: The fine-tuning process is modeled as a Markov decision process, whose states... for The feature vector output by the time-frequency domain feature extraction unit at each time step The basic parameter vector of the currently running preset physiotherapy mode splicing, that is Its actions For the basic parameter vector Adjustment amount of one or more continuous parameters Reward function For composite functions: In the formula, Let be the state similarity function. The penalty term is for the range of parameter adjustment. Inappropriate calculations based on real-time user feedback or abnormal physiological signals , , These are the weighting coefficients; Learning and Decision-Making: An Actor-Critic network architecture is trained using a proximal policy optimization algorithm. The Critic network outputs the probability distribution of the adjustment action given the state. Evaluate the value of the current state; whenever fine-tuning is needed, the model adjusts the settings based on the current state. Actions are sampled from the Actor network. The action is executed and the updated parameters are applied to drive the physiotherapy module, which then observes the new state in the next time period. And calculate the reward function. The network parameters are continuously updated using this sequence of data. , .

[0010] Furthermore, the specific method for the policy synthesizer to construct new policies using generative algorithms includes the following steps: S1. Input representation: State probability distribution User history preference profile And therapeutic efficacy data from groups with similar current user status obtained from the cloud platform. Perform embedding encoding and concatenate the encoded vectors into a high-dimensional context vector: In the formula, , , These are the corresponding embedding encoding functions. This is a vector concatenation operation; S2. Sequence Generation Model: A generative model based on a Transformer decoder architecture is employed, with training data consisting of state-policy pairs annotated by experts or validated historically. This model uses context vectors... As the initial input, a strategy sequence is generated autoregressively through a multi-head self-attention mechanism and a position feedforward network, including: the selected subset of physiotherapy execution modules, the start-up sequence and duration of each module, the intensity change curve of each module throughout the treatment, and the coordination rules between modules; S3. Strategy Verification and Output: The generated strategy sequence undergoes a feasibility check via a lightweight strategy verification module, including parameter range verification, module mutual exclusion check, and estimated energy consumption assessment. After verification, the strategy sequence is parsed into a structured target physiotherapy strategy data object that can be executed by the strategy generation module and delivered to the drive control module.

[0011] Furthermore, the drive control module includes: a multi-channel synchronous controller for receiving a target physiotherapy strategy, which defines the intensity timing curves of one or more physiotherapy execution modules within a predetermined physiotherapy cycle; for the mechanical stimulation module, the synchronous controller generates a high-resolution pulse width modulation signal with a duty cycle proportional to the intensity timing curve to linearly control the torque output of the micro motor; for the phototherapy module, the synchronous controller generates a constant current drive signal and multiple PWM dimming signals to independently control the illuminance and flicker frequency of different wavelength light-emitting elements.

[0012] Furthermore, the multi-channel synchronous controller is embedded with conflict detection and arbitration logic. When the target physiotherapy strategy requires two or more physiotherapy modules to act on the same physical area, the arbitration logic rearranges the timing of the control signals or weights and attenuates the intensity according to predefined priority rules and safety time intervals to ensure the safety of the synergistic effect.

[0013] Furthermore, the intelligent physiotherapy control system also includes a user adaptive learning module for recording triplet data for each physiotherapy session: <user initial state> The target physical therapy strategy adopted Post-session state change It constructs a user-personalized effect prediction model, which is implemented using a neural network, with the user's initial state as input. With the adopted target physical therapy strategy The feature encoding outputs the predicted state changes. By comparing predicted values Compared with actual value The prediction error is calculated and used as a loss function. The parameters of the effect prediction model are continuously updated using the backpropagation algorithm. The updated model is called by the policy generation module to evaluate the expected effect of the candidate policies, so as to prioritize the policy with better prediction effect during policy matching or generation.

[0014] Furthermore, the intelligent control unit communicates with a mobile terminal application via Bluetooth. The mobile terminal application includes: a physiotherapy data dashboard displaying physiological and physiotherapy parameters in time-series charts; a virtual expert interaction interface guiding users through initial calibration and preference settings; a efficacy tracking report evaluating long-term effectiveness based on preset indicators; and a community-driven pattern sharing platform allowing users to upload, download, and rate custom physiotherapy patterns created or verified by other users. The intelligent control unit can receive digitally signed physiotherapy pattern update packages and incremental update files of the algorithm model from the mobile terminal application or a cloud server, enabling OTA upgrades of the physiotherapy pattern library and state analysis model.

[0015] The beneficial effects of this invention are as follows: 1. Realize real-time adaptive physiotherapy based on physiological state: The physiological parameter sensing module collects multi-dimensional physiological signals from the user in real time, and the intelligent control unit analyzes and makes decisions, so that the physiotherapy strategy can dynamically adapt to the user's real-time state changes, get rid of the limitations of fixed programs, and significantly improve the accuracy and targeting of physiotherapy. 2. Enhance the intelligence level of state recognition and strategy generation: By adopting multi-source data fusion, state classification model and advanced strategy generation mechanism (such as reinforcement learning fine-tuning and generative algorithm synthesis), it can more accurately understand complex physiological states and generate highly personalized and even innovative collaborative physiotherapy strategies, realizing full-chain intelligence from "perception" to "decision". 3. Ensure the safety and effectiveness of multi-modal collaborative physiotherapy: Through multi-channel synchronous control and conflict arbitration logic within the drive control module, it is ensured that when multiple physiotherapy modules work together, their control signals are safely scheduled in terms of timing and intensity, avoiding mutual interference or overstimulation between modules, thus improving therapeutic efficacy while ensuring user safety.

[0016] Other advantages, objectives, and features of the invention will be set forth in the following description and will be apparent to those skilled in the art in some respects, or may be learned by practice of the invention. The objectives and other advantages of the invention can be realized and obtained through the following description. Attached Figure Description

[0017] To make the objectives, technical solutions, and beneficial effects of this invention clearer, the following figures are provided for illustration: Figure 1 This is a schematic diagram of the overall system framework and hardware connections according to an embodiment of the present invention; Figure 2 This is a flowchart of the core algorithm of the intelligent control unit in an embodiment of the present invention; Figure 3 This is a flowchart illustrating the online fine-tuning process of the parameter optimizer according to an embodiment of the present invention. Figure 4 This is a flowchart of the strategy synthesizer in an embodiment of the present invention. Detailed Implementation

[0018] like Figure 1-4 As shown, the present invention provides an intelligent physiotherapy control system for a physiotherapy device, used for physiotherapy control of the physiotherapy comb body, comprising: The physiotherapy execution module, integrated into the physiotherapy comb body, includes a massage head for mechanical stimulation, a light-emitting element for phototherapy, a feeding unit for introducing substances, and an ultrasonic head for ultrasonic action. A physiological parameter sensing module is installed on the body of the physiotherapy comb to collect at least one physiological parameter of the user in real time. The intelligent control unit, which is communicatively connected to the physiological parameter sensing module and the physiotherapy execution module, includes: a state analysis module for analyzing the user's current physiological state based on the collected physiological parameters; a strategy generation module for matching or generating a target physiotherapy strategy from a preset physiotherapy mode library based on the current physiological state, wherein the target physiotherapy strategy includes at least the setting of working parameters for at least one execution element in the physiotherapy execution module; and a drive control module for generating control commands to drive the physiotherapy execution module to work according to the target physiotherapy strategy.

[0019] In this design, the feeding element includes a feeding tube. Both the massage head and the feeding tube can serve as the comb teeth of the therapeutic comb body. The difference is that the feeding tube is a fixed tube with a liquid storage chamber at one end and a ball bearing at the other end (the outlet) to allow the nutrient solution to flow out from the feeding unit. By contacting the scalp with the feeding tube and moving it, the ball bearing rolls along the scalp, allowing medication / nutrient solution to be applied to the scalp while minimizing contact with the hair. The massage head is a movable solid column made of elastic material. The control system controls the massage head to provide mechanical stimulation massage, including vibration, pressing, and kneading. The phototherapy unit consists of light-emitting elements (such as red-light LEDs) arrayed on the same side of the therapeutic comb as the comb teeth. The red light emitted by these elements irradiates the scalp, improving the scalp environment and stimulating hair follicle growth. The ultrasonic head is connected to an ultrasonic generator. When in use, the ultrasonic generator is turned on to provide ultrasonic therapy to the target area. This system acquires real-time physiological data from users through sensors (physiological parameter sensing modules) integrated on the comb. The state analysis module of the intelligent control unit interprets this data to determine the user's state (such as relaxation, tension, fatigue, etc.). Subsequently, the strategy generation module selects or dynamically generates a set of physiotherapy strategies containing specific working parameters (such as intensity, light intensity, frequency, etc.) from a preset plan library based on this state. Finally, the drive control module converts the strategy into precise control commands to drive the physiotherapy execution modules (such as massage heads, light-emitting elements, and ultrasonic heads) to work.

[0020] This solution represents a leap from a "fixed procedure" to a "personalized and adaptive" physiotherapy process. The system forms a closed loop of "perception-analysis-decision-execution," enabling it to provide appropriate physiotherapy plans based on the user's current specific physiological condition, thereby significantly improving the targeting, effectiveness, and user experience of physiotherapy.

[0021] In one embodiment of the present invention, the physiological parameter sensing module includes: An optical sensing unit includes at least one photoplethysmography (PPG) sensor for acquiring photoplethysmography (PPG) signals by detecting minute changes in blood volume in subcutaneous capillaries. An electrophysiological sensing unit includes at least one pair of skin conductance (GSR) electrodes for measuring changes in the conductivity of the skin surface to assess sympathetic nerve excitability; The motion sensing unit includes a six-axis inertial measurement unit (IMU) for monitoring head motion acceleration and angular velocity to identify the user's activity state and head posture during the physiotherapy process; The state analysis module includes: a multi-source data fusion unit, used to perform time alignment, filtering and noise reduction, and normalization preprocessing on asynchronous heterogeneous data from optical sensing unit, electrophysiological sensing unit, and motion sensing unit; and a time-frequency domain feature extraction unit, used to calculate the SDNN and RMSSD indices in the time domain and the LF (low frequency) and HF (high frequency) power spectral densities in the frequency domain from the preprocessed PPG signal, extract the SCR frequency and amplitude from the GSR signal, and calculate the motion energy and static ratio from the IMU signal. The state classifier uses a lightweight machine learning model trained on labeled user state data. This model is constructed using a feedforward neural network or support vector machine. The input is the feature vector output by the time-frequency domain feature extraction unit, and the output is the probability distribution of the user being in a preset state. The preset states include a deep relaxation state, a mild tension state, a cognitive fatigue state, and a physiological excitement state.

[0022] In this scheme, the SDNN index represents global variability, reflecting the overall variation of heart rate throughout the day or over a longer period. It is the gold standard for assessing overall cardiovascular health, stress load, and recovery capacity; a high value generally indicates better adaptability and recovery. The RMSSD index represents short-term variability / vagal tone, reflecting short-term, instantaneous changes in heart rate. It is the most direct time-domain indicator for assessing parasympathetic (vagal) activity; a high value indicates effective relaxation and recovery, and is often used to assess the recovery speed after acute stress. The SCR frequency represents the number of skin conductance responses per unit time; a larger SCR amplitude indicates a greater activation intensity of the autonomic nervous system triggered by the corresponding stimulus. The state analysis module first performs timestamp alignment, filtering, denoising, and normalization on the raw signals with different sampling rates and dimensions through a multi-source data fusion unit. The time-frequency domain feature extraction unit extracts key features from the processed signal. For example, it calculates time-domain (e.g., SDNN) and frequency-domain (e.g., LF / HF) indices of heart rate (HR) and heart rate variability (HRV) from PPG signals; it extracts amplitude and frequency features of skin conductance level (SCL) and skin conductance response (SCR) from GSR signals; and it calculates posture angle, motion amplitude, and frequency from IMU signals. The state classifier is a pre-trained machine learning model (e.g., Support Vector Machine (SVM), Random Forest, or Neural Network) that takes the extracted multidimensional feature vectors as input and outputs a probability distribution representing the probability that the user is currently in a predefined state such as "deep relaxation," "mild tension," "cognitive fatigue," or "physiological excitement."

[0023] This solution details the physiological parameter sensing and state analysis process, employing three types of sensors: PPG sensors (optical), GSR electrodes (electrophysiological), and IMU (motion measurement unit) to collect multimodal physiological signals from three dimensions: blood flow, skin conductance, and head movement. The state analysis module first performs fusion preprocessing on the multi-source data, then extracts key physiological features in the time and frequency domains (such as heart rate variability (HRV) and skin conductance response (SCR). Finally, these feature vectors are input into a pre-trained lightweight machine learning model (such as a neural network), outputting the probability distribution of the user's various preset states. This solution transforms raw, multimodal sensor signals into reliable and interpretable user state probabilities. This automated and refined analysis pipeline, based on machine learning, from low-level signals to high-level semantic states, surpasses simple threshold judgment and significantly improves the accuracy and systematic nature of state recognition.

[0024] In one embodiment of the present invention, the strategy generation module includes: a strategy decision engine, configured to receive a user state probability distribution from a state classifier and perform the following operations: if the matching degree between the state corresponding to the highest probability in the probability distribution and the adapted state of a preset mode in the physiotherapy mode library is higher than a first threshold, then the preset mode is directly called as the target physiotherapy strategy; if the matching degree is lower than the first threshold but higher than a second threshold, then a parameter optimizer is started, taking the user's real-time physiological characteristics as input and historical physiotherapy effect data as a reference, and fine-tuning at least one running parameter of the preset mode online through a reinforcement learning model driven by a reward function; if the matching degree is lower than the second threshold, then a strategy synthesizer is triggered, based on the state probability distribution, the user's historical preference profile, and the group efficacy data obtained through the cloud platform, using a generative algorithm to construct a new strategy containing the synergistic temporal and intensity parameters of multiple physiotherapy modules as the target physiotherapy strategy.

[0025] In this scheme, the policy generation module's policy decision engine receives the probability distribution output by the state classifier. Its decision-making process is divided into three levels: Direct matching: If the probability of a certain preset physiological state (such as "deep relaxation") is much higher than that of other states (above the first threshold), and there is a preset mode in the physiotherapy mode library that is specifically for that state, then the mode is directly called. Parameter optimization: If the user's state matches any preset state at a moderate level (between the first and second thresholds), it indicates that the user's state is special or in a transition period. At this time, the parameter optimizer is activated. It uses the user's real-time characteristics and historical efficacy data as a reference to fine-tune the parameters of the closest preset mode to adapt to real-time needs. Strategy synthesis: When the user's state has a low match rate with all preset states (below the second threshold), it indicates that a new situation not covered in the library has been encountered. The strategy synthesizer is triggered. It combines the current state, the user's personal historical preferences, and similar population data obtained from the cloud to create a brand new, customized physiotherapy strategy using generative algorithms.

[0026] This solution establishes a three-tiered flexible decision-making mechanism of "matching-optimization-creation," which not only ensures decision-making efficiency in common situations but also endows the system with creativity and adaptability to cope with complex and unknown situations, greatly enhancing the system's generalization ability and user satisfaction.

[0027] In one embodiment of the present invention, the specific construction and operation of the reinforcement learning model includes: Model definition: The fine-tuning process is modeled as a Markov decision process, whose states... Positioned as The feature vector output by the time-frequency domain feature extraction unit at each time step The basic parameter vector of the currently running preset physiotherapy mode splicing, that is Its actions Defined as a vector of fundamental parameters Adjustment amount of one or more continuous parameters Reward function For composite functions: In the formula, Let be the state similarity function. The penalty term is for the range of parameter adjustment. Inappropriate calculations based on real-time user feedback or abnormal physiological signals , , These are the weighting coefficients; Learning and Decision-Making: An Actor-Critic network architecture is trained using a proximal policy optimization algorithm. The Critic network outputs the probability distribution of the adjustment action given the state. Evaluate the value of the current state; whenever fine-tuning is needed, the model adjusts the settings based on the current state. Actions are sampled from the Actor network. The action is executed and the updated parameters are applied to drive the physiotherapy module, which then observes the new state in the next time period. And calculate the reward function. The network parameters are continuously updated using this sequence of data. , .

[0028] This solution discloses the working mechanism of the reinforcement learning model in the parameter optimizer. The implementation of this model consists of two stages: offline training and online fine-tuning. Offline Training Phase: Environment Simulation: To train the model, a virtual "user-physiotherapy system" simulation environment needs to be constructed first. This environment receives a physiotherapy strategy (action) and simulates changes in the user's state (new state) and feedback (reward) based on a physiological model; state Positioning as The feature vector output by the time-frequency domain feature extraction unit at each time step The basic parameter vector of the currently running preset physiotherapy mode splicing; action Defined as a vector of fundamental parameters Adjustment amount of one or more continuous parameters Reward function For composite functions ;in, The function calculates the cosine similarity or the reciprocal of the Euclidean distance between the current state and the target state (such as the feature vector of the "deeply relaxed" state); The penalties have been significantly adjusted to maintain treatment stability; The function quantifies user discomfort by analyzing real-time physiological signals (such as a sudden increase in GSR) or reserved user feedback interfaces (such as a discomfort button on the app); Network and Algorithm: An Actor-Critic architecture is adopted. Actor network (policy network) A multilayer perceptron is used, with the state as the input. The output is a probability distribution defined in the action space (such as the mean and variance of a Gaussian distribution). Critic Network (Value Network) Similarly, it is a multilayer perceptron, with the input being the state. The output is the expected long-term cumulative reward for that state. The Proximal Policy Optimization (PPO) algorithm is used, leveraging a large amount of interaction data sampled from the simulation environment. , , , The two networks are jointly trained with the goal of maximizing the cumulative reward. After training, the Actor network will have the ability to adaptively adjust its parameters based on the state.

[0029] Online fine-tuning phase: On a real device, when the policy decision engine determines that parameter optimization is needed, the pre-trained PPO model is loaded. The current real-world state is then... Input an Actor network, and the network outputs an adjustment action. (Parameter fine-tuning amount). After the system applies this adjustment and drives the physiotherapy module to work for a period of time, it collects new physiological signals and obtains a new state. And calculate instant rewards These real-world interaction data can be used to further fine-tune the model online (continuous learning) to make it more adaptable to individual user characteristics.

[0030] This solution utilizes reinforcement learning, enabling the system to automatically explore and learn the optimal parameter adjustment strategy through continuous interaction with the user. This achieves truly personalized, adaptive, and fine-tuned adjustments, balancing therapeutic efficacy and comfort.

[0031] In one embodiment of the present invention, the specific method for the policy synthesizer to construct a new policy using a generative algorithm includes the following steps: S1. Input Data Preparation and Representation: Preparing the state probability distribution... User history preference profile And therapeutic efficacy data from groups with similar current user status obtained from the cloud platform. Each through an embedding function Encode the vectors and concatenate them to form a comprehensive high-dimensional context vector C: In the formula, , , These are the corresponding embedding encoding functions. This is a vector concatenation operation; S2. Construct a strategy framework based on the sequence generation model: The context vector obtained in step S1... The input is fed into a pre-trained generative model based on a Transformer decoder architecture, which is trained on a large number of expert-annotated or historically validated "user physiological state-physiotherapy strategy" data pairs. As initial conditions, information is processed through its internal multi-head self-attention mechanism and position feedforward network, and a structured policy sequence is generated in an autoregressive manner. This policy sequence defines the complete framework of the target physical therapy strategy through specific tokens, including: Module selection: Specify a subset of the physiotherapy modules to be used in this treatment, such as "massage head and light-emitting element combination"; Timing planning: Define the start time, duration, and relative timing relationship between each selected module, represented by relative timecode; Intensity curve definition: Defines the intensity change trajectory for each selected module throughout the entire physiotherapy treatment, described in the form of a piecewise function or a sequence of key control points; Collaboration rule setting: Define the collaborative working logic and constraints between multiple modules, such as "the ultrasonic head can only be activated when the massage head is in low-frequency vibration mode"; S3. Strategy Verification and Structured Output: The strategy sequence generated in step S2 is sent to a lightweight strategy verification module for rapid feasibility checking. The checks include: whether the operating parameters of each module are within the hardware safety range, whether there are mutual exclusion conflicts in the control instructions of different modules, and whether the estimated total energy consumption is within the range allowed by the device's battery capacity. If the verification passes, the strategy sequence is parsed into a structured target physiotherapy strategy data object that can be recognized and executed by the strategy generation module. If the verification fails, an error message is fed back, and it may trigger a return to step S2 for regeneration or adjustment. Finally, the verified structured strategy data object is delivered to the drive control module for execution.

[0032] This solution leverages the creative capabilities of generative artificial intelligence, enabling the system to break through the limitations of a preset pattern library when faced with entirely new or complex user states. It automatically synthesizes safe, reasonable, and personalized physiotherapy plans, greatly enhancing the system's intelligence ceiling and its ability to cope with unknown situations.

[0033] In one embodiment of the present invention, the drive control module includes: a multi-channel synchronous controller for receiving a target physiotherapy strategy, which defines the intensity timing curves of one or more physiotherapy execution modules within a predetermined physiotherapy cycle; for the mechanical stimulation module, the synchronous controller generates a high-resolution pulse width modulation signal with a duty cycle proportional to the intensity timing curve to linearly control the torque output of the micro motor; for the phototherapy module, the synchronous controller generates a constant current drive signal and multiple PWM dimming signals to independently control the illuminance and flicker frequency of different wavelength light-emitting elements.

[0034] In this solution, the multi-channel synchronous controller is the final executor of the strategy. It receives the target physiotherapy strategy as a structured data object, which defines the intensity-time curve of each physiotherapy channel (corresponding to an actuator) over a period of time. The controller discretizes this curve into high-frequency control commands. For massage heads (e.g., driven by micro-motors), the controller generates PWM signals and linearly controls the average voltage of the motor by adjusting the duty cycle, thereby precisely controlling its vibration frequency and intensity. For light-emitting elements (e.g., LEDs), the controller generates constant current drives to ensure stable light sources, while simultaneously using additional PWM channels to independently adjust the brightness (dimming) and on / off frequency of different colored LEDs (for photobiological modulation). This achieves precise, synchronous, and digital control of multimodal physiotherapy methods, ensuring that complex physiotherapy strategies can be accurately translated into physical effects, providing a key hardware guarantee for the implementation of intelligent decision-making.

[0035] In one embodiment of the present invention, the multi-channel synchronous controller is embedded with conflict detection and arbitration logic. When the target physiotherapy strategy requires two or more physiotherapy modules to act on the same physical area, the arbitration logic rearranges the timing of the control signals or weights and attenuates the intensity according to predefined priority rules and safety time intervals to ensure the safety of the synergistic effect.

[0036] This solution incorporates a safety arbitration mechanism. Within the multi-channel synchronous controller, conflict detection and arbitration logic continuously monitor all upcoming control commands. When control commands from two or more modules (such as the ultrasound head and massage head) overlap in time and space (acting on the same scalp area), the arbitration logic is activated. It makes decisions based on a pre-defined rule base (e.g., "Ultrasound and high-frequency vibration should avoid simultaneous application to the same point," "Thermotherapy and cold therapy should be spaced at least 30 seconds apart"). The decision may result in: timing reordering: adjusting the activation time of a module to introduce a safety interval; intensity attenuation: reducing the instantaneous intensity of one or more modules according to priority; command rejection: in extreme cases, prohibiting the issuance of a command and reporting an error.

[0037] This solution provides a crucial security layer, preventing harmful interference between modules due to improper strategy generation or sensor errors, ensuring the safety of multimodal collaborative physiotherapy, and serving as the cornerstone of reliable system operation.

[0038] In one embodiment of the present invention, the intelligent physiotherapy control system further includes a user adaptive learning module for recording triplet data for each physiotherapy session: <user initial state> The target physical therapy strategy adopted Post-session state change It constructs a user-personalized effect prediction model, which is implemented using a neural network, with the user's initial state as input. With the adopted target physical therapy strategy The feature encoding outputs the predicted state changes. By comparing predicted values Compared with actual value The prediction error is calculated and used as a loss function. The parameters of the effect prediction model are continuously updated using the backpropagation algorithm. The updated model is called by the policy generation module to evaluate the expected effect of the candidate policies, so as to prioritize the policy with better prediction effect during policy matching or generation.

[0039] This solution enables the system to have personalized learning capabilities, allowing it to better understand the response characteristics of specific users as usage time increases, thereby providing increasingly accurate physiotherapy suggestions and achieving an evolution from "general intelligence" to "personalized intelligence".

[0040] In one embodiment of the present invention, the intelligent control unit communicates with a mobile terminal application via Bluetooth. The mobile terminal application includes: a physiotherapy data dashboard displaying physiological and physiotherapy parameters in time-series charts; a virtual expert interaction interface guiding users to complete initial calibration and preference settings; a efficacy tracking report evaluating long-term use effects based on preset indicators; and a community-driven pattern sharing platform allowing users to upload, download, and rate custom physiotherapy patterns created or verified by other users. The intelligent control unit can receive digitally signed physiotherapy pattern update packages and algorithm model incremental update files from the mobile terminal application or a cloud server, enabling OTA upgrades of the physiotherapy pattern library and state analysis model.

[0041] This solution expands the system's interaction with the outside world. The intelligent control unit connects to the user's mobile terminal application via Bluetooth. This app provides data visualization (physiotherapy data dashboard), personalized setting guidance (virtual expert interaction interface), long-term effect evaluation (treatment tracking report), and a user community (pattern sharing platform). Users can manually select patterns, set preferences, and view historical reports through the app. More importantly, the system achieves OTA upgrades through this channel: developers can package new physiotherapy patterns (after testing and verification) into "pattern update packages," or package optimized state classification models, reinforcement learning models, etc., into "algorithm model incremental update files." After ensuring security through digital signatures, these files are pushed to the user's physiotherapy comb via a cloud server or directly through the app. The intelligent control unit receives and verifies these update packages, writes them to storage, and completes the upgrade of the physiotherapy pattern library and core algorithms.

[0042] Finally, it should be noted that the above preferred embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail through the above preferred embodiments, those skilled in the art should understand that various changes can be made to it in form and detail without departing from the scope defined by the claims of the present invention.

Claims

1. An intelligent physiotherapy control system for a physiotherapy device, used for the physiotherapy control of the physiotherapy comb body, characterized in that, include: The physiotherapy execution module, integrated into the physiotherapy comb body, includes a massage head for mechanical stimulation, a light-emitting element for phototherapy, a feeding unit for introducing substances, and an ultrasonic head for ultrasonic action. A physiological parameter sensing module is installed on the body of the physiotherapy comb to collect at least one physiological parameter of the user in real time. The intelligent control unit, which is communicatively connected to the physiological parameter sensing module and the physiotherapy execution module, includes: a state analysis module for analyzing the user's current physiological state based on the collected physiological parameters; a strategy generation module for matching or generating a target physiotherapy strategy from a preset physiotherapy mode library based on the current physiological state, wherein the target physiotherapy strategy includes at least the setting of working parameters for at least one execution element in the physiotherapy execution module; and a drive control module for generating control commands to drive the physiotherapy execution module to work according to the target physiotherapy strategy.

2. The intelligent physiotherapy control system for the physiotherapy equipment according to claim 1, characterized in that: The physiological parameter sensing module includes: an optical sensing unit for acquiring photoplethysmography (PPG) signals from the scalp or neck region; an electrophysiological sensing unit for measuring skin conductance signals; and a motion sensing unit for monitoring head movement and posture signals. The state analysis module includes: a multi-source data fusion unit, used for time alignment, filtering, noise reduction, and normalization preprocessing of asynchronous heterogeneous data from optical sensing units, electrophysiological sensing units, and motion sensing units; a time-frequency domain feature extraction unit, used to extract features characterizing the user's physiological state from the preprocessed signal, including heart rate variability indicators calculated from photoplethysmography (PPG) signals, skin conductance response features extracted from skin conductance signals, and motion state features calculated from motion posture signals; and a state classifier, used to input the extracted feature vectors into a preset state classification model, which outputs the probability distribution of the user being in various preset physiological states, including deep relaxation, mild tension, cognitive fatigue, and physiological excitement.

3. The intelligent physiotherapy control system for the physiotherapy equipment according to claim 2, characterized in that: The strategy generation module includes a strategy decision engine, which receives the user state probability distribution from the state classifier and performs the following operations: if the matching degree between the state corresponding to the highest probability in the probability distribution and the adapted state of a preset mode in the physiotherapy mode library is higher than a first threshold, then the preset mode is directly called as the target physiotherapy strategy; if the matching degree is lower than the first threshold but higher than the second threshold, then the parameter optimizer is started, taking the user's real-time physiological characteristics as input and historical physiotherapy effect data as a reference, and fine-tuning at least one running parameter of the preset mode online through a reinforcement learning model driven by a reward function; if the matching degree is lower than the second threshold, then the strategy synthesizer is triggered, and based on the state probability distribution, the user's historical preference profile, and the group efficacy data obtained through the cloud platform, a new strategy containing the synergistic temporal and intensity parameters of multiple physiotherapy modules is constructed using a generative algorithm as the target physiotherapy strategy.

4. The intelligent physiotherapy control system of the physiotherapy equipment according to claim 3, characterized in that: The specific construction and operation methods of the reinforcement learning model include: Model definition: The fine-tuning process is modeled as a Markov decision process, whose states... for The feature vector output by the time-frequency domain feature extraction unit at each time step The basic parameter vector of the currently running preset physiotherapy mode splicing, that is Its actions For the basic parameter vector Adjustment amount of one or more continuous parameters Reward function For composite functions: ， In the formula, Let be the state similarity function. The penalty term for the parameter adjustment range. Inappropriate calculations based on real-time user feedback or abnormal physiological signals , , These are the weighting coefficients; Learning and Decision-Making: An Actor-Critic network architecture is trained using a proximal policy optimization algorithm. The Critic network outputs the probability distribution of the adjustment action given the state. Evaluate the value of the current state; whenever fine-tuning is needed, the model adjusts the settings based on the current state. Actions are sampled from the Actor network. The action is executed and the updated parameters are applied to drive the physiotherapy module, which then observes the new state in the next time period. And calculate the reward function. The network parameters are continuously updated using this sequence of data. , .

5. The intelligent physiotherapy control system of the physiotherapy equipment according to claim 3, characterized in that: The specific method for the policy synthesizer to construct new policies using generative algorithms includes the following steps: S1. Input representation: State probability distribution User history preference profile And therapeutic efficacy data from groups with similar current user status obtained from the cloud platform. Perform embedding encoding and concatenate the encoded vectors into a high-dimensional context vector: ， In the formula, , , These are the corresponding embedding encoding functions. This is a vector concatenation operation; S2. Sequence Generation Model: A generative model based on a Transformer decoder architecture is employed, with training data consisting of state-policy pairs annotated by experts or validated historically. This model uses context vectors... As the initial input, a strategy sequence is generated autoregressively through a multi-head self-attention mechanism and a position feedforward network, including: the selected subset of physiotherapy execution modules, the start-up sequence and duration of each module, the intensity change curve of each module throughout the treatment, and the coordination rules between modules; S3. Strategy Verification and Output: The generated strategy sequence undergoes a feasibility check via a lightweight strategy verification module, including parameter range verification, module mutual exclusion check, and estimated energy consumption assessment. After verification, the strategy sequence is parsed into a structured target physiotherapy strategy data object that can be executed by the strategy generation module and delivered to the drive control module.

6. The intelligent physiotherapy control system of the physiotherapy equipment according to claim 3, characterized in that: The drive control module includes: a multi-channel synchronous controller for receiving a target physiotherapy strategy, which defines the intensity timing curves of one or more physiotherapy execution modules within a predetermined physiotherapy cycle; for the mechanical stimulation module, the synchronous controller generates a high-resolution pulse width modulation signal with a duty cycle proportional to the intensity timing curve to linearly control the torque output of the micro motor; for the phototherapy module, the synchronous controller generates a constant current drive signal and multiple PWM dimming signals to independently control the illuminance and flicker frequency of different wavelength light-emitting elements.

7. The intelligent physiotherapy control system for the physiotherapy equipment according to claim 6, characterized in that: The multi-channel synchronous controller has embedded conflict detection and arbitration logic. When the target physiotherapy strategy requires two or more physiotherapy modules to act on the same physical area, the arbitration logic rearranges the timing of the control signals or weights and attenuates the intensity according to predefined priority rules and safety time intervals to ensure the safety of the synergistic effect.

8. The intelligent physiotherapy control system of the physiotherapy equipment according to claim 1, characterized in that: The intelligent physiotherapy control system also includes a user adaptive learning module for recording triplet data for each physiotherapy session: <user initial state> The target physical therapy strategy adopted Post-session state change It constructs a user-personalized effect prediction model, which is implemented using a neural network, with the user's initial state as input. With the adopted target physical therapy strategy The feature encoding outputs the predicted state changes. By comparing predicted values Compared with actual value The prediction error is calculated and used as a loss function. The parameters of the effect prediction model are continuously updated using the backpropagation algorithm. The updated model is called by the policy generation module to evaluate the expected effect of the candidate policies, so as to prioritize the policy with better prediction effect during policy matching or generation.

9. The intelligent physiotherapy control system for the physiotherapy equipment according to claim 1, characterized in that: The intelligent control unit communicates with a mobile terminal application via Bluetooth. The mobile terminal application includes: a physiotherapy data dashboard displaying physiological and physiotherapy parameters in time-series charts; a virtual expert interaction interface guiding users to complete initial calibration and preference settings; a efficacy tracking report evaluating long-term use effects based on preset indicators; and a community-driven pattern sharing platform allowing users to upload, download, and rate custom physiotherapy patterns created or verified by other users. The intelligent control unit can receive digitally signed physiotherapy pattern update packages and algorithm model incremental update files from the mobile terminal application or a cloud server, enabling OTA upgrades of the physiotherapy pattern library and state analysis model.

Images