Artificial intelligence-based electromagnetic control eyelid driving device and driving method

Abstract

This invention belongs to the fields of biomedical engineering, flexible electronics, and artificial intelligence-assisted rehabilitation technology. It provides an artificial intelligence-based electromagnetically controlled eyelid actuation device and method, including real-time acquisition of multimodal physiological signals from both sides of the patient's eyes; calculation of bilateral blinking intention probabilities using a hybrid AI prediction model; and generation of a driving command when multi-frame consistency conditions are met and both sides are synchronized. A controllable magnetic field is generated based on the driving command, driving both eyelids to perform opening and closing movements via a non-contact flexible actuation method using magnetohydrodynamic push-pull force. Simultaneously, based on a dual-layer control mechanism combining feedforward and feedback, the magnetohydrodynamic push-pull force of both eyelids is independently fine-tuned to compensate for differences in eyelid movement, ensuring consistent eyelid closure amplitude. This invention achieves highly sensitive, low-latency, and robust blinking intention recognition and auxiliary driving function execution, exhibiting good biocompatibility, wearing comfort, and clinical application potential.

Description

Technical Field

[0001] This application relates to the fields of biomedical engineering, flexible electronics technology, and artificial intelligence rehabilitation assistance technology, and in particular to an artificial intelligence-based electromagnetic control eyelid driving device and driving method. Background Technology

[0002] Currently, clinical treatment and assistive techniques for individuals with eyelid dysfunction or blinking inability have significant limitations, mainly in the following aspects:

[0003] 1. Existing treatments are highly invasive and lack dynamic adjustment capabilities: Traditional eyelid function correction mainly relies on surgery (such as levator palpebrae superioris shortening and frontalis muscle flap suspension) or mechanical orthotics. While surgery can improve eyelid position to some extent, it is inherently invasive and irreversible, requiring a long recovery period and making it difficult to dynamically adjust according to the patient's condition or individual needs post-surgery. Non-invasive mechanical orthotics, on the other hand, mostly provide static support or simple mechanical linkages. Their rigid drive modes cannot simulate the physiological rhythm and fine dynamics of natural eyelid opening and closing, making it difficult to meet the complex and ever-changing needs of daily activities.

[0004] 2. Existing assistive devices suffer from non-physiological control logic and poor individual adaptability: Even with the emergence of some active assistive devices in recent years, their core control strategies still have fundamental flaws. On the one hand, their movement patterns are often based on preset fixed rhythms or simple commands, failing to simulate the natural, smooth, and coordinated movement patterns of the human eye, resulting in mechanical and stiff movements that severely affect patient comfort and long-term compliance. On the other hand, these devices generally lack the ability to accurately recognize the user's intention to blink, making it impossible to achieve intuitive "thought-driven" control. For patients with impaired binocular function, achieving synchronous and natural binocular coordinated movement is particularly difficult due to the lack of reference to signals from the healthy side and the inability to respond to the differentiated needs of the affected side.

[0005] 3. Systemic Deficiency in Physiological Signal Monitoring and Closed-Loop Control Capabilities: Current technologies face two major bottlenecks in constructing biomimetic closed-loop control systems. First, at the signal acquisition level, the muscle groups driving eyelid movement (such as the orbicularis oculi) are delicate muscles, generating extremely weak electromyographic (EMG) signals and accompanying micro-deformation signals in the skin. Traditional rigid sensors struggle to achieve stable, high-sensitivity multimodal acquisition of these signals while ensuring non-invasive, flexible wear. Second, at the control level, although camera-based eye trackers can track eye movements, they cannot directly acquire the physiological state signals of the eyelid dynamic units (muscles and skin), resulting in a lack of crucial real-time feedback. This lack of end-to-end closed-loop capability prevents the device from adaptively adjusting to the actual working state of the muscles and external interference, ultimately limiting its potential for achieving high-precision, low-latency auxiliary effects.

[0006] In summary, current technologies have significant shortcomings in terms of biomimicry in functional reconstruction, real-time control, individualized adaptability, and physiological closed-loop control capabilities. Summary of the Invention

[0007] In view of the shortcomings of the prior art, the present invention provides an electromagnetically controlled eyelid driving device and driving method based on artificial intelligence. It utilizes multimodal physiological signal fusion and hybrid AI prediction model for active prediction, which can achieve highly sensitive, low-latency, and robust blinking intention recognition and auxiliary driving function execution. It has good biocompatibility, wearing comfort and clinical application potential.

[0008] To achieve the above and related objectives, the present invention adopts the following technical solution:

[0009] The first aspect of this invention provides an artificial intelligence-based electromagnetic control method for driving eyelids, comprising the following steps:

[0010] Step S100: Real-time acquisition of multimodal physiological signals around the patient's bilateral eyes, including electromyographic signals and skin surface strain signals, through non-invasive sensing via skin contact.

[0011] Step S200: Extract local temporal features of multimodal physiological signals through a hybrid AI prediction model, analyze the temporal dependence and bilateral synchronization of multimodal physiological signals, calculate the bilateral blinking intention probability, and generate driving instructions when the multi-frame consistency condition and bilateral synchronization are met.

[0012] In step S300, a controllable magnetic field is generated based on the driving command, and the bilateral eyelids are driven to perform opening and closing actions through a non-contact flexible driving method of magnetohydrodynamic push and pull force. At the same time, based on a dual-layer control mechanism combining feedforward and feedback, the magnetohydrodynamic push and pull force of the bilateral eyelids is independently fine-tuned according to real-time multimodal physiological signals to compensate for the difference in the movement of the bilateral eyelids and make the closing amplitude of the bilateral eyelids consistent.

[0013] Furthermore, in step S200, the hybrid AI prediction model is a hybrid architecture combining convolutional neural networks and long short-term memory networks; the hybrid AI prediction model also includes a multimodal fusion layer, which uses an attention mechanism to weightedly integrate the features of multimodal physiological signals.

[0014] Further, in step S200, when the multi-frame consistency condition is met and both sides are synchronized, the generation of the driving instruction includes: when the probability of blinking intention on both sides exceeds the set threshold and synchronization is achieved within the preset time window, and when blinking intention is determined to exist in at least 2 to 3 consecutive analysis time frames, the driving instruction is generated.

[0015] Furthermore, in step S300, the dual-layer control mechanism combining feedforward and feedback includes: outputting the initial values ​​of the dual-side drive currents based on the strength of the drive command and the pre-calibrated mechanical model; and acquiring multimodal physiological signals in real time during the opening and closing action to independently fine-tune the drive currents in real time.

[0016] Furthermore, step S300 also includes simultaneously executing a safety management strategy and an action rhythm management and calibration strategy during the opening and closing action. The safety management strategy includes dual real-time monitoring and feedback control of drive current and temperature. The action rhythm management and calibration strategy includes performing action rhythm management and action minimum interval time control, and simultaneously completing online self-calibration.

[0017] A second aspect of the present invention provides an artificial intelligence-based electromagnetically controlled eyelid actuation device, comprising a wearable main body, and further comprising:

[0018] The flexible sensor module is integrated into the wearable body and is used to collect multimodal physiological signals around the patient's eyes in real time through non-invasive sensing via skin contact, including electromyographic signals and skin surface strain signals.

[0019] The algorithm control module, which communicates with the flexible sensor model and is located inside the wearable body, is used to extract the local temporal features of multimodal physiological signals through a hybrid AI prediction model, analyze the temporal dependence and bilateral synchronization of multimodal physiological signals, calculate the probability of bilateral blinking intention, and generate driving instructions when the multi-frame consistency condition and bilateral synchronization are met.

[0020] The electromagnetic drive module, which communicates with the flexible sensor module and the algorithm control module, is used to generate a controllable magnetic field based on drive commands and drive the bilateral eyelids to perform opening and closing actions through a non-contact flexible drive method of magnetohydrodynamic push and pull force. At the same time, based on a dual-layer control mechanism combining feedforward and feedback, the magnetohydrodynamic push and pull force of the bilateral eyelids is independently fine-tuned according to real-time multimodal physiological signals to compensate for the difference in the movement of the bilateral eyelids and make the closing amplitude of the bilateral eyelids consistent.

[0021] Furthermore, the flexible sensor module includes a strain sensor array and an electrode array integrally formed on a flexible elastomer substrate using liquid metal screen printing technology, wherein the strain sensor array has a serpentine or wavy liquid metal circuit structure.

[0022] Furthermore, the electromagnetic drive module includes an electromagnet assembly configured to control the drive current of the electromagnet assembly via pulse width modulation to generate a gradient magnetic field; it also includes a current monitoring unit and a temperature sensor.

[0023] Furthermore, the electromagnetic drive module also includes an eyelid contact unit and a magnetofluid unit. The eyelid contact unit includes a sealed microcavity, and the magnetofluid unit is encapsulated within the sealed microcavity. The magnetofluid unit includes a magnetofluid containing iron oxide nanoparticles.

[0024] Furthermore, the outer wall of the sealed microcavity is provided with a limiting structure to restrict excessive flow of magnetofluid.

[0025] The beneficial technical effects of this invention are as follows:

[0026] This invention is intended to assist individuals with eyelid dysfunction or blinking impairment in blinking, or to reconstruct blinking function. It utilizes flexible electronics technology to achieve stable, non-invasive monitoring of multimodal physiological signals, combines a hybrid AI prediction model for high-precision active prediction of bilateral blinking intentions, and employs electromagnetic actuation to achieve precise, low-latency closed-loop drive.

[0027] This invention utilizes a flexible sensing module that adheres non-invasively to the skin and a magnetofluid-based non-contact actuation, avoiding surgical trauma and infection risks, achieving complete non-invasiveness, and improving treatment safety and patient acceptance. Based on a hybrid AI prediction model, this invention generates actuation commands by analyzing multimodal physiological signals. This allows for real-time, adaptive, and precise adjustment of eyelid opening and closing movements according to the patient's actual intentions and the external environment, demonstrating good dynamic adjustment capabilities.

[0028] This invention enables physiological and personalized natural coordinated movement. The driving commands of this invention originate from the patient's own neuromuscular signals (EMG and strain), and the resulting eyelid movements are directly triggered by the patient's neural intentions. Its dynamic process highly simulates natural blinking, with smooth and natural movements, significantly improving user experience and wearing compliance.

[0029] The hybrid AI prediction model of this invention has learning capabilities and can adapt to the differences in signal characteristics among different patients. Through multi-frame consistency, bilateral synchronization judgment, and a two-layer control mechanism combining feedforward and feedback, it can intelligently coordinate eye movements, enabling natural and synchronized opening and closing movements even for patients with bilateral dysfunction.

[0030] This invention integrates flexible electronics and artificial intelligence algorithms to achieve highly biomimetic, real-time, and precise assistance for patients with bilateral eyelid dysfunction. Its advantages include: highly robust intention recognition, effectively overcoming the problems of lack of healthy-side reference and weak signal in bilateral lesions; precise synchronization and difference compensation, achieving precise temporal synchronization of eye movements and individualized independent compensation for amplitude differences; and non-invasiveness and comfort, with flexible electronics and magnetohydrodynamic actuation ensuring good biocompatibility, low power consumption, and comfortable wear.

[0031] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and do not limit this application. Attached Figure Description

[0032] The accompanying drawings, incorporated in and forming part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application. It is obvious that the drawings described below are merely some embodiments of this application, and those skilled in the art can obtain other drawings based on these drawings without inventive effort. In the drawings:

[0033] Figure 1 This is a flowchart of the electromagnetic control eyelid driving method based on artificial intelligence in this application;

[0034] Figure 2 This is a schematic diagram of the structure of the electromagnetically controlled eyelid driving device based on artificial intelligence in this application. Detailed Implementation

[0035] Unless otherwise defined, all technical and / or scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. It should be understood that certain features of the invention (described in the context of separate embodiments for clarity) may also be provided in a single embodiment. Conversely, multiple features of the invention (described in the context of a single embodiment for brevity) may also be provided separately or in any suitable combination or, where appropriate, in any other described embodiment of the invention. Certain features described in the context of various embodiments will not be considered essential features of those embodiments unless the embodiment is inoperable without those elements. The invention is further illustrated below by specific examples; however, it should be noted that the specific process conditions and results described in the embodiments of the invention are merely illustrative and should not be construed as limiting the scope of protection of the invention. All equivalent changes or modifications made in accordance with the spirit and essence of the invention should be covered within the scope of protection of the invention.

[0036] Please see Figure 1 The flowchart of the artificial intelligence-based electromagnetic control eyelid driving method of this application is described in detail below:

[0037] Step S100: Real-time acquisition of multimodal physiological signals around the patient's bilateral eyes, including electromyographic signals and skin surface strain signals, through non-invasive sensing via skin contact.

[0038] Specifically, the multimodal physiological signals of this application may also include electrooculography (EOG) signals. This application can acquire multimodal physiological signals in real time using a flexible sensor that is non-invasively attached to the patient's periocular skin. The flexible sensor is equipped with a flexible electrode array for acquiring electromyography (EMG) signals to directly detect the bioelectrical activity generated by the orbicularis oculi muscle under nerve innervation; it is also equipped with a flexible strain sensor array for acquiring skin surface strain signals to detect the minute stretching and deformation of the periocular skin during blinking.

[0039] Specifically, this application can also preprocess multimodal physiological signals, such as amplifying electromyographic signals and using filters to remove power frequency interference and high-frequency noise, thereby improving the signal-to-noise ratio. This application places flexible sensors symmetrically around the patient's left and right eyes to simultaneously acquire bilateral multimodal physiological signals.

[0040] Step S200: Extract local temporal features of multimodal physiological signals through a hybrid AI prediction model, analyze the temporal dependence and bilateral synchronization of multimodal physiological signals, calculate the bilateral blinking intention probability, and generate driving instructions when the multi-frame consistency condition and bilateral synchronization are met.

[0041] Specifically, the hybrid AI prediction model in this application is a hybrid architecture combining convolutional neural networks and long short-term memory networks. This application employs a dual-channel input hybrid architecture, utilizing CNNs to extract local temporal features of multimodal physiological signals. For example, CNNs can be used to identify a specific electrical pulse peak in electromyography (EMG) signals, or subtle deformations of the skin during rapid stretching in skin surface strain signals. LSTMs are used to capture the temporal dependencies and synchronicity of bilateral multimodal physiological signals. Since an effective blinking action is not instantaneous but consists of a series of temporally continuous events such as intention initiation, muscle contraction, and eyelid closure, LSTMs are specifically designed to learn long-term dependencies in such time-series data.

[0042] Specifically, the hybrid AI prediction model of this application also includes a multimodal fusion layer, which uses an attention mechanism to weightedly integrate the features of multimodal physiological signals to effectively decouple the central blinking intention from the residual motor signals of the diseased muscles. This application utilizes the attention mechanism to automatically assess which signal features are more important for making a correct judgment at a specific moment. For example, during the intention initiation phase, more attention may be paid to the features of electromyographic signals; while during the action execution phase, more reliance may be placed on strain signals. By dynamically assigning appropriate weights to the features of different signals, weighted integration is achieved, thereby making the predictions of the hybrid AI prediction model more accurate and robust.

[0043] Specifically, the hybrid AI prediction model of this application also includes a training strategy based on transfer learning. It first pre-trains using normal human data, then performs transfer learning and fine-tuning using weakly supervised patient data (based on joint detection of button intent and signal peaks), enhancing the model's adaptability to real pathological signals from patients. This application can also implement data augmentation by intentionally adding noise, time distortion, and random occlusion to the training data to ensure the model has strong robustness and resists interference from the real environment.

[0044] Specifically, in this application, when the multi-frame consistency condition is met and both sides are synchronized, the generation of the driving instruction includes: generating the driving instruction when the probability of blinking intention on both sides exceeds a set threshold and synchronization is achieved within a preset time window, and when blinking intention is determined to exist for at least 2 to 3 consecutive analysis time frames. This application can effectively filter out instantaneous noise caused by signal interference or accidental muscle twitching through multi-frame consistency judgment, ensuring that the generated action is a continuous and genuine intention; through bilateral synchronization judgment, it can prevent unilateral false triggering and ensure natural and coordinated action.

[0045] In step S300, a controllable magnetic field is generated based on the driving command, and the bilateral eyelids are driven to perform opening and closing actions through a non-contact flexible driving method of magnetohydrodynamic push and pull force. At the same time, based on a dual-layer control mechanism combining feedforward and feedback, the magnetohydrodynamic push and pull force of the bilateral eyelids is independently fine-tuned according to real-time multimodal physiological signals to compensate for the difference in the movement of the bilateral eyelids and make the closing amplitude of the bilateral eyelids consistent.

[0046] Specifically, in response to a drive command, this application generates a specific control signal that acts on an electromagnet module integrated on a wearable carrier, causing it to generate a magnetic field of controllable strength. The magnetofluid encapsulated in the eyelid contact unit deforms or displaces under the influence of the magnetic field, generating a push-pull force. This push-pull force is transmitted through a flexible medium, thereby driving the eyelid to complete the opening and closing action without contact.

[0047] Specifically, the dual-layer control mechanism combining feedforward and feedback includes: outputting initial values ​​of the bilateral driving currents based on the strength of the driving command and a pre-calibrated mechanical model; and acquiring multimodal physiological signals in real time during the opening and closing action to independently fine-tune the driving currents in real time. More specifically, the feedforward control includes: before executing the opening and closing action, rapidly calculating the initial values ​​of the driving currents (IL0, IR0) for both eyelids based on the strength of the driving command and a pre-calibrated force-current-displacement model. This mechanical model is established based on a large amount of experimental data, defining the relationship between the driving current and the expected eyelid movement amplitude / velocity, providing the basic framework for the action through feedforward control, and ensuring rapid response. Feedback control includes: during the execution of the action, real-time acquisition of multimodal physiological signals such as electromyography signals around the eyes and skin surface strain signals as feedback. The control algorithm compares these real-time signals with the expected values. If a deviation is detected (for example, due to individual differences or muscle fatigue, the movement of one eyelid is slow), the current of the driving electromagnets of the left and right eyes will be independently fine-tuned to dynamically compensate for the difference in movement and ensure the synchronicity of the movement of both eyelids and the consistency of the closing amplitude.

[0048] Specifically, this application also includes simultaneously executing a safety management strategy and an action rhythm management and calibration strategy during the opening and closing action. The safety management strategy includes dual real-time monitoring and feedback control of drive current and temperature. The action rhythm management and calibration strategy includes performing action rhythm management and control of the minimum action interval time, and simultaneously completing online self-calibration.

[0049] More specifically, this application's safety management strategy sets current limits to prevent coil overload; simultaneously, temperature sensors monitor the temperature of key parts of the wearable carrier in real time, ensuring that surface temperature rise is strictly controlled within a safe range (e.g., not exceeding 2°C) to avoid the risk of low-temperature burns; if the monitored value exceeds the limit, immediate intervention will be implemented, such as reducing the current or pausing the drive. To simulate natural blinking and prevent false triggering, this application manages the rhythm of blinking movements, for example, by enforcing a minimum interval time for each movement (≥300 ms~500 ms), ensuring that non-physiological rapid continuous blinking does not occur, thus making the movement more natural and effectively filtering out invalid muscle twitching signals, preventing continuous false triggering and movement crowding, and ensuring the natural continuity of blinking movements. This application has learning and adaptive capabilities; with increased usage time, it can automatically fine-tune its internal mechanical model and control parameters based on long-term feedback data to adapt to subtle changes in the patient's wearing position and electrode impedance, continuously maintaining the consistency and accuracy of the movements.

[0050] like Figure 2 As shown, this application also provides an artificial intelligence-based electromagnetically controlled eyelid actuation device, including a wearable main body, and further comprising:

[0051] The flexible sensor module is integrated into the wearable body and is used to collect multimodal physiological signals around the patient's eyes in real time through non-invasive sensing via skin contact, including electromyographic signals and skin surface strain signals.

[0052] The algorithm control module, which communicates with the flexible sensor model and is located inside the wearable body, is used to extract the local temporal features of multimodal physiological signals through a hybrid AI prediction model, analyze the temporal dependence and bilateral synchronization of multimodal physiological signals, calculate the probability of bilateral blinking intention, and generate driving instructions when the multi-frame consistency condition and bilateral synchronization are met.

[0053] The electromagnetic drive module, which communicates with the flexible sensor module and the algorithm control module, is used to generate a controllable magnetic field based on drive commands and drive the bilateral eyelids to perform opening and closing actions through a non-contact flexible drive method of magnetohydrodynamic push and pull force. At the same time, based on a dual-layer control mechanism combining feedforward and feedback, the magnetohydrodynamic push and pull force of the bilateral eyelids is independently fine-tuned according to real-time multimodal physiological signals to compensate for the difference in the movement of the bilateral eyelids and make the closing amplitude of the bilateral eyelids consistent.

[0054] Specifically, the wearable device in this application can be an eyeglass frame structure. For binocular dysfunction, the device adopts a symmetrical dual-channel design, with an independent flexible sensor module and electromagnetic drive module arranged in the left and right eyelid areas respectively.

[0055] Specifically, this application's flexible sensor module is non-invasively attached to the skin around the human eye. It includes a strain sensor array and an electrode array integrally formed on a flexible elastomer substrate using liquid metal screen printing technology. The strain sensor array has a serpentine or wavy liquid metal circuit structure. This application uses a biocompatible, stretchable, and moderately viscous elastomer material (such as TPU film) as the flexible elastomer substrate. The sensor and flexible interconnect circuit are integrally formed using liquid metal (such as gallium indium alloy) screen printing technology. The strain sensor array, with a serpentine or wavy liquid metal circuit structure, is arranged in the upper eyelid, lower eyelid, and area above the zygomatic arch. Its resistance changes can sensitively reflect the amplitude of micro-deformation of the skin, thereby distinguishing different movement patterns such as natural blinking and rapid blinking. The electrode array (EMG) is placed at key locations in the levator palpebrae superioris muscle or orbicularis oculi muscle to collect electromyographic (EMG) signals, which are more likely to reflect blinking commands issued by the central nervous system.

[0056] Specifically, the electromagnetic drive module (actuator module) of this application is a key execution module for realizing the active and precise opening and closing of the eyelids. It includes an electromagnet assembly, which is configured to control the drive current of the electromagnet assembly through pulse width modulation to generate a gradient magnetic field. More specifically, the electromagnet module of this application uses a high-permeability iron core and enameled copper wire winding, and is integrated inside the lens frame. Through a PWM-modulated constant current drive circuit, the magnetic field strength (e.g., in the range of 10mT~60 mT) can be dynamically and precisely adjusted. The driving mechanism of the electromagnet module is that under the action of the controllable magnetic field generated by the electromagnet, the magnetohydrodynamic unit generates a precise push-pull force (e.g., 0.5mN~1.5mN), which is sufficient to drive the eyelids to partially or completely close, with a target response time of <150 ms.

[0057] Specifically, the electromagnetic drive module of this application further includes an eyelid contact unit and a magnetofluid unit. The eyelid contact unit includes a sealed microcavity composed of a biocompatible elastic film, and the magnetofluid unit is encapsulated within the sealed microcavity of a micron-sized elastomer film. The magnetofluid unit includes a magnetofluid containing iron oxide nanoparticles, and this unit is flexibly attached to the affected eyelid. The outer wall of the sealed microcavity is provided with a limiting structure to restrict excessive flow of the magnetofluid. This limiting structure can be a limiting rib to prevent the magnetofluid from spraying or leaking under the action of a strong magnetic field. More specifically, the eyelid contact unit of this application can be detachably attached to the patient's eyelid skin surface through a medical adhesive layer.

[0058] Specifically, the electromagnetic drive module of this application also includes a current monitoring unit and a temperature sensor. Based on the feedback from the current monitoring unit and the temperature sensor, this application can control the temperature rise of the main body surface of the mirror frame within a safe threshold (such as not exceeding 2°C).

[0059] Specifically, the algorithm control module of this application is the core of the device to achieve intelligent closed-loop operation. It is deployed on a low-power embedded AI chip to realize local inference. The algorithm control module includes a hybrid AI prediction model, which processes signals from the flexible sensor module and generates drive instructions for the electromagnetic drive module by running the hybrid AI prediction model.

[0060] The present invention will be described in detail below through specific examples and embodiments. It should also be understood that the following embodiments are only for specific illustration of the present invention and should not be construed as limiting the scope of protection of the present invention. Any non-essential improvements and adjustments made by those skilled in the art based on the above description of the present invention are within the scope of protection of the present invention. The specific process parameters, etc., in the following examples are merely examples within a suitable range; that is, those skilled in the art can make appropriate selections within the appropriate range based on the description herein, and are not intended to be limited to the specific values ​​in the examples below.

[0061] Example 1

[0062] The device in this embodiment includes a frame and also includes:

[0063] Flexible sensing module: It uses a thermoplastic polyurethane (TPU) film with a thickness of 20 to 50 micrometers as a substrate, and integrates a serpentine or wavy strain sensor array and an electromyography (EMG) signal electrode array using liquid metal screen printing technology, and then encapsulates it using thermopressing technology. This array is non-invasively placed in the upper and lower eyelids, and can collect multimodal physiological signals such as EMG (sampling rate 100Hz) and strain (sampling rate 200Hz) in real time, and distinguish blinking patterns by the resistance changes of the strain sensor.

[0064] The electromagnetic drive module consists of a magnetohydrodynamic unit of iron oxide (Fe3O4) encapsulated in a 10μm~20μm elastomer film microcavity (nanoparticle size controlled at 10nm~15nm) and a frame electromagnet module. The electromagnet uses pulse-width modulation (PWM) constant current drive, generating a push-pull force of 0.5mN~1.5mN under a magnetic field strength of 15mT~60mT, sufficient to drive the eyelid. The target response time is less than 150ms, and the surface temperature rise is controlled within 0.2℃. Performance testing of the actuator (including force-current relationship fitting and 100,000-cycle durability testing) is conducted using a micro-force sensor and high-speed camera.

[0065] Algorithm control module: Employs a low-power embedded artificial intelligence (AI) chip, such as the ARM Cortex-M series, integrating multi-channel low-noise acquisition circuitry and a small lithium battery (six to eight hours of battery life). The hybrid AI prediction model is a hybrid architecture combining convolutional neural networks (CNN) and long short-term memory networks (LSTM). It integrates electromyography (EMG), strain (and optional electrooculography) features through a multimodal fusion layer that introduces an attention mechanism. A strategy of pre-training on normal subjects combined with fine-tuning through patient transfer learning is employed, with accuracy, recall, and F1 score as the main indicators for cross-validation evaluation.

[0066] The above embodiments are merely illustrative of the principles and effects of the present invention and are not intended to limit the invention. Any person skilled in the art can modify or alter the above embodiments without departing from the spirit and scope of the present invention. Therefore, all equivalent modifications or alterations made by those skilled in the art without departing from the spirit and technical concept disclosed in the present invention should still be covered by the claims of the present invention.

Claims

1. An artificial intelligence-based electromagnetically controlled eyelid driving device, characterized in that, Including the wearable main body, it also includes: A flexible sensor module, which is integrated into the wearable body, is used to collect multimodal physiological signals around the patient's eyes in real time through non-invasive sensing via skin contact, including electromyographic signals and skin surface strain signals. An algorithm control module, communicatively connected to the flexible sensor module and housed within the wearable body, is used to extract local temporal features of the multimodal physiological signals using a hybrid AI prediction model, analyze the temporal dependence and bilateral synchronicity of the multimodal physiological signals, calculate the probability of bilateral blinking intention, and generate driving instructions when multi-frame consistency conditions and bilateral synchronization are met. The hybrid AI prediction model is a hybrid architecture combining convolutional neural networks and long short-term memory networks, employing a dual-channel input architecture. The convolutional neural network extracts the local temporal features of the multimodal physiological signals, while the long short-term memory network captures the temporal dependence and synchronicity of the bilateral multimodal physiological signals. The hybrid AI prediction model also includes a multimodal fusion layer, which uses an attention mechanism to weightedly integrate the features of the multimodal physiological signals to effectively decouple central blinking intention from residual motion signals of diseased muscles. The electromagnetic drive module, which is communicatively connected to the flexible sensor module and the algorithm control module, is used to generate a controllable magnetic field based on drive commands and drive both eyelids to perform opening and closing actions through a non-contact flexible drive method of magnetohydrodynamic push-pull force. At the same time, based on a dual-layer control mechanism combining feedforward and feedback, the magnetohydrodynamic push-pull force of both eyelids is independently fine-tuned according to the real-time multimodal physiological signals to compensate for the difference in the movement of both eyelids and make the closing amplitude of both eyelids consistent.

2. The electromagnetically controlled eyelid driving device according to claim 1, characterized in that, The flexible sensor module includes a strain sensor array and an electrode array integrally formed on a flexible elastomer substrate using liquid metal screen printing technology. The strain sensor array has a serpentine or wavy liquid metal circuit structure.

3. The electromagnetically controlled eyelid driving device according to claim 1, characterized in that, The electromagnetic drive module includes an electromagnet assembly configured to control the drive current of the electromagnet assembly via pulse width modulation to generate a gradient magnetic field; it also includes a current monitoring unit and a temperature sensor.

4. The electromagnetically controlled eyelid driving device according to claim 3, characterized in that, The electromagnetic drive module further includes an eyelid contact unit and a magnetofluid unit. The eyelid contact unit includes a sealed microcavity, and the magnetofluid unit is encapsulated within the sealed microcavity. The magnetofluid unit includes a magnetofluid containing iron oxide nanoparticles.

5. The electromagnetically controlled eyelid driving device according to claim 4, characterized in that, The outer wall of the sealed microcavity is provided with a limiting structure to restrict the excessive flow of the magnetofluid.

6. The electromagnetically controlled eyelid driving device according to claim 1, characterized in that, When the multi-frame consistency condition is met and both sides are synchronized, the driving instruction is generated as follows: when the probability of blinking intention on both sides exceeds the set threshold and synchronization is achieved within the preset time window, and when blinking intention is determined to exist in at least 2 to 3 consecutive analysis time frames, the driving instruction is generated.

7. The electromagnetically controlled eyelid driving device according to claim 1, characterized in that, The dual-layer control mechanism combining feedforward and feedback includes: outputting initial values ​​of the dual-side drive currents based on the strength of the drive command and the pre-calibrated mechanical model; and acquiring the multimodal physiological signals in real time during the opening and closing action to independently fine-tune the drive currents in real time.

8. The electromagnetically controlled eyelid driving device according to claim 7, characterized in that, During the opening and closing action, a safety management strategy and an action rhythm management and calibration strategy are executed simultaneously. The safety management strategy includes dual real-time monitoring and feedback control of drive current and temperature. The action rhythm management and calibration strategy includes action rhythm management and minimum action interval control, and online self-calibration is completed simultaneously.

Images