Wearable knee joint self-powered sensing system and use method

Abstract

The invention discloses a wearable knee joint self-powered sensing system and a use method, relates to the technical field of motion monitoring, and solves the technical problem that a motion detection sensor integrating light weight and a self-powered mode is lacked in the prior art. The motor comprises a stator part and a rotor part which are rotationally arranged, a center shaft penetrating through the stator part is arranged on the rotor part, and the stator part and the rotor part are rotationally connected through the center shaft; and a motion state monitoring module and a self-powered module are sequentially and coaxially arranged at the connecting part of the stator part and the rotor part, so that the rotation angle and direction of the knee joint are accurately detected, and key data support is provided for subsequent motion analysis and health monitoring.

Description

Technical Field

[0001] This invention belongs to the field of motion monitoring technology, specifically relating to a wearable knee joint self-powered sensing system and its usage method. Background Technology

[0002] With the rapid development of artificial intelligence and Internet of Things technologies, smart wearable electronic devices have been widely used in sports rehabilitation, health monitoring, and other fields. However, most existing devices rely on external power supplies, which not only limits portability but also increases user costs and maintenance burdens. In addition, sensors in traditional wearable devices also suffer from high energy consumption, complex structures, and poor adaptability, making it difficult to meet the needs of long-term, efficient, and stable monitoring. Currently, there are wearable devices that use pneumatic-electric coupling drive, such as using air pressure to drive the knee joint structure and a motor to drive the hip joint structure during operation, and using a motor to reduce the burden during the swinging phase of walking. However, these devices still require external motor drive, increasing the dependence on external power supplies. At the same time, their relatively complex structure also limits the system's lightweight and portability.

[0003] Therefore, there is currently a lack of motion detection sensors that combine lightweight design and self-powered operation. Summary of the Invention

[0004] To address the aforementioned technical problems, this invention provides a wearable knee joint self-powered sensing system and its usage method. Based on the principles of triboelectric generation and electrostatic induction, the motion state monitoring module can output signals with binary characteristics. Through in-depth analysis of these signals, the rotation angle and direction of the knee joint can be accurately detected, providing crucial data support for subsequent motion analysis and health monitoring. Combining deep learning-based identity recognition and motion state recognition methods, and through comprehensive analysis of the output signals of the triboelectric nanogenerator, accurate user identification can be achieved, as well as effective detection of various motion states such as Parkinson's disease detection, fall detection, sit-stand alternation training, and balance training.

[0005] The technical solution adopted in this invention is as follows:

[0006] A wearable knee joint self-powered sensing system includes a stator and a rotor that are rotatably arranged. A central shaft is provided through the stator on the rotor and the stator and rotor are rotatably connected by the central shaft.

[0007] A motion status monitoring module is coaxially installed at the connection between the stator and the rotor.

[0008] The motion state monitoring module includes a friction layer and a friction block, which are respectively disposed on the stator and the central shaft. The friction layer is electrically connected to a control module.

[0009] Using the above technical solution, based on the principles of triboelectric generation and electrostatic induction, the motion state monitoring module can output signals with binary characteristics. Through in-depth analysis of these signals, the rotation angle and direction of the knee joint can be accurately detected, providing key data support for subsequent motion analysis and health monitoring. Combining deep learning-based identity recognition and motion state recognition methods, through comprehensive analysis of the output signals of the triboelectric nanogenerator, accurate identification of the user's identity can be achieved, as well as effective detection of various motion states such as Parkinson's disease detection, fall detection, sit-stand alternation training, and balance training.

[0010] Preferably, the friction layer includes a sensing channel formed by several copper foil rings, a rotating rod is fixedly arranged on the central axis, the two ends of the rotating rod are symmetrical with respect to the central axis, a plurality of friction blocks are arranged and cooperate with the sensing channel, and the plurality of friction blocks are respectively arranged on the side of the rotating rod close to the sensing channel, and the number of the two ends is symmetrical with respect to the central axis.

[0011] Using the above technical solution, when the central shaft rotates, the rotation of the central shaft drives the rotation rod to rotate, and the rotation of the rotation rod drives the friction block to rotate. The friction block interacts with the copper foil ring and generates an electrical signal based on the principles of triboelectric charging and electrostatic induction. The friction blocks are symmetrically arranged at both ends of the rotation rod, which reduces the signal deviation caused by unilateral force or poor contact. This allows for more accurate detection of the rotation state of the central shaft, making the detection results more stable and reliable, thereby improving the accuracy of monitoring the knee joint movement state.

[0012] Preferably, the sensing channel is composed of a three-channel copper foil with a concentric circle structure, consisting of an outer ring copper foil, a middle ring copper foil, and an inner ring copper foil from the outside to the inside, and the three-channel copper foil is fixed to the outer side of the stator.

[0013] The friction blocks are configured in groups of six, with three blocks in each group. The two groups of friction blocks are symmetrical about the central axis. The distance between two adjacent friction blocks corresponds to the positions of the outer, middle and inner copper foils, respectively. Each friction block has PTFE on its outer surface.

[0014] Using the above technical solution, the sensing channel employs a concentric circle structure of three-channel copper foil, namely, outer ring copper foil, middle ring copper foil, and inner ring copper foil. This design increases the dimension of signal detection. When PTFE sweeps across the three-channel copper foil, the copper foil of different channels contacts or separates from the PTFE at the corresponding ring position, generating three independent channel electrical signals. Through comprehensive analysis of these three channel electrical signals, the rotation angle and direction of the knee joint can be determined more accurately, improving the resolution and accuracy of angle detection. Setting six friction blocks, with every three symmetrically arranged around the central axis, and the distance between two adjacent friction blocks matching the corresponding copper foil position, further optimizes signal generation and detection. This layout ensures that appropriate friction blocks contact the copper foil at different rotation angles, generating stable and recognizable signals. For other electrical signals, PTFE is applied to the outer surface of each friction block. PTFE has excellent triboelectric properties, which can enhance the generation of electrical signals, improve the signal strength and stability, and thus improve the detection sensitivity and reliability of the entire motion state monitoring module. By analyzing the electrical signals of the three channels, a negative voltage peak is used to represent binary "1" and no peak is used to represent binary "0". Eight sensing zones from 15° to 120° are designed with 15° intervals, so that each sensing zone is represented by a unique three-bit binary code. This binary encoding method makes the signal have strong anti-interference ability and is not easily affected by electromagnetic interference and environmental noise. It can more accurately reflect the rotation angle range of the knee joint and provide a reliable data foundation for subsequent signal processing and motion state analysis.

[0015] Preferably, the outer, middle, and inner copper foils are configured as two sets of centrally symmetrical angle range partitions, which correspond to the range of human knee movement. Each set of angle range partitions consists of eight different sensing partitions, each with the same angle, and each sensing partition includes a portion of the outer, middle, and inner copper foils in different arrangements.

[0016] The above technical solution sets the outer, middle, and inner copper foil rings as two sets of centrally symmetrical angle range partitions, which correspond to the range of motion of the human knee. This design fully considers the actual range of motion of the human knee, enabling more comprehensive and accurate monitoring of the knee joint's movement within its normal range of motion, thus improving the targeting and effectiveness of the monitoring. Each angle range partition consists of eight different sensing partitions, each with the same angle and containing copper foil portions arranged in different combinations. Based on the principles of triboelectricity and electrostatic induction, the angle range is determined by whether the three PTFE contacts on the rotor are in contact with the copper foil of the three channels, forming a unique three-bit binary code. This design ensures that angle detection is unaffected by the initial position of the rotor, exhibiting strong anti-interference capabilities. Even in complex electromagnetic environments or in the presence of environmental noise, it can accurately detect the rotation angle of the knee joint, providing stable and reliable angle data for sports rehabilitation and health monitoring.

[0017] Preferably, the stator and rotor are connected by a motion state monitoring module and a self-powered module coaxially arranged in sequence; the self-powered module includes a ratchet and pawl unit, a planetary gear acceleration unit and an energy conversion unit arranged in sequence, and the energy conversion unit is electrically connected to an energy storage module.

[0018] The ratchet and pawl unit includes several fourth links, pawls, and a gear ring. The ends of the fourth links are uniformly fixed in a ring along the central axis. The number of pawls is the same as the number of fourth links. The end of each fourth link away from the central axis is hinged to the end of the pawl. One side of the gear ring is unidirectionally engaged with several pawls. The gear ring is rotatably connected to the central axis.

[0019] By employing the above technical solution, a ratchet and pawl unit and a planetary gear acceleration unit are used to collect the negative work generated during human walking and convert it into electrical energy to power the entire system. This design not only achieves effective energy recovery and utilization, reducing dependence on external power sources, but also helps to slow down muscle movement, reduce the body's metabolic rate, and improve the comfort and sustainability of human movement by recovering negative work. Using a half-wave electromagnetic generator to recover negative work during human walking, torque tests show that in one gait cycle of a 70kg adult, the root mean square value of the torque generated by the knee joint is 18.15 N·m. The maximum torque generated by the system during operation is only 0.05 N·m, accounting for 0.28% of the knee joint torque. This hardly hinders normal knee joint movement and places minimal burden on the human body, demonstrating practical feasibility and superiority.

[0020] The ratchet and pawl unit is uniformly fixed in a ring along the central axis by several fourth links, with pawls hinged at the ends. The gear ring and pawl engage in one-way rotation. When the fourth link rotates in the forward direction, the pawl engages with the gear ring, driving the gear ring to rotate. When the fourth link rotates in the reverse direction, the pawl disengages from the gear ring, thus achieving unidirectional rotation. This unidirectional rotation design can effectively convert the reciprocating motion of the knee joint into unidirectional rotation, providing a stable power input for subsequent energy conversion and motion state monitoring, and ensuring the directionality and effectiveness of energy harvesting and motion detection.

[0021] Preferably, the planetary gear acceleration unit includes a plurality of first gears. The side of the gear ring away from the pawl can mesh with the plurality of first gears. A second fixed disk is fixed to the stator via a first connecting rod. The second fixed disk is located at the end away from the stator. A bracket is fixedly provided on one side of the second fixed disk. The bracket is disposed between the gear ring and the second fixed disk. The plurality of first gears are rotatably connected to the side of the bracket away from the second fixed disk via a rotating shaft. A second gear meshes at the center formed by the plurality of first gears. The second gear is rotatably connected to the central shaft. The end of the second gear away from the gear ring is fixed to the energy conversion unit.

[0022] Using the above technical solution, in the planetary gear acceleration unit, the rotation of the gear ring drives the rotation of multiple first gears, and the first gears then drive the rotation of second gears. Through this gear transmission structure, the rotational speed is increased fourfold. This acceleration design can effectively increase the input speed of the energy conversion unit, enabling the energy conversion unit to convert mechanical energy into electrical energy more efficiently, thereby improving the energy conversion efficiency of the entire self-powered module, providing a more sufficient power supply for the system, and enhancing the system's self-powered capability.

[0023] Preferably, the energy conversion unit includes a first fixed disk, on the side of the first fixed disk near the second fixed disk, a plurality of magnets are disposed, and the plurality of magnets are coaxially and evenly spaced with the central axis; and on the side of the second fixed disk near the magnets, a plurality of coils are disposed.

[0024] Using the above technical solution, the energy conversion unit sets multiple magnets on the first fixed disk and multiple coils on the second fixed disk. Utilizing the principle that magnets cut magnetic field lines, it efficiently converts the negative work generated during human walking into electrical energy. The collected electrical energy is stored in an energy storage container through an energy management circuit to power the system. Simultaneously, the voltage signal generated by the three-channel triboelectric nanogenerator is transmitted to the signal processing circuit via lead wires. After amplification, filtering, and digitization, it is transmitted to the microcontroller and analyzed using an LSTM deep learning model to achieve functions such as identity recognition and motion state detection. This energy conversion and signal processing flow design organically combines energy collection, storage, and signal analysis and processing, realizing the system's self-powered and intelligent monitoring functions, and providing a complete technical solution for the practical application of wearable knee joint self-powered sensing systems.

[0025] A method for using a wearable self-powered knee joint sensing system, comprising the following steps:

[0026] Step 1: Wear the stator on the thigh and the rotor on the calf, aligning the central axis with the part of the knee that rotates; the knee joint movement causes the motion state monitoring module to generate a three-channel triboelectric voltage signal, which is collected by the Arduino microcontroller and transmitted to the control system; the control system performs data preprocessing on the three-channel triboelectric voltage signal to generate a target training set and a target test set;

[0027] Step 2: Train the deep learning model using the target training set from Step 1, and test the model's target performance using the target test set;

[0028] Step 3: Optimize the relevant parameters of the deep learning model to improve the corresponding monitoring performance.

[0029] The above technical solution provides a system usage method. By wearing the stator on the thigh and the rotor on the calf, the central axis is aligned with the knee's rotational position, ensuring accurate monitoring of knee joint movement. Knee joint movement generates a three-channel triboelectric voltage signal from the motion monitoring module, which is collected by an Arduino microcontroller and transmitted to the control system. The control system preprocesses the signal to generate a target training set and a target test set. The target training set is then used to train a deep learning model, and the target test set is used to test the model's performance. Finally, the deep learning model parameters are optimized. This usage method standardizes the system's operation process, ensuring accurate data collection, model training, and performance optimization. This enables precise monitoring and analysis of knee joint movement, providing reliable data support and decision-making basis for sports rehabilitation and health management.

[0030] Preferably, in step 1, data preprocessing is performed through an input gate, a forget gate, and an output gate.

[0031] The above technical solution employs input gates, forget gates, and output gates during data preprocessing, based on the operation of Long Short-Term Memory (LSTM) networks. The input gate controls the input of new information, the forget gate determines whether to retain or discard past information, and the output gate determines the output information. Through the coordinated work of these three gates, the three-channel triboelectric voltage signals can be effectively filtered, integrated, and processed to remove noise and redundant information, extracting valuable features for subsequent deep learning model training and motion state analysis. This improves the quality and usability of the data, contributing to enhanced training performance and accuracy of motion state detection in deep learning models.

[0032] Preferably, step 3, which optimizes the parameters of the identity recognition deep learning model, is as follows:

[0033] Step A01: Consider the impact of slice step size on model performance;

[0034] Step A02: Taking into account both accuracy and runtime metrics, select the optimal step size;

[0035] Step A03: Investigate the impact of individual sample length on model accuracy and processing time to determine the optimal sample length.

[0036] Using the above technical solution, specific steps are proposed for optimizing the parameters of the deep learning model for identity recognition. First, the impact of the slice step size on model performance is considered. The slice step size determines the granularity of the data processed by the model. A suitable slice step size can balance the model's computational load and its ability to capture data features. Then, considering both accuracy and runtime metrics, the optimal step size is selected. This improves the model's efficiency and reduces the waste of computational resources while ensuring accuracy. Finally, the impact of single sample length on model accuracy and processing time is studied to determine the optimal sample length. A suitable sample length allows the model to better learn patterns and features in the data, while avoiding negative impacts on model performance from excessively long or short samples. Through this series of parameter optimization steps, the performance of the deep learning model for identity recognition can be significantly improved, enabling it to more accurately and efficiently identify user identities, providing strong support for personalized applications and security management of the system.

[0037] In summary, due to the adoption of the above technical solution, the beneficial effects of the present invention are:

[0038] In sports rehabilitation and health monitoring applications, deep learning models are used to analyze the signals output by a three-channel triboelectric nanogenerator to monitor movement states such as identity recognition, Parkinson's disease detection, fall detection, sit-stand alternation training, and balance training. For example, the gait characteristics of Parkinson's disease patients differ significantly from those of normal individuals; by analyzing the three-channel triboelectric signals, Parkinson's disease can be detected. When a user falls, the output signal of the three-channel triboelectric nanogenerator will exhibit irregular signals, thus enabling fall detection. Attached Figure Description

[0039] The present invention will be described by way of example and with reference to the accompanying drawings, wherein:

[0040] Figure 1 This is a schematic diagram of the structure of a wearable knee joint self-powered sensing system according to the present invention;

[0041] Figure 2 This is an exploded three-dimensional structural diagram of the motion state monitoring module and the self-powered module in this invention;

[0042] Figure 3 This is a three-dimensional structural diagram of the motion state monitoring module in this invention;

[0043] Figure 4 This is a schematic diagram of the three-channel copper foil in this invention;

[0044] Figure 5 This is a schematic diagram of the three-channel copper foil partitioning in this invention;

[0045] Figure 6 This is a schematic diagram of the ratchet and pawl unit in the self-powered sensing system of the present invention;

[0046] Figure 7 This is a schematic diagram of the planetary gear acceleration unit in the self-powered sensing system of this invention;

[0047] Figure 8 This is an exploded structural diagram of the energy conversion unit in the self-powered sensing system of the present invention;

[0048] Figure 9 This is a flowchart of the digital processing of electrical signals from the motion state monitoring module in this invention;

[0049] Figure 10 This is a schematic diagram illustrating the optimization of the deep learning model for identity recognition in this invention;

[0050] Figure 11 This is a schematic diagram of the training accuracy of the identity recognition model and the confusion matrix of five user identities in this invention;

[0051] Figure 12This is a schematic diagram illustrating the optimization of the deep learning model for motion state recognition in this invention;

[0052] Figure 13 This is a schematic diagram of the training accuracy of the motion state recognition model and the confusion matrix of six motion state recognition methods in this invention.

[0053] Figure Labels

[0054] 1-First fixing part, 2-Second fixing part, 3-First connecting rod, 4-Central shaft, 5-Second connecting rod, 6-Third connecting rod, 7-Friction block, 8-Rotating rod, 9-Pawl, 10-Gear ring, 11-Friction layer, 1101-Outer ring copper foil, 1102-Middle ring copper foil, 1103-Inner ring copper foil, 12-Fourth connecting rod, 13-First gear, 14-Second gear, 15-Bracket, 16-Magnet, 17-Coil, 18-Second fixing plate, 19-First connecting shaft, 20-Second connecting shaft, 21-First fixing plate. Detailed Implementation

[0055] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions in the embodiments of this application will be clearly and completely described below. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0056] Example 1

[0057] A wearable knee joint self-powered sensing system, see attached figure. Figures 1-2 The device includes a stator and a rotor that rotate relative to each other. A central shaft 4 is provided through the stator on the rotor, and the stator and rotor are rotatably connected by the central shaft 4. A motion state monitoring module is coaxially arranged at the connection point of the stator and rotor. The motion state monitoring module includes a friction layer 11 and a friction block 7, which are respectively disposed on the stator and the central shaft 4. The friction layer 11 is electrically connected to a control module. Based on the principles of triboelectricity and electrostatic induction, the motion state monitoring module can output signals with binary characteristics. Through in-depth analysis of these signals, the rotation angle and direction of the knee joint can be accurately detected, providing key data support for subsequent motion analysis and health monitoring. Combining deep learning-based identity recognition and motion state recognition methods, through comprehensive analysis of the output signals of the triboelectric nanogenerator, accurate identification of the user's identity and effective detection of various motion states such as Parkinson's disease detection, fall detection, sit-stand alternation training, and balance training can be achieved.

[0058] The stator includes a first fixing part 1, and a second connecting rod is provided at the lower part of the first fixing part 1. The bottom end of the second connecting rod is a disc-shaped structure for fixing copper foil. The rotor includes a second fixing part 2, and a third connecting rod 6 is provided at the upper part of the second fixing part 2. The third connecting rod 6 is rotatably connected to the disc-shaped structure through a central shaft 4. The first fixing part 1 and the second fixing part 2 are straps or shell structures that can be worn on the user's legs.

[0059] The ratchet has 24 teeth and the pawl has a thickness of 3.5 mm.

[0060] Example 2

[0061] In this embodiment, refer to the appendix. Figure 3 The friction layer 11 includes a sensing channel formed by several copper foil rings. A rotating rod 8 is fixed on the central shaft 4 by pins. The two ends of the rotating rod 8 are symmetrical with respect to the central shaft 4. Multiple friction blocks 7 are provided and cooperate with the sensing channels. Several friction blocks 7 are respectively provided on the side of the rotating rod 8 near the sensing channel, and the number at both ends is symmetrical with respect to the central shaft 4. When the central shaft 4 rotates, the rotation of the central shaft 4 drives the rotating rod 8 to rotate, and the rotation of the rotating rod 8 drives the friction blocks 7 to rotate. The friction blocks 7 interact with the copper foil rings and generate electrical signals based on the principles of triboelectric charging and electrostatic induction. The symmetrical arrangement of friction blocks 7 at both ends of the rotating rod 8 reduces signal deviation caused by unilateral force or poor contact, and can more accurately detect the rotation state of the central shaft 4, making the detection results more stable and reliable, thereby improving the accuracy of knee joint movement monitoring.

[0062] The copper foil ring is made of copper foil, which is made directly from purchased finished copper foil tape, and is about 0.3mm thick.

[0063] In this embodiment, refer to the appendix. Figures 4-5 The sensing channel is composed of a three-channel copper foil with a concentric circle structure, consisting of an outer ring copper foil 1101, a middle ring copper foil 1102, and an inner ring copper foil 1103 from the outside to the inside. The three-channel copper foil is attached and fixed to the outer side of the stator.

[0064] The friction blocks 7 are configured as six, with three blocks forming a group. The two groups of friction blocks 7 are symmetrical about the central axis 4. The distance between two adjacent friction blocks 7 corresponds to the positions of the corresponding outer ring copper foil 1101, middle ring copper foil 1102, and inner ring copper foil 1103. Each friction block 7 has PTFE attached to its outer surface. The sensing channel uses a concentric circle structure of three-channel copper foil, namely outer ring copper foil 1101, middle ring copper foil 1102, and inner ring copper foil 1103. This design increases the dimension of signal detection. When PTFE sweeps across the three-channel copper foil, the copper foil in different channels contacts or separates from the PTFE at the corresponding ring position, generating three independent channel electrical signals. Through comprehensive analysis of these three channel electrical signals, the rotation angle and direction of the knee joint can be determined more accurately, improving the resolution and accuracy of angle detection. The six friction blocks 7, with three blocks forming a group, are symmetrical about the central axis 4, and the distance between two adjacent friction blocks 7 corresponds to the positions of the corresponding copper foil. The phase-matching design further optimizes signal generation and detection. This layout ensures that appropriate friction blocks 7 can contact the copper foil at different rotation angles, generating stable and identifiable electrical signals. PTFE is applied to the outer surface of each friction block 7. PTFE has excellent triboelectric properties, enhancing signal generation, strength, and stability, thereby improving the detection sensitivity and reliability of the entire motion state monitoring module. By analyzing the electrical signals from the three channels, a negative voltage peak represents binary "1," and the absence of a peak represents binary "0." Eight sensing zones, ranging from 15° to 120°, are designed with 15° intervals. Each sensing zone is represented by a unique three-bit binary code. This binary encoding method gives the signal strong anti-interference capabilities, making it less susceptible to electromagnetic interference and environmental noise. It can more accurately reflect the rotation angle range of the knee joint, providing a reliable data foundation for subsequent signal processing and motion state analysis.

[0065] Among them, the outer ring copper foil 1101 has an outer diameter of 85mm and an inner diameter of 67mm, the middle ring copper foil 1102 has an outer diameter of 65mm and an inner diameter of 47mm, and the inner ring copper foil 1103 has an outer diameter of 45mm and an inner diameter of 23mm.

[0066] In this embodiment, refer to the appendix. Figure 5The outer copper foil 1101, middle copper foil 1102, and inner copper foil 1103 are configured as two sets of centrally symmetrical angle range partitions. These angle range partitions correspond to the range of human knee movement, i.e., 15° to 120°. Each set of angle range partitions consists of eight different sensing partitions, each with the same angle. Each sensing partition includes a portion of the outer copper foil 1101, middle copper foil 1102, and inner copper foil 1103 arranged in different combinations. The outer copper foil 1101, middle copper foil 1102, and inner copper foil 1103 are configured as two sets of centrally symmetrical angle... The design employs a range-divided system, with each zone corresponding to the range of motion of the human knee. This design fully considers the actual range of motion of the human knee, enabling more comprehensive and accurate monitoring of the knee joint's movement within its normal range of motion. This improves the targeting and effectiveness of the monitoring. Each angle range zone consists of eight different sensing zones, each with the same angle and containing copper foil sections arranged in different combinations. Based on the principles of triboelectric charging and electrostatic induction, the angle range is determined by whether the three PTFE contacts on the rotor are in contact with the copper foil of the three channels, forming a unique three-bit binary code. This design ensures that angle detection is unaffected by the initial position of the rotor, exhibiting strong anti-interference capabilities. Even in complex electromagnetic environments or in the presence of environmental noise, it can accurately detect the rotation angle of the knee joint, providing stable and reliable angle data for sports rehabilitation and health monitoring.

[0067] These eight sensing areas correspond to eight different angle ranges, and are represented by a unique three-bit binary code: 101, 001, 110, 010, 100, 000, 111, 011.

[0068] Example 3

[0069] In this embodiment, refer to the appendix. Figure 6The stator and rotor are connected by a motion monitoring module and a self-powered module, which are coaxially arranged in sequence. The self-powered module includes a ratchet and pawl unit, a planetary gear acceleration unit, and an energy conversion unit arranged in sequence. The energy conversion unit is electrically connected to an energy storage module. The ratchet and pawl unit includes several fourth links 12, pawls 9, and a gear ring 10. The ends of the fourth links 12 are uniformly fixed in a ring along the central axis 4. The number of pawls 9 is the same as the number of fourth links 12. The end of each fourth link 12 away from the central axis 4 is hinged to the end of a pawl 9. One side of the gear ring 10 is unidirectionally engaged with several pawls 9. The gear ring 10 is rotatably connected to the central axis 4. By using the ratchet and pawl unit and the planetary gear acceleration unit, the negative work generated during human walking is collected and converted into electrical energy to power the entire system. This design not only achieves effective energy recovery and utilization, reducing dependence on external power sources, but also helps to slow down muscle movement, reduce the human metabolic rate, and improve the comfort of human movement during exercise. The system exhibits high efficiency and sustainability. It utilizes a half-wave electromagnetic generator to recover the negative energy generated during walking. Torque tests show that in one gait cycle of a 70kg adult, the root mean square value of the torque generated by the knee joint is 18.15 N·m. The maximum torque generated by the system during operation is only 0.05 N·m, accounting for 0.28% of the knee joint torque. This barely hinders normal knee joint movement and places minimal burden on the human body, demonstrating practical feasibility and superiority. The ratchet and pawl unit, via several fourth links 12 along the center... The spindle 4 is fixed in a uniform ring shape, with a pawl 9 hinged to its end. The gear ring 10 engages with the pawl 9 in one direction. When the fourth link 12 rotates in the forward direction, the pawl 9 engages with the gear ring 10, driving the gear ring 10 to rotate. When the fourth link 12 rotates in the reverse direction, the pawl 9 disengages from the gear ring 10, thus achieving unidirectional rotation. This unidirectional rotation design can effectively convert the reciprocating motion of the knee joint into unidirectional rotation, providing a stable power input for subsequent energy conversion and motion state monitoring, and ensuring the directionality and effectiveness of energy harvesting and motion detection.

[0070] In this embodiment, refer to the appendix. Figure 7The planetary gear acceleration unit includes multiple first gears 13. The side of the gear ring 10 away from the pawl 9 can mesh with the multiple first gears 13. A second fixed disk 18 is fixed to the stator via a first connecting rod 3. The second fixed disk 18 is located at the end away from the stator. A bracket 15 is fixedly mounted on one side of the second fixed disk 18. The bracket 15 is disposed between the gear ring 10 and the second fixed disk 18. The multiple first gears 13 are rotatably connected to the side of the bracket 15 away from the second fixed disk 18 via a rotating shaft. A second gear 13 is meshed at the center of the multiple first gears 13. Gear 14, the second gear 14 is rotatably connected to the central shaft 4, and the end of the second gear 14 away from the gear ring 10 is fixed to the energy conversion unit; in the planetary gear acceleration unit, the rotation of the gear ring 10 drives multiple first gears 13 to rotate, and the first gears 13 then drive the second gear 14 to rotate. Through this gear transmission structure, the rotational speed is increased fourfold. This acceleration design can effectively increase the input speed of the energy conversion unit, enabling the energy conversion unit to convert mechanical energy into electrical energy more efficiently, improving the energy conversion efficiency of the entire self-powered module, providing a more sufficient power supply for the system, and enhancing the system's self-powered capability.

[0071] The first connecting rod 3 is fixed to the second fixed plate 18 via the first connecting shaft 19;

[0072] The bracket 15 is fixed to the second fixed plate 18 via the first connecting shaft 19.

[0073] The first gear 13 is set to 4, with a gear module of 1mm and a transmission ratio of 1:4. The gear ring 10 has 80 teeth and a pitch circle diameter of 80mm. The first gear 13 has 30 teeth and a pitch circle diameter of 30mm. The second gear 14 has 20 teeth and a pitch circle diameter of 20mm.

[0074] In this embodiment, refer to the appendix. Figure 8The energy conversion unit includes a first fixed disk 21, on which multiple magnets 16 are arranged coaxially and evenly spaced with the central axis 4. Multiple coils 17 are arranged on the side of the second fixed disk 18 near the magnets 16. The energy conversion unit efficiently converts the negative work generated during human walking into electrical energy by using the principle of magnets 16 cutting magnetic field lines, through the multiple magnets 16 on the first fixed disk 21 and the multiple coils 17 on the second fixed disk 18. The collected electrical energy is stored in an energy storage container through an energy management circuit to power the system. Simultaneously, the voltage signal generated by the three-channel triboelectric nanogenerator is transmitted to a signal processing circuit via lead wires. After amplification, filtering, and digitization, the signal is transmitted to a microcontroller and analyzed using an LSTM deep learning model to achieve functions such as identity recognition and motion state detection. This energy conversion and signal processing flow design organically combines energy collection, storage, and signal analysis and processing, realizing the system's self-powered and intelligent monitoring functions, providing a complete technical solution for the practical application of wearable knee joint self-powered sensing systems.

[0075] Note that the first link 3, the second link 5, the third link 6, the fourth link 12, the pawl 9, the gear ring 10, the first gear 13, the second gear 14, the bracket 15, and the second fixing plate 18 are all manufactured using photosensitive resin as raw material through 3D printing technology. They have the advantages of simple manufacturing and low cost. In other embodiments, they can be adapted to be replaced with other materials.

[0076] Both the copper foil ring and the PTFE patch are directly attached to the surface of the 3D printed part using finished copper foil tape and PTFE tape.

[0077] Among them, magnet 16 is model NdFeB-N35, there are 6 of them, with a diameter of 20mm and a thickness of 5mm;

[0078] The coil 17 has a wire diameter of 0.2mm, 1500 turns, a quantity of 6, an outer diameter of 20mm, an inner diameter of 2mm, a thickness of 5mm, an internal resistance of 16.6Ω, and a magnet-coil gap of 3mm.

[0079] Example 4

[0080] A method for using a wearable knee joint self-powered sensing system is described in the appendix. Figure 9 The above-mentioned wearable knee joint self-powered sensing system includes the following steps:

[0081] Step 1: Wear the stator on the thigh and the rotor on the calf, aligning the central axis 4 with the knee's rotation point; knee joint movement causes the motion monitoring module to generate a three-channel triboelectric voltage signal, which is collected by the Arduino microcontroller and transmitted to the control system; the control system preprocesses the three-channel triboelectric voltage signal to generate a target training set and a target test set, with a ratio of 9:1.

[0082] Step 2: Train the deep learning model using the target training set from Step 1, and test the model's target performance using the target test set;

[0083] Step 3: Optimize the relevant parameters of the deep learning model to improve the corresponding monitoring performance.

[0084] A system usage method is provided, in which the stator is worn on the thigh and the rotor on the calf, aligning the central axis 4 with the knee's rotational position, ensuring accurate monitoring of knee joint movement. Knee joint movement generates a three-channel triboelectric voltage signal from the motion monitoring module, which is collected by an Arduino microcontroller and transmitted to the control system. The control system preprocesses the signal to generate a target training set and a target test set. The target training set is then used to train a deep learning model, and the target test set is used to test the model's performance. Finally, the deep learning model parameters are optimized. This usage method standardizes the system's operation process, ensuring accurate data collection, model training, and performance optimization, thereby achieving precise monitoring and analysis of knee joint movement status and providing reliable data support and decision-making basis for sports rehabilitation and health management.

[0085] In this embodiment, in step 1, data denoising, enhancement, and slicing preprocessing are performed through input gate, forget gate, and output gate.

[0086] LSTM is based on the RNN network structure, with the addition of a state vector c. t Simultaneously, a gating mechanism is introduced, using gating units to control the forgetting and refreshing of information. In LSTM, the input vector is x. t There are two state vectors c. t and h t (c t h is the internal state vector. t (Represents the output vector). LSTM deep learning models can utilize three gating units (forget gate g). f Input gate g i and output gate g o This is used to control the flow of internal information.

[0087] Gating is a method for controlling the flow of information, which is achieved by controlling the opening and closing of a valve. In LSTM, the degree of valve opening and closing is represented by a gating value vector g. The gating value is compressed to the interval [0,1] by the activation function σ(g), and its mathematical expression is:

[0088]

[0089] When σ(g) = 0, the valve is completely closed; when σ(g) = 1, the valve is fully open. This gating mechanism effectively controls the flow rate of data.

[0090] The three gate units in the LSTM model are illustrated in the figure below, and their corresponding definitions are as follows:

[0091] (1) Forgotten Gate

[0092] The forget gate operates on the state c of the memory unit. t-1 Forgetting gates selectively forget information in memory units, deciding which parts to discard and which to retain. The formula for forgetting gates is:

[0093] g f =σ(w f [h t-1 ,x t ]+b f (2)

[0094] Where σ is the activation function, w f and b f h represents the weight of the forget gate. t-1 It is the output vector of the previous unit, x t The input vector.

[0095] (2) Input Gate

[0096] The input gate also operates on the cell state, determining what new information to store in that cell state. The formula for the input gate is:

[0097] g i =σ(w i [h t-1 ,x t ]+b i (3)

[0098] Among them, w i and b i The weight of the forget gate.

[0099] (3) Output gate

[0100] The function of the output gate is to determine what the final output is; its formula is expressed as:

[0101] g o =σ(w o [h t-1 ,x t ]+b o (4)

[0102] Among them, w o and b o The weight of the forget gate.

[0103] The data preprocessing process utilizes input gates, forget gates, and output gates, based on the operation of Long Short-Term Memory (LSTM) networks. The input gate controls the input of new information, the forget gate determines whether to retain or discard past information, and the output gate determines the output information. Through the coordinated work of these three gates, the three-channel triboelectric voltage signals can be effectively filtered, integrated, and processed to remove noise and redundant information, extracting valuable features for subsequent deep learning model training and motion state analysis. This improves the quality and usability of the data, contributing to enhanced training performance and accuracy of motion state detection in deep learning models.

[0104] In this embodiment, step 3, which optimizes the parameters of the identity recognition deep learning model, is as follows:

[0105] Step A01: Consider the impact of slice step size on model performance;

[0106] Step A02: Taking into account both accuracy and runtime metrics, select the optimal step size;

[0107] Step A03: Investigate the impact of individual sample length on model accuracy and processing time to determine the optimal sample length.

[0108] See attached document Figure 10 A schematic diagram illustrating the optimization of a deep learning model for identity recognition: (a) detection performance under different step lengths; (b) detection performance under different sample lengths. In the identity recognition task, it can be observed that as the step length increases, the training accuracy, testing accuracy, and model running time of the model decrease. When the sample length is 100, the accuracy is close to the optimal value, and the processing time is relatively short. Therefore, this value is chosen as the optimal sample length. (See attached diagram.) Figure 11(a) shows the training accuracy and loss of the identity recognition model; (b) shows the confusion matrix for five user identities, where U1 to U5 represent the five target users for identity recognition. As the number of iterations increases, the training loss of the model steadily decreases, the training accuracy gradually improves, and the final model prediction accuracy reaches 99.68%. The confusion matrix for the five user identities fully demonstrates the high recognition accuracy of the LSTM deep learning model, which also shows that the design of the three-channel triboelectric nanogenerator can effectively sense subtle motion changes and can effectively distinguish the motion characteristics of the five users.

[0109] See attached document Figure 12 (a) shows the model detection performance analysis for different step lengths; (b) shows the model detection performance analysis for different sample lengths. A schematic diagram of the optimization of the deep learning model for motion state recognition is presented. In the motion state detection task, three-channel electrical signals from the user's clothing were collected for 240 seconds. Data sets were generated based on different sliding step lengths and window lengths. First, the impact of the slicing step length on model performance was considered. It was found that as the step length increased, the training accuracy, testing accuracy, and model running time of the model decreased. Considering both accuracy and running time, 5 was selected as the optimal step length. Next, the impact of the length of a single sample on the model's accuracy and processing time was studied. When the sample length was 130, the accuracy was close to the optimal value, and the processing time was short. Therefore, this value was selected as the optimal sample length.

[0110] See attached document Figure 13 (a) Training accuracy and loss of the motion state recognition model; (b) Confusion matrix of six motion states (showing walking, climbing stairs, obstacle avoidance, right turn, left turn, and jogging; M1 to M6 represent the six recognized motion states, namely: walking, climbing stairs, obstacle avoidance, right turn, left turn, and jogging); As the number of iterations increases, the training loss of the model steadily decreases, the training accuracy gradually improves, and the model eventually converges, with a prediction accuracy of 99.96%; The confusion matrix of six motion states reveals the excellent motion recognition capability of the three-channel triboelectric nanogenerator. This will help with training monitoring and rehabilitation assessment for people in rehabilitation. These six movement states are typical movements in walking training during rehabilitation training for lower limb movement disorders. The specific procedures for detecting movement states such as Parkinson's disease detection, fall detection, sit-stand alternation training, and balance training are not shown here. Their principle is motion recognition. By further collecting and expanding motion data, this model can be trained to achieve the recognition and classification of more types of movement states. The recognition of movement states such as Parkinson's disease detection, fall detection, sit-stand alternation training, and balance training are all based on this principle of motion recognition.

[0111] The confusion matrix is ​​a two-dimensional array, where each element C is a 2D array. ijThis represents the number of samples with the actual label i that were predicted as label j. The confusion matrix provides a clear view of the model's performance across different categories. For example, diagonal elements represent the number of samples correctly predicted by the model; off-diagonal elements represent the number of samples incorrectly predicted by the model.

[0112] The confusion matrix can be used to calculate various performance metrics, such as accuracy. Accuracy is the proportion of samples correctly predicted by the model out of the total number of samples.

[0113]

[0114] Where N is the number of categories.

[0115] This paper proposes specific steps for optimizing the parameters of a deep learning model for identity recognition. First, it considers the impact of slice step size on model performance, as the slice step size determines the granularity of data processing. A suitable slice step size can balance the model's computational load and its ability to capture data features. Then, by comprehensively considering accuracy and runtime metrics, the optimal step size is selected. This improves model efficiency and reduces wasted computational resources while ensuring accuracy. Finally, the paper investigates the impact of single sample length on model accuracy and processing time to determine the optimal sample length. A suitable sample length allows the model to better learn patterns and features in the data, while avoiding negative impacts on model performance from excessively long or short samples. Through this series of parameter optimization steps, the performance of the deep learning model for identity recognition can be significantly improved, enabling it to more accurately and efficiently identify user identities, providing strong support for personalized applications and security management in the system.

[0116] It should be noted that:

[0117] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A wearable knee joint self-powered sensing system, characterized in that, It includes a stator and a rotor that rotate relative to each other, and a central shaft (4) is provided through the stator on the rotor. The stator and the rotor are rotatably connected by the central shaft (4). A motion status monitoring module is coaxially installed at the connection between the stator and the rotor. The motion state monitoring module includes a friction layer (11) and a friction block (7). The friction layer (11) and the friction block (7) are respectively disposed on the stator and the central shaft (4). The friction layer (11) is electrically connected to a control module.

2. The wearable knee joint self-powered sensing system according to claim 1, characterized in that, The friction layer (11) includes a sensing channel formed by several copper foil rings. A rotating rod (8) is fixedly installed on the central shaft (4). The two ends of the rotating rod (8) are symmetrical with the central shaft (4). Multiple friction blocks (7) are provided and cooperate with the sensing channel. Several friction blocks (7) are respectively arranged on the side of the rotating rod (8) close to the sensing channel, and the number of the two ends is symmetrical with respect to the central shaft (4).

3. A wearable knee joint self-powered sensing system according to claim 2, characterized in that, The sensing channel is composed of a three-channel copper foil with a concentric circle structure, consisting of an outer copper foil (1101), a middle copper foil (1102), and an inner copper foil (1103) from the outside to the inside. The three-channel copper foil is fixed to the outer side of the stator. The friction blocks (7) are set to six, with three blocks forming a group. The two groups of friction blocks (7) are symmetrical about the central axis (4). The distance between two adjacent friction blocks (7) corresponds to the position between the corresponding outer copper foil (1101), middle copper foil (1102), and inner copper foil (1103). The outer surface of each friction block (7) is provided with PTFE.

4. A wearable knee joint self-powered sensing system according to claim 3, characterized in that, The outer copper foil (1101), middle copper foil (1102), and inner copper foil (1103) are configured as two sets of centrally symmetrical angle range partitions. The angle range partitions correspond to the range of human knee movement. Each set of angle range partitions consists of eight different sensing partitions. Each sensing partition has the same angle, and each sensing partition includes a portion of the outer copper foil (1101), middle copper foil (1102), and inner copper foil (1103) in different arrangements.

5. A wearable knee joint self-powered sensing system according to claim 1, characterized in that, The stator and rotor are connected by a motion state monitoring module and a self-powered module, which are coaxially arranged in sequence. The self-powered module includes a ratchet and pawl unit, a planetary gear acceleration unit, and an energy conversion unit arranged in sequence. The energy conversion unit is electrically connected to an energy storage module. The ratchet and pawl unit includes several fourth links (12), pawls (9), and a gear ring (10). The ends of the fourth links (12) are uniformly fixed in a ring along the central axis (4). The number of pawls (9) is the same as that of the fourth links (12). The end of each fourth link (12) away from the central axis (4) is hinged to the end of the pawl (9). One side of the gear ring (10) is unidirectionally engaged with several pawls (9). The gear ring (10) is rotatably connected to the central axis (4).

6. A wearable knee joint self-powered sensing system according to claim 5, characterized in that, The planetary gear acceleration unit includes multiple first gears (13). The side of the gear ring (10) away from the pawl (9) can mesh with multiple first gears (13). A second fixed disk (18) is fixed on the stator by a first connecting rod (3). The second fixed disk (18) is located at the end away from the stator. A bracket (15) is fixedly provided on one side of the second fixed disk (18). The bracket (15) is located between the gear ring (10) and the second fixed disk (18). Multiple first gears (13) are rotatably connected to the side of the bracket (15) away from the second fixed disk (18) through a rotating shaft. A second gear (14) meshes at the center formed by multiple first gears (13). The second gear (14) is rotatably connected to the central shaft (4). The end of the second gear (14) away from the gear ring (10) is fixed to the energy conversion unit.

7. A wearable knee joint self-powered sensing system according to claim 6, characterized in that, The energy conversion unit includes a first fixed disk (21), on the side of the first fixed disk (21) near the second fixed disk (18) a plurality of magnets (16) are arranged, and the plurality of magnets (16) are arranged coaxially and evenly spaced with the central shaft (4); the second fixed disk (18) is provided with a plurality of coils (17) on the side of the second fixed disk (18) near the magnets (16).

8. A method of using a wearable knee joint self-powered sensing system, characterized in that, The wearable knee joint self-powered sensing system according to any one of claims 1 to 7 includes the following steps: Step 1: Wear the stator on the thigh and the rotor on the calf, so that the central axis (4) is aligned with the part of the knee that rotates; the knee joint movement causes the motion state monitoring module to generate a three-channel triboelectric voltage signal, which is collected by the Arduino microcontroller and transmitted to the control system; the control system performs data preprocessing on the three-channel triboelectric voltage signal to generate a target training set and a target test set; Step 2: Train the deep learning model using the target training set from Step 1, and test the model's target performance using the target test set; Step 3: Optimize the relevant parameters of the deep learning model to improve the corresponding monitoring performance.

9. A method of using a wearable knee joint self-powered sensing system according to claim 8, characterized in that, In step 1, data preprocessing is performed using an input gate, a forget gate, and an output gate.

10. A method of using a wearable knee joint self-powered sensing system according to claim 8, characterized in that, Step 3 involves optimizing the parameters of the identity recognition deep learning model as follows: Step A01: Consider the impact of slice step size on model performance; Step A02: Taking into account both accuracy and runtime metrics, select the optimal step size; Step A03: Investigate the impact of individual sample length on model accuracy and processing time to determine the optimal sample length.

Images