Method for controlling motion behavior of artificial joint

Abstract

The invention relates to a method for controlling the movement behavior of an artificial joint having an upper part and a lower part which is mounted on the upper part so as to be pivotable about a pivot axis, between which a device for influencing the pivoting capacity or pivoting of the upper part relative to the lower part is arranged, the device is coupled to a control unit, a rule set is stored in the control unit, and the control unit activates, deactivates or modulates the device on the basis of input values for the rule set in order to influence the pivoting or pivoting capability, wherein sensor values of at least one sensor arranged on the upper part or the lower part, detected during use of the artificial joint, are provided to at least one machine learning-based estimation method, the estimation method continuously calculates an estimated value for a dynamic or kinematic parameter or for an expected dynamic or kinematic parameter from the sensor values and provides the estimated value as an input value to the rule set and serves as a criterion for activation, deactivation or modulation in the rule set.

Description

Field of the Invention

[0001] The present invention relates to a method for controlling the movement behavior of an artificial joint, in particular an artificial knee joint, which joint has an upper part and a lower part pivotally mounted relative thereto about a pivot axis, and between the upper and lower parts there is arranged a device for influencing the pivoting ability or pivoting of the upper part relative to the lower part, which device is coupled to a control unit in which a rule set is stored, and the control unit activates, deactivates or modulates the device based on input values of the rule set to influence the pivoting movement or pivoting ability. The rule set stores methods and control parameters for controlling the artificial joint. Background Art

[0002] Artificial joints, in particular artificial knee joints, are used in prostheses and orthoses. Prostheses functionally, and sometimes also in appearance, replace missing limbs. Orthoses are applied to limbs to guide, restrict and possibly influence the movement of natural limbs. Orthoses have an orthotic joint located between or forming part of the upper and lower parts. Both the upper and lower parts have fixing means for fixing the orthosis to the limb. Prostheses are provided with fixing means for fixing the prosthesis to the stump or the patient. Between the upper and lower parts of the artificial joint there are arranged corresponding devices (such as dampers or actuators) for influencing the pivoting movement or pivoting ability, which devices are coupled to a control unit by means of which the behavior of the damper or actuator can be activated, deactivated or changed. The damper can for example be a purely passive device, such as a linear hydraulic device, a rotary hydraulic device or a magnetorheological damper. A mechanical brake can also influence the pivoting ability or pivoting of the upper part relative to the lower part. As actuators, in particular electric motors and other energy storage devices are meant, which can initiate or assist movement, or can also resist movement to slow down pivoting. By means of appropriate control, locking of the joint can also be achieved by the actuator, thereby eliminating the pivoting ability.

[0003] The control unit activates, deactivates or modulates the device for influencing the pivoting movement or pivoting ability, for example based on sensor data transmitted to the control unit. The sensors are arranged on the artificial joint or its attachments (such as a prosthesis socket, a distal prosthesis component or an orthosis strut). The sensors can also be arranged on the assisted limb or the contralateral limb.

[0004] For example, state machines stored in the control unit are used to control the resistance change. A rule set can contain multiple state machines that are dynamically activated according to the situation. Based on the sensor data, it can be deduced what state the prosthetic or orthotic device is in and how the adjustment device (such as a valve) must be activated or deactivated to produce a specific motion behavior. For example, in a hydraulic resistance device, the valve is fully or partially closed to change the flow channel connection of the flow cross-section, thereby affecting the corresponding movement of the joint. EP549855B1 describes a prosthetic knee joint controller with a state machine.

[0005] DE102020111535A1 discloses a method for controlling at least one actuator in an orthopedic device, which has an electronic control unit. The control unit is coupled to the actuator and at least one sensor and has at least one electronic processor for processing sensor data. At least one state machine is stored in the control unit, in which the states of the orthopedic device and the state transitions of the actuator are defined. In addition, a classification is also stored in the control unit, in which the sensor data and / or states are automatically classified during the classification process. The classification process and the state machine can be used in combination. Based on the classification and the state, the way of activating or deactivating the actuator is determined as a control signal.

[0006] CN113520683A discloses a lower limb prosthetic control system and its control method, in which the prosthetic has a prosthetic knee joint control motor, an ankle joint control motor, a connecting rod and a housing. The gait information of a healthy person under different conditions is detected by an inertial measurement unit (IMU), and a training data set is created therefrom. A neural network model is established and trained using the collected data in a simulation environment. The conventional neural network model is implemented in the control unit into the lower limb prosthetic. When the control unit receives the input signal from the IMU, it issues an action instruction to the lower limb joint considering the trained network model.

[0007] The article "Knee Angle Estimation based on IMU data and Artificial NeuralNetworks", Bennett et al., 29 thAs known from the Southern Biomedical Engineering Conference, 2013, pages 111 and 112, the measurement of knee joint angle is crucial for gait analysis. The measurement can be carried out by an IMU, but due to non-direct measurement, parameters such as angle, gait phase, and standing symmetry can only be estimated. This study explored how to use artificial neural networks to estimate knee joint angle based on accelerometer and gyroscope data. It was found that the acceleration sensor is the most effective sensor, and when an IMU is arranged above and below the knee respectively, the artificial neural network performs best.

[0008] Angle sensors, such as knee joint angle sensors, require installation space inside or outside the artificial joint and must be installed, wired, and calibrated separately. The control unit requires additional structural design to receive and process angle data, which increases system complexity. Summary of the Invention

[0009] The task of the present invention is to provide a method by which, even in the absence of an angle sensor, it is possible to identify with sufficient precision when and how the pivoting or pivoting ability must be influenced, while keeping the operating cost as low as possible.

[0010] The above task is solved by a method having the features of the independent claims. Advantageous embodiments and improvements of the present invention are disclosed in the dependent claims, the description, and the drawings.

[0011] The present invention relates to a method for controlling the movement behavior of an artificial joint, which joint has an upper component and a lower component pivotally supported on the upper component about a pivot axis, and between the upper and lower components there is arranged a device for influencing the pivoting ability or pivoting of the upper component relative to the lower component, which device is coupled to a control unit in which a rule set is stored, and the control unit activates, deactivates or modulates the device based on input values of the rule set to influence the pivoting movement or pivoting ability, characterized in that sensor values detected during the use of the artificial joint (e.g., from one IMU arranged on the upper or lower component or multiple IMUs arranged on the upper and lower components) are provided to at least one machine learning-based estimation method (MLSV) (e.g., an artificial neural network), which method calculates an estimated value of a kinetic or kinematic parameter (e.g., joint angle) or an estimated value of an expected kinetic or kinematic parameter based on the said sensor data, and provides this estimated value as an input value (e.g., for a joint angle signal) to the rule set and uses it as a criterion for activating, deactivating or modulating the device for influencing the pivoting ability or pivoting in the data set. In one embodiment, it is provided that multiple identical or different machine learning-based estimation methods (MLSV) determine multiple estimated values for different kinetic or kinematic parameters and input them to the rule set. Based on data from sensors (e.g., IMU or multiple IMUs) arranged on the upper or lower component or associated components and limbs of the artificial joint, the joint angle and, if necessary, another value or other values derived therefrom are continuously calculated using this or these machine learning-based estimation methods (MLSV). Without the need to position a direct angle sensor at the joint (e.g., a knee joint angle sensor that requires wiring and calibration), the proposed method generates a virtual joint angle sensor that provides a calculated or estimated value of the joint angle (especially the knee joint angle). The virtual sensor in the said machine learning-based estimation method (MLSV) is based on the evaluation of data from one or more sensors (especially one IMU or multiple IMUs), wherein the machine learning-based estimation method (MLSV) is especially trained for the calculation of the joint angle. The method is not limited to artificial joint devices of the lower limbs, especially artificial knee or ankle joints, and can also be used for other artificial joints, such as hip joints or joints of the upper limbs, such as elbow joints, shoulder joints or wrist joints.

[0012] A machine learning-based estimation method (MLSV) can be implemented, for example, as an artificial neural network (KNN), such as a multi-layer perceptron, or include a KNN. Alternatively, black-box models and regression methods are also applicable, where the internal model and calculation parameters are optimized using pre-acquired data during the training process. An application case of MLSV is to estimate the knee joint angle based on IMU data installed on the upper or lower component. Training data can be obtained by using a system that has both a knee joint angle sensor and an IMU simultaneously. Using these data to train MLSV, the knee joint angle is estimated from the IMU data by using the data of an additional knee joint angle sensor as the "ground truth". Thus, the "ground truth" contains the reference result of the training process, in which MLSV learns to estimate the reference result only based on the existing source data (such as IMU data). Once MLSV is trained, it can be used as a virtual sensor to estimate the knee joint angle. In this way, the physical angle sensor can be replaced. Alternatively, MLSV can also be trained with an "unsupervised learning" method.

[0013] This method of calculating or determining the joint angle or knee joint angle without a direct joint angle sensor is helpful for anti-stumble protection in an artificial knee. Especially when the flexion resistance and extension resistance can be adjusted separately, in an artificial knee controlled by a control unit with a microprocessor, anti-stumble protection can be achieved, that is, increasing the flexion resistance at the moment of the movement reversal in the swing phase. This moment can also be reliably determined by the calculated or estimated value of the knee joint angle or knee joint angular velocity calculated by MLSV. In addition, the movement reversal in the swing phase is a clear and specific situation, which is relatively easy to identify, so the inaccuracies in the calculation or estimation are easily compensated or irrelevant.

[0014] Therefore, the joint angle input value can be provided without being directly detected by an angle sensor, thus greatly reducing the operation cost as well as the installation and calibration cost.

[0015] In one implementation, the machine learning-based estimation method continuously determines multiple estimated values for one or more kinematic or dynamic parameters. Therefore, for example, not only the joint angle can be calculated, but also the load, load history, acceleration, the spatial orientation of the component relative to gravity, etc. can be calculated without directly measuring the required parameters.

[0016] In one embodiment, the sensor values are determined by at least one IMU, for example, for kinematic or dynamic parameters used to estimate the joint angle between the upper part and the lower part by MLSV. Multiple identical or different MLSVs can also be used to determine multiple estimated values of different kinematic or dynamic parameters, which are then used as the basis for further processing. If there are multiple estimated values for different kinematic or dynamic parameters, they are provided as input parameters to the rule set.

[0017] In one implementation, additional sensor data and / or state variables are provided as input values to the rule set, so these additional sensor data and / or state variables are used as criteria for activating, deactivating, or modulating a device for influencing the pivoting ability or pivoting of the upper part relative to the lower part. This is particularly advantageous in safety-critical situations because a plausibility check based on the additional sensor data can compensate for or counteract possible errors in the MLSV. Thus, safety-critical activation, deactivation, or modulation is not performed solely based on the results of the machine learning-based estimation method (MLSV) in one implementation form, but is ensured by other parameters, measurements, or calculations.

[0018] To improve the accuracy of the calculated values and thus the quality of the input to the rule set, the MLSV is supplied with pre-acquired sensor data from a database and trained. The pre-determined data stored in the database helps the MLSV estimate the probability of a specific situation occurring. The database can be continuously updated and coupled with the MLSV to expand the data basis and improve accuracy.

[0019] In one implementation, the database is updated during operation to achieve real-time optimization. In addition, for the refinement and update of the MLSV, inactive phases can be utilized, such as during charging or when not in use, during which the system can be updated based on additional data. This data can be collected during operation, so it contains patient-specific information. Alternatively, the data can be provided centrally by the manufacturer, for example, via the Internet as a "Field Update".

[0020] In an improved solution, the expected joint angle is provided to the rule set with a lead time between 0.001 seconds and 1 second. In this way, the control can intervene in advance, but not prematurely. The MLSV can predict the expected behavior of the upper part relative to the lower part and the development of the current situation, enabling it to respond more quickly to possible actual changes. This prediction is particularly feasible and more accurate because the MLSV can predict the probability of future movements or future movement behaviors with high precision based on database data. Based on these probabilities or estimates, the control unit issues or prepares corresponding instructions in advance to set the corresponding behavior, that is, to set the pivoting movement or pivoting ability of the upper part relative to the lower part about the pivot. From the sensor values of the IMU, the probability of the movement situation or joint state is calculated in the control unit or the MLSV and then provided as an input value to the rule set. Therefore, the MLSV constitutes a virtual sensor that precisely determines variables (such as the knee joint angle) and then transmits this variable as an input quantity to the control unit. In addition to the knee joint angle, other variables such as force, torque, ground conditions, or the slope of the ground can also be calculated or predicted. Therefore, these variables are also calculated by the MLSV with the corresponding probabilities of occurrence and used as input quantities for the control unit. Then these input quantities are processed in their respective rule sets and used as the basis for changing or influencing the pivoting resistance of the artificial joint.

[0021] In one embodiment, the sensor values from the previous time step or period can also be input into the machine learning-based estimation method to increase the accuracy of determining the estimated values, especially to provide the MLSV with movement history information.

[0022] In one embodiment, the machine learning-based estimation method is based on a Gaussian process, which enables the MLSV to provide, in addition to the estimated values of kinematic or dynamic parameters, information on the confidence intervals of the corresponding estimated values, which will be used in subsequent processing. For example, the confidence intervals of the corresponding estimated values are used as criteria for activation, deactivation, or modulation and may be transmitted to the rule set together with the respective determined estimated values. This probabilistic method uses kernel functions that provide, in addition to the estimated values, relevant confidence intervals, that is, an indication of the estimation quality. These additional information is particularly important for the control of orthopedic devices because it can determine the weight of the influence of the estimated values in the rule set on the control of the artificial joint. A small confidence interval corresponds to good estimation accuracy, so the estimated value has a high degree of credibility. Therefore, compared with an estimated value with a larger confidence interval and lower reliability, this estimated value can participate in the control with a higher weight.

[0023] Estimated values of ground conditions and ground slope information are particularly advantageous for controlling an artificial ankle joint or a prosthetic foot. For example, the foot position at the end of the swing phase can be modulated according to the estimated ground slope. To optimize behavior on stairs, it is helpful to estimate the height difference to be overcome. The artificial knee joint can adjust the resistance or assistive behavior according to these parameters. For example, when walking downhill, increasing flexion resistance is provided as the slope increases. When walking uphill, the assistive torque can be adapted to the height difference to be overcome. All this additional information that helps optimize the assistive performance of the artificial joint can be determined as estimated values by a correspondingly trained MLSV. Therefore, it may make sense to implement multiple MLSVs based on different methods in artificial joint control to estimate different parameters (P). For example, ground condition parameters can be estimated using KNN, while joint angles can be estimated by a method based on Gaussian processes. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Embodiments of the present invention will be described in detail below with reference to the drawings. In the drawings:

[0025] Figure 1 A schematic diagram showing a leg prosthesis;

[0026] Figure 2 A schematic diagram showing a leg prosthesis in a flexed position;

[0027] Figure 3 A flowchart is shown;

[0028] Figure 4 A knee joint angle estimation diagram is shown;

[0029] Figure 5 A schematic diagram showing a leg orthosis;

[0030] Figure 6 A parameter regression schematic diagram is shown; and

[0031] Figure 7 A schematic diagram showing a prosthetic ankle joint. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0032] In Figure 1 the schematic diagram shows a prosthetic knee joint that is part of a leg prosthesis. The prosthetic knee joint has an upper component 10 and a lower component 20, which are pivotally mounted on each other about a pivot 15. A prosthetic foot 60 is provided at the distal end of the lower component 20. In the embodiment of the prosthetic leg as Figure 1 shown, a prosthetic cylinder or other device for accommodating the thigh stump or fixing to the human body is arranged or formed on the upper component 10.

[0033] Between an upper part 10 and a lower part 20, there is arranged a device 30 for influencing the pivoting ability or pivoting of the upper part 10 relative to the lower part 20, which device is configured as a linearly acting hydraulic damper. In the illustrated embodiment, the hydraulic damper has a hydraulic chamber or cylinder arranged or formed in a housing or base body 31. A piston 32 is movably mounted in the cylinder. The piston 32 can move along the longitudinal extension direction of the cylinder and is fixed to a piston rod 33 protruding from the housing or base body 31. The piston 32 divides the cylinder into chambers, which are fluid-technically interconnected by hydraulic lines. The base body 31 or the housing can be pivotally supported at a fixed point 23 on the lower part 20 to prevent the piston 32 from jamming during the pivoting movement of the upper part 10 relative to the lower part 20. One end of the piston rod 33 remote from the piston 32 is fixed to the upper part 10, and in the illustrated embodiment, it is fixed at an upper fixed point 21 on a bracket for increasing the distance from the pivot 15. During flexion, the piston 32 is pressed downwards, so the volume of the flexion chamber decreases, and correspondingly, the volume of the extension chamber increases (minus the volume of the retracted piston rod 33). In order to generate pressure in one of the chambers, an electric motor can be provided in the housing 31, which drives a pump (not shown) to apply pressure to the hydraulic fluid in one of the two chambers and thus move the piston 32 in the cylinder in one or the other direction. This causes the orthotic device in the form of a prosthetic leg to perform flexion or extension movements. The electric motor for driving the pump is an optional solution, which can be used in combination with the linear damper 30 in one embodiment. In principle, no driver or motor is required in a passive prosthetic knee joint. Another alternative for the device 30 is to use a rotary damper (especially a rotary hydraulic device), a magnetorheological resistance device, or an electric motor in generator operating mode instead of a linear damper (especially a linear hydraulic device). A combination of several of the above resistance devices or drivers can also be implemented in one design. Instead of a hydraulic damper, the device 30 can also include a linear actuator, a rotary driver, or a combination of the technologies.

[0034] An actuator 34 is arranged inside or on the housing 31, which actuator is coupled to at least one regulating valve 35, and through which regulating valve the hydraulic resistance in the device 30 can be changed. The actuator 34 is coupled to a control unit 40, which activates, deactivates, or modulates the actuator 34 based on sensor values, so as to be able to provide adaptively adjusted resistance and possible hydraulic locking. When the device 30 is implemented as a magnetorheological resistance device, the resistance is changed by activating, deactivating, or modulating the magnetic field, and the actuator 34 is then an electromagnet or a magnetic coil. If the device 30 further includes an active driver, the control unit 40 also provides instructions for actively outputting energy through the driver.

[0035] Sensors 50 are arranged on both the upper part 10 and the lower part 20 for detecting the spatial orientation of the lower part 20 or the upper part 10. In particular, the sensor 50 for detecting the spatial orientation is only arranged on the upper part 10. By means of this sensor 50 (configured as an IMU (inertial measurement unit)), during the use of the prosthetic knee joint, the spatial angle or the absolute angle relative to a fixed spatial orientation (such as the direction of gravity) is determined. In addition to detecting the spatial orientation, the IMU as the sensor 50 can also detect other status data, in particular status data related to the artificial knee joint. The status data includes in particular position, angular attitude, velocity, acceleration, force and their history or changes. The determined spatial angle or other status quantities of the upper part 10 and / or the lower part 20 are transmitted as input quantities to the control unit 40. The control unit 40 modulates, activates or deactivates the actuator 34 to change its flow resistance, viscosity, braking force or the force against buckling movement in the device 30 designed as a hydraulic damper. In order to be able to drive the actuator 34, an energy storage, in particular in the form of a storage battery, is assigned to it. The energy storage can be arranged directly next to the actuator 34 or at other locations of the orthotic device where there is more available space or which is more advantageous for weight distribution considerations. In addition to the electromechanical drive unit with a storage battery or battery as the energy storage, in some embodiments, a mechanical energy storage such as a spring or a flywheel is also provided.

[0036] In addition, the control unit 40 can be arranged on the prosthetic limb and at least one other sensor 50 can be arranged on the prosthetic foot 60. All sensors arranged on the prosthetic limb or orthosis are coupled to the control unit 40 and their sensor values are used as the basis for controlling the actuator 34 of the device 30 (when configured as a damper) or as an input signal for controlling a motor (when configured as a motor). For the case of magnetorheological damping, the sensor values are used to control the magnetic field or its change. Based on the sensor data (in particular the spatial orientation and position data, load, orientation, acceleration and / or deformation data of other components), the actuator 34 is controlled, for example, to reduce or increase the pivoting resistance, limit the end stops and / or generate or assist the relative movement between the upper part 10 and the lower part 20.

[0037] In Figure 2In the schematic view, a prosthetic knee joint in a flexed position is shown, which includes an upper component 10, a lower component 20, and a device 30 arranged therebetween for influencing the pivoting ability or pivoting of the upper component 10 relative to the lower component 20. The upper component 10 can pivot relative to the lower component 20 about a pivot axis 15 against the resistance of the device 30 and has an IMU as a sensor 50. Here, the joint angle α (knee joint angle α in the illustrated embodiment) between the upper component 10 and the lower component 20 changes. In the illustrated embodiment, the joint angle α is measured between the upper component 10 and the lower component 20 on the front side. In the fully extended position, the joint angle α is 0°, and as flexion increases, the joint angle α increases and corresponds to the pivoting angle. In addition, at least one sensor 50 is arranged on the prosthetic foot 60 of the lower component 20, which is also coupled to the control unit 40 and detects the load applied to the foot component 60 or the pivoting or position of the foot component 60 in space or relative to the lower component 20.

[0038] Figure 3A flowchart of the control is shown. Sensor values from an IMU serving as sensor 50 (such as acceleration values and / or the orientation in space of an IMU arranged on the upper part 10 and / or the lower part 20 of an orthopedic joint device, such as a prosthetic knee joint or an orthotic knee joint) are transmitted to a machine learning-based estimation method (MLSV). The MLSV can be trained not only for regression tasks to estimate output variables or probabilities but also for classification tasks. The sensor values from the IMU 50 are evaluated within the MLSV for possible knee joint angles α. The evaluation within the machine learning-based estimation method (MLSV) results in a calculated value or an estimated value K, which is processed in the control unit 40 instead of a direct knee joint angle sensor signal. The estimated value K serves as an input value or an input variable for a rule set stored in the control unit 40. In addition to the estimated value K, other sensor values from the sensor 55 that are not evaluated by the MLSV can also be provided to the control unit 40 with the rule set. The control unit 40 itself determines control instructions for the actuator based on the input values to increase or decrease the resistance, or to switch on or off the drive. Within the MLSV, the probability of the current situation of the user can also be calculated, or a classification value corresponding to a specific situation can be calculated. Thus, the estimated value K of the knee joint angle α is not directly measured by a knee joint angle sensor, but is preferably determined by the artificial intelligence of the MLSV based on one or more sensor values of the IMU arranged on the upper part 10 or the lower part 20. For this purpose, in the artificial intelligence, in an artificial neural network KNN or other MLSV, data in a database is used to train the artificial intelligence or the MLSV. In addition to the estimated value K of the knee joint angle α, other variables or input values of the control unit 40, such as forces or torques that occur or are expected to occur during the movement process, can be calculated or estimated based on the sensor values from the IMU and possibly other sensors transmitted to the MLSV. Due to the evaluation of the sensor data within the MLSV, predictions or probabilities regarding the future course of variables or characteristic quantities can be made, because the MLSV is evaluated based on previous data and data courses. Using the data courses and probability calculations, the further course of the artificial joint variable data curve can be observed and predicted with corresponding probabilities. However, the greater the time span to be predicted or expected, the lower the probability of the prediction occurring. Therefore, the lead time for the MLSV to provide the control unit 40 with the expected joint angle value α or other expected variables is limited. In particular, this limit is between 0.001 seconds and 1 second, on the one hand to enable a perceivable effect of the preset, and on the other hand to avoid excessive changes in the settings of the device 30 that do not conform to the actual conditions at the joint. By using the MLSV and the estimated value K or the calculated value determined based on the course data measured on other orthopedic devices, future movement processes and loads can be estimated and used to improve the control.The MLSV can also be supplied with additional sensor signals different from the IMU signals to improve the accuracy of predicting or calculating the variables to be determined.

[0039] Using this method, the control can operate without a direct joint angle sensor, and the rule set in the control unit 40 is provided only by data from one IMU or multiple IMUs.

[0040] Figure 4 The result of estimating or calculating the knee joint angle α using the MLSV (here an artificial neural network) is shown. The curve αP (dashed line) predicted by the MLSV is compared with the directly measured knee joint angle αR (solid line). The two curves show basically the same course. For the knee joint angle αP calculated only based on the IMU raw data (i.e., acceleration and gyroscope values), there is a deviation from the actual value αR directly measured using a knee joint angle sensor between the maximum values of small knee joint angles α. As the gait course progresses, the two curves αP and αR get closer and closer. In addition, by incorporating additional sensor signals (such as spatial orientation), the result can be further improved. Especially when determining significant events (such as the movement reversal of the calf at the end of the swing phase in an orthotic or prosthetic knee joint), the accuracy determined by the MLSV is sufficient. In this case, the indirect determination without a direct knee joint angle sensor is advantageous because additional sensors can be omitted.

[0041] Figure 5 A schematic diagram of an orthosis is shown, and its basic structure corresponds to that of the Figure 1 prosthesis. The connection of the artificial joint to the leg is achieved through the connecting devices 101, 201 in this case. Figure 1 The implementation form is different from that of the Figure 5 implementation form in that, according to Figure 5 , an additional driver 70 is provided, where the electric motor 70 (possibly through a gearbox) is coupled to a pulley. Then, according to the rotation direction of the motor 70, flexion or extension of the knee joint can be initiated or assisted through a wedge belt or a toothed belt. An implementation form with an electric motor driver 70 through a mechanical power transmission device and a parallel damping through a hydraulic damper 30 can also be applied to a prosthetic knee joint. As described for Figure 1 , in the orthosis, the resistance device can also be constituted by a motor (for example, in the generator mode).

[0042] The direct mechanical coupling of the electric motor as the resistance device 30 to the upper part 10 and the lower part 20 can be achieved by a power transmission device, for example by a lead screw drive, so that instead of the piston rod 33, the lead screw is retracted or extended from the housing 31 by the rotation of a lead screw nut driven by the electric motor. In another embodiment, the electric motor as the resistance device is coupled to the upper part 10 and the lower part 20 by a gearbox device (such as a planetary gearbox) to cause or brake and affect the displacement of the upper part 10 relative to the lower part 20.

[0043] Furthermore, a control unit 40 and at least one IMU as a sensor 50 are arranged on the prosthetic or orthotic device. The angle between the upper part 10 and the lower part 20 can be determined by evaluating the sensor data of two spatial orientation sensors or IMUs 50. All sensors arranged on the prosthetic or orthotic device are coupled to the control unit 40, and their sensor values are used as the basis for controlling the actuator 34 of the resistance device 30 (when configured as a damper), or as an input signal for the motor control of the electric motor 70 (when the resistance device 30 is configured as a motor). In the case of magnetorheological damping, the sensor values are used to control the magnetic field or its change. Based on the sensor data, in particular the spatial orientation and / or angular position and position data, load, direction, acceleration and / or deformation data of other components, the actuator 34 is driven or the electric motor 70 is activated, deactivated or modulated, for example to reduce or increase the pivot resistance, limit the end stops and / or generate or assist the relative movement between the upper part 10 and the lower part 20.

[0044] Figure 6 The parametric regression results plotted on the Y-axis are shown exemplarily in solid lines, and the input values are plotted on the X-axis. The areas ±σ and ±2σ plotted around the solid line represent the respective confidence intervals. The input data provided in the form of training data are plotted as points on the solid line. Within the training data range, the estimation quality is better and the confidence interval is smaller, indicating that the quality of the calculated data improves with the increase in the amount of training data.

[0045] Figure 7 A prosthetic ankle joint is schematically shown, where the upper part 10 is the shank tube and the lower part 20 is the prosthetic foot. Sensors 50 are arranged on both the lower part 20 and the upper part 10, and these sensors are coupled to the control unit 40 to correspondingly affect the actuator 30 or the resistance device through the control unit. For example, the resistance of dorsiflexion and / or plantarflexion is set according to the estimated values of the kinetic or kinematic parameters. The estimated values are supplied to the rule set and can also relate to the ground slope, for example.

Claims

1. A method for controlling the movement behavior of an artificial joint, the artificial joint having an upper part (10) and a lower part (20) pivotally mounted on the upper part about a pivot (15), and means (30) for influencing the pivoting ability or pivoting of the upper part (10) relative to the lower part (20) being arranged between the upper part and the lower part, the means being coupled to a control device (40) in which a rule set is stored and the control device activates, deactivates or modulates the means (30) based on input values for the rule set to influence pivoting or pivoting ability, characterized in that, The sensor values detected during use of the artificial joint of at least one sensor (50) arranged on the upper component (10) or the lower component (20) are provided to at least one machine learning-based estimation method (MLSV), which continuously calculates an estimated value (K) for a kinetic or kinematic parameter (α) or for an expected kinetic or kinematic parameter (α) based on the sensor values and provides the estimated value (K) as an input value to the rule set and uses it as a criterion for activation, deactivation or modulation in the rule set.

2. The method according to claim 1, characterized in that, The machine learning-based estimation method (MLSV) continuously determines multiple estimated values (K) for multiple kinematic or kinetic parameters (α).

3. The method according to claim 1 or 2, characterized in that, The sensor values are determined by at least one IMU.

4. The method according to any one of the preceding claims, characterized in that, The kinetic or kinematic parameter (α) includes or represents the joint angle between the upper component (10) and the lower component (20).

5. The method according to claim 4, characterized in that, The joint angle input value can be provided without directly detecting the joint angle (α) through an angle sensor.

6. The method according to any one of the preceding claims, characterized in that, Additional sensor data and / or state variables are provided as input values to the rule set and used as criteria for activation, deactivation or modulation.

7. The method according to claim 6, characterized in that Safety-critical activation, deactivation or modulation is not performed solely based on the results of the machine learning-based estimation method (MLSV).

8. The method according to any one of the preceding claims, characterized in that, The machine learning-based estimation method (MLSV) is supplied and trained with pre-acquired sensor data from a database.

9. The method according to claim 8, wherein The machine learning-based estimation method (MLSV) is performed with a defined data set.

10. The method according to any one of the preceding claims, characterized in that The machine learning-based estimation method (MLSV) is updated during operation with current data obtained during that operation.

11. The method according to any one of the preceding claims, characterized in that, The machine learning-based estimation method (MLSV) includes an artificial neural network (KNN).

12. The method according to any one of the preceding claims, characterized in that The machine learning-based estimation method (MLSV) is based on a regression method or a parametric black box model.

13. The method according to any one of the preceding claims, characterized in that, The expected joint angle (α) is provided to the rule set with a lead time between 0.001 seconds and 1 second.

14. The method according to any one of the preceding claims, characterized in that, In the machine learning-based estimation method (MLSV), the probability of the motion situation or joint state is calculated based on the sensor values and provided as an input value to the rule set.

15. The method according to any one of the preceding claims, characterized in that The sensor values from previous time steps are also provided to the machine learning-based estimation method (MLSV).

16. The method according to any one of the preceding claims, characterized in that, The machine learning-based estimation method (MLSV) is based on a Gaussian process and provides information about the confidence interval (Kla) of the corresponding estimated value (α).

17. The method according to claim 16, wherein In addition to the determined estimated value (α), the relevant confidence interval (Kla) is also provided to the rule set and used as a criterion for activation, deactivation or modulation.

18. The method according to claim 17, wherein Multiple estimated values (α) together with the relevant confidence intervals (Kla) are transmitted to the rule set.

19. The method according to any one of the preceding claims, characterized in that, Multiple estimated values (α) for different kinematic or kinetic parameters (P) are provided as input parameters to the rule set.

Images