System and method for compensating characteristics of an actuator to improve torque control and performance

Abstract

Various aspects of the present disclosure provide systems and methods for providing torque control. The systems and methods comprise determining, based at least in part on a difference between an input encoder position data and an output encoder position data, an actuator as entering a backlash zone, and while the actuator is within the backlash zone, generating an adjusted position of the output encoder based at least in part on an output from a backlash estimation unit until the actuator is determined to exit the backlash zone.

Description

SYSTEM AND METHOD FOR COMPENSATING CHARACTERISTICS OF ANACTUATOR TO IMPROVE TORQUE CONTROL AND PERFORMANCECROSS-REFERENCE

[0001] This application claims the benefit of U.S. Provisional Application No. US 63 / 669,606, filed July 10, 2024, which is incorporated herein by reference in its entirety.BACKGROUND

[0002] It is known in the art to use actuators in robotic systems. One conventional type of actuator is a cycloidal actuator. Cycloidal actuators, also known as cycloidal drives, are mechanical devices that may be used to convert input rotational motion into different output rotational motion, often with a reduction in speed and a corresponding increase in torque. The core components of a cycloidal actuator include an input shaft, a high-speed cycloidal disk (or cam), needle bearings, and an output shaft. The input shaft is connected to an eccentric cam or cycloidal disk that rotates within a fixed actuator housing. As the cycloidal disk rotates, the lobes of the disk engage with needle bearings that are mounted in a stationary ring gear, creating a rolling motion. The output shaft is connected to a set of pins that engage with holes on the cycloidal disk. As the disk rotates, the pins follow the motion of the holes, causing the output shaft to rotate at a reduced speed compared to the input shaft. The reduction ratio is determined by the number of lobes on the cycloidal disk and the number of pins or rollers in the ring gear. Cycloidal actuators may be compact in size, provide a reasonable torque output, and may handle shock loads. Thea actuators are commonly used in applications such as robotics, industrial machinery, and automation systems, where precise motion control and high torque are important.

[0003] Notwithstanding the above, conventional cycloidal actuators have several drawbacks, particularly concerning torque control. The conventional cycloidal actuators have an unacceptable level of backlash, or the amount of play between components of the actuator, when the direction of motion is reversed, and may be an issue in these types of actuators. The excessive backlash may lead to inaccuracies in positioning and control, making it challenging to achieve precise torque control. Further, conventional cycloidal actuators experience torque ripple, which is the periodic variation in torque output, and is a common issue in cycloidal drives. The torque ripple may cause vibrations and noise in the actuator and the coupled device, adversely affecting the performance and reliability of the actuator, especially in applications requiring smooth and consistent torque output. Still further, achieving precise torque control in conventional cycloidal actuators may be challenging due to the non-linear relationship betweenthe input and output motion. The inherent design of the actuator may result in variable torque output, making it difficult to maintain consistent and accurate torque control, especially in dynamic or rapidly changing load conditions.SUMMARY

[0004] The present disclosure provides an actuation system that employs one or more compensation methods that are designed to enhance the open-loop torque control performance of actuators, such as a cycloidal actuator.

[0005] In an aspect, the present disclosure provides methods and systems for providing torque control. The torque control comprises determining, based at least in part on a difference between an input encoder position data and an output encoder position data, an actuator as entering a backlash zone. The actuator is within the backlash zone, generating an adjusted position of the output encoder based at least in part on an output from a backlash estimation unit until the actuator is determined to exit the backlash zone.

[0006] In some embodiments, the actuator is determined to exit the backlash zone based at least in part a difference between the adjusted position of the output encoder and the input encoder position. In some embodiments, the actuator is determined to exit the backlash zone when the difference is close to zero. In some embodiments, the output from the backlash estimation unit comprises one or more of a backlash amplitude or a backlash center. In some embodiments, the adjusted position of the output encoder is generated by subtracting the backlash amplitude from the output encoder position data as a preload adjustment. In some embodiments, the adjusted position of the output encoder is generated by subtracting the backlash center from the output encoder position data as a center adjustment. In some embodiments, the output is generated based on data stored in a lookup table. In some embodiments, an entry in the lookup table comprises a backlash center and a corresponding difference between an input encoder position data and an output encoder position data. In some embodiments, an entry in the lookup table comprises a backlash amplitude and a corresponding difference between an input encoder position data and an output encoder position data. In some embodiments, the data are collected while rotating a motor of the actuator in full range in both counterclockwise and clockwise directions. In some embodiments, the output from the backlash estimation unit is generated based at least in part on the difference between the input encoder position data and the output encoder position data. In some embodiments, the torque control further comprises upon determining a target torque command is close to zero, controlling a movement of a motor of the actuator through the backlash zone to drive the motor to a backlash center position. In some embodiments, upon determining the actuator exiting the backlash zone,applying a preload adjustment to the output encoder position data. In some embodiments, the preload adjustment is generated using a smooth transition function. In some embodiments, the preload adjustment is generated by taking the target torque and a backlash amplitude from the backlash estimation unit as input. In some embodiments, the torque control comprises an openloop control without feedback sensor data.

[0007] In an aspect, the present disclosure provides methods and systems for providing torque control. The torque control comprises a memory storing computer-executable instructions, one or more processors configured to execute the computer-executable instructions to perform.

[0008] In an aspect, the present disclosure provides methods and systems for providing open-loop torque control. The open-loop torque control comprises receiving a torque control command, adjusting the torque control command based at least in part on a hysteresis compensation, wherein the hysteresis compensation is generated by applying a torque processing model selected from a first model corresponding to torque ramping up and a second model corresponding to torque ramping down, and controlling an actuator using the adjusted torque control command in an open-loop torque control.

[0009] In some embodiments, one or more of the first model or the second model comprises a Dahl model. In some embodiments, the torque processing model is selected based at least in part on a hysteresis state. In some embodiments, the hysteresis state comprises a positive or negative torque ramping state. In some embodiments, the open-loop torque control further comprises adjusting the torque control command by applying a filtering algorithm to the torque control command when transitioning between the first model and the second model.

[0010] In an aspect, the present disclosure provides methods and systems for providing open-loop torque control. The open-loop torque control comprises receiving a target torque command, receiving an input encoder position data and an output encoder position data and generating a backlash compensation based at least in part on an estimated backlash, wherein the estimated backlash is generated based at least in part on the input encoder position data and the output encoder position data, generating a torque compensation based at least in part on a dynamic nonlinear current-to-torque relationship and the target torque command, generating a hysteresis compensation based at least in part on a torque processing model selected from a first model corresponding to torque ramping up and a second model corresponding to torque ramping down and the target torque command, and processing, by a controller, the target torque command, the backlash compensation, the torque compensation and the hysteresis compensation to generate an adjusted torque command for controlling a motor.

[0011] In some embodiments, the dynamic nonlinear current-to-torque relationship is established using a plurality of current measurements and torque measurements. In some embodiments, the dynamic nonlinear current-to-torque relationship is defined by a dynamic Torque Constant (Kt) estimated for the motor. In some embodiments, the Kt estimated for the motor is stored in a table. The open-loop torque control further comprises switching to a Transparent Backdriving Mode upon determining the target torque command is close to zero. In some embodiments, upon switching to the Transparent Backdriving Mode, driving the motor to a backlash center position. In some embodiments, the motor is controlled using an open-loop torque control. In some embodiments, the first model or the second model comprises a Dahl model. In some embodiments, the torque processing model is selected based at least in part on a hysteresis state. In some embodiments, the hysteresis state comprises a positive or negative torque ramping state.

[0012] In an aspect, the present disclosure provides methods and systems for providing open-loop torque control. The open-loop torque control comprises a memory storing computerexecutable instructions; one or more processors configured to execute the computer-executable instructions to perform.

[0013] In an aspect, the present disclosure provides methods and systems for adaptive control of an actuator system. The adaptive control of an actuator system comprises monitoring actuator state with encoders, temperature, and current sensors, predicting behavior using Model Predictive Control, adjusting control inputs in real-time for torque optimization, and adapting to changes via machine learning (e.g., neural networks).

[0014] In some embodiments, multi-sensor data drives dynamic parameter adjustments.

[0015] In an aspect, the present disclosure provides methods and systems for a fault detection system for actuators. The fault detection system for actuators comprises sensors monitoring temperature, vibration, and current, a multi-sensor fusion module for health profiling, machine learning for anomaly detection and failure prediction, and a feedback module for diagnostic alerts.

[0016] In some embodiments, predictive models analyze real-time sensor data.

[0017] In some cases, certain applications such as robotic applications may require improved torque control and / or other performance of actuators. The methods and systems herein may allow for actuators with improved precision, accuracy and / or performance metrics (e.g., torque / position accuracy, responsiveness, efficiency, etc.) that may be utilized in various applications including, but not limited to, robotic systems and non-robotic systems wherein improved torque control is desired. In some cases, the control algorithms and methods herein may be integrated or applied to any existing actuator system without requiring replacement ofthe entire actuator. For instance, controller may be applied to any existing actuator system using any transmission layout to provide or improve its torque control operation. The present disclosure describes compensation methods that expand the number and type of actuators suitable by characterizing these inefficiencies and embedding compensation approaches in an actuation control scheme for use with the robotic actuator or any other applications.

[0018] In an aspect, the present disclosure provides methods and systems for non-linear dynamics compensation. The non-linear dynamics compensation comprises estimating and compensating for non-linear actuator dynamics that affect a robotic actuator’s torque application on a load. This involves determining and correcting any non-linear behavior of the actuator to ensure more accurate torque application.

[0019] In an aspect, the present disclosure provides methods and systems for accurate torque estimation. The accurate torque estimation comprises estimating the resulting torque applied from an actuator on a load. This is achieved through advanced algorithmic techniques that account for various inefficiencies and inaccuracies inherent in traditional torque estimation methods.

[0020] In an aspect, the present disclosure provides methods and systems for backlash control. The backlash control comprises controlling an actuator while operating within backlash. This feature ensures that the actuator may operate smoothly and accurately even when mechanical play (backlash) is present in the system, preventing instability and improving overall control performance.

[0021] In an aspect, the present disclosure provides methods and systems for hysteresis compensation. The hysteresis compensation comprises compensating for hysteresis losses by creating separate models for different hysteresis states (positive and negative torque ramping) and implementing a filtering technique to transition between the models.

[0022] In an aspect, the present disclosure provides methods and systems for dynamic torque constant (Kt) estimation. The dynamic torque constant (Kt) estimation comprises dynamically estimating a torque constant (Kt) for each actuator, accounting for non-linearities and variations in the actuator’s performance, leading to more accurate torque estimation.

[0023] In an aspect, the present disclosure provides methods and systems for transparent backdriving mode. The transparent backdriving mode comprises an operational mode where the motor of the actuator is driven to the center of the backlash when the desired torque is near zero, minimizing the inertia the load feels from the actuator.

[0024] In an aspect, the present disclosure provides methods and systems for real-time adaptability. The real-time adaptability comprises continuously adjusting control signals in realtime to adapt to changes in operating conditions, ensuring consistent performance.

[0025] In an aspect, the present disclosure provides methods and systems for cost and complexity reduction. The cost and complexity reduction comprises avoiding expensive components such as torque cells by using sensorless torque control methods, reducing the overall cost and complexity of the actuator system.

[0026] In an aspect, the present disclosure provides methods and systems for improved torque bandwidth. The improved torque bandwidth comprises enhancements in torque control algorithms lead to improved torque bandwidth, allowing the actuator to respond more quickly and accurately to changes in load and command signals.

[0027] In an aspect, the present disclosure provides methods and systems for noise and vibration reduction. The noise and vibration reduction comprises compensating for unwanted characteristics such as friction and cogging, the system reduces noise and vibration, leading to smoother and quieter operation.

[0028] Additional aspects and advantages of the present disclosure will become readily apparent to those skilled in this art from the following detailed description, wherein only illustrative embodiments of the present disclosure are shown and described. As will be realized, the present disclosure is capable of other and different embodiments, and its several details are capable of modifications in various obvious respects, all without departing from the disclosure. Accordingly, the drawings and description are to be regarded as illustrative in nature, and not as restrictive.INCORPORATION BY REFERENCE

[0029] All publications, patents, and patent applications mentioned in this specification are herein incorporated by reference to the same extent as if each individual publication, patent, or patent application was specifically and individually indicated to be incorporated by reference. To the extent publications and patents or patent applications incorporated by reference contradict the disclosure contained in the specification, the specification is intended to supersede and / or take precedence over any such contradictory material.BRIEF DESCRIPTION OF THE DRAWINGS

[0030] The novel features of the invention are set forth with particularity in the appended claims. A better understanding of the features and advantages of the present invention will be obtained by reference to the following detailed description that sets forth illustrative embodiments, in which the principles of the invention are utilized, and the accompanying drawings (also “Figure” and “FIG.” herein), of which:

[0031] FIG. 1 illustrates an exemplary actuation system.

[0032] FIG. 2 illustrates an exemplary actuator.

[0033] FIG. 3 illustrates an exemplary backlash and pitch line.

[0034] FIGs. 4A and 4B illustrate an exemplary backlash quantification.

[0035] FIG. 5 illustrates an exemplary backlash compensation unit.

[0036] FIG. 6 illustrates an exemplary current and torque relationship.

[0037] FIG. 7 illustrates an exemplary hysteresis curve.

[0038] FIG. 8 illustrates an exemplary hysteresis compensation.

[0039] FIG. 9 illustrates an exemplary Dahl model.DETAILED DESCRIPTION

[0040] While various embodiments of the invention have been shown and described herein, it will be obvious to those skilled in the art that such embodiments are provided by way of example only. Numerous variations, changes, and substitutions may occur to those skilled in the art without departing from the invention. It should be understood that various alternatives to the embodiments of the invention described herein may be employed.

[0041] Whenever the term “at least,” “greater than,” or “greater than or equal to” precedes the first numerical value in a series of two or more numerical values, the term “at least,” “greater than” or “greater than or equal to” applies to each of the numerical values in that series of numerical values. For example, greater than or equal to 1, 2, or 3 is equivalent to greater than or equal to 1, greater than or equal to 2, or greater than or equal to 3.

[0042] Whenever the term “at most,” “up to,” “no more than,” “less than,” or “less than or equal to” precedes the first numerical value in a series of two or more numerical values, the term “no more than,” “less than,” or “less than or equal to” applies to each of the numerical values in that series of numerical values. For example, less than or equal to 3, 2, or 1 is equivalent to less than or equal to 3, less than or equal to 2, or less than or equal to 1.

[0043] The present disclosure provides compensation techniques to improve the applied torque accuracy and torque bandwidth of robotic actuators. The system and method of the present disclosure is shown for example in FIGS. 1-9. Those skilled in the art will appreciate that the present disclosure may be implemented in a number of different applications and embodiments and is not specifically limited in its application to the particular embodiment depicted herein.

[0044] As herein, the terms “open loop torque control,” “quasi direct drive,” “direct drive,” and “torque control” refer to any torque control methods used to estimate or control the torque applied to a load by an actuator, with or without a sensing element, whether the load is connected to the output or fixed and the effective load is the actuator itself.

[0045] In some embodiments, "target torque" may be defined as the specific torque value that an actuator aims to achieve during operation. The term “target torque” may also be referred to as "desired torque" which are utilized interchangeably throughout the specification.

[0046] As used herein, the term “actuator” refers to a mechanical device that converts energy into motion and is responsible for moving or controlling a mechanism or system. The actuators may operate using various types of energy sources, including electric, hydraulic, pneumatic, thermal energy, and cycloidal actuators. The electric actuator uses electrical energy to produce motion. The electric actuators may be either linear or rotary, depending on the type of motion required. The hydraulic actuators utilize fluid pressure to generate motion and are typically used in heavy machinery due to their ability to produce high force and precise control. Examples include hydraulic cylinders and hydraulic motors. The pneumatic actuators use compressed air to create motion. Examples include pneumatic cylinders and air-driven motors. The thermal actuators rely on temperature changes to produce motion. The cycloidal actuators, also known as cycloidal drives, convert rotational motion into a different rotational motion with a reduction in speed and an increase in torque.

[0047] Unlike non-robotic systems where motors are designed for general motion or positioning tasks (e.g., valves, conveyors, machine tools) which optimized for constant speed or constant force applications, in many robotic applications, actuators (e.g., robotic actuators) may be used to have high precision, low latency, and high bandwidth (for response) and may have improved torque control such as a low-speed, high-torque output with precise modulation (e.g., for joint control). For instance, in robotic applications, a system for controlling or measuring the torque applied to the load may be used to achieve a specific torque, which is referred to as torque control, and requires additional sensors, such as torque cells. Sensorless Torque Control, also referred to as Open-loop torque control, direct drive actuation, or quasi-direct drive actuation, may estimate the measured and applied torque using characteristics of the motor and the measured current applied using a motor controller without the use of additional sensors, such as torque cells.

[0048] Sensorless torque control may reduce cost, reduces actuator packaging, improves robustness, and typically improves torque bandwidth. The overall performance of these openloop torque-controlled actuators may be dependent on the relationship between the applied current to the motor and the resulting torque applied to the load. Friction, hysteresis, back- driving torque, and other phenomena of actuators may negatively affect this relationship, making many cycloidal actuators and associated systems unsuitable for open-loop torque control.

[0049] Conventional methods may be employed to address current-to-torque variability, applied torque hysteresis, and backlash in the cycloidal actuators. A conventional approach fortorque control actuation may employ a closed-loop torque control. This may use a method to measure the torque applied to the load via a strain gauge, associated deflection measurements, or other methods. Once a sensing method is applied, a feedback controller may be used to calculate an error term, which may be fed into a controller that applies current to reduce the error. The use of additional torque sensing elements may be cost-prohibitive, impose additional design constraints for overload protection, and affects the overall performance of the torque control system.

[0050] In order to address some of these drawbacks, a series elastic actuation technique may be employed as an alternative to the strain-based torque sensing technique, where a spring element with a known spring constant K in series with the actuator load path may be employed. The deflection of the K value may be used to estimate torque output, which may then be used as feedback for a torque controller. Due to the spring element in the load path, the deflection may result in lower torque bandwidth.

[0051] In some embodiments, the actuator may comprise a motor, a gearbox, an input / output encoder, a controller, a gearbox, a transmission unit, or any combination thereof. In some embodiments, the motor may be selected from a DC motor, an AC motor, or a stepper motor. In some embodiments, the motor may further include brushless motors or any combination thereof. In some embodiments, the gearbox may be configured to provide various gear ratios to optimize torque and speed. In some embodiments, the gearbox may include a spur gear, a planetary gear, or a worm gear, or any combination thereof. In some embodiments, the input / output encoder may be utilized to provide position feedback to the controller. In some embodiments, the encoder may be an incremental encoder, an absolute encoder, or a quadrature encoder, or any combination thereof. In some embodiments, the controller may be configured to process signals from the encoder and control the operation of the motor accordingly. In some embodiments, the controller may employ a microcontroller, a digital signal processor, or a field-programmable gate array, or any combination thereof. In some embodiments, the actuator may be configured to operate in a closed-loop control system. In some embodiments, the closed-loop system may enhance precision and responsiveness by continuously adjusting the motor's operation based on feedback from the encoder. In some embodiments, the actuator may be housed in a casing made from materials such as aluminum, steel, or plastic, or any combination thereof. In some embodiments, the actuator may further include additional components such as limit switches, thermal sensors, or communication interfaces, or any combination thereof. In some embodiments, the actuator configuration may be optimized for specific applications, including robotics, automation, or automotive systems.

[0052] Traditional actuators may suffer from backlash and hysteresis. These issues may lead to inaccuracies in torque estimation and control, impacting overall performance and typically limiting the maximum effective gear ratio. To address backlash, low-backlash gearboxes may be used, but these gearboxes typically increase friction and hysteresis, which affect the control of torque applied to the load of the actuator. To circumvent these issues, a torque sensing method may be used.

[0053] Existing approaches may attempt to minimize backlash through mechanical design improvements, such as tighter tolerances and preloading mechanisms, or specific transmission selection, such as a backlash free harmonic drive. While these methods may reduce backlash to a limited degree, they often add complexity and cost to the actuator design or add some negative effect on performance. Hysteresis is a phenomenon where the value of a physical property lags behind changes in the effect causing it, such as when the output of a motor lags behind the input due to friction, imperfect gear ratios, and structural deflection. Hysteresis compensation may involve complex analysis of the underlying cause. The conventional solutions may rely on extensive empirical data and are often error prone. Due to the complexity and inaccuracy of modeling, this problem may be solved with closed-loop torque feedback.

[0054] Further, the torque constant (Kt) is a parameter that determines the relationship between the current supplied to the motor and the resulting torque of the cycloidal actuator. Conventional techniques may use a constant Kt value, which does not account for variations in the actuator’s performance due to manufacturing tolerances, wear, operating conditions, or the material properties of the motor. This may lead to inaccurate torque estimations and application and suboptimal control of the actuator.

[0055] The disclosure pertains to an actuation system 10, as shown in FIG. 1, which comprises an actuator having a motor 20 and a transmission 25, an input encoder 40 on the motor side of the actuator, an output encoder 50 on the load or transmission side of the actuator, a load 30 that is coupled to the output of the actuator, a controller 60 that communicates with the actuator by sending control signals to the actuator and receiving state information from the actuator, and compensation units 80, 90, 100, 110 designed to enhance the performance of the actuator. The controller 60 may be operably coupled to the actuator. The controller 60 may receive a torque command signal from an external source, such as from the compensation units, and may use the compensation units and the state information of the actuator to apply a specific current to the motor 20 to achieve a desired torque output that may be applied on the load 30.

[0056] The backlash estimation unit 80 and the backlash compensation unit 90 may be employed to address backlash issues with the actuator. The torque compensation unit 110 may be employed to address torque estimation inaccuracy issues, and the hysteresis compensationunit 100 may be employed to address torque application phase delays related to hysteresis, and unwanted nonlinear characteristics, including, but not limited to, cogging, friction, and stiction. The compensation units help improve torque estimation and accuracy, actuator control and stability, and thus reducing vibration and noise in the actuator and load, prolong service life of the actuator, expand the type and number of actuators suitable for open loop torque control, and / or improve feed-forward torque calculation used in closed loop torque control applications.

[0057] The motor 20 converts electrical energy into mechanical torque. The transmission 25 transmits power from the motor to the load, often including gearboxes. The load 30 is the object or system to which the actuator applies force or torque. The input encoder 40 may be a sensor that measures the rotor position of the actuator with respect to a stator position, which is connected to a fixed portion of the actuator. The output encoder 50 may also be a sensor that measures the load position with respect to the fixed component of the transmission 25. The controller 60 may process and generate control signals that sends commands to the actuator and receives feedback and associated information from the actuator to adjust performance. The estimation and compensation units may be configured to correct or mitigate unwanted characteristics in the actuator.

[0058] An example of a suitable actuator is shown for example in FIG. 2. In some embodiments, the actuator may be a cycloidal actuator. The actuator includes the motor 20, the transmission, 25. and may be coupled to the load 30. The torque or force is generated using the motor 20, which is housed in a stationary enclosure. The present disclosure provides various algorithms and methods to determine a relationship between the torque or force generated by the motor 20 and the effective output felt or received by the load 30 after passing through the transmission 25. In some cases, the relationship may be determined taking into account dynamic characteristics with fine-grained, adaptive continuous estimation. The transmission 25, or gearbox, may comprise an input, an output, and a stationary or fixed component. Both the motor 20 and the transmission 25 typically have unwanted characteristics that impact the relationship between the torque or force generated by the motor and the effective torque output. The motor 20 may include the input encoder 40 that senses the rotor position with respect to the stator position, which is connected to a fixed portion of the actuator. The transmission 25 may include the output encoder 50 that senses the load position with respect to the fixed component of the transmission 25.

[0059] Typical robotic actuators may exhibit several unwanted characteristics that affect open-loop torque application bandwidth and accuracy, including friction losses, backlash, and cogging. The friction losses include torque loss from internal frictions in the mechanical components of the actuator, such as bearings, transmission components, and unexpected orunwanted rubbing, and the like. The hysteresis may be a nonlinear lag in torque response due to the state of the actuator and the physical or material properties of the actuator. The backlash is the play between the mechanical components of the actuator that leads to a disconnected load path from the motor to the actuator load. The cogging may refer to magnetic forces between the stator steel and rotor magnets that generate an unwanted torque on the rotor.

[0060] The actuator of the actuation system 10 generates an input movement or motion and the motor 20 generates torque based on the current supplied to the motor. The transmission 25 of the actuator transmits output torque from the motor 20 to the load 30. The output encoder 50 measures the load’s position or rotation. The controller 60 may receive a torque command from an outside source or from the torque compensation unit 110. The controller 60 receives the torque signal and the actuator’s current state from the actuator, and the torque compensation unit 110 compensates to adjust the motor’s current to achieve the desired output torque on the load. The backlash estimation unit 80 estimates the backlash in the actuator and the backlash compensation unit 90 activates when the actuator is within the backlash region, ensuring smooth transitions and accurate torque application. The hysteresis compensation unit 100 may employ a plurality of models to compensate for hysteresis losses, improving control accuracy.

[0061] The present disclosure offers advanced compensation techniques to enhance the performance of open-loop torque-controlled actuators by mitigating unwanted characteristics to ensure precise torque application and improved torque bandwidth, thereby advancing open-loop torque control performance. By addressing issues such as backlash, torque variability, and hysteresis, the system of the present disclosure improves torque estimation accuracy, control stability, and overall actuator efficiency.

[0062] As shown for example in FIG. 3, backlash refers to the movement or “play” in the load path between the motor 20 and the load 30, where the motor 20 and the load 30 are able to move independently from each other due to mechanical phenomenon. This is a typical characteristic of many robotic transmissions, such as planetary gearboxes, but may be caused by other phenomena, or mechanical components, such as bearings or machining tolerances within the actuator. This phenomenon negatively affects the precision and performance of the actuator. When the motor 20 is operating within the backlash region, the generated motor torque or force is no longer coupled to the load 30. The backlash then travels through, until the mechanical components of the actuator are once again engaged. This travel distance is a function of the actuator state and may be quantified as a backlash width defining the play travel distance at a specified position. The backlash amplitude is defined as half of the backlash width. The present disclosure provides a system and method to quantify the backlash and to operate within the backlash region. The actuator backlash may be quantified using a plurality of backlash datacollection techniques. According to one technique, as shown in FIG. 4A, the actuator is externally preloaded in a counterclockwise direction using a controlled torque, such that the actuator is against a boundary of the backlash. The motor is then driven under a slow and stiff velocity. The actuator is then moved through a full range of motion in each direction (e.g., positive and negative directions). The differences between the input and output encoder positions are then recorded at various points throughout the range of motion. Repeat the movement back and forth to ensure consistent readings. The actuator is then preloaded in a clockwise direction using a controlled torque, and the data is gathered as described in the counterclockwise direction. Again, the actuator is rotated through the full range of motion, and the input and output encoder positions are recorded. The difference between the input encoder position and the output encoder position are recorded during the counterclockwise and clockwise preloads at each point. These will be referred to as the counterclockwise and clockwise encoder difference signals. The backlash estimation unit 80 may determine the backlash center / setpoint and backlash amplitude by averaging the encoder difference signals to determine the center / setpoint of the backlash and this information may be store in the backlash table. The backlash estimation unit 80 may then calculate the amplitude of backlash, i.e., the difference between the setpoint and one of the difference signals, and store this information in the backlash table. The amplitude may be the same in both directions because it is defined from the center / setpoint signal. An example of the backlash table is shown in FIG. 4B. The illustrated backlash lookup table correspond to a full rotation of the actuator output and is stored in the backlash estimation unit 80. The backlash estimation unit 80 thus estimates and quantifies the amount of backlash by measuring the difference between the position of the input encoder 40 and the output encoder 50 under controlled preloading conditions. The estimated backlash information generated by the backlash estimation unit 80 may be stored in a suitable storage device, and may be arranged in any selected format, such as in a backlash table. The backlash estimation unit 80 may then generate an output signal that includes information associated with the amount of backlash in the actuator.

[0063] The backlash information is conveyed to the backlash compensation unit 90. As shown in FIG. 5, the backlash compensation unit 90 may include a preload adjustment unit 92 that adjusts the position of the output encoder 40 by subtracting the amplitude of the backlash at a particular point. The backlash compensation unit 90 also includes a backlash center adjustment unit 94 that adjusts the position of the output encoder position by subtracting the backlash center location received from the backlash estimation unit 80. The difference between the adjusted output encoder position and the input encoder position is determined. A position controller 96 tracks the delta signal to ensure the actuator exits the backlash zone in a controlled manner. Thisis accomplished with a mathematical fit expressed as a discrete transfer function. The backlash compensation unit 90 may determine whether the actuator is in backlash by using the actuator’s output encoder 40, the state of motion of the actuator, and the information in the backlash table. The backlash compensation unit 90 may determines if the actuator is entering or exiting the backlash region and compensates accordingly. Physically, the actuator is entering backlash if the mechanical components within the actuator disengage, and it remains in backlash until rotated enough to re-engage. Identifying this engagement is challenging, as the mechanical engagement is not directly observable. The backlash compensation unit 90 may track the output position, the applied torque sign, and the measured encoder distance traveled since the change in torque sign to determine when entering and exiting the backlash region. The backlash amplitude defines the width (e.g., two times the amplitude) of the backlash region at a particular location.

[0064] When the motor is within the backlash region, the first output encoder position is subtracted with the amplitude of the backlash at that point. This step is called the preload adjustment. For example, while rotating in a certain direction, the actuator is at one end of the backlash. A preload adjustment may adjust the motor / actuator position from the end of the backlash to the center. After the preload adjustment, the output encoder position signal is further subtracted with the backlash center location from the backlash estimation unit 80. This is called the backlash center adjustment unit 94. The difference between the output encoder - adjusted for preload direction and centered to account for backlash as outlined in the previous paragraphs - and the input encoder is calculated and referred to as delta. The delta signal is fed into the position controller unit 96. The position controller unit 96 tracks the delta signal, thereby exiting the backlash zone in a controlled manner. After the actuator crosses the backlash and re-engages, the delta signal settles close to at least about zero; the delta signal is generally near but not exactly zero due to the deflection in the output shafts. The delta signal may be at least about -0.1, 0, or +0.1. This method ensures a smooth and stable transition within the backlash region.

[0065] When the controller 60 receives a desired torque near zero, the motor 20 enters into a “transparent backdriving mode” in which the motor 20 is driven to the center of the backlash. This keeps the actuator in a “disconnected state” where a feedback controller in the backlash compensation unit is using the input and output encoders to operate in the backlash, lowering the inertia the load feels from the actuator. When transitioning across the backlash zone, the backlash compensation unit 90 uses a mathematical function, such as a sigmoid function, with a desired torque and the backlash amplitude from the backlash estimation unit 80 as inputs to help smoothen the output of the preload adjustment unit 92. This ensures that the backlash compensation unit 90 does not go unstable for sudden and small changes in torque direction. The backlash compensation unit 90 may generate a backlash compensation signal that is conveyed tothe controller 60. The controller in turn may generate a backlash control signal that is received by the motor 20 and may be used to compensate for the backlash. Specifically, when the actuator is operating within the backlash region, the backlash control signal generated by the controller 60 stabilizes the motor velocity and the control position. The controller 60 via the backlash compensation unit 90 may dynamically adjust the motor based on the backlash data generated by the backlash estimation unit 80 to ensure smooth and accurate movement of the motor 20 through backlash. The backlash compensation reduces instabilities in the actuator, improves back drivability of the actuator, and enhances motor control performance. In summary, the backlash estimation and compensation process provides for continuously monitoring the output encoder position to identify the backlash crossing when the torque sign changes, adjust the output encoder position by subtracting the amplitude of the backlash at that point to move from the end of the backlash to the center, further adjusting the output encoder position by subtracting the backlash center location from the backlash estimation unit 80, determining the difference between the adjusted output encoder position and the input encoder position, referred to as the delta signal, using the position controller to track the delta signal to ensure smooth and stable transition within the backlash region, and when the desired torque is near zero, drive the motor to the center of the backlash to minimize the inertia the load feels from the actuator. The desired torque may be at least about -0.1, 0, or +0.1. Further, the process includes employing a mathematical function, such as a sigmoid function, with desired torque and backlash amplitude to compute a desired position within the backlash during transitions.

[0066] The actuation system 10 of the present disclosure may also perform a dynamic torque processing technique via the torque compensation unit 110. The illustrated torque compensation unit 110 may employ a torque data collection technique by monitoring, tracking, and recording motor current and resulting actuator torque at various levels and use a mathematical approach to create a processing method that captures non-linearities in the current to torque relationship. The dynamic torque processor may estimate the output torque based on the measured current and the state of the actuator or calculates the needed motor current to achieve a desired torque using the model and the actuator state. This approach provides more accurate torque estimation than conventional methods. The actuator, and specifically the motor 20, may have a torque constant (Kt) associated therewith that is defined as the relationship between the current supplied to the motor and the resulting torque generated thereby. The torque constant Kt is typically supplied by manufacturers as a fixed value defining the torque generated per ampere of current at a measured temperature up to a specified current. The effective relationship between current and torque is not constant, due to the material properties of the motor, the friction of bearings, and the inertia of the system moved by the rotor, among others. The estimation of the torque applied to the loadfrom the motor given the measured current is a key component of open loop torque control actuation. The first quantifiable component is the relationship between the torque generated in the motor and the resulting torque at the motor’s output prior to the transmission input. This relationship is affected by various phenomena such as manufacturing tolerances, material or physical properties of the stator or rotor magnets, the saturation of the stator steel, the temperature of the stator, or the temperature of the magnets. The issues with the conventional torque constants include variability, non-linearity, and environmental factors. With regard to variability, the torque constant may vary due to differences in materials, assembly processes, and operational wear. The relationship between current and torque is typically not linear. Using a single constant value fails to capture this non-linearity, leading to errors in torque estimation. Temperature changes, load variations, and other environmental conditions may alter the torque characteristics of the motor, making the constant Kt value less reliable. To address these issues, a more dynamic approach to torque processing is used, one that may adapt to the varying conditions and non-linearities inherent in the system.

[0067] The torque compensation unit 110 may be configured to collect the measured motor current and the resulting actuator torque, at various currents across the minimum and maximum motor currents, and then store the information in a table. The torque compensation unit 110 may employ a processing technique to capture the non-linearity between the active current and the resulting torque into a mathematical algorithm that may be used in real time to accurately estimate the resulting torque. An example method to measure the current to torque relationship may include connecting a torque cell between the actuator output and a fixed ground, applying a known current to the motor and use the torque cell to measure the resulting torque, repeating the measurements at various current levels to gather sufficient data, comparing the measured current to the output torque to calculate the relationship, and using a processing technique to generate a dynamic relationship. The processor accounts for non-linearities and variations in the actuator’s performance. The torque compensation unit 110 may then estimate the output torque of the actuator or use it to calculate the needed current to achieve the desired torque. FIG. 6 illustrates the conventional approach of using a single torque constant Kt to model the relationship between current and torque. There is noticeable nonlinearity in the residual error (center plot) vs the modeled torque constant Kt by the torque compensation unit 110 instead. The nonlinearity from the linear Kt case is essentially removed, and the magnitude of the error is reduced.

[0068] The actuation system 10 of the present disclosure may also compensate for the hysteresis associated with the actuator via the hysteresis compensation unit 100. The hysteresis compensation unit 100 may gather data to identify hysteresis losses affecting torque output, and record discrepancies between desired and measured torque. The hysteresis compensation unit100 establishes separate torque processing models for different hysteresis states (positive and negative torque ramping) and uses a filtering technique to transition between these models. The hysteresis compensation unit 100 may perform the compensation by continuously updating the motor control commands to account for hysteresis, ensuring accurate torque estimation and reducing lag. This method improves precision, control stability, and efficiency by effectively compensating for hysteresis losses.

[0069] Hysteresis in torque-controlled actuators is a phenomenon where there is a lag or difference in the torque output when the input torque changes direction. This lag is caused by various factors such as internal friction, material properties, mechanical backlash, and magnetic hysteresis. Torque hysteresis negatively affects the performance of actuators because it introduces inaccuracies in torque control, making it difficult to achieve precise and responsive movements. This may lead to inefficiencies and reduced performance in applications requiring high precision, such as robotics. Hysteresis varies depending on the actuator’s characteristics and operating conditions. An example hysteresis curve is shown in FIG. 7. Toque hysteresis is problematic since it reduces the precision of torque control, making it challenging to achieve exact positioning and movement. The lag in response may cause instability in control systems, leading to oscillations or unsteady behavior. Inefficiencies may arise as the system constantly compensates for the delayed response, wasting energy and reducing overall performance. Continuous compensation for hysteresis may accelerate wear and tear on mechanical components, shortening the actuator’s lifespan. The hysteresis compensation unit 100 may help reduce one or more of these effects and may help further reduce this lag. The hysteresis compensation unit 100 may calculate the data by applying a sinusoidal torque input to an actuator. A torque cell is then used to measure the torque at the output. It may be observed that the error between the desired and measured torque varies when the direction of torque changes. There is a asymmetry in the output torque depending on the sign of the rate of change of torque (ramping the torque up vs down).

[0070] In some cases, the problem of torque hysteresis is solved using a Dahl model. The Dahl model is a mathematical approach used to describe and compensate for friction-induced hysteresis in mechanical systems. In the context of friction compensation. It provides a way to model and predict the hysteresis behavior, allowing for better control and compensation in openloop torque tracking. An example of the hysteresis compensation working is shown in FIG. 8, and an example of the Dahl model is shown in FIG. 9. The Dahl model is a phenomenological model that describes the nonlinear frictional force in mechanical systems, particularly those exhibiting hysteresis behavior due to internal friction, backlash, or material compliance. The model introduces an internal state variable z(t) that represents the displacement of the internalfriction element, rather than just the position of the actuator. This state is governed by the rate of change of displacement and reflects the history-dependent friction force.

[0071] The hysteresis losses may be identified by conducting testing to identify hysteresis losses affecting the torque output, recording the torque output at various positions and currents, noting any discrepancies due to hysteresis, establish two separate torque processing models (one for each side of the hysteresis), and the torque processing models reflect the different torque outputs depending on the direction of torque. The hysteresis compensation unit 100 may implement a filtering algorithm to switch between the two torque processing models based on the current operating state. Smoothly transition between the torque processing models to avoid abrupt changes and ensure accurate torque estimation. The hysteresis compensation unit 100 applies the filtered torque processing models to adjust the motor control commands. Continuously update the compensation during operation to account for changing conditions. By implementing these compensation algorithms, the disclosure enhances the performance of quasidirect drive actuators, providing more accurate torque estimation and improved control stability.

[0072] The actuation system 10 of the present disclosure integrates backlash processing, dynamic torque processing, and hysteresis processing into a single cohesive system 10. This comprehensive approach enhances overall actuator performance, providing accurate torque estimation, improved control stability, and reduced vibration and noise.

[0073] The actuation system may comprise a plurality of compensation units to improve open-loop torque control performance of robotic actuators. The actuation system may include the backlash estimation unit 80 and the backlash compensation unit 90 that employ a method to measure actuator backlash by quantifying the individual backlash width at each output position using a preloading method. It also describes a processing and control technique to operate within the backlash to avoid instabilities, improve backdrivability, and enhance control performance, turning backlash into a desirable feature for low backdrivability. Conventional methods for handling backlash often use mechanical tolerances or interference fits that reduce performance or add a cost. Oversized or interference-fit parts may eliminate backlash but increase internal friction as parts rub against each other. High-precision parts may also reduce backlash, but the complexity of machining will increase cost. Software approaches typically assume a fixed backlash width and ignore joint velocities when operating within the backlash using heuristics.

[0074] The system of the present disclosure also employs the torque compensation unit 110 that employs a processing algorithm to dynamically estimate the torque constant (Kt) for each actuator. This approach accounts for non-linearities and variations in the actuator’s performance, providing a more accurate torque estimation than the conventional use of a constant Kt value. Conventional systems use a constant Kt value, which does not account for variations due tomanufacturing tolerances, wear, material properties, and operating conditions, leading to less accurate torque estimations. The torque processing method allows for non-linearities to the Kt value, improving torque estimation accuracy and torque application.

[0075] The system also employs a hysteresis compensation unit 100 that compensates for hysteresis losses by creating two separate Kt processors for different hysteresis states (positive torque ramping and negative torque ramping) and implementing a filtering algorithm to transition between these models. Typically, external torque references are used, such as load cells or torque cells, which are used in feedback loops to mitigate any non-linearities or inefficiencies in actuator output torque. This approach is expensive and adds additional complexity and cost to the actuator. The dual processing approach with a filtering algorithm provides an effective solution to estimating the output torque, allowing for accurate torque estimation and compensation in real-time without the complexity of external load sensing.

[0076] In one aspect, the present disclosure relates to an adaptive control algorithm designed for actuator systems. The systems and methods herein may optimize torque control under varying operational conditions. In some embodiments, conventional actuator control systems may rely on fixed parameters, which may not adapt well to changing conditions. In some embodiments, this lack of adaptability can result in suboptimal performance, particularly in applications experiencing varying loads, temperatures, or wear over time. In some embodiments, the present disclosure provides an adaptive control algorithm that may dynamically adjust control parameters in real-time using sensor feedback. In some embodiments, the algorithm may optimize torque control by leveraging a unique integration of Model Predictive Control (MPC) with machine learning techniques, such as neural networks or reinforcement learning.

[0077] In some embodiments, the system may create a self-tuning mechanism that adapts to the actuator's evolving behavior. This system continuously adapts to an actuator’s real -world behavior, accounting for time-varying conditions such as temperature drift, load changes, and mechanical wear to ensure high-precision torque control. In some embodiments, the system may monitor the actuator state via multiple sensors. In some embodiments, the sensors may include encoders for position and velocity measurement, temperature sensors for assessing thermal conditions, and current sensors for load monitoring, or any combination thereof.

[0078] In some embodiments, the MPC may be employed to predict future actuator behavior based on current and historical data. In some embodiments, the MPC may adjust control inputs, such as motor current, to achieve optimal torque output. In some embodiments, the machine learning component may perform several functions. In some embodiments, the machine learning may learn actuator-specific dynamics over time, enabling the system to understand the unique behavior of the actuator. In some embodiments, the machine learning may adapt to changes suchas wear, temperature shifts, or load variations, or any combination thereof. In some embodiments, the machine learning may continuously refine control precision, enhancing the system's overall performance. Unlike traditional fixed-parameter methods, the adaptive control algorithm may ensure consistent torque performance in dynamic settings. In some embodiments, the system may automatically adjust control parameters without requiring manual intervention, thereby reducing the need for user input and facilitating seamless operation.

[0079] In one aspect, the present disclosure uniquely combines MPC and machine learning for real-time torque optimization. In some embodiments, the self-tuning capability of the system may leverage multi-sensor data to enhance adaptability, ensuring reliable torque control under varying operational conditions. In some embodiments, the adaptive control algorithm may enhance torque accuracy and stability. In some embodiments, the adaptive control algorithm may eliminate a need for manual recalibration. In some embodiments, the adaptive control algorithm may provide reliable performance in robotics / automation.

[0080] A method for adaptive control of an actuator system, comprising: monitoring actuator state with a plurality of sensors comprising an encoder, a temperature sensor, and a current sensor; predicting a future behavior of the actuator using Model Predictive Control (MPC); adjusting a control input in real-time for torque optimization based at least in part on the predicted future behavior; and refining the control input using a machine learning model based at least in part on the monitored actuator state. As described above, a Model Predictive Control (MPC) framework is employed to anticipate the actuator’s future response and optimize the control signal accordingly. The Model Predictive Control (MPC) framework may use a discretetime dynamic model of the actuator (state-space, grey -box, or learned) to simulate future actuator behavior over a fixed time horizon HHH, such as 50-200 ms in duration. Based on the predicted future behavior of the actuator, the MPC computes the optimal control input (e.g., current setpoint) for the next time step. This value is immediately issued to the motor controller. The system updates at a high control frequency (e.g., 1-5 kHz) to ensure rapid responsiveness and smooth motion. In some cases, to improve performance over time and adapt to nonlinear or time-varying system dynamics, the system includes a machine learning-based refinement layer that modifies or augments the control signal computed by the MPC. The machine learning model (e.g., neural network, Gaussian process, or reinforcement learning agent) is trained on historical and real-time sensor data to capture dynamics not well-modeled by the MPC (e.g., friction hysteresis, load-dependent behavior, wearing / aging effects). The inputs to the machine learning model may include the sensor data (e.g., position, velocity from the encoder, current, temperature reading) and / or prior control inputs and errors). The output of the ML model may be a delta value or scaling factor applied to the MPC command to minimize torque error andimprove tracking.

[0081] In one aspect, the present disclosure relates to a fault detection and diagnostic system embedded in actuator systems. The systems and methods herein may utilize real-time monitoring to enhance reliability and safety by detecting anomalies and predicting potential failures. In some embodiments, traditional actuator systems may lack integrated fault detection mechanisms, which may lead to unexpected failures and costly downtime. In some embodiments, the ability to continuously monitor key parameters is essential for maintaining system health and ensuring optimal performance. In one aspect, the present disclosure provides a fault detection and diagnostic system that may monitor key parameters such as temperature, vibration, and current draw in real-time. In some embodiments, the system may leverage multi-sensor fusion and machine learning techniques, including anomaly detection and predictive models, for continuous health assessment without the need for external tools.

[0082] In some embodiments, the system may integrate various sensors to track critical parameters. In some embodiments, the sensors may include temperature sensors to monitor overheating risks, vibration sensors to assess mechanical wear, and current draw sensors to identify electrical issues, or any combination thereof. In some embodiments, the multi-sensor fusion may build a unified actuator health profile by combining data from the various sensors. In some embodiments, this comprehensive profile may provide a holistic view of the actuator's operational status, enabling more accurate diagnostics. In some embodiments, the machine learning component may process the sensor data to perform several functions. In some embodiments, the machine learning may identify anomalies, such as vibration spikes indicative of bearing wear. In some embodiments, the machine learning may predict failures through timeseries analysis, allowing for proactive measures to be taken before issues arise. In some embodiments, the system may deliver actionable diagnostics, including maintenance alerts that inform operators of necessary interventions.

[0083] In some embodiments, the fault detection and diagnostic system may operate continuously, enabling preemptive maintenance strategies. In some embodiments, this continuous operation may enhance system reliability by addressing potential issues before they escalate into significant failures. In some embodiments, the present disclosure provides a fully integrated diagnostics solution within the actuator system itself. In some embodiments, the combination of multi-sensor fusion with machine learning may offer predictive insights that traditional systems may lack. In some embodiments, the early detection of issues may boost the reliability of actuator systems, thereby reducing the likelihood of unexpected failures. In some embodiments, predictive maintenance facilitated by the system may cut operational costs by optimizing maintenance schedules and resource allocation. In some embodiments, the ability toprevent failures may enhance safety in applications where actuator systems are employed.

[0084] The multi-sensor fusion may comprise processing sensor data streams by a sensor fusion module, which combines the heterogeneous signals into a coherent actuator health vector. The sensor fusion module may employ various sensor fusion methods such as Kalman fdtering or Bayesian inference to correlate signals with uncertainty management, Principal Component Analysis (PCA) or autoencoders to reduce dimensionality and detect inter-sensor correlations, or signal synchronization to handle timing mismatches between sensors. This beneficially provides a unified representation of the actuator's internal state, which enhances fault visibility beyond what any single sensor may provide.

[0085] The fused sensor data may then be processed by a ML model-based Anomaly Detection Algorithms (e.g., Isolation Forest, One-Class SVM, LSTM Autoencoders). The ML model may be trained on normal operational baselines to detect out-of-distribution behavior such as unexpected vibration frequencies, abnormal temperature rise under nominal load, and / or increased current under no-load conditions. Once an anomaly is detected, the system classifies the fault based on feature patterns (e.g., spike in Z-axis vibration + current surge = bearing wear), allowing for targeted diagnostics.

[0086] In one aspect, the present disclosure introduces a set of advanced compensation units configured to enhance the performance of open-loop torque-controlled actuators. By addressing critical issues such as backlash, torque variability, and hysteresis losses, the disclosure improves torque estimation and control stability. The methods may comprise creation of a backlash processor, dynamic torque processing, and effective hysteresis processing with dual torque processing models and a filtering algorithm. These integrated techniques provide a comprehensive solution, enabling more accurate torque estimation and smoother, more stable actuator operation.

[0087] In one aspect, the present disclosure may offer substantial advantages, including enhanced control stability and precision and real-time adaptability. This disclosure provides a robust solution for compensating unwanted characteristics in actuators, ensuring precise torque application and improved torque bandwidth, thereby advancing open-loop torque control performance. Because of the modular and comprehensive nature of the compensation units making up this controller, it is very versatile. The compensation units of this controller tackle actuator imperfections on an isolated level. The compensators found within this controller fix imperfections found in most existing actuators. Thus, this controller could potentially be applied to any existing actuator system using any transmission layout to enable or improve its torque control operation.

[0088] While preferred embodiments of the present invention have been shown and described herein, it will be obvious to those skilled in the art that such embodiments are provided by way of example only. It is not intended that the invention be limited by the specific examples provided within the specification. While the invention has been described with reference to the aforementioned specification, the descriptions and illustrations of the embodiments herein are not meant to be construed in a limiting sense. Numerous variations, changes, and substitutions will now occur to those skilled in the art without departing from the invention. Furthermore, it shall be understood that all aspects of the invention are not limited to the specific depictions, configurations or relative proportions set forth herein which depend upon a variety of conditions and variables. It should be understood that various alternatives to the embodiments of the invention described herein may be employed in practicing the invention. It is therefore contemplated that the invention shall also cover any such alternatives, modifications, variations or equivalents. It is intended that the following claims define the scope of the invention and that methods and structures within the scope of these claims and their equivalents be covered thereby.

Claims

CLAIMSWHAT IS CLAIMED IS:

1. A method for providing torque control, the method comprising: determining, based at least in part on a difference between an input encoder position data and an output encoder position data, an actuator as entering a backlash zone; and while the actuator is within the backlash zone, generating an adjusted position of the output encoder based at least in part on an output from a backlash estimation unit until the actuator is determined to exit the backlash zone.

2. The method of claim 1, wherein the actuator is determined to exit the backlash zone based at least in part a difference between the adjusted position of the output encoder and the input encoder position.

3. The method of claim 2 or any of the preceding claims, wherein the actuator is determined to exit the backlash zone when the difference is close to zero.

4. The method of claim 1, wherein the output from the backlash estimation unit comprises one or more of a backlash amplitude or a backlash center.

5. The method of claims 1 or 4, wherein the adjusted position of the output encoder is generated by subtracting the backlash amplitude from the output encoder position data as a preload adjustment.

6. The method of claims 1 or 4, wherein the adjusted position of the output encoder is generated by subtracting the backlash center from the output encoder position data as a center adjustment.

7. The method of claim 1, wherein the output is generated based on data stored in a lookup table.

8. The method of claims 1 or 7, wherein an entry in the lookup table comprises a backlash center and a corresponding difference between an input encoder position data and an output encoder position data.

9. The method of claims 1 or 7, wherein an entry in the lookup table comprises a backlash amplitude and a corresponding difference between an input encoder position data and an output encoder position data.

10. The method of claims 1 or 7, wherein the data are collected while rotating a motor of the actuator in full range in both counterclockwise and clockwise directions.

11. The method of claim 1, wherein the output from the backlash estimation unit is generated based at least in part on the difference between the input encoder position data and the output encoder position data.

12. The method of claim 1, further comprising upon determining a target torque command is close to zero, controlling a movement of a motor of the actuator through the backlash zone to drive the motor to a backlash center position.

13. The method of claims 1 or 12, wherein upon determining the actuator exiting the backlash zone, applying a preload adjustment to the output encoder position data.

14. The method of claims 1, 12 or 13, wherein the preload adjustment is generated using a smooth transition function.

15. The method of claims, 1, 12 or 13, wherein the preload adjustment is generated by taking the target torque and a backlash amplitude from the backlash estimation unit as input.

16. The method of claim 1, wherein the torque control comprises an open-loop control without feedback sensor data.

17. A system for providing torque control, the system comprising: a memory storing computer-executable instructions; one or more processors configured to execute the computerexecutable instructions to perform any of the claims 1-16.

18. A method for providing open-loop torque control, the method comprising:(a) receiving a torque control command;(b) adjusting the torque control command based at least in part on a hysteresis compensation, wherein the hysteresis compensation is generated by applying a torque processing model selected from a first model corresponding to torque ramping up and a second model corresponding to torque ramping down; and(c) controlling an actuator using the adjusted torque control command in an open-loop torque control.

19. The method of claim 18, wherein one or more of the first model or the second model comprises a Dahl model.

20. The method of claim 18, wherein the torque processing model is selected based at least in part on a hysteresis state.

21. The method of claims 18 or 20, wherein the hysteresis state comprises a positive or negative torque ramping state.

22. The method of claims 18 or 20, further comprising adjusting the torque control command by applying a filtering algorithm to the torque control command when transitioning between the first model and the second model.

23. A system for providing open-loop torque control, the system comprising: a memory storing computer-executable instructions; one or more processors configured to execute the computer-executable instructions to perform any of the claims 18-22.

24. A method for providing open-loop torque control, the method comprising:(a) receiving a target torque command;(b) receiving an input encoder position data and an output encoder position data and generating a backlash compensation based at least in part on an estimated backlash, wherein the estimated backlash is generated based at least in part on the input encoder position data and the output encoder position data;(c) generating a torque compensation based at least in part on a dynamic nonlinear current-to-torque relationship and the target torque command;(d) generating a hysteresis compensation based at least in part on a torque processing model selected from a first model corresponding to torque ramping up and a second model corresponding to torque ramping down and the target torque command; and(e) processing, by a controller, the target torque command, the backlash compensation, the torque compensation and the hysteresis compensation to generate an adjusted torque command for controlling a motor.

25. The method of claim 24, wherein the dynamic nonlinear current-to-torque relationship is established using a plurality of current measurements and torque measurements.

26. The method of claim 24, wherein the dynamic nonlinear current-to-torque relationship is defined by a dynamic Torque Constant (Kt) estimated for the motor.

27. The method of claims 24 or 26, wherein the Kt estimated for the motor is stored in a table.

28. The method of claim 24, further comprising switching to a Transparent Backdriving Mode upon determining the target torque command is close to zero.

29. The method of claims 24 or 28, wherein upon switching to the Transparent Backdriving Mode, driving the motor to a backlash center position.

30. The method of claim 24, wherein the motor is controlled using an open-loop torque control.

31. The method of claim 24, wherein the first model or the second model comprises a Dahl model.

32. The method of claims 24 or 31, wherein the torque processing model is selected based at least in part on a hysteresis state.

33. The method of claims 24, 31 or 32, wherein the hysteresis state comprises a positive or negative torque ramping state.

34. A system for providing open-loop torque control, the system comprising: a memory storing computer-executable instructions; one or more processors configured to execute the computer-executable instructions to perform any of the claims 24-33.

35. A method for adaptive control of an actuator system, comprising:(a) monitoring actuator state by collecting sensor data from an encoder, a temperature sensor, or a current sensor;(b) predicting a future behavior of the actuator using Model Predictive Control;(c) adjusting a control input in real-time for torque optimization based at least in part on the predicted future behavior; and(d) refining the control input using a machine learning model by taking at least the sensor data as input to adapt to changes.

36. A fault detection system for an actuator system, comprising;(a) a plurality of sensors configured to collect sensor data comprising a temperature, a vibration or current of the actuator system;(b) a multi-sensor fusion module configured to process the sensor data to generate a health profiling of the actuator system; and(c) taking the health profiling as input by a machine learning model to output an anomaly detection result.

Images