A low-torque servo module fault diagnosis and fault-tolerant control method and system

Abstract

The application provides a low-torque servo module fault diagnosis and fault-tolerant control method and system, which comprises the following steps: constructing a servo projection control model; inputting state variables and output error data of the model into a preset state perception unit to generate a state estimation signal; fitting a weight adjustment factor of the preset unit through the output error data to generate a weight dynamic adjustment value; analyzing estimation error and input-output residual of the model according to the state estimation signal and the weight dynamic adjustment value to generate a fault influence estimation signal; and generating an abnormal state compensation instruction by linearly combining the fault influence estimation signal and the state estimation signal to dynamically control the running state of the low-torque servo module. The application can realize real-time perception of control deviation of the low-torque servo module, realize closed-loop compensation at the instruction level through coupling of residual mapping and projection compensation mechanism, and effectively improve the fluency in the human-computer interaction process.

Description

Technical Field

[0001] This application relates to the field of servo module technology, and in particular to a method and system for fault diagnosis and fault-tolerant control of low-torque servo modules. Background Technology

[0002] In the current field of intelligent fitness and sports assistance, AI-powered fitness equipment and multifunctional sports equipment are gradually becoming key carriers for enhancing personalized training effects and user interaction experiences. Especially in scenarios requiring high-precision motion feedback, low-latency response control, and low-load interaction, low-torque servo modules, due to their compact structure, fast response, and low energy consumption, have been widely used in devices such as intelligent rowing machines, strength training equipment, and rehabilitation robotic arms. These devices collect real-time data on the user's force output, limb position, and speed changes, sending continuous adjustment control commands to the low-torque servo module to achieve assisted movement guidance, force matching, and safety limitations, thereby achieving efficient and personalized training goals.

[0003] However, in actual use, due to the high uncertainty and complexity of user movement behavior, such as sudden changes in movement rhythm, short bursts of strong force, or slight posture deviations, servo modules often experience problems such as state perception lag, control response drift, or failure to identify minor faults when operating under low load or low torque. Existing technologies generally rely on static models or empirical rules to handle state prediction, anomaly identification, and command compensation for low-torque servo modules. These technologies are difficult to effectively adapt to the varied and personalized human-computer interaction modes in fitness equipment, and are particularly unable to promptly identify and compensate for control residuals under non-steady-state conditions during intelligent training processes with high control precision requirements and rapid changes in movement rhythm. Summary of the Invention

[0004] This application provides a method and system for fault diagnosis and fault-tolerant control of low-torque servo modules, which is used to solve the problem that related technologies are difficult to effectively adapt to the changing and personalized human-computer interaction modes in fitness equipment.

[0005] The first aspect of this application provides a method for fault diagnosis and fault-tolerant control of a low-torque servo module, the method comprising: Based on the structural parameters and control signals of the low-torque servo module, a servo projection control model is constructed. The state variables and output error data of the servo projection control model are input into a preset state perception unit to generate a state estimation signal. Key response features are extracted from the output error data, and the weight adjustment factor of the state perception unit is fitted according to the error change rate of the key response features to generate a dynamic weight adjustment value. Based on the state estimation signal and the weight dynamic adjustment value, analyze the estimation error and input-output residual of the servo projection control model to generate a fault impact estimation signal; By linearly combining the fault impact estimation signal and the state estimation signal, an abnormal state compensation command is generated to dynamically control the operating state of the low torque servo module.

[0006] Optionally, in the first implementation of the first aspect of this application, the step of constructing a servo projection control model based on the structural parameters and control signals of the low-torque servo module includes: A system description matrix is ​​generated by collecting the structural parameters of the low-torque servo module. Based on the expected input command, actual response current and position feedback value in the control signal, a set of state variables is generated, and the system description matrix is ​​combined with the set of state variables to construct a state space expression structure. In the state-space representation structure, orthogonal subspace channels are divided based on the cross-coupling relationship between the voltage control component and the torque output component; By setting mapping weights in the orthogonal subspace channels, the projection path of the control command to the state response is determined; A servo projection control model is constructed based on the mapping weights and the projection path.

[0007] Optionally, in a second implementation of the first aspect of this application, the step of inputting the state variables and output error data of the servo projection control model to a preset state sensing unit to generate a state estimation signal includes: Construct the current state prediction vector based on the state variables output by the servo projection control model; The state prediction vector is compared with the encoder feedback signal and current sensing data of the low torque servo module to generate real-time output error data. The state prediction vector and the output error data are synchronously input into a preset state sensing unit, and the observation error signal is extracted by comparing the difference. Based on the deviation between the observation error signal and the state estimate value at the previous moment, and combined with the preset dynamic adjustment weights, the state prediction vector is weighted and iterated to generate the state estimate signal at the current moment.

[0008] Optionally, in a third implementation of the first aspect of this application, the step of extracting key response features from the output error data and fitting the weight adjustment factor of the state perception unit according to the error change rate of the key response features to generate a dynamic weight adjustment value includes: By performing sliding time window processing on the output error data, the key error interval of dynamic response change within the target operating cycle is extracted; Based on the combined trend of position error, current deviation and speed drift in the critical error interval, a response feature vector of the unsteady behavior of the low torque servo module is generated. By calculating the difference between the change trends of the response feature vector in adjacent time segments, a set of error change rate parameters is generated. The weight adjustment factor is weighted and fitted based on the error change rate parameter set, and the dynamic adjustment value of the weight is output.

[0009] Optionally, in the fourth implementation of the first aspect of this application, the step of analyzing the estimation error and input-output residual of the servo projection control model based on the state estimation signal and the weight dynamic adjustment value to generate a fault impact estimation signal includes: By comparing the state estimation signal with the historical state response trajectory of the servo projection control model, a model state deviation index is generated. Based on the model state deviation index and the weight dynamic adjustment value, a residual analysis matrix is ​​constructed, and the coupling offset and transient mutation amount are extracted according to the residual analysis matrix. The coupling offset and the transient change are projected onto the preset input residual subspace and output error subspace, respectively, to generate the input residual vector and the output residual vector; By performing cross-correlation deconstruction on the input residual vector and the output residual vector, a fault impact estimation signal corresponding to the multivariate disturbance effect of the low torque servo module is generated.

[0010] Optionally, in the fifth implementation of the first aspect of this application, the step of dynamically controlling the operating state of the low-torque servo module by linearly combining the fault impact estimation signal and the state estimation signal to generate an abnormal state compensation command includes: A state fusion matrix is ​​constructed based on the state estimation signal and the fault impact estimation signal, and the state fusion matrix is ​​subjected to time-series normalization processing to generate a state vector group. By inputting the state vector group into the preset structure adjustment unit for projection channel analysis and state weight decoupling, servo command interference adjustment parameters are generated; Based on the servo command interference adjustment parameters and the parameter change trend of the state vector group, the abnormal state compensation component at the current moment is determined. The abnormal state compensation component is vector-level superimposed with the corresponding real-time state parameters in the state estimation signal to output the abnormal state compensation command.

[0011] Optionally, in a sixth implementation of the first aspect of this application, the method further includes: Based on the dynamic difference between the voltage response component and the target servo behavior in the servo projection control model, a voltage residual sequence matrix is ​​generated within the target period. By performing time-series clustering decomposition on the voltage residual sequence matrix, voltage residual component groups are extracted; The voltage residual component group is jointly analyzed with the input residual vector to generate a voltage residual mapping table; Based on the voltage residual mapping table, a residual compensation vector consistent with the current servo control channel is extracted and vector-superimposed with the original servo control command to generate a voltage residual compensation command.

[0012] A second aspect of this application provides a low-torque servo module fault diagnosis and fault-tolerant control device, which is used to implement a low-torque servo module fault diagnosis and fault-tolerant control method. The low-torque servo module fault diagnosis and fault-tolerant control device includes: The module is used to build a servo projection control model based on the structural parameters and control signals of the low-torque servo module. The generation module is used to input the state variables and output error data of the servo projection control model into the preset state sensing unit to generate a state estimation signal. The fitting module is used to extract key response features from the output error data and fit the weight adjustment factor of the state perception unit according to the error change rate of the key response features to generate a dynamic weight adjustment value. The analysis module is used to analyze the estimation error and input-output residual of the servo projection control model based on the state estimation signal and the weight dynamic adjustment value, and generate a fault impact estimation signal. The control module is used to generate an abnormal state compensation command by linearly combining the fault impact estimation signal and the state estimation signal, and to dynamically control the operating state of the low torque servo module.

[0013] A third aspect of this application provides an electronic device, including a memory and a processor, wherein the processor is configured to execute a computer program stored in the memory, and when the processor executes the computer program, it implements the steps of the low-torque servo module fault diagnosis and fault-tolerant control method provided in the first aspect of this application.

[0014] The fourth aspect of this application provides a computer-readable storage medium storing a computer program thereon. When the computer program is executed by a processor, it implements the steps of the low-torque servo module fault diagnosis and fault-tolerant control method provided in the first aspect of this application.

[0015] In summary, the fault diagnosis and fault-tolerant control method and system for a low-torque servo module provided in this application constructs a servo projection control model based on the structural parameters and control signals of the low-torque servo module. The state variables and output error data of the servo projection control model are input to a preset state sensing unit to generate a state estimation signal. Key response features are extracted from the output error data, and the weight adjustment factor of the state sensing unit is fitted based on the error change rate of the key response features to generate a dynamic weight adjustment value. The estimation error and input-output residuals of the servo projection control model are analyzed based on the state estimation signal and the dynamic weight adjustment value to generate a fault impact estimation signal. An abnormal state compensation command is generated by linearly combining the fault impact estimation signal and the state estimation signal to dynamically control the operating state of the low-torque servo module. This application can perceive the control offset of the low-torque servo module in the non-steady-state response process in real time, and achieves command-level closed-loop compensation through coupled residual mapping and projection compensation mechanisms, effectively improving the smoothness of the human-computer interaction process. Attached Figure Description

[0016] Figure 1 A flowchart illustrating the low-torque servo module fault diagnosis and fault-tolerant control method provided in this application embodiment; Figure 2 A schematic diagram of the program module of the low torque servo module fault diagnosis and fault-tolerant control device provided in the embodiments of this application; Figure 3 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0017] To make the inventive objectives, features, and advantages of this application more apparent and understandable, the technical solutions in the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0018] To address the challenge that related technologies struggle to effectively adapt to the diverse and personalized human-computer interaction modes in fitness equipment, this application provides a method for fault diagnosis and fault-tolerant control of low-torque servo modules, such as... Figure 1This is a flowchart illustrating the low-torque servo module fault diagnosis and fault-tolerant control method provided in this embodiment. The low-torque servo module fault diagnosis and fault-tolerant control method includes the following steps: Step 110: Construct a servo projection control model based on the structural parameters and control signals of the low-torque servo module.

[0019] Specifically, based on the structural parameters and control signals of the low-torque servo module, a servo projection control model is constructed. This process mainly relies on the basic structural characteristics of each actuator joint in the device, such as stiffness, moment of inertia, and drive response curves. Combined with the voltage control quantity, desired position trajectory, and load disturbance feedback in the input signals, a state-space representation model is established to describe the dynamic behavior of the system. Through an orthogonal subspace partitioning strategy, the multi-channel coupling relationship between voltage input and torque output is separated, thereby constructing a projection model that maps the control signal to the state response. This lays the foundation for dynamic control in subsequent state estimation and fault analysis.

[0020] In one optional implementation of this embodiment, a system description matrix is ​​generated by collecting the structural parameters of the low-torque servo module; a set of state variables is generated based on the expected input command, actual response current, and position feedback value in the control signal, and the system description matrix and the set of state variables are combined to construct a state space expression structure; in the state space expression structure, orthogonal subspace channels are divided based on the cross-coupling relationship between the voltage control component and the torque output component; by setting mapping weights in the orthogonal subspace channels, the projection path of the control command on the state response is determined; and a servo projection control model is constructed based on the mapping weights and the projection path.

[0021] Specifically, in this embodiment, a system description matrix is ​​generated by collecting the structural parameters of the servo module. These structural parameters include physical and electromechanical characteristics of the motor, such as stator inductance, resistance, moment of inertia, back electromotive force constant, and transmission ratio. These characteristics collectively determine the dynamic response of the system under the influence of control signals. The system description matrix is ​​a mathematical expression constructed using these parameters, used to characterize the mapping relationship between input signals (such as voltage) and system states (such as speed, position, and current). For example, when a user performs a high-frequency weight-bearing stretching exercise using a fitness device, the servo motor needs to quickly respond to the repeatedly changing target torque. This matrix is ​​used to describe how the control system transitions from input voltage to actual output torque and speed. After obtaining the system description matrix, a set of state variables needs to be generated based on the control signals to reflect the current actual operating state of the device. The control signals include the expected input command (i.e., the command voltage or current corresponding to the user's training action target), the actual response current of the motor (collected by a Hall sensor), and the position feedback value (provided in real time by the encoder). By combining these signals, a multi-dimensional set of system state variables can be extracted, such as rotor angle, angular velocity, and current vector. Subsequently, the set of state variables is combined with the system description matrix to construct a state-space representation structure. This structure is a mathematical model of a dynamic system, formally represented by state equations and output equations, which respectively describe the internal state evolution and how it affects the external observable output. This achieves the integration of control logic and system physical behavior, allowing subsequent control strategies to infer the system state in real time based on the model's behavior. For example, when a fitness device assists a user in performing variable-speed squats, the system needs to predict the user's load trend in a certain posture and actively adjust the output torque to prevent instability. This state-space structure is the key basis for prediction. Then, within this state-space representation structure, the coupling relationship between the voltage control component and the torque output component needs to be identified, and orthogonal subspace channels are divided accordingly. The voltage control component refers to the control voltage applied by the system through the inverter, while the torque output component is the torque ultimately transmitted to the load (i.e., the user's joint) to generate mechanical motion. In servo control, since voltage regulation directly affects the current, and current and torque have a linear relationship, there is a physical coupling between the two. In AI-powered fitness scenarios, such as when a user performs a slow-start abdominal crunch exercise, the system needs to maintain the initial output torque within physiologically tolerable limits. In this case, the cross-influence between voltage control and the actual output torque must be handled through spatial orthogonalization to avoid control redundancy or response delay. The purpose of dividing the system into orthogonal subspace channels is to mathematically orthogonalize the influence of control inputs, allowing different control dimensions to act on the system along different spatial directions, thereby enhancing the independence and clarity of the control strategy. Mapping weights are set within these orthogonal channels to determine the specific projection path of the control command onto the system's state response.Mapping weights are parameters pre-set or adaptively adjusted based on experience or data-driven methods, used to quantify the influence of control commands on state variables of a particular channel. For example, in running-assisted training, if a user's center of gravity shift is detected, causing gait instability, the system needs to increase the mapping weight of the lateral stabilization channel to guide the servo module to respond more to lateral control inputs, thereby maintaining the symmetry of the user's movements. The existence of these weights allows the control system to dynamically allocate control resources for different biomechanical needs, improving the intelligence of human-machine collaboration. Finally, a servo projection control model is constructed by combining the above mapping weights and channel projection paths. This model comprehensively considers the input structure of control signals, the dynamic characteristics of system response, and the interaction between control channels, and is a servo modeling method under multi-input multi-output conditions. In AI fitness equipment, this model can achieve precise prediction and tracking control of the module state under complex training movements, enabling the motor to not only accurately reflect the user's movement intentions but also adjust torque output in a timely manner under sudden disturbances (such as posture changes or gravity shifts), achieving highly coordinated assisted control. Through this model, nonlinear dynamic changes in fitness training can be effectively addressed, ensuring the real-time response of the equipment and the safety of movement guidance.

[0022] Step 120: Input the state variables and output error data of the servo projection control model into the preset state sensing unit to generate a state estimation signal.

[0023] Specifically, the state variables and output error data of the servo projection control model are input into a preset state sensing unit to generate a state estimation signal. In AI-powered fitness motion control, this process compares the predicted state variables such as position, current, and velocity with the actual collected sensor feedback information, extracts the error between the current system operation and the ideal state, and uses an embedded weight adjustment mechanism to perform weighted correction on the predicted state based on the error signal to obtain a dynamic estimation result of the current state. This method can track minute offsets in the operation of the servo module in real time, providing a high-resolution state basis for anomaly detection.

[0024] In one optional implementation of this embodiment, the step of inputting the state variables and output error data of the servo projection control model into a preset state sensing unit to generate a state estimation signal includes: constructing a state prediction vector for the current moment based on the state variables output by the servo projection control model; comparing the state prediction vector with the encoder feedback signal and current sensing data of the low-torque servo module to generate real-time output error data; synchronously inputting the state prediction vector and output error data into the preset state sensing unit, and extracting the observation error signal by comparing the difference; and performing weighted iteration on the state prediction vector based on the deviation between the observation error signal and the state estimation value at the previous moment, combined with a preset dynamic adjustment weight, to generate the state estimation signal for the current moment.

[0025] Specifically, in the state estimation process, a state prediction vector for the current moment is constructed based on the state variables output by the servo projection control model. These state variables refer to the dynamic information within the system at the current moment, including position, velocity, current, and angular acceleration, which originate from the inferred values ​​output by the controller after modeling the servo system. The state prediction vector is a structured combination of variables formed on this basis, used to predict the system's trajectory in the next control cycle. For example, when a user performs lat pulldown training on a smart strength training device, the system infers its load curve, motor response, and muscle tension change trends based on the model, generating a prediction vector containing target current, displacement, and acceleration. This prediction vector is used to estimate the module's next force control behavior in the current posture. The aforementioned state prediction vector is then compared with the actually acquired encoder feedback signal and current sensor data to generate real-time output error data. The encoder feedback signal mainly reflects the motor shaft position and speed, serving as a key basis for evaluating execution accuracy; the current sensor data provides real-time information on the system's current current consumption, directly reflecting the difference between the load state and the electromagnetic drive. By comparing the model's predicted values ​​with the actual detected values ​​dimension by dimension, the current prediction accuracy and response error can be calculated. For example, if the predicted angular displacement is 45 degrees, but the encoder feedback value is 47 degrees, it can be determined that the module has a hysteresis bias. If the current prediction is 1.2A and the actual measurement is 1.6A, it indicates that the current load fluctuation may exceed the model's fitting range. Obtaining this error data is of great significance for identifying tension asymmetry, load imbalance, or user posture deviations during training. After obtaining the output error data, the state prediction vector and the above error signal need to be synchronously input into the preset state perception unit, and the observation error signal is extracted by comparing the difference. The state perception unit is a state assessment module with online learning capabilities, which integrates error filtering, feedback fusion, and deviation gain calculation mechanisms. This unit identifies potential control offset trends by comparing the difference between the state prediction and the actual response, and the extracted observation error signal reflects the degree of deviation between the system state and the desired trajectory. For example, in the operation of fitness equipment, if the user suddenly changes the exercise rhythm during training, the state perception unit can quickly detect this sudden behavior, map the deviation into a high-dimensional observation error vector, and provide a basis for dynamic adjustment of the control system. To ensure the temporal continuity and disturbance resistance of the state estimation results, it is necessary to perform weighted iteration on the state prediction vector based on the deviation between the observation error signal and the state estimate value at the previous moment, combined with preset dynamically adjusted weights, to generate the state estimation signal at the current moment. Dynamically adjusted weights refer to a parameter configuration used to balance the confidence level between model prediction and sensor feedback; the weights dynamically change with observation error, system state jump rate, and load stability.In fitness equipment, if the user's strength output is stable, the system can increase the weight of the model's predictions, reducing the interference of short-term sensor errors on the control system. Conversely, if the user experiences sudden changes in speed or posture, the system will automatically decrease the weight of the model's predictions, increasing reliance on sensor feedback. Through this weighted iterative process, the generated state estimation signal not only maintains an accurate grasp of the current state but also possesses the ability to respond quickly to short-term disturbances and potential anomalies. This enables AI-powered fitness equipment to continuously ensure the safety and training effectiveness of the exercise process under various postures, rhythms, and loads.

[0026] Step 130: Extract key response features from the output error data, and fit the weight adjustment factor of the state perception unit according to the error change rate of the key response features to generate a dynamic weight adjustment value.

[0027] Specifically, key response features are extracted from the output error data, and the weight adjustment factor of the state perception unit is fitted based on the error change rate of these key response features to generate a dynamic weight adjustment value. During this process, a sliding time window is used to capture sudden dynamic responses such as uneven user force application and impedance fluctuations, identifying non-steady-state behavior feature vectors. Then, by performing differential analysis on the trend of feature changes over time, the error change rate is calculated, and a fitting model is constructed to dynamically correct the perception weight parameters, helping the model to respond more accurately to abnormal disturbances during high-frequency actions.

[0028] In one optional implementation of this embodiment, the step of extracting key response features from the output error data and fitting the weight adjustment factor of the state perception unit according to the error change rate of the key response features to generate a dynamic weight adjustment value includes: extracting the key error interval of dynamic response change within the target operating cycle by performing sliding time window processing on the output error data; generating a response feature vector of the unsteady behavior of the low torque servo module according to the joint trend of position error, current deviation and speed drift in the key error interval; generating an error change rate parameter set by calculating the difference between the change trends of the response feature vector in adjacent time segments; and weighting the weight adjustment factor based on the error change rate parameter set to output a dynamic weight adjustment value.

[0029] Specifically, in this embodiment, to ensure that the device maintains a smooth, safe, and high-precision response under user-controlled operation or sudden load changes, it is necessary to dynamically identify the unsteady behavior of the servo module and adaptively adjust the error weights. First, by processing the output error data using a sliding time window, the key error intervals of dynamic response changes within the target operating cycle can be extracted. A sliding time window is a time-series data processing mechanism. Its core lies in setting a fixed-length time window and performing sliding sampling in unit time steps, thereby achieving continuous observation of the time-varying characteristics in the data sequence. In a fitness scenario, when a user performs push-up assisted training, frequent fluctuations in limb strength cause abrupt changes in the position error and current response of the servo module. Sliding window processing can accurately capture these intervals with significant time-domain changes, further used to identify the unsteady response behavior of the servo system. Based on the joint trend information of position error, current deviation, and speed drift contained in this key error interval, a response feature vector of the low-torque servo module can be generated. This feature vector is a multi-dimensional time-varying descriptive parameter set, reflecting the comprehensive dynamic characteristics of the module under the influence of external disturbances or internal modeling errors. Position error represents the deviation between the target position and the actual position; current deviation represents the difference between the input current command and the actual driving current; and speed drift reflects the gradual deviation between the module's movement speed and the ideal speed trajectory. Taking a smart rowing machine as an example, when a user pulls the slide, the large instantaneous acceleration causes current fluctuations and speed deviations. By combining the above three indicators, a key feature vector that fully describes the current unsteady response can be constructed, thus providing a basis for subsequent error change trend judgment. By calculating the difference between the change trends of the above response feature vector in adjacent time segments, an error change rate parameter set can be generated. The essence of this operation is to perform first-order difference processing on the feature vector sequence to extract its time derivative information, so as to characterize the evolution rate of key physical quantities over time. For example, in a smart load training system, if the current deviation increases from 0.1A to 0.4A in a continuous sampling period, while the speed drift trend remains positive, it indicates that the current module is in a high-load output state, and its system dynamics have deviated from the stable control boundary. Through difference processing, a quantitative error growth rate index can be obtained, providing a mathematical basis for subsequent weight adjustment. A weighted fitting model is constructed based on the error change rate parameter set, and the weight adjustment factors are matched using this model to ultimately output the dynamic adjustment value of the weights. The weighted fitting model is a multidimensional linear or nonlinear mapping structure, whose main purpose is to establish the relationship between error evolution characteristics and the control weight adjustment mechanism. This model can employ multiple regression, kernel function regression, or a residual-weighted neural network. By modeling the correlation between the error growth rate and historical state estimation errors, the system can adjust the influence weights of different feedback paths in the controller in real time according to the current dynamic performance.For example, in squat assist devices, when a user suddenly increases the pace of their movements, causing a momentary increase in the servo module's speed accompanied by a positional shift, the weighted fitting model can map the rate of change of position error into an increased adjustment factor. This enhances the influence of position channel feedback on the overall state estimation, enabling rapid response and compensation for sudden disturbances. Therefore, this process not only effectively identifies the unsteady-state behavior of low-torque servo modules but also dynamically matches optimal state-aware control parameters, thereby improving the control stability and intelligent response capabilities of AI-powered fitness equipment in multi-motion scenarios.

[0030] Step 140: Analyze the estimation error and input / output residuals of the servo projection control model based on the state estimation signal and the weight dynamic adjustment value, and generate the fault impact estimation signal.

[0031] Specifically, based on the state estimation signal and the dynamic adjustment value of the weights, the estimation error and input-output residuals of the servo projection control model are analyzed to generate a fault impact estimation signal. During the operation of the fitness equipment, this processing logic compares the current state estimation result with the equipment's historical control trajectory to identify possible abnormal signal sources in motion control, such as motor micro-jamming and control delay. By constructing a residual analysis matrix, the deviation between the system's input disturbance and output response is separated, thereby extracting fault feature vectors and realizing the identification and modeling of potential disturbances.

[0032] In one optional implementation of this embodiment, the step of generating a fault impact estimation signal by analyzing the estimation error and input-output residuals of the servo projection control model based on the state estimation signal and the weight dynamic adjustment value includes: generating a model state deviation index by comparing the state estimation signal with the historical state response trajectory of the servo projection control model; constructing a residual analysis matrix based on the model state deviation index and the weight dynamic adjustment value, and extracting coupling offset and transient mutation amount based on the residual analysis matrix; projecting the coupling offset and transient mutation amount to a preset input residual subspace and output error subspace respectively to generate an input residual vector and an output residual vector; and generating a fault impact estimation signal corresponding to the multivariate disturbance effect of the low torque servo module by performing cross-correlation deconstruction on the input residual vector and the output residual vector.

[0033] Specifically, in this embodiment, a model state deviation index can be generated by comparing the current state estimation signal with the historical state response trajectory of the servo projection control model. The state estimation signal represents the real-time prediction of the internal motion state of the device, while the historical state response trajectory records the device's motion performance under normal or expected operating conditions. By comparing the deviation between the two, the difference between the current state and the ideal state of the device can be captured, thereby reflecting potential anomalies or faults. For example, when a smart fitness device performs repetitive movements, if the joint position or torque output of a certain action differs significantly from the historical normal trajectory, the deviation index can reveal possible control misalignment or mechanical jamming in the servo module. Based on the above model state deviation index and the dynamic adjustment value of the weights, a residual analysis matrix is ​​constructed. The residual analysis matrix is ​​a mathematical structure that integrates multi-dimensional residual information to reveal the complex coupling relationship between the input and output of the servo system. The dynamic adjustment value of the weights reflects the system's sensitivity adjustment to error response. Combining the two can effectively highlight key residual signals in abnormal behavior, especially the nonlinear coupling effect between multiple variables. In the context of smart fitness equipment, sudden changes in equipment load or uneven force applied by the user manifest abnormal signals as coupling offsets and transient mutations in the matrix. These parameters help distinguish between normal fluctuations and genuine fault signs. By extracting coupling offsets and transient mutations from the residual analysis matrix, the spatiotemporal characteristics of abnormal signals can be more clearly identified. Coupling offsets reflect the persistent deviation between different state variables, while transient mutations describe short-term, high-amplitude abnormal changes. These two types of information are crucial for fault diagnosis because they reveal the imbalance between the force and response of the servo module, thus affecting the smoothness and accuracy of the equipment's movements. For example, during rapid, continuous, repetitive movements on fitness equipment, a surge in transient mutations may indicate momentary motor overload or sensor signal interference. The coupling offsets and transient mutations are projected onto preset input residual subspaces and output error subspaces, respectively, generating input residual vectors and output residual vectors. The input residual subspace focuses on the error distribution between control commands and actual input signals, while the output error subspace focuses on the deviation between sensor feedback and the actual response of the equipment. Orthogonal decomposition breaks down complex residual information into different subspaces, facilitating the identification of fault modes caused by input control failure and output execution anomalies. For example, the input residual vector may reveal interference or distortion of the command signal, while the output residual vector may indicate wear in mechanical parts or sensor failure. Finally, cross-correlation decomposition of the input and output residual vectors generates a fault impact estimation signal reflecting the effects of multivariate disturbances in the low-torque servo module. Cross-correlation decomposition analyzes the correlation and temporal synchronization between two residual vectors, revealing the causal relationship and abnormal coupling characteristics between control input and output response.In AI-powered fitness equipment, this fault impact estimation signal effectively reflects the combined effect of multi-source disturbances on equipment performance, helping to distinguish between single sensor errors and system-level faults. For example, this signal can identify sudden load changes caused by variations in user weight and control command deviations caused by electrical noise, allowing for adjustments to control strategies to ensure smooth and safe training movements. Overall, this method, through multi-dimensional data fusion and residual analysis, provides accurate fault diagnosis and fault-tolerant control guarantees for the servo modules of intelligent fitness equipment.

[0034] Step 150: By linearly combining the fault impact estimation signal and the state estimation signal, an abnormal state compensation command is generated to dynamically control the operating state of the low torque servo module.

[0035] Specifically, in this embodiment, an abnormal state compensation command is generated by linearly combining the fault impact estimation signal and the state estimation signal to dynamically control the operating state of the low-torque servo module. In AI fitness equipment, this control command can be used to correct joint commands in real time. For example, when the user's joint load changes in the opposite direction due to the inertia of the movement, the servo control output is adjusted according to the current state fusion result, so that the equipment can maintain the smoothness and following accuracy of the movement while ensuring safety, and achieve more natural motion assistance and feedback control.

[0036] In one optional implementation of this embodiment, the step of dynamically controlling the operating state of the low-torque servo module by linearly combining the fault impact estimation signal and the state estimation signal to generate an abnormal state compensation command includes: constructing a state fusion matrix based on the state estimation signal and the fault impact estimation signal, and performing time-series normalization processing on the state fusion matrix to generate a state vector group; generating servo command interference adjustment parameters by inputting the state vector group into a preset structure adjustment unit for projection channel analysis and state weight decoupling; determining the abnormal state compensation component at the current moment based on the servo command interference adjustment parameters and the parameter change trend of the state vector group; and superimposing the abnormal state compensation component with the corresponding real-time state parameters in the state estimation signal at the vector level to output the abnormal state compensation command.

[0037] Specifically, in this embodiment, the current state estimation signal and the fault impact estimation signal are integrated to construct a state fusion matrix. The state estimation signal represents the real-time operating state of the system based on sensor feedback and model prediction, while the fault impact estimation signal reflects the impact of various abnormal disturbances and potential faults experienced by the system during operation. Fusion of these two types of information results in a more comprehensive and accurate description of the system state. The state fusion matrix uses a matrix format to structurally manage multidimensional data, enabling a unified expression of state information from different sources in the spatiotemporal dimensions. For example, in fitness equipment, the state estimation of a servo motor reflects the actual position and speed of the motor during different training movements, while the fault estimation reflects deviations caused by sudden load changes or sensor anomalies. Fusion of these two aspects helps to comprehensively capture abnormal features during the movement process. Temporal normalization is applied to the state fusion matrix to unify the dimensions of different time points and data dimensions, eliminating the influence of differences in numerical ranges. Normalization not only promotes the stability of subsequent algorithms but also ensures that various types of state information have equal weight and comparability during the fusion process. For AI-powered fitness equipment, the state parameters of the servo module during the training cycle may vary significantly due to differences in movement amplitude and speed. Temporal normalization processing can avoid amplification or masking of anomalies caused by differences in movement intensity, thus ensuring the accuracy of state analysis. This processing generates a state vector set containing normalized fused state data from multiple time points, forming a continuous description of the motion trajectory and abnormal behavior. The state vector set is then input into a pre-defined structural adjustment unit. This unit features projection channel resolution and state weight decoupling. Projection channel resolution maps the high-dimensional state vector to several mutually orthogonal low-dimensional subspace channels through a linear transformation, each channel corresponding to specific dynamic characteristics or anomaly patterns. Weight decoupling separates the contribution weights of different state components, identifying and isolating the influencing components caused by servo command interference, thereby generating servo command interference adjustment parameters. This factor is essentially an adjustment parameter reflecting the intensity and direction of interference, revealing the specific influence path of the control signal on the state response. In AI-powered fitness equipment, servo motors need to respond quickly to changes in human movement, requiring frequent adjustments to control commands. High-frequency interference or errors in these commands can lead to disjointed movements or distorted postures. This adjustment factor can precisely quantify and compensate for this impact, ensuring smooth and accurate movement execution. Based on the servo command interference adjustment parameters and the changing trends of parameters in the state vector group, a linear response function model is constructed. This model uses linear algebra and system identification theory to characterize how interference factors affect the response of state variables over time. The linear response function maps the influence of the interference adjustment factor to the state parameter space, forming a prediction of the compensation components for abnormal states. Through this model, the magnitude and trend of deviations caused by interference in the servo module within the current movement cycle can be estimated in real time, providing a theoretical basis for compensation.For example, in a repetitive fitness movement, if the interference factor shows an increasing trend, the linear response model can predict future state deviations in advance, thereby adjusting the control strategy to prevent the movement from going out of control. Finally, the calculated abnormal state compensation component is vector-level superimposed with the corresponding real-time state parameters in the state estimation signal to output an abnormal state compensation command. Vector-level superposition means applying the compensation component as a vector offset to the current state in the multi-dimensional state space, correcting the motion trajectory of the servo module and making it return to the expected trajectory. This operation realizes model-based dynamic compensation, enabling AI fitness equipment to continuously adjust the servo module response, counteract the influence of interference, and improve the smoothness and accuracy of the movement. For example, when the fitness equipment performs a rapid arm swing, the compensation command can promptly correct the position error caused by changes in motor load, ensuring the stability of the movement amplitude and rhythm, thereby improving the user experience and training effect.

[0038] In one optional implementation of this embodiment, a voltage residual sequence matrix within the target period is generated based on the dynamic difference between the voltage response component in the servo projection control model and the target servo behavior; the voltage residual sequence matrix is ​​decomposed by temporal clustering to extract voltage residual component groups; the voltage residual component groups are jointly analyzed with the input residual vector to generate a voltage residual mapping table; based on the voltage residual mapping table, a residual compensation vector consistent with the current servo control channel is extracted and vector-superimposed with the original servo control command to generate a voltage residual compensation command.

[0039] Specifically, in this embodiment, the voltage response component refers to the voltage control signal output by the control model within a specific time period, while the target servo behavior is the ideal execution trajectory defined by the input of the external servo task. The dynamic difference refers to the sequence of differences between the voltage response component and the target servo voltage behavior at the same time, reflecting the deviation trend between the system response and the desired control. Periodically sampling this dynamic difference yields a voltage offset data set covering multiple time points within the target response period. This set is organized into a multi-dimensional time series matrix, namely the voltage residual sequence matrix. Each column of this matrix represents the time residual data of a certain voltage channel, and each row represents the voltage residual of all channels at a certain time. After the voltage residual sequence matrix is ​​constructed, it can be structured using time-series clustering decomposition. Time-series clustering decomposition analyzes the behavioral similarity and synchronicity between multiple channels in the voltage residual sequence on the time axis, grouping voltage channels with approximately time-varying patterns together using time-series trajectory distance metrics (such as Dynamic Time Warping (DTW) or Euclidean distance). The process outputs multiple residual component sets with high correlation or dynamic coupling characteristics. Each component set represents the response residual characteristics of the voltage control system under a certain type of behavior. For example, under low-speed, high-torque conditions, some voltage channels may exhibit synchronous upward offset, and the component set formed by these components represents the residual aggregation trend under this condition. After obtaining multiple voltage residual component sets, they need to be further analyzed jointly with the input residual vector. The input residual vector is defined as the sequence of differences between the system control input and feedback input, such as the deviation between the desired drive current and the measured current. The joint analysis process constructs a joint residual correlation matrix to match the voltage residual component sets with the input residual vector in the time dimension, analyzing their statistical correlation, hysteresis response mode, and energy distribution synergy. Based on this, the correspondence between voltage channels and input disturbances is extracted, and a voltage residual mapping table is constructed. This mapping table is a relational structure used to describe the contribution of servo system input disturbances to each voltage residual channel. Each element in the table represents the residual response intensity between a certain input disturbance and a specific voltage channel. For example, if an input disturbance occurs in the servo module due to a sudden change in shaft load, it can be observed that this disturbance causes significant amplitude fluctuations in the 3rd and 5th voltage channels, respectively. Therefore, in the mapping table, the weights of the disturbance entries in the 3rd and 5th channels will be significantly higher than those in other channels. Based on the voltage residual mapping table, a residual compensation vector consistent with the current servo control channel can be further extracted. By identifying channel entries in the mapping table that highly match the current channel number or response behavior, their corresponding residual patterns are extracted and subjected to time-domain weighted averaging to generate a voltage compensation vector consistent with the channel's dynamic behavior. This compensation vector is then superimposed on the original servo control command at the vector level, i.e., a compensation component is injected into the original command to adjust the channel voltage output to more closely approximate the target servo behavior.The voltage residual compensation command generated in this way can actively apply a compensation signal at the control layer based on the coupling relationship between historical offset characteristics and input residuals, thereby reducing the impact of abnormal disturbances on servo behavior and enhancing the system's stable operation capability.

[0040] According to the fault diagnosis and fault-tolerant control method for a low-torque servo module provided in this application, a servo projection control model is constructed based on the structural parameters and control signals of the low-torque servo module. The state variables and output error data of the servo projection control model are input to a preset state sensing unit to generate a state estimation signal. Key response features are extracted from the output error data, and the weight adjustment factor of the state sensing unit is fitted based on the error change rate of the key response features to generate a dynamic weight adjustment value. The estimation error and input-output residuals of the servo projection control model are analyzed based on the state estimation signal and the dynamic weight adjustment value to generate a fault impact estimation signal. An abnormal state compensation command is generated by linearly combining the fault impact estimation signal and the state estimation signal to dynamically control the operating state of the low-torque servo module. By implementing this application, a servo projection control model is constructed, and a self-adjusting state sensing mechanism and dynamic weight adjustment process are introduced. This not only enables multi-dimensional estimation of fault impact but also allows for the dynamic generation of abnormal state compensation commands based on the state estimation signal and weight change trends, effectively improving the operational stability of the low-torque servo module under rework conditions.

[0041] Figure 2 This application provides a low-torque servo module fault diagnosis and fault-tolerant control device, which can be used to implement the low-torque servo module fault diagnosis and fault-tolerant control method in the aforementioned embodiments. Figure 2 As shown, the fault diagnosis and fault-tolerant control device for the low-torque servo module mainly includes: Module 10 is used to construct a servo projection control model based on the structural parameters and control signals of the low-torque servo module. The generation module 20 is used to input the state variables and output error data of the servo projection control model into the preset state sensing unit to generate a state estimation signal. The fitting module 30 is used to extract key response features from the output error data and fit the weight adjustment factor of the state perception unit according to the error change rate of the key response features to generate a dynamic weight adjustment value. Analysis module 40 is used to analyze the estimation error and input / output residuals of the servo projection control model based on the state estimation signal and the weight dynamic adjustment value, and to generate a fault impact estimation signal. The control module 50 is used to generate abnormal state compensation commands by linearly combining the fault impact estimation signal and the state estimation signal, and to dynamically control the operating state of the low torque servo module.

[0042] In one optional implementation of this embodiment, the construction module is specifically used to: generate a system description matrix by collecting the structural parameters of the low-torque servo module; generate a set of state variables based on the expected input command, actual response current, and position feedback value in the control signal, and combine the system description matrix with the set of state variables to construct a state space expression structure; in the state space expression structure, divide orthogonal subspace channels based on the cross-coupling relationship between the voltage control component and the torque output component; determine the projection path of the control command on the state response by setting mapping weights in the orthogonal subspace channels; and construct a servo projection control model based on the mapping weights and the projection path.

[0043] In one optional implementation of this embodiment, the generation module is specifically used to: construct a state prediction vector for the current moment based on the state variables output by the servo projection control model; compare the state prediction vector with the encoder feedback signal and current sensing data of the low-torque servo module to generate real-time output error data; synchronously input the state prediction vector and the output error data to a preset state sensing unit, and extract the observation error signal by comparing the difference; based on the deviation between the observation error signal and the state estimate value at the previous moment, and combined with a preset dynamic adjustment weight, perform weighted iteration on the state prediction vector to generate the state estimate signal for the current moment.

[0044] In one optional implementation of this embodiment, the fitting module is specifically used to: extract the key error interval of dynamic response change within the target operating cycle by performing sliding time window processing on the output error data; generate a response feature vector of the unsteady behavior of the low torque servo module based on the joint trend of position error, current deviation and speed drift in the key error interval; generate an error change rate parameter set by calculating the difference between the change trends of the response feature vector in adjacent time segments; and perform weighted fitting on the weight adjustment factor based on the error change rate parameter set to output the weight dynamic adjustment value.

[0045] In one optional implementation of this embodiment, the analysis module is specifically used to: generate a model state deviation index by comparing the difference between the state estimation signal and the historical state response trajectory of the servo projection control model; construct a residual analysis matrix based on the model state deviation index and the dynamic adjustment value of the weights, and extract the coupling offset and transient mutation amount according to the residual analysis matrix; project the coupling offset and transient mutation amount to the preset input residual subspace and output error subspace respectively to generate input residual vector and output residual vector; and generate a fault influence estimation signal corresponding to the multivariate disturbance influence of the low torque servo module by performing cross-correlation deconstruction on the input residual vector and output residual vector.

[0046] In one optional implementation of this embodiment, the control module is specifically used to: construct a state fusion matrix based on the state estimation signal and the fault impact estimation signal, and perform time-series normalization processing on the state fusion matrix to generate a state vector group; generate servo command interference adjustment parameters by inputting the state vector group into a preset structure adjustment unit for projection channel analysis and state weight decoupling; determine the abnormal state compensation component at the current moment based on the servo command interference adjustment parameters and the parameter change trend of the state vector group; and perform vector-level superposition of the abnormal state compensation component and the corresponding real-time state parameters in the state estimation signal to output the abnormal state compensation command.

[0047] In an optional embodiment of this example, the control module is further configured to: generate a voltage residual sequence matrix within the target period based on the dynamic difference between the voltage response component in the servo projection control model and the target servo behavior; extract voltage residual component groups by performing time-series clustering decomposition on the voltage residual sequence matrix; jointly analyze the voltage residual component groups with the input residual vector to generate a voltage residual mapping table; and extract a residual compensation vector consistent with the current servo control channel based on the voltage residual mapping table, and vector superimpose it with the original servo control command to generate a voltage residual compensation command.

[0048] According to the fault diagnosis and fault-tolerant control device for a low-torque servo module provided in this application, a servo projection control model is constructed based on the structural parameters and control signals of the low-torque servo module. The state variables and output error data of the servo projection control model are input to a preset state sensing unit to generate a state estimation signal. Key response features are extracted from the output error data, and the weight adjustment factor of the state sensing unit is fitted based on the error change rate of the key response features to generate a dynamic weight adjustment value. The estimation error and input-output residuals of the servo projection control model are analyzed based on the state estimation signal and the dynamic weight adjustment value to generate a fault impact estimation signal. An abnormal state compensation command is generated by linearly combining the fault impact estimation signal and the state estimation signal to dynamically control the operating state of the low-torque servo module. Through the implementation of this application, a servo projection control model is constructed, and a self-adjusting state sensing mechanism and dynamic weight adjustment process are introduced. This not only enables multi-dimensional estimation of fault impact but also allows for the dynamic generation of abnormal state compensation commands based on the state estimation signal and weight change trends, effectively improving the operational stability of the low-torque servo module under rework conditions.

[0049] According to the scheme provided in this application Figure 3 An electronic device is provided as an embodiment of this application. This electronic device can be used to implement the low-torque servo module fault diagnosis and fault-tolerant control method in the foregoing embodiments, and mainly includes: The system includes a memory 301, a processor 302, and a computer program 303 stored in the memory 301 and executable on the processor 302. The memory 301 and the processor 302 are connected via communication. When the processor 302 executes the computer program 303, it implements the low-torque servo module fault diagnosis and fault-tolerant control method described in the foregoing embodiments. The number of processors can be one or more.

[0050] The memory 301 can be a high-speed random access memory (RAM) or a non-volatile memory, such as a disk storage device. The memory 301 is used to store executable program code, and the processor 302 is coupled to the memory 301.

[0051] Furthermore, embodiments of this application also provide a computer-readable storage medium, which may be disposed in the electronic device described in the above embodiments, and the computer-readable storage medium may be as described above. Figure 3 The memory in the illustrated embodiment.

[0052] The computer-readable storage medium stores a computer program that, when executed by a processor, implements the low-torque servo module fault diagnosis and fault-tolerant control method described in the foregoing embodiments. Furthermore, the computer-readable storage medium can also be a USB flash drive, external hard drive, read-only memory (ROM), RAM, magnetic disk, or optical disk, or any other medium capable of storing program code.

[0053] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0054] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0055] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A low-torque servo module fault diagnosis and fault-tolerant control method, characterized in that, include: Based on the structural parameters and control signals of the low-torque servo module, a servo projection control model is constructed. The state variables and output error data of the servo projection control model are input into a preset state perception unit to generate a state estimation signal. The system extracts key response features from the output error data and fits the weight adjustment factor of the state perception unit based on the error change rate of the key response features to generate a dynamic weight adjustment value. Specifically, this includes: extracting key error intervals of dynamic response changes within the target operating cycle by processing the output error data through a sliding time window; generating a response feature vector of the unsteady-state behavior of the low-torque servo module based on the joint trend of position error, current deviation, and speed drift within the key error interval; generating an error change rate parameter set by calculating the difference between the change trends of the response feature vector in adjacent time segments; and weighting the weight adjustment factor based on the error change rate parameter set to output a dynamic weight adjustment value. Based on the state estimation signal and the weight dynamic adjustment value, analyze the estimation error and input-output residual of the servo projection control model to generate a fault impact estimation signal; By linearly combining the fault impact estimation signal and the state estimation signal, an abnormal state compensation command is generated to dynamically control the operating state of the low torque servo module.

2. The low-torque servo module fault diagnosis and fault-tolerant control method according to claim 1, characterized in that, The step of constructing a servo projection control model based on the structural parameters and control signals of the low-torque servo module includes: A system description matrix is ​​generated by collecting the structural parameters of the low-torque servo module. Based on the expected input command, actual response current and position feedback value in the control signal, a set of state variables is generated, and the system description matrix is ​​combined with the set of state variables to construct a state space expression structure. In the state-space representation structure, orthogonal subspace channels are divided based on the cross-coupling relationship between the voltage control component and the torque output component; By setting mapping weights in the orthogonal subspace channels, the projection path of the control command to the state response is determined; A servo projection control model is constructed based on the mapping weights and the projection path.

3. The low-torque servo module fault diagnosis and fault-tolerant control method according to claim 1, characterized in that, The step of inputting the state variables and output error data of the servo projection control model into a preset state sensing unit to generate a state estimation signal includes: Construct the current state prediction vector based on the state variables output by the servo projection control model; The state prediction vector is compared with the encoder feedback signal and current sensing data of the low torque servo module to generate real-time output error data. The state prediction vector and the output error data are synchronously input into a preset state sensing unit, and the observation error signal is extracted by comparing the difference. Based on the deviation between the observation error signal and the state estimate value at the previous moment, and combined with the preset dynamic adjustment weights, the state prediction vector is weighted and iterated to generate the state estimate signal at the current moment.

4. The low-torque servo module fault diagnosis and fault-tolerant control method according to claim 1, characterized in that, The step of analyzing the estimation error and input-output residuals of the servo projection control model based on the state estimation signal and the weight dynamic adjustment value, and generating a fault impact estimation signal, includes: By comparing the state estimation signal with the historical state response trajectory of the servo projection control model, a model state deviation index is generated. Based on the model state deviation index and the weight dynamic adjustment value, a residual analysis matrix is ​​constructed, and the coupling offset and transient mutation amount are extracted according to the residual analysis matrix. The coupling offset and the transient change are projected onto the preset input residual subspace and output error subspace, respectively, to generate the input residual vector and the output residual vector; By performing cross-correlation deconstruction on the input residual vector and the output residual vector, a fault impact estimation signal corresponding to the multivariate disturbance effect of the low torque servo module is generated.

5. The low-torque servo module fault diagnosis and fault-tolerant control method according to claim 1, characterized in that, The step of dynamically controlling the operating state of the low-torque servo module by linearly combining the fault impact estimation signal and the state estimation signal to generate an abnormal state compensation command includes: A state fusion matrix is ​​constructed based on the state estimation signal and the fault impact estimation signal, and the state fusion matrix is ​​subjected to time-series normalization processing to generate a state vector group. By inputting the state vector group into the preset structure adjustment unit for projection channel analysis and state weight decoupling, servo command interference adjustment parameters are generated; Based on the servo command interference adjustment parameters and the parameter change trend of the state vector group, the abnormal state compensation component at the current moment is determined. The abnormal state compensation component is vector-level superimposed with the corresponding real-time state parameters in the state estimation signal to output the abnormal state compensation command.

6. The low-torque servo module fault diagnosis and fault-tolerant control method according to claim 4, characterized in that, The method further includes: Based on the dynamic difference between the voltage response component and the target servo behavior in the servo projection control model, a voltage residual sequence matrix is ​​generated within the target period. By performing time-series clustering decomposition on the voltage residual sequence matrix, voltage residual component groups are extracted; The voltage residual component group is jointly analyzed with the input residual vector to generate a voltage residual mapping table; Based on the voltage residual mapping table, a residual compensation vector consistent with the current servo control channel is extracted and vector-superimposed with the original servo control command to generate a voltage residual compensation command.

7. A fault diagnosis and fault-tolerant control device for a low-torque servo module, characterized in that, The low-torque servo module fault diagnosis and fault-tolerant control device is used to implement the low-torque servo module fault diagnosis and fault-tolerant control method according to claim 1, wherein the low-torque servo module fault diagnosis and fault-tolerant control device comprises: The module is used to build a servo projection control model based on the structural parameters and control signals of the low-torque servo module. The generation module is used to input the state variables and output error data of the servo projection control model into the preset state sensing unit to generate a state estimation signal. The fitting module is used to extract key response features from the output error data and fit the weight adjustment factor of the state perception unit according to the error change rate of the key response features to generate a dynamic weight adjustment value. The analysis module is used to analyze the estimation error and input-output residual of the servo projection control model based on the state estimation signal and the weight dynamic adjustment value, and generate a fault impact estimation signal. The control module is used to generate an abnormal state compensation command by linearly combining the fault impact estimation signal and the state estimation signal, and to dynamically control the operating state of the low torque servo module.

8. An electronic device, comprising: Includes memory and processor, of which: The processor is used to execute computer programs stored in the memory; When the processor executes the computer program, it implements the steps in the low-torque servo module fault diagnosis and fault-tolerant control method according to any one of claims 1 to 6.

9. A computer readable storage medium having stored thereon a computer program, characterized in that, When the computer program is executed by the processor, it implements the steps in the low-torque servo module fault diagnosis and fault-tolerant control method according to any one of claims 1 to 6.

Images