A servo motor disorder elimination control method

Abstract

The application discloses a servo motor disorder elimination control method, and belongs to the technical field of servo motors.The application comprises time sequence verification and guardianship based on a causal diagram model, data reconstruction and fault-tolerant control based on a deep state estimator, and physical phase self-calibration based on a high-frequency impedance ellipse.The application can identify and isolate abnormal data before the abnormal data affect a control system through a causal diagram model and a space-time label, thereby changing passive processing into active defense.A state estimator based on deep learning can generate more accurate fault-tolerant control quantities than traditional linear prediction, thereby ensuring that the performance of the system does not degrade during time sequence disorder.Rotary high-frequency vector injection and impedance ellipse analysis technology can decouple phase errors, parameter errors and other types of problems, realize comprehensive online self-calibration and self-tuning, and make the system always maintain optimal performance.

Description

Technical Field

[0001] This invention relates to the field of servo motor technology, and specifically to a servo motor disorder elimination control method. Background Technology

[0002] The "out-of-order" problem in servo systems—the mismatch between the spatiotemporal sequence of control commands and the actual state of the physical system—has become a bottleneck restricting reliability improvements. This type of problem mainly manifests in two ways: 1. Timing Disorder: In digital controllers based on multi-tasking operating systems, high-priority interrupts, network communication blockages, or software anomalies can cause the execution cycle of the control loop to become disordered. For example, the current loop may incorrectly combine speed and position commands from different calculation cycles. This "spatiotemporal mismatch" in control quantities can cause high-frequency jitter or low-frequency wandering of motor torque. Traditional control architectures based on fixed frequencies cannot detect and correct such errors. 2. Phase Disorder: Deviations in the initial rotor position identification, encoder installation errors, or mechanical deformation after long-term operation can cause a fixed or slowly varying deviation between the rotor spatial phase calculated by the controller and the actual phase. This deviation disrupts the accuracy of magnetic field orientation in vector control, causing the servo system to always operate in a suboptimal state, resulting in decreased torque output efficiency, increased noise, and sluggish dynamic response.

[0003] Existing solutions are mostly passive: when timing disorder causes significant system anomalies, a watchdog circuit is used for reset and restart; for phase disorder, manual calibration or initial position identification procedures after shutdown are relied upon. These methods cannot meet the needs of continuous production and highly autonomous equipment. To solve the above problems, this invention provides a servo motor disorder elimination control method. Summary of the Invention

[0004] To address the aforementioned technical shortcomings, the present invention aims to provide a servo motor out-of-order control method. Through a causal graph model and spatiotemporal tags, abnormal data can be identified and isolated before it affects the control system, transforming passive processing into proactive defense. A deep learning-based state estimator can generate more accurate fault-tolerant control quantities than traditional linear prediction, ensuring that the system's performance does not degrade during timing out-of-order periods. Rotating high-frequency vector injection and impedance elliptic analysis techniques can decouple various problems such as phase errors and parameter errors, achieving comprehensive online self-calibration and self-tuning. This solves the technical problems mentioned in the background section.

[0005] To solve the above-mentioned technical problems, the present invention adopts the following technical solution: The present invention provides a servo motor disorder elimination control method, including the following: Time-series verification and safeguards based on cause-effect graph model: Step S11: Control flow cause-effect graph modeling; Step S12: Generation and transmission of spatiotemporal consistency tags; Step S13: Model-based temporal anomaly detection and isolation; Data reconstruction and fault-tolerant control based on depth state estimator: Step S21: Construction and online learning of the deep state estimator; Step S22: Seamless data reconstruction and control quantity integration. When the timing guardian isolates abnormal data, the data reconstruction mechanism is triggered. Physical phase self-calibration based on high-frequency impedance ellipse: Step S31: Rotating high-frequency voltage vector injection and full-frequency response acquisition; Step S32: Dynamic impedance ellipse fitting and feature extraction; Step S33: Multi-dimensional error decoupling and real-time compensation.

[0006] In this embodiment, by using a causal graph model and spatiotemporal labels, abnormal data can be identified and isolated before it affects the control system, transforming passive processing into proactive defense. The state estimator based on deep learning can generate more accurate fault-tolerant control quantities than traditional linear prediction, ensuring that the system's performance does not degrade during time-series disorder. Rotating high-frequency vector injection and impedance elliptic analysis techniques can decouple multiple problems such as phase error and parameter error, achieving comprehensive online self-calibration and self-tuning, so that the system always maintains optimal performance.

[0007] Preferably, the control flow cause-effect graph modeling in step S1 is as follows: Establish a cause-effect graph model G = (V, E) for the servo system control task; Wherein, the node set V = {P, S, C} represents the position loop, velocity loop, and current loop tasks, respectively; the directed edge set E represents the data dependency relationship between tasks, that is, P→S: the position loop output is the velocity loop input, S→C: the velocity loop output is the current loop input.

[0008] Preferably, the generation and transmission of the spatiotemporal consistency label in step S12 are as follows: Generate a global spatiotemporal label Λ(n) for each control cycle n.<t_n,s_n,H(n)> ; Where t_n is the precise timestamp, s_n is the monotonically increasing sequence number, and H(n) is the system state hash value for the current period and the previous k periods, such as the characteristic value of instructions and outputs. Any data packet passed from one task to the next must carry the spatiotemporal tag Λ of its generation time.

[0009] Preferably, the model-based temporal anomaly detection and isolation in step S13 is as follows: Set a timing guardian at the execution entry point of tasks S and C: Causality check: Verify whether the label Λ_input of the input data matches the expected output label of the predecessor task of the current task. For example, the velocity loop checks whether its input comes from the most recently completed position loop task. Spatiotemporal continuity check: Verify whether s_n is continuous and whether the system state evolution represented by H(n) is smooth and conforms to dynamic laws. Once an anomaly is detected, the data packet is immediately marked as "spatiotemporally contaminated data" and isolated to prevent it from entering the control algorithm. Preferably, the construction and online learning of the depth state estimator in step S21 are as follows: The depth state estimator takes a historical normal state sequence X_{nk:n-1} = [x(nk), ..., x(n-1)] as input, where x can include position, velocity, current, etc.; and outputs the state prediction value x̂(n) for the current period. The depth state estimator adopts a long short-term memory network or a temporal convolutional network structure, which can continuously fine-tune its parameters online to adapt to the dynamic characteristics of the controlled object.

[0010] Preferably, the seamless data reconstruction and control quantity fusion in step S22 are detailed below: Input the historical normal state sequence X_{nk:n-1} = [x(nk), ...,x(n-1)] into the deep state estimator to generate the state prediction value x̂(n); The following fusion strategy is used to generate control quantities: u(n) = α u_fb(x̂(n)) + (1-α) u_ff(x̂(n)); Where u_fb is the feedback control variable based on the predicted state, u_ff is the feedforward control variable based on the predicted state, and α is the fusion weight dynamically adjusted according to the prediction uncertainty. This process achieves a seamless and smooth switch from normal feedback control to predictive fault-tolerant control.

[0011] Preferably, the rotating high-frequency voltage vector injection and full-frequency response acquisition in step S31 are detailed as follows: Unlike traditional single-axis injection, this invention injects a high-frequency voltage vector ΔU_h with constant amplitude, frequency ω_h, and rotating at an angular velocity ω_r into the dq coordinate system; ΔU_dh = V_h cos(ω_h t +ω_r t); ΔU_qh = V_h sin(ω_h t +ω_r t); The high-frequency current responses ΔI_dh and ΔI_qh of the dq axis are acquired synchronously.

[0012] Preferably, the dynamic impedance ellipse fitting and feature extraction in step S32 are detailed below: Within one rotation period T_r = 2π / ω_r, the high-frequency current response [ΔI_dh, ΔI_qh] is plotted as an elliptic scatter plot; the center position, major axis / minor axis length (L_max, L_min) and tilt angle φ of the elliptic scatter plot are extracted by an elliptic fitting algorithm.

[0013] Preferably, the multi-dimensional error decoupling and real-time compensation in step S33 are detailed below: The offset of the ellipse center in the elliptic scatter plot mainly characterizes the zero-point drift error of the measurement circuit. The ellipse axis ratio (L_max / L_min) of the elliptic scatter plot reflects the saturation degree of the motor magnetic circuit and can be used to correct the inductance parameters L_d and L_q online. The tilt angle φ of the elliptic scatter plot directly corresponds to the rotor spatial phase error that needs to be compensated. ; according to Dynamic compensation is performed on the rotor position angle used for feedback and transformation: θ_corrected = θ_measured + .

[0014] The beneficial effects of this invention are as follows: This invention uses a causal graph model and spatiotemporal labels to identify and isolate abnormal data before it affects the control system, transforming passive processing into proactive defense. Based on a deep learning-based state estimator, it can generate more accurate fault-tolerant control quantities than traditional linear prediction, ensuring that the system's performance does not degrade during time-series disorder. Rotating high-frequency vector injection and impedance elliptic analysis techniques can decouple multiple problems such as phase error and parameter error, achieving comprehensive online self-calibration and self-tuning, enabling the system to always maintain optimal performance. Attached Figure Description

[0015] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0016] Figure 1 A flowchart of a servo motor disorder elimination control method provided in an embodiment of the present invention. Detailed Implementation

[0017] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0018] Example: Figure 1 As shown, the present invention provides a servo motor disorder elimination control method, including the following: Time-series verification and safeguards based on cause-effect graph model: Step S11: Control flow cause-effect graph modeling; Step S12: Generation and transmission of spatiotemporal consistency tags; Step S13: Model-based temporal anomaly detection and isolation; Data reconstruction and fault-tolerant control based on depth state estimator: Step S21: Construction and online learning of the deep state estimator; Step S22: Seamless data reconstruction and control quantity integration. When the timing guardian isolates abnormal data, the data reconstruction mechanism is triggered. Physical phase self-calibration based on high-frequency impedance ellipse: Step S31: Rotating high-frequency voltage vector injection and full-frequency response acquisition; Step S32: Dynamic impedance ellipse fitting and feature extraction; Step S33: Multi-dimensional error decoupling and real-time compensation.

[0019] In this embodiment, by using a causal graph model and spatiotemporal labels, abnormal data can be identified and isolated before it affects the control system, transforming passive processing into proactive defense. The state estimator based on deep learning can generate more accurate fault-tolerant control quantities than traditional linear prediction, ensuring that the system's performance does not degrade during time-series disorder. Rotating high-frequency vector injection and impedance elliptic analysis techniques can decouple multiple problems such as phase error and parameter error, achieving comprehensive online self-calibration and self-tuning, so that the system always maintains optimal performance.

[0020] Furthermore, in step S1, the control flow cause-effect graph is modeled as follows: Establish a cause-effect graph model G = (V, E) for the servo system control task; Wherein, the node set V = {P, S, C} represents the position loop, velocity loop, and current loop tasks, respectively; the directed edge set E represents the data dependency relationship between tasks, that is, P→S: the position loop output is the velocity loop input, S→C: the velocity loop output is the current loop input.

[0021] Furthermore, the generation and transmission of the spatiotemporal consistency label in step S12 are as follows: Generate a global spatiotemporal label Λ(n) for each control cycle n.<t_n,s_n,H(n)> ; Where t_n is the precise timestamp, s_n is the monotonically increasing sequence number, and H(n) is the system state hash value for the current period and the previous k periods, such as the characteristic value of instructions and outputs. Any data packet passed from one task to the next must carry the spatiotemporal tag Λ of its generation time.

[0022] Furthermore, the model-based temporal anomaly detection and isolation in step S13 is as follows: Set a timing guardian at the execution entry point of tasks S and C: Causality check: Verify whether the label Λ_input of the input data matches the expected output label of the predecessor task of the current task; for example, the velocity loop checks whether its input comes from the most recently completed position loop task; Spatiotemporal continuity check: Verify whether s_n is continuous and whether the evolution of the system state represented by H(n) is smooth and conforms to the laws of dynamics. Furthermore, the construction and online learning of the deep state estimator in step S21 are detailed below: The depth state estimator takes a historical normal state sequence X_{nk:n-1} = [x(nk), ..., x(n-1)] as input, where x can include position, velocity, current, etc.; and outputs the state prediction value x̂(n) for the current period. The depth state estimator adopts a long short-term memory network or a temporal convolutional network structure, which can continuously fine-tune its parameters online to adapt to the dynamic characteristics of the controlled object.

[0023] Furthermore, the seamless data reconstruction and control quantity fusion in step S22 are detailed below: Input the historical normal state sequence X_{nk:n-1} = [x(nk), ...,x(n-1)] into the deep state estimator to generate the state prediction value x̂(n); The following fusion strategy is used to generate control quantities: u(n) = α u_fb(x̂(n)) + (1-α) u_ff(x̂(n)); Where u_fb is the feedback control quantity based on the predicted state, u_ff is the feedforward control quantity based on the predicted state, and α is the fusion weight dynamically adjusted according to the prediction uncertainty; this process realizes a seamless and smooth switch from normal feedback control to predictive fault-tolerant control.

[0024] Furthermore, the details of the rotating high-frequency voltage vector injection and full-frequency response acquisition in step S31 are as follows: A high-frequency voltage vector ΔU_h with constant amplitude, frequency ω_h, and rotating at an angular velocity ω_r is injected into the dq coordinate system. ΔU_dh = V_h cos(ω_h t +ω_r t); ΔU_qh = V_h sin(ω_h t +ω_r t); The high-frequency current responses ΔI_dh and ΔI_qh of the dq axis are acquired synchronously.

[0025] Furthermore, the dynamic impedance ellipse fitting and feature extraction in step S32 are detailed below: Within one rotation period T_r = 2π / ω_r, the high-frequency current response [ΔI_dh, ΔI_qh] is plotted as an elliptic scatter plot; the center position, major axis / minor axis length (L_max, L_min) and tilt angle φ of the elliptic scatter plot are extracted by an elliptic fitting algorithm.

[0026] Furthermore, the multi-dimensional error decoupling and real-time compensation in step S33 are detailed below: The offset of the ellipse center in the elliptic scatter plot mainly characterizes the zero-point drift error of the measurement circuit. The ellipse axis ratio (L_max / L_min) of the elliptic scatter plot reflects the saturation degree of the motor magnetic circuit and can be used to correct the inductance parameters L_d and L_q online. The elliptic tilt angle φ of the elliptic scatter plot directly corresponds to the rotor spatial phase error that needs to be compensated. ;according to Dynamic compensation is performed on the rotor position angle used for feedback and transformation: θ_corrected = θ_measured + .

[0027] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A servo motor disorder elimination control method, characterized in that, Includes the following: Time-series verification and safeguards based on cause-effect graph model: Step S11: Control flow cause-effect graph modeling; Step S12: Generation and transmission of spatiotemporal consistency tags; Step S13: Model-based temporal anomaly detection and isolation; Data reconstruction and fault-tolerant control based on depth state estimator: Step S21: Construction and online learning of the deep state estimator; Step S22: Seamless data reconstruction and control quantity integration. When the timing guardian isolates abnormal data, the data reconstruction mechanism is triggered. Physical phase self-calibration based on high-frequency impedance ellipse: Step S31: Rotating high-frequency voltage vector injection and full-frequency response acquisition; Step S32: Dynamic impedance ellipse fitting and feature extraction; Step S33: Multi-dimensional error decoupling and real-time compensation.

2. The servo motor disorder elimination control method as described in claim 1, characterized in that, In step S1, the control flow cause-effect graph is modeled as follows: Establish a cause-effect graph model G = (V, E) for the servo system control task; Wherein, the node set V = {P, S, C} represents the position loop, velocity loop, and current loop tasks, respectively; the directed edge set E represents the data dependency relationship between tasks, that is, P→S: the position loop output is the velocity loop input, S→C: the velocity loop output is the current loop input.

3. The servo motor disorder elimination control method as described in claim 1, characterized in that, The generation and transmission of the spatiotemporal consistency label in step S12 are as follows: Generate a global spatiotemporal label Λ(n) for each control cycle n.<t_n,s_n,H(n)> ; Where t_n is the precise timestamp, s_n is the monotonically increasing sequence number, and H(n) is the system state hash value for the current period and the previous k periods.

4. The servo motor disorder elimination control method as described in claim 1, characterized in that, The model-based temporal anomaly detection and isolation in step S13 is as follows: Set a timing guardian at the execution entry point of tasks S and C: Causality check: Verify whether the label Λ_input of the input data matches the expected output label of the predecessor task of the current task; Spatiotemporal continuity check: Verify whether s_n is continuous and whether the evolution of the system state represented by H(n) is smooth and conforms to the laws of dynamics.

5. The servo motor disorder elimination control method as described in claim 1, characterized in that, The construction and online learning of the deep state estimator in step S21 are detailed below: The deep state estimator takes the historical normal state sequence X_{nk:n-1} = [x(nk), ..., x(n-1)] as input and outputs the state prediction value x̂(n) for the current period.

6. The servo motor disorder elimination control method as described in claim 1, characterized in that, The seamless data reconstruction and control quantity fusion in step S22 are detailed below: Input the historical normal state sequence X_{nk:n-1} = [x(nk), ...,x(n-1)] into the deep state estimator to generate the state prediction value x̂(n); The following fusion strategy is used to generate control quantities: u(n) =α u_fb(x̂(n)) + (1-α) u_ff(x̂(n)); Where u_fb is the feedback control quantity based on the predicted state, u_ff is the feedforward control quantity based on the predicted state, and α is the fusion weight dynamically adjusted according to the prediction uncertainty.

7. The servo motor disorder elimination control method as described in claim 1, characterized in that, Step S31, involving the rotation of high-frequency voltage vector injection and full-frequency response acquisition, is detailed below: A high-frequency voltage vector ΔU_h with constant amplitude, frequency ω_h, and rotating at an angular velocity ω_r is injected into the dq coordinate system. ΔU_dh = V_h cos(ω_h t +ω_r t); ΔU_qh = V_h sin(ω_h t +ω_r t); The high-frequency current responses ΔI_dh and ΔI_qh of the dq axis are acquired synchronously.

8. The servo motor disorder elimination control method as described in claim 1, characterized in that, Step S32, dynamic impedance ellipse fitting and feature extraction, is detailed below: Within one rotation period T_r = 2π / ω_r, the high-frequency current response [ΔI_dh, ΔI_qh] is plotted as an elliptic scatter plot; the center position, major axis / minor axis length (L_max, L_min) and tilt angle φ of the elliptic scatter plot are extracted by an elliptic fitting algorithm.

9. The servo motor disorder elimination control method as described in claim 1, characterized in that, The multi-dimensional error decoupling and real-time compensation in step S33 are detailed below: The tilt angle φ of the elliptic scatter plot directly corresponds to the rotor spatial phase error that needs to be compensated. ; according to Dynamic compensation is performed on the rotor position angle used for feedback and transformation: θ_corrected = θ_measured + .

Images