Servo driver control method

Abstract

The invention relates to the technical field of servo drivers, in particular to a servo driver control method, and the technical scheme comprises the steps: carrying out the online identification of load inertia and external disturbance through a self-adaptive observer, generating a dynamic compensation parameter, constructing a multi-model parallel self-adaptive observer group, and carrying out the online identification of the load inertia and external disturbance through the self-adaptive observer group, a dynamic confidence evaluation mechanism is designed, disturbance dynamic compensation is realized, parameter sudden change and slow change scenes can be taken into consideration, the real-time performance of dynamic response can be ensured, compensation parameters are input into a fuzzy neural network controller, an optimal control quantity is generated in combination with a preset control target, a multi-source compensation parameter input channel is constructed, and the optimal control quantity is generated. A dynamic input weighting module is designed, fuzzy rule confidence is updated through an online strategy gradient algorithm to generate an optimized control quantity, the working condition adaptability can be enhanced, a dynamic target can be dealt with, finally, a sliding mode variable structure algorithm is adopted to carry out high-frequency buffeting suppression on the control quantity, and a final driving signal is output to a power module. And the control precision of the servo driver is improved.

Description

Technical Field

[0001] This invention relates to the field of servo drive technology, and in particular to a servo drive control method. Background Technology

[0002] Servo drives precisely control current and voltage to drive servo motors, thereby achieving precise control of position, speed, and torque. At its core is a digital signal processor (DSP), capable of executing complex control algorithms to achieve digital, networked, and intelligent control. For power devices, intelligent power modules are typically used, integrating drive circuitry and various protection functions. Currently, servo drive control methods using a single observer cannot handle both abrupt and gradual parameter changes. Fixed-weight fusion strategies result in lag in dynamic response and lack a mechanism to verify the reliability of parameter estimation. Static fusion of compensation parameters leads to poor adaptability to operating conditions, and the fixed fuzzy rule base cannot cope with dynamic targets, resulting in low control accuracy for servo drives.

[0003] In view of this, we propose a servo drive control method to solve the existing problems. Summary of the Invention

[0004] The purpose of this invention is to provide a servo drive control method to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a servo driver control method, comprising the following steps:

[0006] S1: Real-time acquisition of motor position signals, speed signals, and load torque signals from the servo system;

[0007] S2: The load inertia and external disturbances are identified online through an adaptive observer, and dynamic compensation parameters are generated.

[0008] S3: Input the compensation parameters into the fuzzy neural network controller and generate an optimized control quantity by combining it with the preset control target;

[0009] S4: The sliding mode variable structure algorithm is used to suppress high-frequency chattering of the control quantity and output the final drive signal to the power module.

[0010] Furthermore, the specific steps of S2 include:

[0011] A1: Construct a multi-model parallel adaptive observer group, which includes a Model Reference Adaptive Observer (MRAO), an Extended State Observer (ESO), and a Recursive Least Squares Observer (RLS). Among them, the Model Reference Adaptive Observer (MRAO) updates the inertia estimate online based on Lyapunov stability theory. The Extended State Observer (ESO) expands external disturbances into system states for joint estimation, while the Recursive Least Squares Observer (RLS) identifies time-varying disturbance parameters through a data-driven approach.

[0012] A2: Design a dynamic confidence assessment mechanism and calculate the output weights of each observer;

[0013] A3: Generate fusion compensation parameters And it is injected into the feedforward control channel to achieve dynamic disturbance compensation, where θ i For the i-th compensation parameter, w i For θ i The corresponding weights, where N is the number of observers.

[0014] Furthermore, the specific steps of S3 include:

[0015] B1: Construct a multi-source compensation parameter input channel to receive the inertia compensation amount ΔJ, disturbance compensation amount Δd, and system state feedback amount x from the adaptive observer;

[0016] B2: Design a dynamic input weighting module to adjust the input weights of each compensation parameter in real time according to system operating conditions. Where, Δθ i Let β be the short-term rate of change of the i-th compensation parameter, and β be the sensitivity coefficient.

[0017] B3: Embed a multi-objective optimization layer in the fuzzy neural network, transforming the preset control objective into a loss function: L=λ1·ITAE+λ2·||u||2+λ3·∑(Δθ-Δθ) ref ) 2 Where Δθ is the short-term rate of change of the compensation parameter, Δθ ref The reference value for compensating for the short-term rate of change of the parameter is u, which is the control variable, ITAE is the error integral criterion, and λ1, λ2, and λ3 are weight values.

[0018] B4: Update the confidence of fuzzy rules using an online policy gradient algorithm to generate the optimized control variable u. opt =FNN(w·Δθ,x), where w is the weight corresponding to Δθ.

[0019] Furthermore, in A1, the inertia update law of the Model Reference Adaptive Observer (MRAO) is: Where η is the adaptive gain, e ω =ω m -ω ref For rotational speed tracking error, K is the eigenvalue of the system stiffness matrix, and ω is the rotational speed tracking error. m For rotational speed, ω ref This is a reference value for the rotational speed.

[0020] Furthermore, in A1, the design of the Extended State Observer (ESO) includes: expanding the total disturbance d(t) into system state x3, and constructing an increased-dimensional state equation: Where h(t) is the system's impulse response function, J is the inertia value, and B is the input matrix; the disturbance is estimated using a nonlinear feedback law: Where β3 is the observer gain.

[0021] Furthermore, in A2, the dynamic confidence assessment mechanism includes: when a mutational load is detected, i.e. The RLS observer weights were temporarily increased to w. RLS ≥0.7, where δ J The deviation is J; in the steady state phase, i.e., ||e ω If ||<ε, then switch to MRAO dominant mode, i.e., w MRAO ≥0.8, where ε is the error threshold.

[0022] Furthermore, S2 also includes a parameter reliability verification module: comparing the theoretical power consumption. Compared with the actual measured power consumption P means If the deviation exceeds the threshold, the observer is reset, where α is the weight value; the consistency of the output distribution of each observer is evaluated by Wasserstein distance, and outlier estimates with deviation > 3σ are removed, where σ is the standard deviation of the data point from the mean.

[0023] Furthermore, in B2, the dynamic input weighting module includes: when a sudden disturbance is detected, i.e., ||Δd||>d th The disturbance compensation weight w d Increase to the preset maximum value w max , where d th This is the threshold for disturbance compensation; in the steady-state phase, i.e., ||e ω ||<ε, activates online identification and adaptive weight adjustment of inertia compensation parameter ΔJ.

[0024] Furthermore, in B3, the implementation of the multi-objective optimization layer includes: constructing the Pareto front solution set and solving for the optimal weight combination using the NSGA-III algorithm. Design a dynamic pruning mechanism for the fuzzy rule base to remove rules with confidence scores below the threshold or redundant contributions.

[0025] Furthermore, in B4, the online policy gradient algorithm is specifically defined as follows: Define the fuzzy rule confidence ρ k These are trainable parameters; the policy gradient is evaluated using Monte Carlo sampling. Where b is the baseline function, π(ρ) k ) represents the policy distribution.

[0026] Compared with the prior art, the beneficial effects of the present invention are:

[0027] This invention uses an adaptive observer to identify load inertia and external disturbances online, generating dynamic compensation parameters. Specifically, it constructs a multi-model parallel adaptive observer group, designs a dynamic confidence evaluation mechanism, and implements dynamic disturbance compensation. This not only accommodates both abrupt and gradual parameter changes but also ensures real-time dynamic response. Furthermore, the compensation parameters are input into a fuzzy neural network controller, which, combined with a preset control objective, generates optimized control quantities. This involves constructing a multi-source compensation parameter input channel, designing a dynamic input weighting module, and updating the fuzzy rule confidence using an online policy gradient algorithm to generate optimized control quantities. This not only enhances adaptability to operating conditions but also addresses dynamic objectives. Finally, a sliding mode variable structure algorithm is used to suppress high-frequency chattering in the control quantities, outputting the final drive signal to the power module, thereby improving the control accuracy of the servo driver. Attached Figure Description

[0028] Figure 1 This is a flowchart illustrating a servo driver control method according to the present invention. Detailed Implementation

[0029] The technical solution of the present invention will be further described below with reference to the accompanying drawings and specific embodiments.

[0030] Example 1

[0031] like Figure 1 As shown, a servo driver control method includes the following steps:

[0032] S1: Real-time acquisition of motor position signals, speed signals, and load torque signals from the servo system;

[0033] S2: The load inertia and external disturbances are identified online through an adaptive observer, and dynamic compensation parameters are generated.

[0034] S3: Input the compensation parameters into the fuzzy neural network controller and generate an optimized control quantity by combining it with the preset control target;

[0035] S4: The sliding mode variable structure algorithm is used to suppress high-frequency chattering of the control quantity and output the final drive signal to the power module.

[0036] The working principle of the servo driver control method based on Embodiment 1 is as follows:

[0037] The specific steps of S2 include:

[0038] A1: Construct a multi-model parallel adaptive observer group, which includes a Model Reference Adaptive Observer (MRAO), an Extended State Observer (ESO), and a Recursive Least Squares Observer (RLS). Among them, the Model Reference Adaptive Observer (MRAO) updates the inertia estimate online based on Lyapunov stability theory. The Extended State Observer (ESO) expands external disturbances into system states for joint estimation, while the Recursive Least Squares Observer (RLS) identifies time-varying disturbance parameters through a data-driven approach.

[0039] A2: Design a dynamic confidence assessment mechanism and calculate the output weights of each observer;

[0040] A3: Generate fusion compensation parameters And it is injected into the feedforward control channel to achieve dynamic disturbance compensation, where θ i For the i-th compensation parameter, w i For θ i The corresponding weights, where N is the number of observers.

[0041] In A1, the inertia update law for the Model Reference Adaptive Observer (MRAO) is: Where η is the adaptive gain, e ω =ω m -ω ref For rotational speed tracking error, K is the eigenvalue of the system stiffness matrix, and ω is the rotational speed tracking error. m For rotational speed, ω ref This is a reference value for the rotational speed.

[0042] Furthermore, in A1, the design of the Extended State Observer (ESO) includes: expanding the total disturbance d(t) into the system state x3, and constructing an increased-dimensional state equation: Where h(t) is the system's impulse response function, J is the inertia value, and B is the input matrix; the disturbance is estimated using a nonlinear feedback law: Where β3 is the observer gain.

[0043] In A2, the dynamic confidence assessment mechanism includes: when a mutational load is detected, i.e. The RLS observer weights were temporarily increased to w. RLS ≥0.7, where δ J The deviation is J; in the steady state phase, i.e., ||e ω If ||<ε, then switch to MRAO dominant mode, i.e., w MRAO ≥0.8, where ε is the error threshold.

[0044] Furthermore, S2 also includes a parameter reliability verification module: comparing the theoretical power consumption. Compared with the actual measured power consumption Pmeans If the deviation exceeds the threshold, the observer is reset, where α is the weight value; the consistency of the output distribution of each observer is evaluated by Wasserstein distance, and outlier estimates with deviation > 3σ are removed, where σ is the standard deviation of the data point from the mean.

[0045] The specific steps of S3 include:

[0046] B1: Construct a multi-source compensation parameter input channel to receive the inertia compensation amount ΔJ, disturbance compensation amount Δd, and system state feedback amount x from the adaptive observer;

[0047] B2: Design a dynamic input weighting module to adjust the input weights of each compensation parameter in real time according to system operating conditions. Where, Δθ i Let β be the short-term rate of change of the i-th compensation parameter, and β be the sensitivity coefficient.

[0048] B3: Embed a multi-objective optimization layer in the fuzzy neural network, transforming the preset control objective into a loss function: L=λ1·ITAE+λ2·||u||2+λ3·∑(Δθ-Δθ) ref ) 2 Where Δθ is the short-term rate of change of the compensation parameter, Δθ ref The reference value for compensating for the short-term rate of change of the parameter is u, which is the control variable, ITAE is the error integral criterion, and λ1, λ2, and λ3 are weight values.

[0049] B4: Update the confidence of fuzzy rules using an online policy gradient algorithm to generate the optimized control variable u. opt =FNN(w·Δθ,x), where w is the weight corresponding to Δθ.

[0050] In B2, the dynamic input weighting module includes: when a sudden disturbance is detected, i.e., ||Δd||>d th The disturbance compensation weight w d Increase to the preset maximum value w max , where d th This is the threshold for disturbance compensation; in the steady-state phase, i.e., ||e ω ||<ε, activates online identification and adaptive weight adjustment of inertia compensation parameter ΔJ.

[0051] In B3, the implementation of the multi-objective optimization layer includes: constructing the Pareto front solution set and solving for the optimal weight combination using the NSGA-III algorithm. Design a dynamic pruning mechanism for the fuzzy rule base to remove rules with confidence scores below the threshold or redundant contributions.

[0052] In B4, the online policy gradient algorithm is specifically defined as follows: Define the fuzzy rule confidence ρk These are trainable parameters; the policy gradient is evaluated using Monte Carlo sampling. Where b is the baseline function, π(ρ) k ) represents the policy distribution.

[0053] The specific steps of S4 include:

[0054] C1: Combining the traditional exponential tendency law with the power tendency law, the expression is: Among them, 0<α<1, ε>0, k>0, Δ d This is a compensation term for the disturbance observer;

[0055] C2: Construct a dual-channel sliding surface, including the primary sliding surface s1=e+c1∫edt and the auxiliary sliding surface. Cooperative control of two sliding surfaces is achieved through a dynamic coupling factor ρ(t);

[0056] C3: An adaptive boundary layer thickness adjustment mechanism is adopted to adjust the saturation function parameter φ(t) in real time according to the system state. Where φ0 is the initial thickness and γ is the adaptive gain;

[0057] C4: Processes the power module switching characteristics through the dead-time compensation module to generate the final drive signal u. final =u smc +u deadtime , where u smc For sliding mode control signal, u deadtime This is the dead time signal.

[0058] In C1, the perturbation observer compensation term Δ d The update law, estimated in real time using the Extended State Observer (ESO), is: Δ d =β1(z1-x1)+β2(z2-x2), where z1 and z2 are the ESO observation states, and β1 and β2 are the observer gain coefficients.

[0059] In C2, the dynamic coupling factor ρ(t) is adjusted by the fuzzy inference engine, with the input variables being the sliding surface distance |s1| and the error rate of change. The output rule is: when |s1|>s th When ρ→1, the principal sliding surface dominates; when |s1|≤s th When ρ→0, the auxiliary sliding surface suppresses chattering; where s th This is the threshold value for the distance between the sliding surfaces.

[0060] The above specific embodiments are merely several preferred embodiments of the present invention. Based on the technical solutions of the present invention and the relevant teachings of the above embodiments, those skilled in the art can make various alternative improvements and combinations to the above specific embodiments.

Claims

1. A servo driver control method, characterized in that, The operating steps include: S1: Real-time acquisition of motor position signals, speed signals, and load torque signals from the servo system; S2: The load inertia and external disturbances are identified online through an adaptive observer, and dynamic compensation parameters are generated. S3: Input the compensation parameters into the fuzzy neural network controller and generate an optimized control quantity by combining it with the preset control target; S4: The sliding mode variable structure algorithm is used to suppress high-frequency chattering of the control quantity and output the final drive signal to the power module.

2. The servo driver control method according to claim 1, characterized in that, The specific steps of S2 include: A1: Construct a multi-model parallel adaptive observer group, which includes a Model Reference Adaptive Observer (MRAO), an Extended State Observer (ESO), and a Recursive Least Squares Observer (RLS). Among them, the Model Reference Adaptive Observer (MRAO) updates the inertia estimate online based on Lyapunov stability theory. The Extended State Observer (ESO) expands external disturbances into system states for joint estimation, while the Recursive Least Squares Observer (RLS) identifies time-varying disturbance parameters through a data-driven approach. A2: Design a dynamic confidence assessment mechanism and calculate the output weights of each observer; A3: Generate fusion compensation parameters And it is injected into the feedforward control channel to achieve dynamic disturbance compensation, where θ i For the i-th compensation parameter, w i For θ i The corresponding weights, where N is the number of observers.

3. The servo driver control method according to claim 1, characterized in that, The specific steps of S3 include: B1: Construct a multi-source compensation parameter input channel to receive the inertia compensation amount ΔJ, disturbance compensation amount Δd, and system state feedback amount x from the adaptive observer; B2: Design a dynamic input weighting module to adjust the input weights of each compensation parameter in real time according to system operating conditions. Where, Δθ i Let β be the short-term rate of change of the i-th compensation parameter, and β be the sensitivity coefficient. B3: Embed a multi-objective optimization layer in the fuzzy neural network, transforming the preset control objective into a loss function: L=λ1·ITAE+λ2·||u||2+λ3·∑(Δθ-Δθ) ref ) 2 Where Δθ is the short-term rate of change of the compensation parameter, Δθ ref The reference value for compensating for the short-term rate of change of the parameter is u, which is the control variable, ITAE is the error integral criterion, and λ1, λ2, and λ3 are weight values. B4: Update the confidence of fuzzy rules using an online policy gradient algorithm to generate the optimized control variable u. opt =FNN(w·Δθ,x), where w is the weight corresponding to Δθ.

4. The servo driver control method according to claim 2, characterized in that: In A1, the inertia update law for the Model Reference Adaptive Observer (MRAO) is: Where η is the adaptive gain, e ω =ω m -ω ref For rotational speed tracking error, K is the eigenvalue of the system stiffness matrix, and ω is the rotational speed tracking error. m For rotational speed, ω ref This is a reference value for the rotational speed.

5. A servo driver control method according to claim 2, characterized in that, In A1, the design of the Extended State Observer (ESO) includes: expanding the total disturbance d(t) into system state x3, and constructing an increased-dimensional state equation: Where h(t) is the system's impulse response function, J is the inertia value, and B is the input matrix; the disturbance is estimated using a nonlinear feedback law: Where β3 is the observer gain.

6. The servo driver control method according to claim 2, characterized in that, In A2, the dynamic confidence assessment mechanism includes: when a mutational load is detected, i.e. The weight of the RLS observer was temporarily increased to w. RLS ≥0.7, where δ J The deviation is J; in the steady state phase, i.e., ||e ω If ||<ε, then switch to MRAO dominant mode, i.e., w MRAO ≥0.8, where ε is the error threshold.

7. The servo driver control method according to claim 2, characterized in that, S2 also includes a parameter reliability verification module: comparing theoretical power consumption. Compared with the actual measured power consumption P means If the deviation exceeds the threshold, the observer is reset, where α is the weight value; the consistency of the output distribution of each observer is evaluated by Wasserstein distance, and outlier estimates with deviation > 3σ are removed, where σ is the standard deviation of the data point from the mean.

8. A servo driver control method according to claim 3, characterized in that, In B2, the dynamic input weighting module includes: when a sudden disturbance is detected, i.e., ||Δd||>d th The disturbance compensation weight w d Increase to the preset maximum value w max , where d th This is the threshold for disturbance compensation; in the steady-state phase, i.e., ||e ω ||<ε, activates online identification and adaptive weight adjustment of inertia compensation parameter ΔJ.

9. A servo driver control method according to claim 3, characterized in that, In B3, the implementation of the multi-objective optimization layer includes: constructing the Pareto front solution set and solving for the optimal weight combination using the NSGA-III algorithm. Design a dynamic pruning mechanism for the fuzzy rule base to remove rules with confidence scores below the threshold or redundant contributions.

10. A servo driver control method according to claim 3, characterized in that, In B4, the online policy gradient algorithm is specifically defined as follows: Define the fuzzy rule confidence ρ k These are trainable parameters; the policy gradient is evaluated using Monte Carlo sampling. Where b is the baseline function, π(ρ k ) represents the policy distribution.

Images