A motor sliding table control parameter optimization method based on an environment perception agent model

Abstract

The application discloses a motor sliding table control parameter optimization method based on an environment perception agent model and relates to the field of motor servo control. The method improves the traditional control method from two aspects of constructing an observable environment quantity and assisting parameter setting of an agent model. First, a motor-ball screw sliding table model considering environment degradation is established, and a sample library covering multiple control parameter pools and multiple degradation levels is constructed. Then, a standard section running with fixed control parameters is inserted as a degradation probe after each working section, and an observable environment quantity is constructed according to the servo driving quantity response of the standard section. Then, an agent model is trained with control parameters, stroke information and the environment quantity as inputs and multiple performance indexes as outputs. Finally, the agent model is used to predict and score the performance of candidate control parameters under threshold constraints, and the optimal control parameter is output. The method can re-set the control parameters under the degradation condition, and takes into account the control accuracy, energy consumption and safety.

Description

Technical Field

[0001] This invention relates to the field of motor servo control, and in particular to a method for optimizing motor slide control parameters based on an environment-aware agent model. Background Technology

[0002] The motor-ball screw slide system is a typical transmission-actuator in automated machining units and CNC feed systems, and its positioning accuracy and tracking performance directly affect machining quality. However, under long-term operation and changing environmental conditions, the system's friction, load, damping, and stiffness will change. Fixed control parameters are difficult to balance control accuracy, energy consumption, and safety under different degradation conditions. To maintain the system's overall performance at a better level, the control parameters need to be retuned according to the system status.

[0003] In the field of servo system parameter tuning, manual experience-based tuning relies on expert knowledge and struggles to cover complex degradation states. Single-condition parameter optimization is mostly geared towards healthy environments and steady-state objects, lacking adaptability to degradation processes. Intelligent optimization algorithms typically require extensive high-precision simulations and experimental evaluations, while reinforcement learning suffers from weak interpretability and difficulty in ensuring training security. Regarding condition monitoring, data-driven methods are primarily used for fault diagnosis, health monitoring, and wear prediction, rarely applying health status further to control parameter retuning. How to use observable degradation states as conditions, mapping them uniformly with travel information and control parameters to multiple performance indicators, and thereby achieving parameter retuning for degradation conditions, remains a pressing problem to be solved. Summary of the Invention

[0004] To address the aforementioned shortcomings, this invention provides a method for optimizing motor slide control parameters based on an environmental perception proxy model, which balances control accuracy, energy consumption, and safety under degraded operating conditions.

[0005] The technical solution adopted by the present invention to achieve the above objectives is as follows:

[0006] A method for optimizing motor slide control parameters based on an environment-aware agent model includes the following steps:

[0007] 1) Establish a motor-ball screw slide model that takes into account environmental degradation;

[0008] 2) Construct a sample library covering multiple control parameter pools and multiple degradation levels;

[0009] 3) Using control parameters, travel information, and environmental quantities from the sample library as inputs and multiple performance indicators as outputs, construct and train the surrogate model.

[0010] 4) Under threshold constraints, the performance of candidate control parameters is predicted using a surrogate model and scored using an evaluation function to output the optimal control parameters;

[0011] 5) Apply the optimal control parameters to the motor slide to optimize the control parameters of the motor slide under degraded conditions.

[0012] The motor-ball screw slide model consists of a motor drive, a position loop, a speed loop, a current loop control loop, voltage / current limiting, and a ball screw-slide mechanical transmission. A degradation factor is constructed to characterize the degree of degradation in an unobservable environment. A degradation-related model is also constructed, consisting of viscous damping, Coulomb friction, and stator resistance. The parameters in the degradation-related model all increase with the degradation factor.

[0013] The degradation factor Specifically:

[0014]

[0015] in, For the maximum degradation, is the time constant.

[0016] The degradation-related model is specifically as follows:

[0017]

[0018]

[0019]

[0020]

[0021] in, , , These represent the viscous damping, Coulomb friction, and stator resistance under healthy conditions, respectively. , , The degradation sensitivity coefficient, For mechanical angular velocity, For smoothing parameters, This is the frictional torque.

[0022] The sample library consists of a control parameter pool and a random combination of degradation levels. Each sample contains several working segments, each followed by a standard segment as a degradation probe. The working segments operate using working control parameters, and the standard segments operate using fixed standard parameters.

[0023] The standard segment consists of a first buffer segment, a forward standard segment, a second buffer segment, and a reverse standard segment connected in sequence. The environmental quantities are constructed from the servo drive responses of the standard segments, specifically:

[0024] Let the absolute average values ​​of the control voltage on the forward and reverse standard sections be respectively , The absolute mean values ​​of the q-axis currents are respectively , Then the environmental quantity is:

[0025]

[0026]

[0027] in, The average control voltage for the standard section. This represents the average q-axis current of the standard segment. , It is a quantity that is asymmetrical in both directions.

[0028] The surrogate model employs a multilayer perceptron, and its input includes five control parameters. Including: position loop proportional gain Velocity feedforward , acceleration feed forward Velocity loop proportional gain Velocity loop integral gain 6 items of itinerary information Including travel length runtime Average speed Sports Mode Direction of movement and short and fast signs and 4 environmental quantities , , , The output includes tracking error. Endpoint error Overshoot Control voltage peak q-axis current peak Average absolute power With cumulative absolute electrical energy .

[0029] Step 4) specifically refers to:

[0030] Using the environmental parameters at the start of each test window as inputs to the degradation state of the next three working segments within that window, a surrogate model predicts the performance indicators of each set of candidate control parameters in each future working segment. These indicators are then scored according to an evaluation function, and the candidate control parameter set with the lowest score is selected. The evaluation function... for:

[0031]

[0032]

[0033]

[0034]

[0035]

[0036] in, The constraint penalty term is 1 when the predictive performance of the candidate control parameter does not meet the normal threshold constraint, and 0 when the constraint is met. This represents the maximum normalized deficit among the three segments. Let be the predicted value of the k-th performance index for the j-th segment by the surrogate model. This represents the normal threshold corresponding to the k-th performance metric. This represents the cost of average control force across three segments.

[0037] The candidate control parameters are generated by random search, Bayesian optimization, particle swarm optimization, or differential evolution.

[0038] A motor slide control parameter optimization system based on an environment-aware agent model includes:

[0039] The simulation model building module is used to build a motor-ball screw slide model that takes into account environmental degradation.

[0040] The sample library construction module is used to build a sample library covering multiple control parameter pools and multiple degradation levels;

[0041] The surrogate model module is used to build and train a surrogate model by taking control parameters, travel information and environmental variables from the sample library as inputs and multiple performance indicators as outputs.

[0042] The parameter tuning module is used to predict the performance of candidate control parameters under threshold constraints using a surrogate model and score them through an evaluation function, and output the optimal control parameters.

[0043] The control execution module is used to apply the optimal control parameters to the motor slide to optimize the control parameters of the motor slide under degraded conditions.

[0044] The present invention has the following beneficial effects and advantages:

[0045] 1. Using a standard segment with fixed control parameters as a degradation probe, an observable environmental quantity is constructed using only its servo drive response. This allows for characterization of the system degradation state without relying on the true value of the degradation factor, and is less affected by the control parameters of the working segment.

[0046] 2. By establishing a mapping relationship between environmental quantities, travel information, control parameters and multiple performance indicators through a surrogate model, it is possible to predict multiple performance indicators of candidate control parameters at a lower cost, avoiding reliance on expensive high-precision simulation and experimental evaluation.

[0047] 3. Under threshold constraints, candidate control parameters are scored and screened by relying on surrogate models and evaluation functions, avoiding the need to test candidate parameters one by one in a large number of real systems. This allows for a trade-off between control accuracy and energy consumption while meeting preset safety constraints.

[0048] 4. By placing degradation state characterization, performance prediction, and parameter selection within the same optimization framework, control parameters can be adjusted specifically according to different degradation conditions, thus possessing a certain degree of adaptability to operating conditions. Attached Figure Description

[0049] Figure 1 This is a system principle block diagram of the present invention;

[0050] Figure 2 This is a schematic diagram of the standard segment degradation probe and environmental quantity structure of the present invention;

[0051] Figure 3 This is a flowchart of the control parameter tuning process based on the surrogate model of the present invention. Detailed Implementation

[0052] The present invention will now be described in further detail with reference to the accompanying drawings and embodiments.

[0053] A method for optimizing motor slide control parameters based on an environment-aware agent model includes the following steps:

[0054] A motor-ball screw slide model considering environmental degradation was established, and a sample library covering multiple control parameter pools and multiple degradation levels was constructed.

[0055] A standard segment running with fixed control parameters is inserted after each working segment as a degradation probe, and an observable environmental quantity is constructed based on the servo drive response of the standard segment.

[0056] The agent model is trained by taking control parameters, travel information, and environmental quantities as inputs and multiple performance indicators as outputs.

[0057] Under threshold constraints, the surrogate model is used to predict and score the performance of candidate control parameters and output the optimal control parameters.

[0058] The optimal control parameters are applied to the motor slide to optimize the control parameters of the motor slide under degraded conditions.

[0059] The proposed model establishes a motor-ball screw slide table that considers environmental degradation, using a degradation factor to characterize the degree of unobservable environmental degradation. This degradation factor increases exponentially with time. Viscous damping, Coulomb friction, and stator resistance all increase with the degradation factor.

[0060]

[0061]

[0062]

[0063]

[0064]

[0065] in, As a degradation factor, For the maximum degradation, It is a time constant; , , These represent the viscous damping coefficient, Coulomb friction coefficient, and initial stator resistance under healthy conditions, respectively. , , The degradation sensitivity coefficient, For mechanical angular velocity, For smoothing parameters, This is the frictional torque.

[0066] The standard segment consists of a first buffer segment, a forward standard segment, a second buffer segment, and a reverse standard segment connected sequentially, and operates using fixed control parameters. The observable environmental quantities are:

[0067]

[0068]

[0069] in, , These are the absolute average control voltages for the forward and reverse standard sections, respectively. , These are the absolute mean values ​​of the q-axis current for the forward and reverse standard segments, respectively. , Characterizes the amount of drive required to complete the standard segment. , It is an asymmetric quantity in both positive and negative states.

[0070] The proxy model takes a 15-dimensional vector as input and outputs a 7-dimensional vector.

[0071]

[0072]

[0073]

[0074]

[0075] in, For the control parameter vector, including the position loop proportional gain Velocity feedforward , acceleration feed forward Velocity loop proportional gain and velocity loop integral gain ; For trip information, including trip length runtime Average speed Sports Mode Direction of movement and short and fast signs ; , , , The environmental quantity is described above. , , , , , , These are tracking error, endpoint error, overshoot, peak control voltage, peak q-axis current, average absolute power, and cumulative absolute energy, respectively. The surrogate model employs a multilayer perceptron.

[0076] The environmental quantities input to the proxy model are filtered. The process is as follows: the environmental quantities of each trajectory are subjected to Hampel robust filtering and moving average in time sequence to obtain the filtered environmental quantities.

[0077] The process of using the surrogate model to predict and score the performance of candidate control parameters under threshold constraints employs the following evaluation function to select the candidate control parameter group with the lowest score:

[0078]

[0079] The candidate control parameters are generated by random search, Bayesian optimization, particle swarm optimization, or differential evolution.

[0080] The present invention also provides a motor slide control parameter optimization system based on an environment-aware surrogate model, comprising: a sample library construction module, an environmental quantity construction module, a surrogate model module, a parameter tuning module, and a control execution module.

[0081] Example

[0082] This invention provides a method for optimizing motor slide control parameters based on an environment-aware agent model, the system principle block diagram of which is shown below. Figure 1 As shown, the main components include: a simulation model of a motor-ball screw-slide table, multi-condition data acquisition, preprocessing and environmental quantity extraction, a state-performance proxy model, control parameter optimization, and simulation window playback verification. Each part is described in detail below.

[0083] I. Motor-Ball Screw Slide Model and Environmental Degradation

[0084] The simulation model in this embodiment consists of a motor drive, position loop, speed loop, current loop control loop, voltage / current limiting, and ball screw-slide table mechanical transmission. Considering environmental degradation during long-term motor operation, a degradation factor that increases exponentially with time is introduced. As shown in equation (1):

[0085] (1)

[0086] in, For the maximum degradation, The time constant is used. This embodiment uses different... Four degradation levels are generated: healthy, mild, moderate, and severe. The degradation factor is only used to construct the simulated degradation environment and is not used as input to the surrogate model.

[0087] Viscous damping, Coulomb friction, and stator resistance all increase with the degradation factor, as shown in equations (2) to (4), and the frictional torque... The expression is shown in equation (5):

[0088] (2)

[0089] (3)

[0090] (4)

[0091] (5)

[0092] in, , , These represent the viscous damping, Coulomb friction, and initial stator resistance under healthy conditions, respectively. , , The degradation sensitivity coefficient, For mechanical angular velocity, For smoothing parameters. As degradation deepens, the frictional torque increases, the system drive margin decreases, and the original tuning parameters tend to be insufficient; at the same time, the increase in stator resistance will reduce the current response capability, and a higher voltage is required to maintain the same current, thereby affecting the voltage peak, current peak and energy consumption indicators. Equations (2) to (5) together constitute the degradation-related model.

[0093] Ignoring flexible deformation, the mechanical motion equations and electrical equations are shown in equations (6) to (8):

[0094] (6)

[0095] (7)

[0096] (8)

[0097] in, For equivalent rotational inertia, The torque constant of the motor is... For q-axis current, For external load torque, For the displacement of the slide table, For the lead screw, For mechanical rotation, This is the gap nonlinear displacement term. For equivalent inductance, This is the q-axis control voltage. This represents the back electromotive force term. Mechanical and electrical equations, along with degradation-related models, jointly describe the impact of environmental degradation on the servo system.

[0098] In this embodiment, the adjustable control parameter is taken as a 5-dimensional vector:

[0099] (9)

[0100] These are the position loop proportional gain, velocity feedforward, acceleration feedforward, velocity loop proportional gain, and velocity loop integral gain, respectively; the position loop integral and differential gains are fixed at 0 in this embodiment.

[0101] II. Multi-condition sample library

[0102] This embodiment constructs a sample library covering multiple control parameter pools and multiple degradation levels. In addition to samples with superior performance, it also includes samples with insufficient control parameters, performance degradation, and samples near the safety boundary. The simulation slide travel range is 0–1000 mm. Each sample contains several working segments, each followed by a standard segment with fixed parameters as a degradation probe. The working segments operate using working control parameters, while the standard segments operate using fixed standard parameters. Based on performance index thresholds, samples can be divided into three categories: normal, degraded but safe, and unsafe, thus providing a data foundation for surrogate model learning and parameter tuning.

[0103] III. Construction of Standard Segment Degenerate Probe and Environmental Quantities

[0104] like Figure 2 As shown, the observable environmental quantities are constructed from the servo drive responses of the fixed-parameter standard segments following each working segment, serving as observable proxy quantities for the system's degradation state. Each standard segment consists of a first buffer segment, a forward standard segment, a second buffer segment, and a reverse standard segment connected sequentially. Let the absolute average control voltages on the forward and reverse standard segments be respectively... , The absolute mean values ​​of the q-axis currents are respectively , The environmental quantities are defined as shown in equations (10) and (11):

[0105] (10)

[0106] (11)

[0107] in, The average control voltage for the standard section. The average q-axis current of the standard section reflects the amount of drive required to complete the standard section under fixed parameters. , The environmental quantities are asymmetrical, reflecting issues such as gaps. For each trajectory, the environmental quantities are sequentially subjected to Hampel robust filtering and moving average, yielding the filtered environmental quantities. Since the same fixed-parameter standard segment is used as a probe under different degradation environments, the environmental quantities can more accurately reflect the system's degradation state and are less affected by the control parameters of the working segment, thus providing more accurate degradation information for the surrogate model.

[0108] IV. State-Performance Proxy Model

[0109] Replacing expensive simulation evaluation with a surrogate model is an effective way to handle high-dimensional and high-cost optimization. In this embodiment, the input of the surrogate model is a 15-dimensional vector, and the output is a 7-dimensional vector, as shown in equations (12) to (14):

[0110] (12)

[0111] (13)

[0112] (14)

[0113] The input includes 5 control parameters. 6 items of itinerary information (journey length) runtime Average speed Sports Mode Direction of movement and short and fast signs ) and 4 environmental quantities , , , The output includes tracking error. Endpoint error Overshoot Control voltage peak q-axis current peak Average absolute power With cumulative absolute electrical energy The surrogate model employs a multilayer perceptron with three hidden layers, SiLU activation, batch normalization followed by Dropout after the first two hidden layers, and the target variable undergoes a non-negative transformation before standardization. It is trained using SmoothL1 loss and the AdamW optimizer with early stopping. Since the overshoot is close to 0 and non-zero values ​​are sparse on most samples, it is treated as a threshold constraint indicator.

[0114] V. Control Parameter Tuning and Verification

[0115] like Figure 3 As shown, the process for tuning the control parameters is as follows: Considering the short duration of the short window and the relatively slow change of the degradation state within the window, only the current environmental quantity measured at the beginning of the window is used as the common degradation state input for each future segment in each test window; for each set of candidate control parameters, the performance index of each future segment is predicted by the surrogate model and the evaluation function is calculated, and the smaller the evaluation function, the better. The evaluation function is shown in Equation (15):

[0116] (15)

[0117] In the formula, where, The penalty term is 1 when the prediction performance of the candidate control parameter does not meet the normal threshold constraint, and 0 when it meets the constraint; valid indicates that the prediction of all future segments of the candidate control parameter is within the normal threshold, as shown in equation (16):

[0118] (16)

[0119] When candidate control parameters do not meet the normal threshold constraints, a large penalty will be imposed and they will be preferentially excluded. The maximum value of the normalized deficit in the three segments is shown in equations (17) and (18):

[0120] (17)

[0121] (18)

[0122] in, This represents the predicted value of the proxy model for the k-th performance index in the j-th segment. This represents the normal threshold corresponding to the k-th performance metric. The average control force cost for the three segments is shown in equation (19):

[0123] (19)

[0124] The candidate control parameters can be generated by random search, Bayesian optimization, particle swarm optimization, or differential evolution. During window-level simulation playback, only the parameter set with the smallest evaluation function is used to return to the simulation model for execution, in order to obtain and verify the actual performance indicators. Therefore, this invention enables retuning of control parameters under degraded conditions based on observable environmental quantities, ensuring safety while also considering control accuracy and energy consumption.

[0125] This embodiment also provides a motor slide control parameter optimization system based on an environment-aware surrogate model, including a sample library construction module, an environmental quantity construction module, a surrogate model module, a parameter tuning module, and a control execution module. Each module is used to implement the corresponding steps of the above method.

[0126] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for optimizing motor slide control parameters based on an environment-aware agent model, characterized in that, Includes the following steps: 1) Establish a motor-ball screw slide model that takes into account environmental degradation; 2) Construct a sample library covering multiple control parameter pools and multiple degradation levels; 3) Using control parameters, travel information, and environmental quantities from the sample library as inputs and multiple performance indicators as outputs, construct and train the surrogate model. 4) Under threshold constraints, the performance of candidate control parameters is predicted using a surrogate model and scored using an evaluation function to output the optimal control parameters; 5) Apply the optimal control parameters to the motor slide to optimize the control parameters of the motor slide under degraded conditions.

2. The method for optimizing motor slide control parameters based on an environment-aware agent model according to claim 1, characterized in that, The motor-ball screw slide model consists of a motor drive, a position loop, a speed loop, a current loop control loop, voltage / current limiting, and a ball screw-slide mechanical transmission. A degradation factor is constructed to characterize the degree of degradation in an unobservable environment. A degradation-related model is also constructed, consisting of viscous damping, Coulomb friction, and stator resistance. The parameters in the degradation-related model all increase with the degradation factor.

3. The method for optimizing motor slide control parameters based on an environment-aware agent model according to claim 2, characterized in that, The degradation factor Specifically: in, For the maximum degradation, is the time constant.

4. The method for optimizing motor slide control parameters based on an environment-aware agent model according to claim 2, characterized in that, The degradation-related model is specifically as follows: in, , , These represent the viscous damping, Coulomb friction, and stator resistance under healthy conditions, respectively. , , The degradation sensitivity coefficient, For mechanical angular velocity, For smoothing parameters, This is the frictional torque.

5. The method for optimizing motor slide control parameters based on an environment-aware agent model according to claim 1, characterized in that, The sample library consists of a control parameter pool and a random combination of degradation levels. Each sample contains several working segments, each followed by a standard segment as a degradation probe. The working segments operate using working control parameters, and the standard segments operate using fixed standard parameters.

6. The method for optimizing motor slide control parameters based on an environment-aware agent model according to claim 5, characterized in that, The standard segment consists of a first buffer segment, a forward standard segment, a second buffer segment, and a reverse standard segment connected in sequence. The environmental quantities are constructed from the servo drive responses of the standard segments, specifically: Let the absolute average values ​​of the control voltage on the forward and reverse standard sections be respectively , The absolute mean values ​​of the q-axis currents are respectively , Then the environmental quantity is: in, The average control voltage for the standard section. This represents the average q-axis current of the standard segment. , It is a quantity that is asymmetrical in both directions.

7. The method for optimizing motor slide control parameters based on an environment-aware agent model according to claim 1, characterized in that, The surrogate model employs a multilayer perceptron, and its input includes five control parameters. Including: position loop proportional gain Velocity feedforward , acceleration feed forward Velocity loop proportional gain Velocity loop integral gain 6 items of itinerary information Including travel length runtime Average speed Sports Mode Direction of movement and short and fast signs and 4 environmental quantities , , , The output includes tracking error. Endpoint error Overshoot Control voltage peak q-axis current peak value Average absolute power With cumulative absolute electrical energy .

8. The method for optimizing motor slide control parameters based on an environment-aware agent model according to claim 1, characterized in that, Step 4) specifically involves: Using the environmental parameters at the start of each test window as inputs to the degradation state of the next three working segments within that window, a surrogate model predicts the performance indicators of each set of candidate control parameters in each future working segment. These indicators are then scored according to an evaluation function, and the candidate control parameter set with the lowest score is selected. The evaluation function... for: in, The constraint penalty term is 1 when the predictive performance of the candidate control parameter does not meet the normal threshold constraint, and 0 when the constraint is met. This represents the maximum normalized deficit among the three segments. Let be the predicted value of the k-th performance index for the j-th segment by the surrogate model. This represents the normal threshold corresponding to the k-th performance metric. This represents the cost of average control force across three segments.

9. The method for optimizing motor slide control parameters based on an environment-aware agent model according to claim 1, characterized in that, The candidate control parameters are generated by random search, Bayesian optimization, particle swarm optimization, or differential evolution.

10. A motor slide control parameter optimization system based on an environment-aware agent model, characterized in that, include: The simulation model building module is used to build a motor-ball screw slide model that takes into account environmental degradation. The sample library construction module is used to build a sample library covering multiple control parameter pools and multiple degradation levels; The surrogate model module is used to build and train a surrogate model by taking control parameters, travel information and environmental variables from the sample library as inputs and multiple performance indicators as outputs. The parameter tuning module is used to predict the performance of candidate control parameters under threshold constraints using a surrogate model and score them through an evaluation function, and output the optimal control parameters. The control execution module is used to apply the optimal control parameters to the motor slide to optimize the control parameters of the motor slide under degraded conditions.

Images