A multi-working-condition calibration method for simulation parameters of an electromechanical system

Abstract

The present application relates to the technical field of simulation model calibration, and particularly relates to a kind of electromechanical system simulation parameter multi-working condition calibration method, comprising obtaining multi-working condition data, and the working condition segment division of multi-working condition data;Initial simulation model of electromechanical system is constructed, and multi-working condition residual matrix is generated;Parameter sensitivity identification is carried out based on the matrix, and parameter working condition contribution result is obtained;Contribution stability analysis is carried out, and parameter layering result is obtained;Calibration is carried out to layering result, and parameter calibration result is obtained;Playback verification is carried out, and parameter application boundary result is obtained;In combination with uncalibratable working condition identification result, multi-working condition calibration result is output.The present application decomposes multi-working condition data into fine-grained working condition segment, is not dependent on the condition of large-scale traversal calibration or experience parameter adjustment, is conducive to improving the calibration accuracy of electromechanical system simulation model under multi-working condition, effectively reduces the risk that different working condition errors mutually restrict each other in traditional method, parameter cross-working condition distortion.

Description

Technical Field

[0001] This invention relates to the field of simulation model calibration technology, and in particular to a multi-condition calibration method for simulation parameters of electromechanical systems. Background Technology

[0002] In the development of integrated joint modules for humanoid robots, the same joint module needs to operate under multiple working conditions. Its simulation model usually needs to describe the electromagnetic response of the motor, the transmission error of the reducer, and the bearing friction at the same time. However, the test data under different working conditions have different sensitivities to the same parameter. Some parameters can reduce the error after calibration under one working condition, but will cause the simulation deviation to increase under another working condition. Therefore, when the data of multiple working conditions are not completely consistent, the parameter sensitivity changes with the working condition, and some parameters have coupling conflicts, the simulation parameters of the electromechanical system are calibrated in layers. This makes it difficult for the calibrated simulation model to maintain physical consistency across working conditions and make limited corrections to errors at the same time.

[0003] Existing technologies typically involve first establishing an electromechanical simulation model of the joint module, then selecting several typical working conditions for bench testing; subsequently, comparing the test data with the simulation output, and using least squares, genetic algorithms, or manual parameter tuning to optimize key parameters, thereby reducing the error between the simulation output and the test curve.

[0004] In humanoid robot joint modules, low-speed, high-torque conditions mainly expose friction, clearance, and reducer elasticity issues, while high-speed, low-torque conditions mainly expose inertia, back electromotive force, drive delay, and sensor phase lag issues. Long-term operation conditions will introduce temperature rise, resulting in changes in resistance, lubrication status, and torque sensor drift. If existing calibration methods directly use all operating condition data to fit a set of parameters, they are prone to incorrectly absorbing the differences in operating conditions caused by temperature, load, and speed into the fixed parameters, resulting in a smaller error in one operating condition and a larger error in another. Summary of the Invention

[0005] This invention addresses the problem in existing electromechanical system simulation parameter calibration techniques when the same joint module operates under multiple conditions. Existing techniques typically combine all operating condition data for optimization or use the overall average residual for parameter fitting, which easily masks local error characteristics under different operating conditions. This leads to cross-condition distortion, where the error decreases after calibration in some conditions but increases in others. Furthermore, existing techniques usually output only one set of optimal parameters, making it difficult to effectively distinguish between stable intrinsic parameters across operating conditions and sensitive parameters that change with the operating conditions. This results in a model that fits well under test conditions but lacks reliability under new operating conditions, exhibiting poor stability of detection results in different application scenarios, and easily leading to parameter interference and the masking of real physical deviations by operating condition fluctuations. Therefore, this invention proposes a multi-condition calibration method for electromechanical system simulation parameters.

[0006] To achieve the above objectives, on the one hand, this invention proposes a multi-condition calibration method for simulation parameters of electromechanical systems, comprising:

[0007] Acquire multi-condition data, divide the multi-condition data into condition segments, and obtain a set of condition segments;

[0008] An initial simulation model of the electromechanical system is constructed, and the initial simulation model is run under each working condition segment of the set of working condition segments to generate a multi-working condition residual matrix.

[0009] Based on the multi-condition residual matrix, parameter sensitivity identification is performed on each simulation parameter in the electromechanical system to obtain the parameter condition contribution results.

[0010] Based on the parameter operating condition contribution results, a contribution stability analysis of each simulation parameter in the electromechanical system is performed to obtain parameter stratification results.

[0011] The parameter stratification results are calibrated to obtain the parameter calibration results;

[0012] The parameter calibration results were replayed for verification to obtain the parameter applicability boundary results.

[0013] The parameter calibration results and parameter applicable boundary results are summarized to obtain multi-condition calibration results.

[0014] Preferably, the multi-condition data is divided into condition segments to obtain a set of condition segments, including:

[0015] The multi-condition data are arranged in the order of sampling points to form a sampling time series;

[0016] The output angular velocity, output torque, output angular acceleration, driver temperature, and output torque abrupt change are obtained point by point from the sampled time series.

[0017] Each sampling point is divided into velocity range, output torque range, acceleration range, commutation state, temperature range, and impact state according to a preset threshold, thus obtaining the sampling point type;

[0018] Intervals with consecutive sampling points of the same type are merged into a single working condition segment. When any dimensional condition changes, the current segment ends and a new segment begins, thus obtaining a set of working condition segments.

[0019] Preferably, based on the multi-condition residual matrix, parameter sensitivity identification is performed on each simulation parameter in the electromechanical system to obtain the parameter condition contribution results, including:

[0020] Obtain the allowable variation range of each simulation parameter in the electromechanical system, and perturb each parameter within its allowable variation range;

[0021] For each parameter disturbance, the disturbance parameters are re-simulated using the initial simulation model, the disturbance residuals corresponding to each working condition segment are calculated, and the disturbance residuals are compared with the output residuals in the multi-working condition residual matrix to obtain the residual changes of the disturbance on each working condition.

[0022] Based on the changes in residuals, determine the degree and direction of the influence of the simulation parameter on the corresponding working condition segment, and generate the parameter working condition contribution result;

[0023] The parameter condition contribution results are used to distinguish between inherent parameters across different operating conditions, condition-sensitive parameters, and parameters to be restricted.

[0024] Preferably, based on the residual changes, the degree and direction of the influence of the simulation parameter on the residuals under different operating conditions are determined, including:

[0025] For each disturbance parameter, calculate the residual change under each operating condition segment;

[0026] Based on the magnitude of the residual change, the sensitivity of each disturbance parameter is calculated by the ratio of the residual change to the parameter change, thereby determining the degree of influence of the simulation parameter on the corresponding working condition segment.

[0027] Based on the positive and negative directions of the sensitivity of the disturbance parameter, the direction of the influence of the simulation parameter on the corresponding working condition segment is determined. A positive sensitivity indicates that an increase in the parameter leads to an increase in the residual, and a negative sensitivity indicates that an increase in the parameter leads to a decrease in the residual. The direction of influence includes positive and negative changes.

[0028] The degree and direction of the influence are recorded as the contribution results of the parameter operating conditions.

[0029] Preferably, based on the parameter operating condition contribution results, a contribution stability analysis is performed on each simulation parameter in the electromechanical system to obtain parameter stratification results, including:

[0030] Extract the degree and direction of influence of each parameter under different working condition segments from the parameter contribution results;

[0031] For each simulation parameter, count the number of working condition segments in which the simulation parameter has a negative influence across all working condition segments, and calculate the proportion of this number to the total number of working condition segments to obtain the consistency ratio value.

[0032] For each simulation parameter, the sensitivity range ratio is calculated based on its degree of influence, and compared with the average sensitivity of the parameter under all operating conditions to obtain the sensitivity fluctuation value.

[0033] Based on the consistency ratio and sensitivity fluctuation values, the simulation parameters are divided into cross-condition inherent parameters, condition-sensitive parameters, or parameters to be restricted, resulting in parameter stratification results.

[0034] Preferably, based on the consistency ratio value and sensitivity fluctuation value, the simulation parameters are divided into cross-condition inherent parameters, condition-sensitive parameters, or parameters to be limited, including:

[0035] When the consistency ratio value is greater than or equal to the first ratio threshold and the sensitivity fluctuation value is less than the preset stability threshold, the simulation parameter is classified as a cross-condition inherent parameter.

[0036] When the consistency ratio value is less than the first ratio threshold and greater than or equal to the second ratio threshold, and the sensitivity fluctuation value is greater than or equal to the preset stability threshold, the simulation parameter is classified as a working condition sensitive parameter.

[0037] When the consistency ratio value is less than the second ratio threshold, the simulation parameter is classified as a parameter to be restricted.

[0038] Preferably, the parameter stratification results are calibrated to obtain parameter calibration results, including:

[0039] The parameter stratification results are obtained, including cross-condition inherent parameters, condition-sensitive parameters, and parameters to be restricted;

[0040] For the parameter items that are classified as cross-condition inherent parameters, the initial simulation model is called and run under all condition segments, and optimization is performed with the goal of minimizing the comprehensive residual of multiple condition segments, thereby obtaining the inherent parameter calibration results;

[0041] Under the condition that the inherent parameter calibration results remain unchanged, the working condition sensitive parameters are calibrated separately for each working condition, thereby obtaining the working condition correction parameter results;

[0042] The calibration results of the inherent parameters and the calibration results of the operating condition correction parameters are output as the parameter calibration results.

[0043] On the other hand, the multi-condition calibration method for simulation parameters also includes:

[0044] Preferably, the parameter calibration results are replayed for verification to obtain the parameter applicability boundary results, including:

[0045] Input the parameter calibration results into the initial simulation model, and select the verification condition to run the simulation.

[0046] The verification residuals of each output quantity are calculated for the verification conditions, and it is determined whether the verification residuals meet the preset acceptance threshold. Thus, each condition segment is divided into the parameter applicable range, the cautious application range, and the inapplicable range.

[0047] The parameters applicable boundary results are formed by summarizing each interval and its corresponding speed range, torque range, temperature range, commutation frequency and impact load range;

[0048] The applicable boundary results of the parameters can be used to guide the reliable invocation of multi-condition simulation models under different conditions, and to identify uncalibrable conditions.

[0049] Preferably, the verification residuals of each output quantity are calculated for the verification condition, and it is determined whether the verification residuals meet the preset acceptance threshold, including:

[0050] Obtain the verification condition segments, substitute the parameter configurations corresponding to each verification condition segment into the initial simulation model to run the simulation, calculate the verification output, and compare the verification output with the experimental output point by point to calculate the verification residual of each output.

[0051] For each verification condition segment, if all verification residuals in that verification condition segment are lower than the corresponding acceptance threshold, then that verification condition segment is classified as the parameter applicable range.

[0052] If the verification residual exceeds the acceptance threshold but the direction of residual change is explained by the condition-sensitive parameters, then the verification condition segment is classified as a cautious application range.

[0053] If the verification residual exceeds the acceptance threshold and the operating condition sensitive parameters need to exceed the allowable range of variation in order to reduce the error, then the verification operating condition segment is classified as an inapplicable interval.

[0054] Preferably, if the condition-sensitive parameter of the current working condition segment has reached the boundary of its allowable variation range, and the output residual of the segment is still greater than the preset residual threshold, the working condition segment is marked as a candidate uncalibrable working condition.

[0055] For a work condition segment marked as a candidate uncalibrable work condition, query its interval category in the parameter application boundary results:

[0056] If the operating condition segment is classified as an inapplicable range, it will be treated as the final uncalibrable operating condition, and an uncalibrable identification result will be generated.

[0057] If the operating condition is classified as an applicable range or a cautiously applicable range, its candidate mark will be removed, and the calibration will be considered successful. No uncalibrable prompt will be generated.

[0058] Compared with the prior art, the present invention has the following beneficial effects:

[0059] I. This invention decomposes multi-condition data into fine-grained condition segments and performs transferability stratification of simulation parameters through contribution stability analysis. This distinguishes simulation parameters into cross-condition inherent parameters and condition-sensitive parameters. To a certain extent, this avoids the problems of traditional methods that mix all condition errors in optimization and incorrectly attribute condition differences to fixed physical parameters. Without relying on large-scale traversal calibration or empirical parameter tuning, it retains the local error characteristics of each condition segment, which helps improve the calibration accuracy of electromechanical system simulation models under multiple conditions. It effectively reduces the risk of mutual constraint between different condition errors and parameter distortion across conditions in traditional unified optimization methods, while also ensuring that the calibration results have clear physical consistency and a basis for judging the applicable boundaries of the conditions.

[0060] Second, by performing multi-condition playback verification on the calibrated parameters, this invention further evaluates the applicable boundaries of the parameters based on the calibration results, effectively solving the problem that existing technologies only output a set of optimal parameters or parameters for different operating conditions. This enables engineers to clearly obtain more suitable simulation parameters for electromechanical systems, improves the predictability and generalization ability of simulation parameters under new operating conditions, and avoids simulation failures caused by parameter overfitting or improper operating condition extrapolation.

[0061] Third, this invention improves the ability to identify the sources of local errors under different speed, torque, temperature and impact conditions by dividing the working condition into segments and analyzing the sensitivity of parameters to the working conditions. On this basis, it merges the results to form a multi-working-condition segmentation, thereby avoiding the cross-working-condition distortion problem caused by mixing working condition differences with model structure errors in the prior art. This is conducive to improving the pertinence and reliability of electromechanical system simulation parameter calibration. Furthermore, by layering parameter transferability and providing prompts for non-calibrable working conditions, it further weakens the interference of working condition differences and model structure deficiencies on parameter calibration, thus making the applicable boundaries of the calibrated simulation parameters under multiple working conditions clearer. Attached Figure Description

[0062] Figure 1 This is a flowchart of the simulation parameter multi-condition calibration method of the present invention;

[0063] Figure 2 This is a comparison diagram of the effects of the present invention and the prior art. Detailed Implementation

[0064] To make the technical means, creative features, objectives, and effects of this invention easier to understand, the invention is further described below with reference to specific embodiments. However, the following embodiments are merely preferred embodiments of this invention and not all of them. Other embodiments obtained by those skilled in the art based on the embodiments described herein without creative effort are all within the protection scope of this invention.

[0065] Example 1:

[0066] To achieve the above objectives, please refer to Figure 1 This invention provides a multi-condition calibration method for simulation parameters of electromechanical systems, comprising:

[0067] Acquire multi-condition data, divide the multi-condition data into condition segments, and obtain a set of condition segments.

[0068] In this embodiment, it should be specifically noted that multi-condition data refers to the time-series measurement data synchronously collected through bench testing when the integrated joint module of the humanoid robot is operating in multiple different working states, specifically including:

[0069] The target electromechanical system is subjected to multi-condition testing, including but not limited to low-speed high-torque condition, high-speed low-torque condition, frequent start-stop condition, impact load condition, continuous temperature rise condition, and forward and reverse reversing condition.

[0070] The duration of each operating condition is determined by the test bench capacity and joint rating parameters to ensure at least three complete motion cycles are covered; if the operating condition is used for temperature rise calibration, it continues until the rate of temperature change is below a preset stability threshold, for example, the temperature change is less than 1°C within 5 consecutive minutes.

[0071] Under various operating conditions, multiple operating condition data are collected, including but not limited to motor current, motor voltage, torque output, joint input speed, joint output angle, output torque, driver temperature, reducer housing temperature, encoder position, control commands, and load torque.

[0072] It's important to know that multi-condition data acquisition is achieved by setting up a dedicated dynamic test bench. Specifically, motor current is acquired through a series sampling resistor and a differential amplifier; motor voltage is measured separately as DC bus voltage and phase voltage; output torque is obtained by either a differential voltage signal provided by a strain gauge torque sensor built into the joint module or through a series-connected external calibrated torque sensor; joint input speed and encoder position are read via bus communication using an absolute encoder integrated within the joint, and the joint input angular velocity is calculated; the joint output angle is directly given by the encoder position, and angular acceleration is calculated based on the encoder position; the driver temperature and reducer housing temperature are measured using NTC thermistors or K-type thermocouples in close contact with the heat sink. The system measures the surface of the plate and housing; control commands are recorded in real time by the host computer software, including timestamps and command values; load torque is acquired from the torque loop feedback data of the loading device or an independent torque sensor; all analog signals, such as current, voltage, torque, and temperature, are conditioned and then input into the analog input channel of the data acquisition card; digital signals, such as encoder and bus data, are synchronously acquired through a dedicated counter / timer module or industrial Ethernet; the data acquisition system uses a unified external clock to achieve multi-channel synchronous sampling, with sampling frequencies set according to signal bandwidth: current and voltage no less than 5kHz, torque and encoder no less than 2kHz, and temperature 10–100Hz; finally, all data are timestamped and stored in HDF5 or CSV format.

[0073] During the multi-condition data acquisition process, the angular velocity, angular acceleration, output torque, temperature, and sensor signals at the joint output end are collected in a continuous time series at a fixed sampling frequency. The measurement data at each sampling moment constitutes a sampling point. Based on the multi-condition data of the sampling points, the multi-condition data can be divided into condition segments by comparing it with the rated speed, rated torque, historical acceleration statistics, and ambient temperature.

[0074] It should be noted that each sampling point is a complete measurement vector of multi-condition data recorded at a uniform sampling frequency at a certain moment, including information such as motor current, voltage, output angle, output torque, and temperature.

[0075] As one implementation method:

[0076] During the development of the integrated joint module for humanoid robots, the module has a rated rotation speed of 120 rpm and a rated torque of 80 Nm. The joint module under test is mounted on a dynamic test bench equipped with a high-dynamic servo motor with a rated torque of 120 Nm as a loading device. The data acquisition card sampling rate is set to 2 kHz. Automated testing is performed in the following order:

[0077] Under low-speed, high-torque conditions, the joint module operates at 15 rpm (approximately 12.5% ​​of rated speed) while the loading device applies a load of 55 Nm (approximately 68.8% of rated torque) for 35 seconds, during which current, voltage, angle, torque, and temperature are recorded every 0.1 seconds. Under high-speed, low-torque conditions, the joint module operates at 100 rpm (approximately 83.3% of rated speed) while the loading device applies a load of 20 Nm (25% of rated torque) for 22 seconds.

[0078] In frequent start-stop conditions, the joint module moves according to a triangular wave velocity curve with a cycle of 0.5 seconds, with a peak speed of 60 rpm, and runs continuously for 60 cycles.

[0079] In the forward and reverse reversal operation, the joint module rotates forward at 60 rpm for 2 seconds and then immediately switches to reverse for 2 seconds, repeating 20 times.

[0080] Under impact load conditions, the joint module runs stably at 40 rpm. The loading device increases the load from 20 Nm to 120 Nm (150% of rated torque) in 10 ms, maintains it for 0.5 seconds, and then unloads it back to 20 Nm. This process is repeated 10 times.

[0081] Finally, a continuous temperature rise test was conducted. The joint module was continuously operated at 60 rpm and 40 Nm. The temperature of the driver and the reducer housing was recorded every minute until the temperature change was less than 1°C for 5 consecutive minutes (this stable condition was reached after about 12 minutes of actual operation).

[0082] All collected data were indexed by timestamps, resampled to 1kHz using linear interpolation, and saved as CSV files.

[0083] Furthermore, the multi-condition data is divided into condition segments to obtain a set of condition segments, specifically including:

[0084] The multi-condition data are arranged in the order of sampling points to form a sampling time series. Each sampling point is traversed, and the sampling point type is determined according to the speed range, output torque range, acceleration range, commutation state, temperature range, and impact state.

[0085] It should be explained that when dividing the working condition segments, each dimension is divided independently and can be combined to form multi-dimensional working condition segments. Among them, the speed range includes low-speed sampling points, medium-speed sampling points, and high-speed sampling points. Specifically, the joint output angular velocity is compared with the rated speed. Sampling points with an output angular velocity less than or equal to 20% of the rated speed are classified as low-speed segments, sampling points with an output angular velocity greater than 20% but less than 70% of the rated speed are classified as medium-speed segments, and sampling points with an output angular velocity greater than or equal to 70% of the rated speed are classified as high-speed segments.

[0086] The output torque range includes high torque sampling points, medium torque sampling points, and low torque sampling points. Specifically, it compares the joint output torque with the rated torque. Sampling points where the joint output torque is less than or equal to 30% of the rated torque are designated as low torque sampling points. Sampling points where the joint output torque is greater than 30% of the rated torque but less than 60% of the rated torque are designated as medium torque sampling points. Sampling points where the joint output torque is greater than or equal to 60% of the rated torque are designated as high torque sampling points.

[0087] The acceleration range includes fast start-stop sampling points and slow smoothing sampling points. Specifically, the joint output angular acceleration is compared with the historical normal acceleration. If the angular acceleration is greater than the 95th percentile of the historical normal acceleration, the sampling point is taken as the fast start-stop sampling point; otherwise, it is taken as the slow smoothing sampling point.

[0088] The commutation state includes forward, reverse, and commutation instant. The specific determination is based on the joint output angular velocity. When the angular velocity is positive, the sampling point is taken as the forward sampling point. When the angular velocity is negative, the sampling point is taken as the reverse sampling point. When the angular velocity changes from positive to negative or from negative to positive, the sampling point is taken as the commutation instant sampling point.

[0089] The temperature range includes low-temperature sampling points, rising-temperature sampling points, and stable high-temperature sampling points. Specifically, the driver temperature is compared with the ambient temperature. When the driver temperature is ≤ ambient temperature + 5℃, the sampling point is determined to be a low-temperature sampling point. When the driver temperature is between ambient temperature + 5℃ and the maximum allowable operating temperature, the sampling point is determined to be a rising-temperature sampling point. When the driver temperature is ≥ maximum allowable operating temperature × 95%, the sampling point is determined to be a stable high-temperature sampling point.

[0090] The impact state includes impact sampling points and stable sampling points. Specifically, the torque mutation amount is obtained by subtracting the output torque of adjacent timestamps. Sampling points whose torque mutation amount exceeds the 95th percentile of the historical normal mutation amount are regarded as impact sampling points, otherwise they are regarded as stable sampling points.

[0091] Intervals with consecutive sampling points that meet the same dimensional conditions are merged into a single working condition segment. The start and end times and statistical characteristics are recorded. When any dimensional condition changes, the current segment ends and a new segment begins, thus obtaining a set of working condition segments. Each segment records the start and end times, a list of working condition labels, and the corresponding working condition segment, such as "low-speed high-torque segment" or "rapid start-stop impact segment".

[0092] An initial simulation model of the electromechanical system is constructed, and the initial simulation model is run under each working condition segment of the set of working condition segments to generate a multi-working condition residual matrix.

[0093] In this embodiment, it should be specifically noted that the details of constructing the initial simulation model of the electromechanical system are as follows:

[0094] An initial simulation model of the integrated joint module is established using a multi-domain unified modeling language. This model includes at least a motor model, a reducer model, a friction model, a sensor model, and a thermal model. The motor model uses the voltage and torque equations of a brushless DC or permanent magnet synchronous motor, with PWM duty cycle or voltage command as input and electromagnetic torque as output. The reducer model includes transmission stiffness, damping coefficient, clearance, and transmission efficiency, and outputs joint-side torque. The friction model uses the LuGre or Stribeck model, including static friction coefficient, Coulomb friction coefficient, and viscous friction coefficient. The sensor model includes encoder quantization error, torque sensor zero drift and temperature drift, and signal delay represented by a first-order inertial element. The thermal model uses a lumped-parameter thermal network, including winding thermal capacity, shell thermal resistance, thermal time constant, and resistance coefficient as a function of temperature.

[0095] This initial simulation model is used to simulate the dynamic response of the electromechanical system under various operating conditions, providing a basis for subsequent residual calculation and parameter calibration.

[0096] Secondly, simulation parameters are extracted based on the initial simulation model, and initial values ​​are assigned to each parameter, including motor parameters, transmission mechanism parameters, sensor parameters, and control parameters. Among them, the motor parameters include motor torque constant, back EMF constant, winding resistance, inductance, and moment of inertia. The torque constant and back EMF constant are preferably derived from the motor's factory calibration data. The winding resistance and inductance are measured using a multimeter and an LCR meter, and the moment of inertia is calculated from the CAD model.

[0097] The transmission mechanism parameters include reduction ratio, transmission stiffness, damping coefficient, and clearance. The reduction ratio is determined by the reducer specification or the number of gear teeth; the transmission stiffness and clearance are taken from the reducer product manual; the damping coefficient is fitted through loading tests.

[0098] The control parameters include current loop proportional gain, current loop integral gain, speed loop proportional gain, speed loop integral gain, and position loop proportional gain, etc., which are read directly from the driver or controller configuration file.

[0099] The sensor parameters include encoder quantization error, torque sensor zero drift, and sensor delay time. The encoder resolution / quantization error is obtained from the datasheet; the torque sensor zero drift is determined by readings at different temperatures under no-load conditions; and the delay time is measured by a step response test.

[0100] Furthermore, the initial simulation model is run under each of the working condition segments in the set of working condition segments to generate a multi-working condition residual matrix, specifically including:

[0101] For all sampling points in each working condition segment, the corresponding test output quantities are extracted, including but not limited to motor response quantities, joint motion quantities, temperature quantities, and sensor signals and control quantities. Among them, motor response quantities include motor current, voltage, and torque; joint motion quantities include output angle, output angular velocity, and output torque; temperature quantities include driver temperature and reducer housing temperature; and sensor signals and control quantities include encoder position, control commands, and load torque. The test output quantities of all sampling points in each working condition segment are statistically processed to form the test output vector sequence of that segment.

[0102] The initial simulation model is input with the same operating conditions as the working condition segment. The simulation is run to obtain the corresponding simulation output quantities, which are then statistically organized to form a simulation output vector sequence.

[0103] It should be explained that since both the experimental sampling points and the simulation sampling points are based on the same time grid, the simulation output and the experimental data can be compared point by point at the same timestamp to calculate the residual of each type of output.

[0104] For the j-th working condition segment and the k-th type of output, let the experimental output vector sequence be y. test (j, k), the simulation output vector sequence is y sim (j, k), then the pointwise residual e(j, k, t) = y test (j, k, t)-y sim (j, k, t), j=1~J, k=1~K, J and K represent the total number of operating conditions and the total number of outputs, respectively. The root mean square error of the point-by-point residuals under each operating condition is calculated, thereby obtaining the output residuals of each operating condition, specifically expressed as:

[0105] ,

[0106] Where N j Let t represent the total number of sampling points contained in the j-th working condition segment, and t represent the time index, i.e., the t-th sampling point within a working condition segment.

[0107] It should be explained that, depending on the type of output quantity, k can be set from 1 to 11. Each k value uniquely corresponds to a type of physical quantity. When calculating the residual matrix under multiple operating conditions, all or some of the output quantity types can be selected according to the calibration target. For example, if only angle, speed, torque, temperature and response delay are considered, then only k values ​​from 1 to 5 are taken, where k=1 corresponds to angle, k=2 corresponds to speed, k=3 corresponds to torque, k=4 corresponds to temperature, and k=5 corresponds to response delay.

[0108] The output residuals within each working condition segment are summarized according to the output quantity category, thus forming a multi-working condition residual matrix with J rows and K columns. The matrix rows correspond to the working condition segments, the matrix columns correspond to the output quantity categories, and the element E(j,k) represents the output residual of the k-th type of output quantity under the j-th working condition segment.

[0109] It should be added that, since the target application scenarios and core performance indicators of electromechanical systems are different, and different application scenarios have different focuses on simulation accuracy, the selection of the output quantities of each parameter in the multi-condition calibration method of simulation parameters needs to be bound to the final use of the product. At this time, the key output quantities of each parameter in each condition segment are selected as the subsequent output quantities. For example, if the joint module is mainly used for position control scenarios, such as robot joint positioning, trajectory tracking, and applications with high repeatability accuracy requirements, then the output angle is taken as the key output quantity. If the joint module is mainly used for torque control scenarios, such as force-controlled assembly, collision detection, and compliant control, then the output torque is taken as the key output quantity.

[0110] This invention generates a multi-condition residual matrix from the output residuals, which can quantify the deviation between the initial simulation model and the actual system output under each condition segment, fully preserving the dynamic information of various output quantities. By analyzing the residual matrix, the contribution of each parameter to the residuals under different conditions can be identified, providing basic data for subsequent parameter sensitivity identification. At the same time, the residual matrix can also be used for contribution stability analysis and parameter stratification, helping to distinguish between inherent parameters across conditions and condition-sensitive parameters. Furthermore, this matrix provides a quantitative basis for parameter calibration and identification of non-calibrable conditions, guides the parameter adjustment range, and ensures the accuracy and reliability of the multi-condition simulation model.

[0111] Based on the multi-condition residual matrix, parameter sensitivity identification is performed on each simulation parameter in the electromechanical system to obtain the parameter condition contribution results.

[0112] In this embodiment, it is important to specifically explain the sensitivity identification performed on each parameter in the initial parameter set. The identification specifically includes:

[0113] Determine the allowable range of variation for each simulation parameter to be identified in the initial parameter set.

[0114] It should be explained that the principles for determining the permissible range of variation are as follows:

[0115] For motor parameters, the allowable variation range of torque constant and back electromotive force constant is determined according to the factory calibration error range of the motor, usually ±5% of the nominal value; the initial value of winding resistance is measured with a multimeter at room temperature, but due to temperature influence and measurement error, its allowable variation range is ±10% of the initial value; the allowable variation range of inductance is ±20% of the value measured by an LCR meter; the rotor moment of inertia of the motor is calculated according to the value given by the CAD model or the motor manual, and the allowable variation range is ±10% to reflect the machining tolerance and material density deviation.

[0116] For transmission mechanism parameters, the allowable variation range of the reduction ratio is within ±1% of the design value, because gear machining errors are usually very small; the initial value of the transmission stiffness is taken from the torsional stiffness given in the reducer product manual, but due to the influence of assembly preload and lubrication conditions, the allowable variation range can be expanded to ±20% of the initial value; the damping coefficient is difficult to measure directly, and the initial value is derived from the average value of test data of the same specification prototype, with an allowable variation range of ±30% of the initial value; the initial value of the clearance is taken from the reducer specification, with an allowable variation range of ±50% of the initial value, because the clearance is greatly affected by wear and assembly.

[0117] For control parameters, the allowable variation range of the control parameters is ±30% of the initial value, because the optimal control parameters differ greatly under different operating conditions, and the controller parameters are allowed to be adjusted within a certain range during actual debugging; however, for systems that have been well tuned, it can be narrowed to ±15%. If a certain control parameter lacks a clear initial value (such as when using an adaptive control algorithm), then the typical value (such as the gain corresponding to the current loop bandwidth) is used as the initial value by default, and the allowable variation range is ±50% of that value.

[0118] For sensor parameters, the allowable variation range of encoder quantization error is set to half to two times its resolution; the initial value of torque sensor zero drift is determined by the average readings at different temperatures under no-load conditions, and the allowable variation range is ±20% of the average value; the initial value of sensor delay time constant is given by the time it takes for the step response test to reach 90% of the steady state, and the allowable variation range is ±25% of the initial value.

[0119] For each parameter to be identified, a one-way perturbation analysis is performed. Let the i-th parameter in the initial parameter set be p. i Its initial value is p 0i The allowed range of variation is [p] mini p maxi Within the allowable range of variation, positive and negative perturbations are applied to the simulation parameter to obtain the corresponding perturbation parameter.

[0120] It should be added that the principle for selecting the perturbation step size is to use 5% of the initial parameter value as the minimum step size. However, when the allowable range of variation is narrow, 1 / 5 of the allowable range width can be used as the step size. To ensure the stability of sensitivity identification, multiple perturbation amplitudes can be selected, such as ±5%, ±10%, and ±15% of the initial parameter value.

[0121] For each disturbance parameter, keep all other parameters fixed at their initial values, substitute the disturbance parameter into the initial simulation model, and run the simulation under each working condition segment to obtain the disturbance output of each working condition segment. Then, subtract the experimental output before the disturbance to obtain the disturbance residual.

[0122] The disturbance residual is compared with the output residual in the multi-condition residual matrix to obtain the residual change. The residual changes of each parameter in the initial parameter set under each disturbance are summarized. Using the parameter change as the independent variable and the residual change as the dependent variable, the sensitivity S of the simulation parameter under the j-th condition segment and the k-th type of output is calculated by difference approximation. i (j, k), specifically represented as:

[0123] ,

[0124] Where ΔE r (j, k) represents the disturbance residual for the j-th operating condition segment and the k-th type of output after the r-th disturbance, p ri -p 0i This represents the difference between the value of the i-th simulation parameter under the r-th perturbation and its initial value, i.e., the parameter change.

[0125] It needs to be explained that, The first-order partial derivative of the perturbation residual with respect to the simulation parameter is given. When the parameter pi changes slightly, the rate and direction of the residual will also change. In practice, since the simulation model usually does not have an explicit analytical expression, the partial derivative cannot be directly calculated. Therefore, the difference approximation is used to replace the differential. This method can quantitatively describe the degree of influence of the simulation parameter change on the residual, thereby obtaining the sensitivity of the simulation parameter.

[0126] It should be added that when multiple perturbation amplitudes are used, the average value of the sensitivity under each perturbation is taken as the final sensitivity. The larger the absolute value of the sensitivity, the stronger the influence of the simulation parameter on the corresponding working condition segment.

[0127] Based on the positive or negative direction of the sensitivity of the disturbance parameter, the direction of the influence of the simulation parameter on the corresponding working condition segment is determined. A positive sensitivity indicates that an increase in the parameter leads to an increase in the residual, and the direction of the influence of the simulation parameter on the corresponding working condition segment is a positive change. A negative sensitivity indicates that an increase in the parameter leads to a decrease in the residual, and the direction of the influence of the simulation parameter on the corresponding working condition segment is a negative change.

[0128] Finally, the degree and direction of influence of each parameter under different operating conditions are summarized to form the parameter operating condition contribution results.

[0129] The influence of each simulation parameter in this invention on different operating conditions and output quantities provides a quantitative basis for subsequent parameter stratification and step-by-step calibration. Specifically, by perturbing each parameter one by one and observing the changes in the residuals, it is possible to determine whether each parameter improves or reduces simulation accuracy under different operating conditions, as well as the intensity of its influence. Through sensitivity identification, this invention solves the problem of blindly optimizing uniformly during calibration in traditional techniques, which easily leads to the erroneous absorption of operating condition differences as fixed parameter errors, resulting in a reduction in error in one operating condition while amplifying error in another. Therefore, this invention lays the necessary technical foundation through stratified calibration, ensuring that the calibration process can both reduce the comprehensive error of multiple operating conditions and maintain the physical consistency of parameters across different operating conditions.

[0130] Based on the parameter operating condition contribution results, a contribution stability analysis is performed on each simulation parameter in the electromechanical system to obtain parameter stratification results.

[0131] In this embodiment, it is important to specifically explain that a contribution stability analysis is performed on the parameter contribution results under different operating conditions to determine the consistency of the influence of each parameter under multiple operating conditions. Based on this, parameter stratification results are obtained, including cross-operating condition inherent parameters, operating condition sensitive parameters, and parameters to be restricted. Among them, cross-operating condition inherent parameters refer to parameters whose contribution direction is consistent under different operating conditions and whose influence on strength stability is stable, such as moment of inertia, reduction ratio, and foundation stiffness. Operating condition sensitive parameters refer to parameters that are only effective within specific speed, torque, temperature, or reversal ranges, such as friction correction coefficient, thermal drift coefficient, and sensor delay compensation. Parameters to be restricted refer to parameters whose contribution direction is contradictory or that lead to the deterioration of residuals under critical operating conditions. These parameters are usually not involved in calibration or are only used as diagnostic criteria. The specific steps are as follows:

[0132] The influence degree and direction of each parameter under different working condition segments are extracted from the parameter contribution results. Statistical analysis is performed on working condition segments in which the contribution direction of each simulation parameter is consistent across all working condition segments.

[0133] It needs to be explained that consistent contribution direction means that for the same simulation parameter, when the same direction of perturbation is applied to the simulation parameter under multiple different operating conditions, the resulting influence direction remains consistent in each operating condition segment. There will be no situation where the residual decreases in some operating conditions and increases in others. For example, if the perturbation of a certain parameter is increased, the perturbation residual in each operating condition segment will decrease. In this case, the influence direction of the simulation parameter in different operating conditions segments is negative, which is called consistent contribution direction.

[0134] Specifically, the number of operating condition segments in which each parameter has a negative influence across all operating condition segments is counted, and the proportion of each parameter to the total number of operating condition segments is calculated to obtain the consistency ratio value Ci, which is specifically expressed as: Ci = Mi / J*100%, where Mi represents the number of operating condition segments in which the i-th simulation parameter has a negative influence across all operating condition segments, and J represents the total number of operating condition segments. This ratio reflects the prevalence of the simulation parameter's positive influence under multiple operating conditions. If the simulation parameter can reduce the residual in more than 70% of the operating condition segments, it is considered that the parameter has good cross-operating condition stability; conversely, if it is only effective in a small number of specific operating condition segments, it indicates that it has strong operating condition dependence.

[0135] The sensitivity of each simulation parameter under different operating conditions is statistically analyzed to obtain the maximum and minimum sensitivity of each simulation parameter across all operating conditions. The difference between the two is used to calculate the sensitivity range ratio of each simulation parameter, and the average sensitivity of each simulation parameter across all operating conditions is calculated. Then, the sensitivity range ratio is compared with the average sensitivity to obtain the sensitivity fluctuation value. The smaller the sensitivity fluctuation value, the more stable the parameter sensitivity under that output quantity, that is, the residual change intensity caused by parameter change is similar under different operating conditions. Larger fluctuations indicate that the parameter's influence on different operating conditions is uneven and the stability is poor.

[0136] It is important to know that the consistency ratio value represents the proportion of the i-th parameter whose contribution direction is consistent across all operating condition segments, reflecting the consistency of the parameter's influence on the residual. The sensitivity fluctuation value represents the degree of fluctuation in the sensitivity of the i-th parameter under different operating condition segments, reflecting the stability of the parameter across operating conditions.

[0137] Furthermore, the simulation parameters are divided into cross-condition inherent parameters, condition-sensitive parameters, or parameters to be constrained, specifically including:

[0138] A first proportional threshold, a second proportional threshold, and a stability threshold are preset. For example, the first proportional threshold can be 70%, the second proportional threshold can be 40%, and the stability threshold can be 0.05.

[0139] When the consistency ratio is greater than or equal to the first ratio threshold and the sensitivity fluctuation is less than the preset stability threshold, it indicates that the influence of the parameter on the multi-condition residual is stable. Such parameters reflect the inherent physical properties of the electromechanical system that do not change with the operating conditions, such as moment of inertia, reduction ratio, and foundation stiffness. They should maintain a uniform value throughout the multi-condition calibration. Therefore, this simulation parameter is classified as a cross-condition inherent parameter.

[0140] When the consistency ratio is less than the first ratio threshold and greater than or equal to the second ratio threshold, and the sensitivity fluctuation value is greater than or equal to the preset stability threshold, it indicates that the parameter has an uneven impact on different working conditions and needs to be corrected separately under different working conditions. This type of parameter is mainly related to factors that change with working conditions, such as friction correction, thermal drift, and sensor delay. It can be corrected to a limited extent for different working conditions while keeping the inherent parameters unchanged. Therefore, this simulation parameter is classified as a working condition sensitive parameter.

[0141] When the consistency ratio is less than the second ratio threshold, it indicates that the parameter will lead to an increase in residuals under most operating conditions, and the calibration is not very meaningful. Such parameters usually indicate that there may be missing parts in the model structure. Therefore, this simulation parameter is classified as a parameter to be restricted.

[0142] The results of the simulation parameter classification are summarized to generate parameter stratification results, including the category, consistency ratio, and sensitivity fluctuation value of each parameter.

[0143] As one implementation method:

[0144] Assume the initial parameter set contains three parameters: parameter A is the overall reduction ratio of the reducer, parameter B is the friction coefficient, and parameter C is the motor torque constant;

[0145] By calculating the sensitivity of each parameter under five operating conditions, the parameter contribution results are as follows: The sensitivity of parameter A under the five operating conditions is 0.8, 0.75, 0.82, 0.79, and 0.81, respectively, and the influence direction is positive in all cases; the sensitivity of parameter B is -0.1, 0.05, -0.08, 0.02, and -0.09, respectively, and the influence direction is alternating between positive and negative; the sensitivity of parameter C is 0.5, 0.48, 0.52, 0.49, and 0.51, respectively, and the influence direction is positive in all cases.

[0146] Based on the above data, the consistency ratio and sensitivity fluctuation of the contribution direction for each parameter are calculated first: the consistency ratio of parameter A is 1.0, and the sensitivity fluctuation is approximately 0.025; the consistency ratio of parameter B is 0.4, and the sensitivity fluctuation is approximately 0.07; the consistency ratio of parameter C is 1.0, and the sensitivity fluctuation is approximately 0.02.

[0147] The first proportional threshold is set to 70%, the second proportional threshold is set to 40%, and the sensitivity fluctuation is set to 0.05. Each parameter is then classified into different strata: if the consistency ratio of parameter A is greater than 70% and the sensitivity fluctuation is less than 0.05, it is determined to be a cross-operating condition inherent parameter; if the consistency ratio of parameter B is less than 70% and the sensitivity fluctuation is greater than 0.05, it is determined to be an operating condition sensitive parameter; if the consistency ratio of parameter C is greater than 70% and the sensitivity fluctuation is less than 0.05, it is determined to be a cross-operating condition inherent parameter.

[0148] This implementation method allows for hierarchical management of simulation parameters based on their contribution to stability and sensitivity fluctuations across operating conditions. Parameters inherent across operating conditions are used for global calibration to ensure the overall consistency of the simulation model under all operating conditions. Condition-sensitive parameters are used for local calibration under specific operating conditions to improve the model's accuracy in those conditions. Parameters with low consistency or high fluctuations can be restricted from adjustment or temporarily left uncalibrated, thereby reducing the risk of parameter interference between different operating conditions and improving the stability and reliability of the multi-condition simulation model.

[0149] The parameter stratification results are calibrated to obtain the parameter calibration results.

[0150] In this embodiment, it should be specifically explained that calibrating the parameter stratification results and obtaining the parameter calibration results specifically includes:

[0151] Obtain the parameter stratification results. The parameters to be constrained are fixed at their initial values ​​during the calibration process and do not participate in any optimization iterations. Set the optimization boundary conditions for all parameters. The optimization boundary conditions are the allowable variation ranges corresponding to the cross-condition inherent parameters and the condition-sensitive parameters. Then, calibrate the cross-condition inherent parameters and the condition-sensitive parameters respectively.

[0152] For details on calibrating the inherent parameters across operating conditions, please refer to the following:

[0153] Only the parameter items that are classified as cross-condition inherent parameters are called, and the condition-sensitive parameters and the parameters to be restricted are temporarily fixed to their respective initial values ​​or default values;

[0154] Global optimization is performed with the goal of minimizing the combined residual of multiple working condition segments;

[0155] Specifically, the comprehensive residual is defined as the weighted sum of all operating condition segments and all output residuals. The default weights are set to equal weights. For example, the weights of angle residuals, torque residuals, and temperature residuals are all 1 / 3. If a product focuses more on torque control accuracy, the weight of the torque residual can be increased accordingly. The optimization algorithm can be least squares, genetic algorithm, or particle swarm optimization. Taking the genetic algorithm as an example: the population size is set to 50, the crossover probability is 0.8, the mutation probability is 0.1, and the maximum number of iterations is 200. In each iteration, the inherent parameter values ​​of each group across operating conditions in the population are substituted into the initial simulation model. The simulation is run under all operating condition segments, the output residuals of each operating condition segment are calculated and weighted, and the comprehensive residual is obtained. After iterative optimization, when the comprehensive residual reaches the minimum value and meets the convergence criterion, the optimization stops, and the optimal value of the current cross-operating condition inherent parameter is output, which is recorded as the inherent parameter calibration result.

[0156] While keeping the inherent parameter calibration results unchanged, the associated condition-sensitive parameters are calibrated separately for different operating condition segments, as detailed below:

[0157] For each operating condition segment, query the parameter stratification results for parameters that are highly sensitive and classified as operating condition sensitive in that segment. Only adjust these parameters, while keeping the remaining parameters (including cross-operating condition inherent parameters and other operating condition sensitive parameters unrelated to that segment) fixed. For example, in the low-speed, high-torque segment, this operating condition mainly exposes friction and clearance issues, and the friction coefficient and clearance are highly sensitive, so only the friction coefficient and clearance parameters are adjusted in the calibration of this segment; temperature parameters are only sensitive in the continuous temperature rise segment, so only the calibration is performed in this segment.

[0158] The calibration objective is to minimize the output residual for this segment, and local optimization algorithms such as gradient descent or pattern search can be used.

[0159] Specifically, taking the pattern search method as an example: set the initial step size to 10% of the allowable range of parameter variation, the shrinkage factor to 0.5, the expansion factor to 2, and the minimum step size to 1% of the initial step size; for each working condition segment, run the optimization separately, and when the output residual under the working condition segment no longer decreases significantly (the residual change is less than 1% for three consecutive iterations) or the maximum number of iterations (default 50 times) is reached, stop the optimization, and record the current parameter value as the working condition correction parameter result for that segment;

[0160] After all operating condition segments have been calibrated, the results of the operating condition correction parameters for all segments are summarized to obtain the set of operating condition correction parameters.

[0161] Summarize all the calibration results obtained from the above two steps and output the parameter calibration results, including the inherent parameter values ​​across working conditions and the working condition correction parameter values ​​corresponding to each working condition segment.

[0162] Example 1 decomposes multi-condition data into fine-grained condition segments and performs transferability stratification of simulation parameters through contribution stability analysis. Without relying on large-scale traversal calibration or empirical parameter tuning, it retains the local error characteristics of each condition segment, which is beneficial to improving the calibration accuracy of electromechanical system simulation models under multiple conditions. It effectively reduces the risk of mutual constraints between errors of different conditions and parameter distortion across conditions in traditional unified optimization methods. At the same time, it can also ensure that the calibration results have clear physical consistency and a basis for judging the applicable boundary of the condition.

[0163] Example 2:

[0164] This embodiment 2 further provides an improved solution based on embodiment 1. Embodiment 1 solved the problem of mutual constraint of errors under multiple operating conditions and parameter distortion across operating conditions, enabling characteristic errors under different speed, torque, temperature, and other operating conditions to be effectively preserved through parameter layering and directional calibration. However, in practical application scenarios, such as impact contact, rapid reversing, and continuous operation under different ambient temperatures, the reliability of the same simulation model often decreases significantly. Existing simulation parameter calibration methods usually only output a set of optimal parameters or output parameters for each operating condition separately, without clearly distinguishing which parameters are stable physical parameters across operating conditions and which parameters are only compensation parameters for a certain operating condition. This makes it difficult for engineers to determine the source of simulation errors. This embodiment introduces a cross-operating condition replay verification and parameter applicability boundary determination mechanism based on embodiment 1. The parameter calibration results are replayed for verification to obtain the parameter applicability boundary results, and an uncalibrable operating condition prompt is output. This achieves the effect of further distinguishing calibrable deviations and model structure deficiencies while retaining the advantages of layered calibration, further solving the defect of insufficient cross-operating condition reliability caused by changes in operating conditions. See below for details:

[0165] The parameter calibration results were replayed for verification to obtain the parameter applicability boundary results.

[0166] In this embodiment, it should be specifically noted that after completing the calibration of inherent parameters across operating conditions and the calibration of condition-sensitive parameters, the obtained parameter calibration results need to be re-injected into the initial simulation model. All measured operating conditions and several combinations of operating conditions are then replayed for verification to check the simulation accuracy of the calibrated model under different operating conditions. Based on this, the applicable operating condition boundaries for the parameters are defined, specifically including:

[0167] The inherent parameters across operating conditions in the calibration results are written into the simulation model as global fixed values. At the same time, the corresponding operating condition correction parameter results are loaded according to different operating condition segments: for each specific operating condition segment, the operating condition sensitive parameters obtained from calibration under that segment are replaced with the default values ​​or initial values ​​in the simulation model.

[0168] It should be explained that since the cross-condition inherent parameters are intrinsic properties of the electromechanical system that do not change with the operating conditions, such as moment of inertia and reduction ratio, they are written into the model as global fixed values ​​to ensure that the model follows the same physical laws under any operating condition, avoiding the logical contradiction that the same hardware is assigned different physical parameters under different operating conditions. On the other hand, the condition-sensitive parameters change with the operating conditions (speed, torque, temperature, commutation, etc.). By loading the correction parameters obtained from their respective calibrations in different operating condition segments, the simulation model can specifically compensate for the errors under specific operating conditions.

[0169] Select the verification conditions for playback verification. The verification conditions include the first type of verification conditions and the second type of verification conditions. The first type of verification conditions are the original operating condition segments that have already been calibrated, such as low speed and high torque, high speed and low torque, frequent start-stop, forward and reverse reversal, and continuous temperature rise conditions, which are used to verify the residual reduction effect of the calibrated model under the same conditions. The second type of verification conditions are combined or transitional operating conditions that have not been calibrated, such as the continuous torque variation condition at medium speed, the variable load operation condition during the process of temperature rising from room temperature to high temperature, and the random motion condition containing multiple reversals, which are used to evaluate the generalization ability of the calibration parameters under new operating conditions.

[0170] It should be added that the input data for the second type of verification condition can come from independent verification tests, or a segment of data that is not used for calibration can be reserved from the original test data.

[0171] Acquire verification condition segments, and for each verification condition segment, substitute the corresponding parameter configuration into the initial simulation model, run the simulation under the same operating conditions as the experiment, and after the simulation is completed, obtain the verification output quantity, and compare the verification output quantity with the experimental output quantity point by point, calculate the verification residual of each output quantity. The verification residual can be obtained by the same calculation method as the output residual. Compare the verification residual with the preset acceptance threshold. The setting of the acceptance threshold depends on the specific purpose of the simulation model: if it is used for the initial selection of controller parameters, it can be relatively relaxed, for example, the root mean square error of the angle does not exceed 5% of the peak-to-peak value of the experiment, the root mean square error of the torque does not exceed 5% of the rated torque, and the temperature error does not exceed 3℃; if it is used for life prediction or safety redundancy verification, the threshold should be tightened, and the above 5%, 5%, and 3℃ are used as the basic acceptance threshold by default.

[0172] It should be added that the verification test segments include segments corresponding to the first type of verification test and segments corresponding to the second type of verification test. The specific methods for obtaining them are as follows:

[0173] The segments corresponding to the first type of verification conditions are directly extracted from the set of condition segments. Since these segments have already undergone load segment division, labeling and merging, and their data have been used in the previous sensitivity analysis and parameter optimization process, they only need to be retrieved according to the original segment boundaries during playback verification, without the need for re-division.

[0174] The second type of verification operation segment starts from the boundaries of the divided operation segment and constructs combined or transitional segments. Specifically, it involves selecting two or more adjacent (or non-adjacent) operation segments, splicing them together in chronological order, and adding a transition process at the splicing point. For example, it involves generating a ramp change in speed or torque through linear interpolation to form a continuous operation segment. For instance, the latter half of a low-speed, high-torque segment can be directly connected to the first half of a high-speed, low-torque segment without inserting a steady-state transition to simulate abrupt changes in operation. Alternatively, the temperature rise segment of a continuous temperature rise segment can be superimposed with the impact segment of an impact load segment to verify the model response under thermal-mechanical coupling conditions. The constructed segments may not have continuous records in the original experimental data, but by recombining the simulation input, the performance of parameters under complex transitional states can be tested in a targeted manner.

[0175] Based on whether the verification residuals under each verification condition segment in the playback verification meet the acceptance threshold, and whether the condition-sensitive parameters are within the allowable range of variation, the applicable condition boundaries for the parameters are defined, specifically including:

[0176] It should be explained that the operating condition correction parameters here refer to the specific values ​​of the set of operating condition correction parameters that are actually written into the simulation model and participate in the simulation calculation when a certain verification operating condition segment is replayed for verification. Specifically, for the first type of verification operating condition, these parameters are directly taken from the results of the operating condition correction parameters, such as the winding resistance temperature coefficient corresponding to the continuous temperature rise segment. For the second type of verification operating condition, these parameters can be estimated by interpolation or nearest neighbor matching from the existing calibration results based on the characteristics of the segment such as speed, torque, and temperature, or they can be uniformly set to the default initial value when it is impossible to estimate.

[0177] All verification test segments are sorted according to continuous variables such as speed, torque, temperature, commutation frequency and impact intensity to establish a multi-dimensional test space;

[0178] For each verification condition segment, if all verification residuals in the verification condition segment are lower than the corresponding acceptance threshold, and the called condition-sensitive parameters do not touch the allowable range of change, then the interval is designated as the parameter applicable interval. Verification residuals include, but are not limited to, angle verification residuals, torque verification residuals, and temperature verification residuals.

[0179] If the verification residual exceeds the acceptance threshold, but the direction of the residual change can be explained by adjusting the operating condition sensitive parameters, then the interval is classified as a cautious application interval, prompting users to pay extra attention when using it;

[0180] It should be explained that during the playback verification process, for a certain verification condition segment, if the residual of the calibrated simulation model under that segment exceeds the acceptance threshold, but the direction of change of the residual has a clear physical correspondence with the adjustment direction of a certain condition sensitive parameter, and the residual can be expected to be reduced to within the threshold by independently adjusting the corresponding condition sensitive parameter under that segment within the allowable range of change, then that interval is classified as a cautious application interval.

[0181] As one implementation method:

[0182] Suppose that in the playback verification under high-speed, low-torque conditions, the simulation output response delay verification residual exceeds the acceptance threshold, specifically manifested as the simulation response being approximately 15ms slower than the experimental result. Based on the parameter contribution results under these conditions, the "sensor delay compensation amount" has been classified as a condition-sensitive parameter. Furthermore, the sensitivity analysis of this parameter under these conditions shows that increasing the compensation amount within the allowable range reduces the response delay verification residual. The current verification residual aligns with the improvement direction achieved by increasing the compensation amount, indicating that the error is not due to missing model structure but simply because the compensation amount used in the current verification has not been precisely calibrated for this condition. Therefore, even though the verification residual for this condition segment temporarily exceeds the acceptance threshold, it is still classified as a "cautious application range," suggesting that if higher simulation accuracy is required under these conditions, the "sensor delay compensation amount" can be supplemented and calibrated separately for this segment without modifying the model structure or recalibrating the inherent parameters across different conditions.

[0183] If the verification residual exceeds the acceptance threshold, and the operating condition sensitive parameters need to exceed the allowable range of variation in order to reduce the error, then the interval is classified as an inapplicable interval.

[0184] For each verification condition segment, upper and lower limits of the applicable range are given. For example, based on the playback verification results, the output is: the applicable speed range is 10% to 90% of the rated speed, the applicable torque range is 20% to 80% of the rated torque, the applicable temperature range is -10℃ to 70℃, the commutation frequency does not exceed 2Hz, and the impact load amplitude does not exceed 120% of the rated torque. These boundary values ​​are extracted from discrete verification points through linear interpolation or fitting. Specifically, the highest and lowest speeds, maximum and minimum torques, etc., that meet all threshold conditions can be found, and they are appropriately extrapolated to adjacent untested points, for example, the extrapolation does not exceed 10% of the test range.

[0185] Finally, the output parameter applicable boundary results are compiled into a structured table, which includes at least the operating condition dimension (speed, torque, temperature, commutation frequency, impact intensity), applicable range, verification residual within the verification operating condition segment, whether the parameter boundary is reached, and recommended use level (parameter applicable range, cautious application range, and inapplicable range). This result, together with the subsequent uncalibrable operating condition prompts, constitutes an important part of the calibration output, helping engineers to determine under which operating conditions the calibrated simulation model can be trusted, and under which operating conditions recalibration or model structure improvement is required.

[0186] Furthermore, during the calibration of the sensitive parameters of the operating condition, if the sensitive parameters of the current operating condition segment have reached the boundary of its allowable range of change, and the output residual of the segment is still greater than the preset residual threshold, such as 50% of the output residual, the specific value is determined according to the product accuracy requirements and engineering experience, then the further calibration of the segment is immediately stopped, and the operating condition segment is marked as a candidate uncalibrable operating condition.

[0187] For a work condition segment marked as a candidate uncalibrable work condition, query its interval category in the parameter application boundary results:

[0188] If the operating condition segment is classified as an inapplicable range, it will be treated as the final uncalibrable operating condition, and an uncalibrable identification result will be generated.

[0189] If the operating condition is classified as an applicable range or a cautiously applicable range, its candidate mark will be removed, and the calibration will be considered successful. No uncalibrable prompt will be generated.

[0190] It should be added that for a section of a work condition that is not marked as a candidate uncalibrable work condition but is classified as an "inapplicable interval" in the parameter applicable boundary results, it should also be identified as the final uncalibrable work condition, because its playback verification residual does not meet the acceptance threshold and cannot be resolved by parameter adjustment, indicating that the cause of the error exceeds the parameter calibration capability.

[0191] It should be explained that for each final uncalibrable operating condition, an uncalibrable prompt result is generated, which includes at least the operating condition identifier, the triggering reason, and suggested measures.

[0192] For example, when the torque error under low-speed, high-torque conditions needs to be reduced by significantly increasing the friction coefficient, but this adjustment leads to a significant deterioration in speed response under high-speed conditions, the system will no longer continue to increase the friction coefficient, but will instead output an uncalibrable warning that "the nonlinear friction model of the reducer may be insufficient"; when the angle residual and torque residual cannot be reduced simultaneously under impact load conditions, an uncalibrable warning that "the structural flexibility or impact contact model may be insufficient" will be output.

[0193] Finally, the parameter calibration results and parameter applicable boundary results are summarized to obtain the multi-condition calibration results.

[0194] This embodiment further evaluates the applicable boundaries of parameters by replaying the verification results, and corrects the parameter calibration results formed in Embodiment 1. This allows the parameter portability decoupling results to further apply to the identification of local operating condition errors and the generation process of cross-operating condition calibration results of the whole machine simulation model, forming a closed-loop processing mechanism from operating condition segment decoupling to parameter hierarchical calibration to cross-operating condition verification and boundary output.

[0195] Depend on Figure 2 It is evident that, compared to existing technologies, this invention demonstrates superior performance in multi-condition error decoupling capability, parameter stability characterization across conditions, parameter differentiation capability, and simulation reliability. Specifically, by segmenting conditions and analyzing parameter condition sensitivity, this invention improves the identification of local error sources under different speed, torque, temperature, and impact conditions. Furthermore, through parameter transferability layering and non-calibrable condition prompts, it further reduces the interference of condition differences and model structure deficiencies on parameter calibration. This results in clearer applicability boundaries for calibrated simulation parameters across multiple conditions and more stable and accurate cross-condition simulation results.

[0196] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the claimed invention.

Claims

1. A multi-condition calibration method for simulation parameters of an electromechanical system, characterized in that, include: Acquire multi-condition data, divide the multi-condition data into condition segments, and obtain a set of condition segments; An initial simulation model of the electromechanical system is constructed, and the initial simulation model is run under each working condition segment of the set of working condition segments to generate a multi-working condition residual matrix. Based on the multi-condition residual matrix, parameter sensitivity identification is performed on each simulation parameter in the electromechanical system to obtain the parameter condition contribution results. Based on the parameter operating condition contribution results, a contribution stability analysis of each simulation parameter in the electromechanical system is performed to obtain parameter stratification results. The parameter stratification results are calibrated to obtain the parameter calibration results; The parameter calibration results were replayed for verification to obtain the parameter applicability boundary results. The parameter calibration results and parameter applicable boundary results are summarized to obtain multi-condition calibration results.

2. The simulation parameter multi-condition calibration method according to claim 1, characterized in that, The multi-condition data is divided into condition segments to obtain a set of condition segments, including: The multi-condition data are arranged in the order of sampling points to form a sampling time series; The output angular velocity, output torque, output angular acceleration, driver temperature, and output torque abrupt change are obtained point by point from the sampled time series. Each sampling point is divided into velocity range, output torque range, acceleration range, commutation state, temperature range, and impact state according to a preset threshold, thus obtaining the sampling point type; Intervals with consecutive sampling points of the same type are merged into a single working condition segment. When any dimensional condition changes, the current segment ends and a new segment begins, thus obtaining a set of working condition segments.

3. The simulation parameter multi-condition calibration method according to claim 1, characterized in that, Based on the multi-condition residual matrix, parameter sensitivity identification is performed on each simulation parameter in the electromechanical system to obtain the parameter condition contribution results, including: Obtain the allowable variation range of each simulation parameter in the electromechanical system, and perturb each parameter within its allowable variation range to obtain the perturbation parameters; The initial simulation model is used to re-simulate each disturbance parameter, calculate the disturbance residual corresponding to each working condition segment, and compare the disturbance residual with the output residual in the multi-working condition residual matrix to obtain the residual change of the disturbance for each working condition. Based on the changes in residuals, determine the degree and direction of the influence of the simulation parameter on the corresponding working condition segment, and generate the parameter working condition contribution result; The parameter condition contribution results are used to distinguish between cross-condition inherent parameters, condition-sensitive parameters, and parameters to be restricted.

4. The simulation parameter multi-condition calibration method according to claim 3, characterized in that, Based on the changes in residuals, determine the degree and direction of the influence of the simulation parameter on the residuals under different operating conditions, including: For each disturbance parameter, calculate the residual change under each operating condition segment; Based on the magnitude of the residual change, the sensitivity of each disturbance parameter is calculated by the ratio of the residual change to the parameter change, thereby determining the degree of influence of the simulation parameter on the corresponding working condition segment. Based on the positive and negative directions of the sensitivity of the disturbance parameter, the direction of the influence of the simulation parameter on the corresponding working condition segment is determined. A positive sensitivity indicates that an increase in the parameter leads to an increase in the residual, and a negative sensitivity indicates that an increase in the parameter leads to a decrease in the residual. The direction of influence includes positive and negative changes. The degree and direction of the influence are recorded as the contribution results of the parameter operating conditions.

5. The simulation parameter multi-condition calibration method according to claim 1, characterized in that, Based on the parameter operating condition contribution results, a contribution stability analysis is performed on each simulation parameter in the electromechanical system to obtain parameter stratification results, including: Extract the degree and direction of influence of each parameter under different working condition segments from the parameter contribution results; For each simulation parameter, count the number of working condition segments in which the simulation parameter has a negative influence across all working condition segments, and calculate the proportion of this number to the total number of working condition segments to obtain the consistency ratio value. The sensitivity range ratio is calculated based on the degree of influence of the simulation parameters, and compared with the average sensitivity of the parameter under all working conditions to obtain the sensitivity fluctuation value. Based on the consistency ratio and sensitivity fluctuation values, the simulation parameters are divided into cross-condition inherent parameters, condition-sensitive parameters, or parameters to be restricted, resulting in parameter stratification results.

6. The simulation parameter multi-condition calibration method according to claim 5, characterized in that, Based on the consistency ratio and sensitivity fluctuation values, the simulation parameters are divided into cross-condition inherent parameters, condition-sensitive parameters, or parameters to be constrained, including: When the consistency ratio value is greater than or equal to the first ratio threshold and the sensitivity fluctuation value is less than the preset stability threshold, the simulation parameter is classified as a cross-condition inherent parameter. When the consistency ratio value is less than the first ratio threshold and greater than or equal to the second ratio threshold, and the sensitivity fluctuation value is greater than or equal to the preset stability threshold, the simulation parameter is classified as a working condition sensitive parameter. When the consistency ratio value is less than the second ratio threshold, the simulation parameter is classified as a parameter to be restricted.

7. The simulation parameter multi-condition calibration method according to claim 1, characterized in that, The parameter stratification results are calibrated to obtain parameter calibration results, including: The parameter stratification results are obtained, including cross-condition inherent parameters, condition-sensitive parameters, and parameters to be restricted; For the parameter items that are classified as cross-condition inherent parameters, the initial simulation model is called and run under all condition segments, and optimization is performed with the goal of minimizing the comprehensive residual of multiple condition segments, thereby obtaining the inherent parameter calibration results; Under the condition that the inherent parameter calibration results remain unchanged, the working condition sensitive parameters are calibrated separately for each working condition, thereby obtaining the working condition correction parameter results; The calibration results of the inherent parameters and the calibration results of the operating condition correction parameters are output as the parameter calibration results.

8. The simulation parameter multi-condition calibration method according to claim 1, characterized in that, The parameter calibration results were replayed for verification to obtain the parameter applicability boundary results, including: Input the parameter calibration results into the initial simulation model, and select the verification condition to run the simulation. The verification residuals of each output quantity are calculated for the verification conditions, and it is determined whether the verification residuals meet the preset acceptance threshold. Thus, each condition segment is divided into the parameter applicable range, the cautious application range, and the inapplicable range. The parameters applicable boundary results are formed by summarizing each interval and its corresponding speed range, torque range, temperature range, commutation frequency and impact load range; The applicable boundary results of the parameters can be used to guide the reliable invocation of multi-condition simulation models under different conditions, and to identify uncalibrable conditions.

9. The simulation parameter multi-condition calibration method according to claim 8, characterized in that, For each output quantity in the verification condition, the verification residual is calculated, and it is determined whether the verification residual meets the preset acceptance threshold. Based on this, each condition segment is divided into a parameter applicable range, a cautiously applicable range, and an inapplicable range, including: Obtain the verification condition segments, substitute the parameter configurations corresponding to each verification condition segment into the initial simulation model to run the simulation, calculate the verification output, and compare the verification output with the experimental output point by point to calculate the verification residual of each output. For each verification condition segment, if all verification residuals in that verification condition segment are lower than the corresponding acceptance threshold, then that verification condition segment is classified as the parameter applicable range. If the verification residual exceeds the acceptance threshold but the direction of residual change is explained by the condition-sensitive parameters, then the verification condition segment is classified as a cautious application range. If the verification residual exceeds the acceptance threshold and the operating condition sensitive parameters need to exceed the allowable range of variation in order to reduce the error, then the verification operating condition segment is classified as an inapplicable interval.

10. The simulation parameter multi-condition calibration method according to claim 8, characterized in that, Also includes: If the sensitive parameters of the current working condition segment have reached the boundary of its allowable range of change, and the output residual of the segment is still greater than the preset residual threshold, then the working condition segment is marked as a candidate uncalibrable working condition. For a work condition segment marked as a candidate uncalibrable work condition, query its interval category in the parameter application boundary results: If the operating condition segment is classified as an inapplicable range, it will be treated as the final uncalibrable operating condition, and an uncalibrable identification result will be generated. If the operating condition is classified as an applicable range or a cautiously applicable range, its candidate mark will be removed, and the calibration will be considered successful. No uncalibrable prompt will be generated.

Images