Servo motor mechanical performance grading evaluation method, system, device and storage medium

Abstract

The present disclosure relates to the technical field of servo motor performance testing and intelligent evaluation, and provides a servo motor mechanical performance grading evaluation method, system, device and storage medium, the method comprising: collecting rated parameters and real-time operating parameters of the servo motor, and calculating evaluation indexes; configuring non-uniform B-spline nodes according to a speed ratio of the servo motor, and generating B-spline base functions adapted to current working conditions; performing forward propagation operation on the B-spline base functions and control point weights through a KAN neural network to obtain a basic weight vector; taking a FAHP judgment matrix corrected by a thermal-mechanical coupling degradation factor as an optimization benchmark, iteratively updating the control point weights of the KAN neural network by using a gradient descent method until a consistency ratio of the FAHP judgment matrix meets a predetermined convergence condition, and outputting an optimal weight vector; and performing weighted summation on normalized values of the evaluation indexes and the optimal weight vector to obtain a final evaluation score for mechanical performance grading.

Description

Technical Field

[0001] This disclosure relates to the field of servo motor performance testing and intelligent evaluation technology, and in particular to a method, system, electronic device and computer-readable storage medium for graded evaluation of the mechanical performance of servo motors. Background Technology

[0002] With the rapid development of industrial automation and high-end motion control technology, servo motors, as core execution units in high-end equipment such as precision manufacturing, industrial robots, and CNC machine tools, require real-time and accurate evaluation of their mechanical performance to ensure machining accuracy, operational stability, and production safety. In the full lifecycle management of servo motors, mechanical performance grading and evaluation is a fundamental step in achieving predictive maintenance, equipment selection optimization, and operating condition adaptation. The accuracy and reliability of the evaluation results directly affect the quality of operation and maintenance decisions for high-end equipment.

[0003] Currently, the evaluation methods for the mechanical performance grading of servo motors mainly follow two technical routes: one is a comprehensive evaluation method based on FAHP (Fuzzy Analytic Hierarchy Process), which constructs a pairwise comparison judgment matrix between indicators, calculates the weight of each evaluation indicator, and then performs a weighted score; the other is a data-driven intelligent evaluation method, which uses models such as neural networks to perform end-to-end performance mapping on motor operating data. In recent years, some studies have attempted to introduce KAN (Kolmogorov-Arnold Networks) into the field of weight optimization in order to improve the nonlinear fitting ability of the weights. In terms of data verification, most existing technologies use unidirectional physical residual correction methods such as power conservation checks to simply filter the measured data collected by sensors.

[0004] In practical industrial applications, the aforementioned technical solutions suffer from inaccurate evaluations due to a lack of adaptability and nonlinear characterization in weight allocation. Specifically, traditional FAHP methods typically pre-determine a fixed judgment matrix based on expert experience, and the weights remain unchanged throughout the evaluation cycle, failing to respond to dynamic changes in motor operating conditions. In actual servo motor operation, when the speed crosses the rated constant torque region and the high-speed field-weakening constant power region, the contribution of each performance index to the overall mechanical performance evaluation differs significantly: the sensitivity of the speed deviation rate increases sharply in the field-weakening region, while the evaluation priority of torque and power in the overload heating region should be correspondingly increased. Existing methods cannot dynamically adjust the non-uniform sampling strategy of weight mapping according to the real-time speed ratio of the motor, resulting in severely insufficient mapping resolution in nonlinearly sensitive regions such as the field-weakening region and the high-load region. Furthermore, existing technologies do not introduce a thermo-mechanical coupling degradation factor, failing to dynamically amplify the judgment scale of torque and power indices relative to speed and load rate indices based on the cumulative thermal effects caused by the load rate deviating from the optimal operating point. This leads to a continuous underestimation of potential risks under overload conditions, resulting in evaluation results that deviate significantly from engineering realities.

[0005] Furthermore, the aforementioned technical solutions lack both virtual and physical constraints and a soft fault-tolerance mechanism, resulting in poor noise resistance and robustness. Specifically, in industrial settings, servo motor sensor data is affected by factors such as electromagnetic interference, temperature drift, and asynchronous signal sampling, often disrupting the physical conservation relationship between power, torque, and speed. Existing evaluation methods either use only unidirectional physical residuals for hard threshold filtering of data as "qualified / unqualified" or rely entirely on data-driven models while ignoring physical mechanism constraints. The former simply rejects input when physical inconsistencies occur, causing evaluation interruption; the latter directly substitutes distorted data into the model, outputting false high or low scores. Neither approach simultaneously constructs a two-way physical constraint system of "residual between measured power and theoretical mechanical power" and "self-consistency residual of rated parameters." This makes it impossible to generate continuous penalty factors to softly correct weighted scores when sensor data exhibits noise, drift, or physical inconsistencies, resulting in a lack of effective fault tolerance and anti-interference capabilities, and ultimately failing to meet the high-reliability evaluation requirements of industrial settings.

[0006] In summary, improving the accuracy of servo motor mechanical performance grading assessment under varying operating conditions and its noise resistance under data distortion conditions are urgent technical problems that need to be solved. Summary of the Invention

[0007] In view of this, the present disclosure provides a method, apparatus, electronic device and computer-readable storage medium for grading and evaluating the mechanical performance of a servo motor, in order to solve the technical problems in the prior art where static and fixed weight allocation leads to inaccurate evaluation under varying operating conditions, and the lack of dual physical constraints and soft fault tolerance mechanisms results in poor noise robustness.

[0008] A first aspect of this disclosure provides a method for grading and evaluating the mechanical performance of a servo motor, comprising: collecting rated parameters and real-time operating parameters of the servo motor; calculating evaluation indicators based on the rated parameters and real-time operating parameters; and performing normalization processing according to the attribute type of each evaluation indicator. The rated parameters include at least rated torque, rated speed, and rated power; the real-time operating parameters include at least average torque, average speed, average output power, and average load rate; and the evaluation indicators include at least torque error rate, speed deviation rate, power achievement rate, and average load rate. The method further comprises: configuring non-uniform B-spline nodes according to the speed ratio of the servo motor; generating B-spline basis functions adapted to the current operating conditions; wherein the speed ratio is the ratio of average speed to rated speed; and when the speed ratio is in different predetermined intervals, the non-uniform B-spline nodes have different control point density distributions on both sides of the rated speed to match the nonlinear distribution characteristics of the motor's external characteristics; and performing forward propagation operations on the B-spline basis functions and control point weights through a KAN neural network to output the initial weight benchmark values ​​of each evaluation indicator, which are then normalized to obtain the basic weight vector. Using the FAHP judgment matrix corrected by the thermo-mechanical coupling degradation factor as the optimization benchmark, the control point weights of the KAN neural network are iteratively updated using the gradient descent method until the consistency ratio of the FAHP judgment matrix meets the predetermined convergence condition, and the optimal weight vector is output. The thermo-mechanical coupling degradation factor is calculated based on the degree to which the average load rate deviates from the predetermined optimal operating point. The correction includes adjusting the judgment scale of the torque and power indices in the FAHP judgment matrix relative to the speed and load rate indices based on the thermo-mechanical coupling degradation factor. The normalized values ​​of the evaluation indices are weighted and summed with the optimal weight vector, and the weighted summation result is corrected with a physical penalty factor to obtain the final evaluation score. The mechanical performance is graded based on the final evaluation score. The physical penalty factor is generated based on the fusion result of the measured power physical residual and the rated parameter self-consistency residual. The measured power physical residual is calculated based on the physical conservation relationship between average torque, average speed and average output power, and the rated parameter self-consistency residual is calculated based on the physical conservation relationship between rated torque, rated speed and rated power.

[0009] A second aspect of this disclosure provides a servo motor mechanical performance grading and evaluation system, comprising: a data acquisition and processing module for acquiring rated parameters and real-time operating parameters of the servo motor, calculating evaluation indicators based on the rated parameters and real-time operating parameters, and performing normalization processing according to the attribute type of each evaluation indicator. The rated parameters include at least rated torque, rated speed, and rated power; the real-time operating parameters include at least average torque, average speed, average output power, and average load rate; and the evaluation indicators include at least torque error rate, speed deviation rate, power achievement rate, and average load rate. A non-uniform B-spline node configuration module is used to configure non-uniform B-spline nodes according to the servo motor's speed ratio, generating B-spline basis functions adapted to the current operating conditions. The speed ratio is the ratio of average speed to rated speed. When the speed ratio is within different predetermined intervals, the non-uniform B-spline nodes have different control point density distributions on both sides of the rated speed to match the nonlinear distribution characteristics of the motor's external characteristics. A KAN neural network weight mapping module is used to perform forward propagation operations on the B-spline basis functions and control point weights through a KAN neural network, outputting initial weight benchmark values ​​for each evaluation indicator, and then normalizing them according to their attribute type. After normalization, a basic weight vector is obtained. The weight iteration optimization module uses the FAHP judgment matrix corrected by the thermo-mechanical coupling degradation factor as the optimization benchmark, and iteratively updates the control point weights of the KAN neural network using gradient descent until the consistency ratio of the FAHP judgment matrix meets a predetermined convergence condition, outputting the optimal weight vector. The thermo-mechanical coupling degradation factor is calculated based on the degree to which the average load rate deviates from the predetermined optimal operating point. Correction includes adjusting the judgment scale of the torque and power indices relative to the speed and load rate indices in the FAHP judgment matrix based on the thermo-mechanical coupling degradation factor. The comprehensive scoring module performs a weighted summation of the normalized values ​​of the evaluation indicators and the optimal weight vector, and corrects the weighted summation result with a physical penalty factor to obtain the final evaluation score. Mechanical performance is graded based on the final evaluation score. The physical penalty factor is generated based on the fusion result of the measured power physical residual and the rated parameter self-consistency residual. The measured power physical residual is calculated based on the physical conservation relationship between average torque, average speed, and average output power, while the rated parameter self-consistency residual is calculated based on the physical conservation relationship between rated torque, rated speed, and rated power.

[0010] A third aspect of this disclosure provides an electronic device including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the method described above.

[0011] A fourth aspect of this disclosure provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the above-described method.

[0012] The beneficial effects of this disclosure embodiment compared with the prior art are as follows: The technical solution of this disclosure embodiment collects the rated parameters and real-time operating parameters of the servo motor, calculates the evaluation index and performs normalization processing; based on the speed ratio, non-uniform B-spline nodes are configured, and the B-spline basis function and control point weights are forward propagated through the KAN neural network to output the initial weight benchmark value; the thermo-mechanical coupling degradation factor is calculated according to the load rate deviation, the scaling of the FAHP judgment matrix is ​​dynamically corrected, and the control point weights are iteratively updated with gradient descent until consistent convergence, outputting the optimal weight vector, so that the weights are optimized in real time with the working conditions, solving the problem of inaccurate evaluation under variable working conditions such as weak magnetic field and overload of traditional static weighting; the normalized index and the optimal weight are weighted and summed, and the measured power conservation is verified based on the torque-speed product, and the self-consistency is verified based on the rated parameters, generating the measured power residual and the rated parameter residual and merging them into a continuous physical penalty factor, which is used to softly correct the weighted score, and finally outputs the performance level and applicable scenario, realizing the fault-tolerant evaluation of data distortion and physical inconsistency, and significantly improving the working condition adaptability and field reliability of the evaluation. Attached Figure Description

[0013] To more clearly illustrate the technical solutions in the embodiments of this disclosure, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this disclosure. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0014] Figure 1 This is a schematic flowchart of the servo motor mechanical performance grading and evaluation method provided in the embodiments of this disclosure; Figure 2 This is a schematic diagram of the truth error RMSE comparison analysis of the embodiments of this disclosure; Figure 3 This is a schematic diagram illustrating the comparative analysis of anti-distortion attenuation rates in embodiments of this disclosure; Figure 4 This is a schematic diagram illustrating the physical violation interception rate comparison analysis of embodiments of this disclosure; Figure 5 This is a schematic diagram of the working condition adaptive sensitivity comparison analysis of the embodiments of this disclosure; Figure 6 This is a schematic diagram of the GUI visualization tool of the servo motor mechanical performance grading and evaluation system provided in this embodiment of the disclosure; Figure 7 This is a schematic diagram of the servo motor mechanical performance grading and evaluation system provided in this embodiment of the disclosure; Figure 8 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this disclosure. Detailed Implementation

[0015] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, so as to provide a thorough understanding of the embodiments of this disclosure. However, those skilled in the art will understand that this disclosure may also be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods have been omitted so as not to obscure the description of this disclosure with unnecessary detail.

[0016] The following will describe in detail, with reference to the accompanying drawings, the servo motor mechanical performance grading and evaluation scheme according to embodiments of this disclosure.

[0017] Figure 1 This is a flowchart illustrating a method for grading and evaluating the mechanical performance of a servo motor, as provided in an embodiment of this disclosure. Figure 1 As shown, the servo motor mechanical performance grading and evaluation method of this disclosure includes: Step S101: Collect the rated parameters and real-time operating parameters of the servo motor, calculate the evaluation index based on the rated parameters and real-time operating parameters, and normalize them according to the attribute type of each evaluation index. The rated parameters include at least rated torque, rated speed and rated power, the real-time operating parameters include at least average torque, average speed, average output power and average load rate, and the evaluation index includes at least torque error rate, speed deviation rate, power achievement rate and average load rate.

[0018] Step S102: Configure non-uniform B-spline nodes according to the speed ratio of the servo motor, and generate B-spline basis functions that adapt to the current working conditions. Here, the speed ratio is the ratio of the average speed to the rated speed. When the speed ratio is in different predetermined ranges, the non-uniform B-spline nodes have different control point density distributions on both sides of the rated speed to match the nonlinear distribution characteristics of the motor's external characteristics.

[0019] Step S103: The B-spline basis function and control point weights are forward propagated using the KAN neural network to output the initial weight benchmark values ​​of each evaluation index, which are then normalized to obtain the basic weight vector.

[0020] Step S104: Using the FAHP judgment matrix corrected by the thermo-mechanical coupling degradation factor as the optimization benchmark, the control point weights of the KAN neural network are iteratively updated using the gradient descent method until the consistency ratio of the FAHP judgment matrix meets the predetermined convergence condition, and the optimal weight vector is output. The thermo-mechanical coupling degradation factor is calculated based on the degree to which the average load rate deviates from the predetermined optimal operating point. The correction includes adjusting the judgment scale of the torque index and power index in the FAHP judgment matrix relative to the speed index and load rate index based on the thermo-mechanical coupling degradation factor.

[0021] Step S105: The normalized values ​​of the evaluation indicators are weighted and summed with the optimal weight vector, and the weighted summation result is corrected with a physical penalty factor to obtain the final evaluation score. The mechanical performance is then graded based on the final evaluation score. The physical penalty factor is generated based on the fusion result of the measured power physical residual and the rated parameter self-consistency residual. The measured power physical residual is calculated based on the physical conservation relationship between average torque, average speed and average output power, and the rated parameter self-consistency residual is calculated based on the physical conservation relationship between rated torque, rated speed and rated power.

[0022] The non-uniform B-spline node configuration forms the geometric basis for subsequent KAN weight mapping—the spatial distribution of the B-spline nodes determines in which operating condition ranges the KAN network achieves higher fitting resolution. If a uniform node configuration is used, performance degradation characteristics cannot be effectively captured in nonlinearly sensitive regions such as the weak magnetic field zone, leading to distorted weight mapping. By differentially configuring the control point density according to different speed ratio ranges, the B-spline basis functions are matched with the nonlinear distribution characteristics of the motor's external properties, providing the KAN network with an optimal input space sampling strategy.

[0023] The role of the KAN neural network is to perform forward propagation operations on the B-spline basis functions representing the spatial location of the working condition and the control point weights as learnable parameters, and output the initial weight baseline values ​​of each index under the current working condition. The core advantage of KAN is that its learnable activation functions are located on the edges rather than on the nodes, which can fit complex nonlinear weight mapping relationships with fewer parameters.

[0024] The FAHP decision matrix serves as the decision-making benchmark for determining the relative importance of various indicators. However, the traditional FAHP matrix is ​​fixed and cannot adapt to changes in operating conditions. The introduction of the thermo-mechanical coupling degradation factor gives the decision matrix operating condition responsiveness—when the load rate deviates from the optimal operating point, it means the motor has entered an overload or overheating state. In this case, the evaluation priority of torque and power indicators should be increased accordingly. This dynamic response is achieved by adjusting the matrix scaling. Using the modified FAHP matrix as the optimization benchmark, the control point weights of KAN are iteratively updated through gradient descent until the matrix consistency ratio converges, ensuring that the weight vector satisfies both the fuzzy consistency mathematical constraints and reflects the true importance ranking of the current operating condition.

[0025] The dual-dimensional physical residual is the core of the noise resistance score. Relying solely on data-driven weighted summations will significantly deviate from the true values ​​when sensor data is distorted. The measured power physical residual is used to verify whether the currently acquired torque, speed, and power satisfy the mechanical power conservation principle (P = T·ω); the rated parameter self-consistency residual is used to verify whether the torque, speed, and power in the nameplate parameters satisfy the same physical relationship, identifying parameter input errors or abnormal intrinsic parameters. The two residuals construct physical constraints from the dimensions of the "real-time data chain" and the "intrinsic parameter chain," respectively. The resulting continuous physical penalty factor is neither a hard threshold rejection (directly reporting an error when there is physical inconsistency) nor a complete ignore, directly substituting distorted data. Instead, it applies a continuous reduction in weighting penalty to the weighted score when physical deviations are large, achieving soft-tolerance assessment while preserving assessment capability and ensuring result reliability.

[0026] In the following text, the servo motor mechanical performance grading and evaluation system will be referred to as the system.

[0027] This disclosure proposes a servo motor mechanical performance grading and evaluation scheme based on the fusion of KAN-FAHP and virtual-real bidirectional physical residuals. Through a standardized, progressive, and closed-loop iterative technical process, it achieves accurate perception of motor performance, adaptive weight optimization based on operating conditions, and robust noise reduction scoring. The complete closed loop includes: offline parameter configuration and real-time data acquisition; adaptive configuration of non-uniform B-spline nodes based on speed ratio for refined analysis of operating condition characteristics; forward propagation of the B-spline basis functions and control point weights using a KAN neural network to output initial weight benchmark values; calculation of the thermo-mechanical coupling degradation factor based on the degree of load rate deviation from the optimal operating point and dynamic reconstruction of the FAHP judgment matrix; iterative updating of the KAN control point weights using gradient descent until consistent convergence; weighted summation of normalized evaluation indicators and optimal weights, fusion of virtual-real bidirectional physical residuals to generate a continuous physical penalty factor for soft correction of the score; intelligent determination of multi-level performance grades; and GUI-based online monitoring and report output for traceability, forming a complete closed-loop technical process.

[0028] This solution addresses the technical shortcomings of traditional evaluation methods—statically fixed weights, poor nonlinear characterization capabilities, lack of physical constraints, and low tolerance for data distortion—through the synergistic effect of three core mechanisms. First, the combination of non-uniform B-spline node configuration and KAN weight mapping enables the system to automatically adjust the sampling distribution based on the speed ratio, providing higher mapping resolution in nonlinearly sensitive regions such as weak magnetic fields, ensuring that the initial weights accurately adapt to the current speed conditions. Second, the thermo-mechanical coupling degradation factor drives the dynamic reconstruction of the FAHP judgment matrix, allowing the weights of indicators such as torque and power to be nonlinearly and adaptively adjusted according to the load rate deviation. This automatically amplifies the judgment scale in the overload heating range, effectively solving the problems of inaccurate evaluation and underestimated overload risk under variable operating conditions such as high-speed weak magnetic fields and overload heating caused by traditional static weighting. Third, the fusion of virtual and real-world bidirectional physical residuals generates a continuous physical penalty factor. When sensor data contains noise, drift, or physical inconsistencies, it automatically performs soft weighting correction on the weighted score, avoiding false high scores or incorrect assessments. Simulation experiments show that its physical violation interception rate under data distortion conditions reaches 96.8%, and its anti-distortion attenuation rate is more than 50% better than the comparison model. The deep integration of these three innovations completely breaks down the technical barriers between physical mechanism modeling and intelligent data assessment, significantly improving the working condition adaptability and industrial field reliability of servo motor mechanical performance classification assessment, and realizing the full-process engineering implementation from theoretical algorithm modeling to direct deployment in the industrial field.

[0029] In step S101, before calculating the evaluation indicators based on the rated parameters and real-time operating parameters, the method further includes: removing input data with a self-consistency deviation of rated parameters exceeding 10%, and removing input data with measured output power exceeding 150% of rated power; and performing normalization processing according to the attribute type of each evaluation indicator, including: torque error rate and speed deviation rate are extremely small indicators, using extremely small normalization method; power achievement rate is an extremely large indicator, using extremely large normalization method; average load rate is a moderate indicator, using moderate normalization method, with its optimal operating point being 70%. And / or, after classifying mechanical performance based on the final evaluation score, the method further includes: triggering a data inconsistency warning when the deviation between the average load rate and the power achievement rate exceeds 20%, triggering a mechanical transmission maintenance prompt when the torque error rate exceeds 10%, triggering a controller parameter mismatch prompt when the speed deviation rate exceeds 5%, and triggering an inefficient operation warning or an overload operation risk warning when the average load rate is below 30% or above 100%, respectively.

[0030] In this scheme, pre-emptive physical constraint verification eliminates contradictory physical data at the source, preventing invalid input from contaminating the assessment model and improving the authenticity and reliability of the input data. Differential normalization ensures that all indicators are within the same numerical range and evaluation dimension before weighted summation, guaranteeing the mathematical rationality of multi-indicator fusion calculations. The post-conformity early warning mechanism forms an independent monitoring link decoupled from the main assessment process. It automatically identifies abnormal states and provides maintenance alerts without interfering with the main assessment process, enabling the system to simultaneously possess assessment and diagnostic functions, thus enhancing the decision support capabilities for industrial field operations and maintenance.

[0031] Step S101 strictly follows the IEC 60034 series of rotating electrical machine performance and test methods standards, completes the standardized acquisition of rated inherent parameters and real-time measured parameters of servo motors, physical constraint verification, derived index calculation and normalization preprocessing, constructs a standardized, verifiable and highly adaptable four-dimensional mechanical performance evaluation feature system, abandons the traditional unconstrained, static and single data acquisition mode, and provides a standardized data benchmark that conforms to the physical mechanism of the motor for subsequent intelligent evaluation models.

[0032] This system synchronously collects the rated parameters and real-time measured operating parameters of the servo motor, constructing a multi-dimensional mechanical performance evaluation input feature system that comprehensively covers the inherent attributes and dynamic operating status of the motor. The rated parameters are obtained based on the equipment nameplate and the IEC 60034 standard, mainly including rated torque, rated speed, and rated power. Real-time measured operating parameters include average torque, average speed, average output power, and average load rate. The load rate can be automatically calculated by the system based on power achievement rate or manually entered, with the valid value strictly limited to the range of 0 to 150% to avoid abnormal data exceeding limits.

[0033] Based on the aforementioned raw data acquisition parameters, and combining the dynamics and power transmission mechanism of the servo motor, four core quantitative evaluation indicators are derived and calculated. These indicators are then categorized into three types—extremely small, extremely large, and moderate—accurately characterizing the overall mechanical performance of the motor. These four core quantitative evaluation indicators include: Torque error rate is a very small indicator. The smaller the value, the better the motor output performance. It is used to characterize the degree of deviation between the actual output torque of the motor and the rated standard torque, and directly reflects the mechanical output stability and torque control accuracy of the servo system. Speed ​​deviation rate is a very small indicator. The smaller the value, the higher the speed control accuracy. It is used to characterize the deviation level between the actual stable operating speed of the motor and the rated nominal speed, and reflects the steady-state performance of the servo speed closed-loop control. Power achievement rate is a very large indicator. The larger the value, the stronger the power output capability. It is used to characterize the degree of matching between the actual output power of the motor and the rated power, and objectively reflects the motor's power output margin and load adaptability. Average load rate is a moderate indicator, with an optimal operating benchmark of 70% load rate. It is used to characterize the matching rationality of the current operating conditions of the motor and can effectively identify adverse operating conditions such as inefficiency under light load, overload under heavy load, and operating conditions deviating from the optimal range.

[0034] To prevent invalid data and non-compliant operating conditions from interfering with the accuracy of the evaluation, a strict physical constraint verification mechanism has been added to this step to ensure the authenticity and rationality of the input data from the source. Specifically, on the one hand, a self-consistency verification of rated parameters is carried out. Based on the inherent physical correspondence between motor speed, torque, and power, the theoretical rated power is obtained by converting rated torque and rated speed, and compared with the rated power on the equipment nameplate, eliminating contradictory rated parameters with deviations exceeding 10%. On the other hand, an over-limit verification of actual operating conditions is carried out, automatically eliminating invalid data that exceeds the rated power by 150%, ensuring that all input data fully conforms to the physical operating laws of the servo motor.

[0035] Based on the industrial servo motor operation and maintenance grading standards, a three-level performance evaluation and early warning threshold system is established, including three performance levels: Level 1 Excellent, Level 2 Good, and Level 3 Qualified, corresponding to comprehensive score ranges of 85 points and above, 70 to 84 points, and below 70 points, respectively. Simultaneously, for the four core evaluation indicators, optimal ideal values ​​and worst-case limits are clearly defined, and quantitative evaluation boundaries are delineated: the optimal value for torque error is 0%, and the worst value is 10%; the optimal value for speed deviation is 0%, and the worst value is 5%; the optimal value for power achievement rate is 100%, and the worst value is 30%; the optimal operating point for load rate is 70%, and the effective adaptable operating range is 30% to 100%.

[0036] Furthermore, targeted differential normalization processing is implemented for the three types of differentiated indicators: extremely small, extremely large, and moderate. All indicator values ​​are uniformly mapped to the standard range of 0 to 1, completely eliminating the differences in dimensions and attributes between different indicators. This enables the fusion calculation of multi-dimensional heterogeneous indicators and provides a unified and standardized data benchmark for subsequent intelligent evaluation and weight iteration optimization of the KAN-FAHP model.

[0037] In step S102, when configuring non-uniform B-spline nodes according to the speed ratio of the servo motor, if the speed ratio is less than or equal to 1.0, 70% of the control points are configured in the low speed to rated speed range and 30% of the control points are configured in the high speed standby range; if the speed ratio is greater than 1.0, 50% of the control points are configured in the low speed to rated speed range and 50% of the control points are configured in the rated speed to current operating speed range.

[0038] Specifically, a speed ratio of 1.0 is the physical inflection point of the servo motor's external characteristics. Below this value, the motor operates in the rated constant torque region, while above it, it enters the high-speed field-weakening constant power region. The nonlinear characteristic distributions of these two regions are drastically different. When the speed ratio is less than or equal to 1.0, the nonlinear performance fluctuations of the motor are mainly concentrated in the low-speed to rated speed range, i.e., the field-weakening transition region. Therefore, 70% of the control points are densely arranged in this region to precisely capture the nonlinear characteristics of this range. When the speed ratio is greater than 1.0, the motor has entered the high-speed field-weakening region, and 50% of the control points need to be densely arranged in the range from the rated speed to the current high-speed operation to enhance the characterization capability of this region.

[0039] The control point density allocation strategy is automatically switched according to the physical operating range of the speed ratio, enabling the B-spline basis function to have higher fitting resolution in the region with the most significant nonlinearity, and to be sparsely arranged in the region with weaker nonlinearity to reduce computational overhead. This achieves adaptive optimization of the sampling resources based on operating conditions. Compared with uniform node configuration, this significantly improves the accuracy of capturing the nonlinear performance degradation characteristics of the weak magnetic transition region and the high-speed weak magnetic region without increasing the total number of control points, providing more accurate input space sampling for subsequent KAN weight mapping.

[0040] The technical solution of this disclosure overcomes the shortcomings of traditional FAHP (Analytic Analog-Fractional Hierarchy Process) methods, such as static weight fixation, poor adaptability to operating conditions, neglect of motor external characteristic nonlinearity, and detachment from physical mechanisms. It integrates four core mechanisms: non-uniform B-spline adaptive node configuration, dynamic matrix reconstruction of thermo-mechanical coupling factors, KAN neural network consistency iterative optimization, and virtual-real bidirectional physical residual noise-resistant scoring. This constructs a multi-algorithm collaborative adaptive intelligent evaluation model that deeply couples data-driven and physical mechanisms. This model can dynamically update evaluation weights and correct evaluation scores based on the real-time speed ratio of the servo motor and load conditions, effectively suppressing evaluation errors caused by data distortion and operating condition disturbances, and achieving accurate quantitative evaluation of motor mechanical performance under complex and variable operating conditions.

[0041] In the model construction process, the initial configuration of basic parameters and core matrices is completed first. The weight values ​​of the four evaluation indicators are constrained to a range of 0.15 to 0.45 to ensure that the weight distribution of each indicator is balanced and reasonable without extreme bias. The basic FAHP judgment matrix corresponding to the four-dimensional indicators is initialized to construct the relative importance evaluation benchmark between each performance indicator, providing an initial benchmark for subsequent dynamic reconstruction. Simultaneously, the core parameters of the KAN neural network are initialized, and the spline control point matrix is ​​set to adapt to the four evaluation indicators. Each indicator corresponds to five sets of B-spline control points to ensure the accuracy of nonlinear fitting. The maximum number of iterations is set to 500, and the gradient descent learning rate is set to 0.02 to balance convergence speed and optimization accuracy. At the same time, based on the external characteristics of servo motor operation, two core operating condition feature parameters are defined: the speed ratio representing the speed operating range and the load rate deviation representing the load condition deviating from the optimal state. Based on this, a non-uniform node adaptive generation rule and a thermo-mechanical coupling degradation factor calculation mechanism are established, completely abandoning the traditional fixed mode of uniform node sampling and static weight allocation.

[0042] To address the differences in nonlinear external characteristics between the rated constant torque operating region and the high-speed field-weakening constant power operating region of a servo motor, an adaptive non-uniform B-spline node allocation strategy is designed. First, the speed ratio parameter is numerically clamped, limiting the effective range to 0.1 to 2.0 to filter out invalid data interference from start-stop moments and extreme abnormal speeds. Then, the control point density is dynamically configured according to the speed ratio range: when the speed ratio is less than or equal to 1.0, the motor is in its rated normal operating range, with 70% of the control points densely distributed in the low-speed to rated speed range to precisely capture the nonlinear performance fluctuations in the field-weakening transition region, and the remaining 30% evenly distributed in the high-speed standby range for overall coverage; when the speed ratio is greater than 1.0, the motor enters the high-speed field-weakening operating region, with 50% of the control points evenly distributed in the low-speed to rated speed range, and the remaining 50% concentrated in the range from rated speed to the current high-speed operating range to enhance the nonlinear characteristic representation capability of the high-speed field-weakening region. Based on the aforementioned adaptive density distribution rules, non-uniform node vectors are generated, and triangular B-spline basis functions adapted to real-time working conditions are constructed to achieve adaptive nonlinear fitting of complex working condition input distribution. Through forward propagation operation of KAN network, combined with the control point weight parameters and B-spline basis function fitting results, the initial weight benchmark values ​​of four performance indicators are calculated. After physical constraint clipping and overall normalization, the initial basic weights of each indicator adapted to the current working condition are output, completing the preliminary matching and initialization of the working condition adaptive weights.

[0043] For four differentiated indicators—torque error rate, speed deviation rate, power achievement rate, and average load rate—precise normalization corrections are achieved by matching extremely small, extremely large, and moderate indicator attributes. This unifies the data value distribution of each indicator, eliminates differences in units and evaluation orientations, and ensures the rationality of multi-indicator fusion calculations. Simultaneously, a multi-dimensional operating condition consistency verification and intelligent early warning mechanism is established: a data inconsistency warning is triggered when the deviation between load rate and power achievement rate exceeds 20%; when the torque error rate exceeds 10%, it indicates potential problems such as wear and abnormal clearance in the mechanical transmission mechanism, suggesting maintenance; when the speed deviation rate exceeds 5%, it indicates poor servo controller parameter adaptation and decreased speed control accuracy; when the load rate is below 30% or above 100%, inefficient operation warnings and overload operation risk warnings are triggered respectively. Furthermore, boundary truncation is applied to out-of-limit normalized values ​​to prevent invalid data from lowering the overall evaluation accuracy.

[0044] In step S103, FAHP consistency iteration is performed through KAN optimization. Specifically, the dynamically reconstructed adaptive FAHP judgment matrix is ​​used as the optimization benchmark, and a matrix consistency ratio of less than 0.02 is set as the high-precision convergence target. The gradient descent method is used to continuously iteratively update the spline control point parameters of the KAN network, achieving high-precision adaptive optimization of the evaluation weights. The iteration process is as follows: First, the initial weights of each performance index under the current working condition are calculated through forward propagation of the KAN network. Combined with the physical constraints of the motor, the weights are pruned and normalized to obtain an effective real-time weight vector. Second, the maximum eigenvalue and consistency index of the FAHP judgment matrix are calculated. Combined with the fixed random consistency index of the fourth-order matrix, the real-time matrix consistency ratio is solved to quantitatively evaluate the rationality of the current weight system. Third, the model loss gradient is solved through numerical approximation methods, the KAN network control point parameters are updated in reverse, and the nonlinear fitting weights are fine-tuned. The system repeatedly executes the above forward calculation, consistency verification and parameter reverse update process until the matrix consistency ratio meets the convergence accuracy requirements or the number of iterations reaches the maximum limit, then terminates the iterative optimization process, and finally outputs the optimal adaptive weight vector that adapts to the current speed, load and thermal state, so as to realize the fully automatic high-precision dynamic adaptation of the weights of the four performance indicators with complex changing operating conditions.

[0045] While performing iterative optimization of KAN weights, the system employs a closed-loop iterative operation logic, cyclically executing the entire process of multi-source feature acquisition, index calculation, adaptive generation of non-uniform B-spline nodes, solution of thermo-mechanical coupling degradation factors, dynamic reconstruction of the FAHP matrix, iterative optimization of KAN weights, index normalization correction, bidirectional physical residual score correction, and operating condition consistency verification warnings. This continuously iterates and optimizes the model weight parameters and evaluation scores. The closed-loop iteration terminates when the FAHP matrix consistency ratio meets the convergence accuracy or reaches the maximum number of iterations. After convergence, the system comprehensively matches the performance of each index component, the adaptively optimized weights, and the physical residual correction results to complete a comprehensive, high-precision quantitative evaluation of the servo motor's mechanical performance.

[0046] Before step S104, the thermo-mechanical coupling degradation factor can be determined as follows: when the deviation of the average load rate from the predetermined optimal operating point is greater than 0, the thermo-mechanical coupling degradation factor is calculated according to the exponential nonlinear rule. The greater the deviation, the greater the thermo-mechanical coupling degradation factor. When the deviation is less than or equal to 0, the thermo-mechanical coupling degradation factor is a predetermined standard value. The correction of the FAHP judgment matrix includes updating the four judgment scales of torque index relative to speed index, torque index relative to load rate index, power index relative to speed index, and power index relative to load rate index according to the thermo-mechanical coupling degradation factor, and synchronously updating the symmetrical position elements according to the matrix symmetry and reciprocity characteristics.

[0047] A deviation greater than 0 indicates that the actual load rate exceeds the optimal operating point (70%), and the motor enters an overload operation state. At this point, mechanical stress concentration and heat loss intensify, significantly increasing the importance of torque and power indicators in evaluating mechanical performance. An exponential nonlinear rule is used to calculate the degradation factor, ensuring that the factor increases faster with greater deviation, thus more aggressively amplifying the weight of core power indicators. The modified scale is specifically divided into four categories: torque relative to speed, torque relative to load rate, power relative to speed, and power relative to load rate—these four scales directly determine the relative priority of torque and power indicators in the comprehensive evaluation. Simultaneously, the symmetrical elements are updated synchronously according to the symmetric and reciprocal characteristics of the FAHP matrix to ensure the mathematical consistency of the matrix.

[0048] The thermo-mechanical coupling degradation factor is calculated using exponential nonlinear rules, resulting in a nonlinear positive correlation between the degree of overload and the magnitude of weight adjustment. Adjustments are gradual when deviations are slight and rapid when deviations are severe, accurately matching the nonlinear deterioration trend of overload heating. Through targeted updates of four key judgment scales and synchronous correction of symmetrical position elements, the FAHP judgment matrix can continuously and dynamically evolve with load conditions. This significantly improves the evaluation priority of torque and power indicators in the overload heating range, completely solving the problems of traditional fixed matrices failing to respond to accumulated thermal effects and underestimating overload risk.

[0049] To address the issues that traditional FAHP fixed weights cannot adapt to motor overload heating conditions and the distortion of torque and power performance evaluation priorities, this disclosure proposes a thermo-mechanical coupling degradation factor based on load rate deviation. This enables real-time dynamic reconstruction of the FAHP judgment matrix, allowing the index weights to adaptively adjust with load conditions and heating degradation states to match the actual impact weights on motor performance under different conditions.

[0050] When the load rate deviation is greater than 0, the motor is in the overload and overheating operating range, the load exceeds the optimal operating point, mechanical stress is concentrated, and heat loss is aggravated. At this time, the thermo-mechanical coupling degradation factor is calculated by an exponential nonlinear amplification rule. The larger the deviation, the faster the factor increases, which significantly increases the evaluation weight of the two core mechanical power indicators, torque and power, highlighting the evaluation priority of mechanical load-bearing capacity and power output stability under overload conditions. When the load rate deviation is less than or equal to 0, the motor is in a light load or rated optimal steady-state condition. The thermo-mechanical coupling degradation factor is fixed at a predetermined standard value, keeping the basic FAHP judgment matrix stable to ensure the smoothness of steady-state condition evaluation.

[0051] Based on real-time calculated thermo-mechanical coupling degradation factors, the internal correlation elements of the FAHP judgment matrix are dynamically updated. The focus is on correcting the relative importance scales between torque and speed, torque and load rate, power and speed, and power and load rate. Simultaneously, elements at symmetrical positions are updated synchronously according to the matrix's symmetric and reciprocal properties, strictly ensuring the consistency and rationality of the judgment matrix. Through this dynamic matrix reconstruction, the evaluation system achieves real-time adaptation to different operating conditions such as overload heating, rated steady-state operation, and light-load operation, completely solving the problem of poor adaptability of traditional fixed-matrix operating conditions.

[0052] The predetermined convergence condition in step S104 can be that the consistency ratio of the FAHP judgment matrix is ​​less than 0.02; the learning rate of the gradient descent method can be 0.02, and the maximum number of iterations can be 500.

[0053] The consistency ratio (CR) of the FAHP judgment matrix is ​​a key indicator for measuring the logical consistency of the matrix; a smaller CR value indicates higher logical self-consistency in weight judgments. The stochastic consistency index (RI) for a fourth-order matrix is ​​a fixed value of 0.90, and the convergence target is set to CR < 0.02, far stricter than the 0.10 threshold typically accepted by traditional FAHP, ensuring higher logical rigor in the output weight vector. A learning rate of 0.02 controls the step size of gradient descent, balancing convergence speed and stability; a maximum of 500 iterations is used as a termination condition to prevent infinite loops in extreme cases.

[0054] By setting a consistency ratio convergence threshold far stricter than that of traditional FAHP (0.02 < 0.10), the logical consistency and mathematical rigor of the weight vector are significantly improved, avoiding weight distribution distortion caused by insufficient matrix consistency. Gradient descent, combined with a suitable learning rate and iteration upper limit, controls computational overhead while ensuring convergence accuracy, enabling the system to stably converge to the optimal weight vector that meets high consistency requirements within a finite number of steps.

[0055] In step S105, before correcting the weighted summation result using a physical penalty factor, the method further includes: truncating the measured power physical residual and the rated parameter self-consistency residual by upper limits, respectively, truncating residuals exceeding a predetermined upper limit to the predetermined upper limit, and then merging and averaging the truncated two to obtain a comprehensive physical deviation level. A physical penalty factor is generated based on the comprehensive physical deviation level; the higher the comprehensive physical deviation level, the smaller the physical penalty factor. Specifically, the measured power physical residual is obtained by calculating the theoretical mechanical power based on the product of average torque and average speed, and calculating the deviation between the average output power and the theoretical mechanical power; the rated parameter self-consistency residual is obtained by calculating the theoretical rated power based on the product of rated torque and rated speed, and calculating the deviation between the rated power and the theoretical rated power.

[0056] The measured power physical residual is calculated by multiplying the average torque and average speed collected in real time to determine the theoretical mechanical power that should be output under the current operating conditions. This residual is compared with the average output power measured by the sensor to obtain the degree of deviation. This residual reflects whether the "real-time data chain satisfies power conservation." The rated parameter self-consistency residual is calculated by multiplying the rated torque and rated speed on the nameplate to determine the theoretical rated power. This residual is compared with the rated power on the nameplate to obtain the degree of deviation. This residual reflects whether the "inherent parameter chain is self-consistent" and can identify errors in nameplate parameter input or abnormalities in equipment parameters. To prevent extreme outliers from dominating the penalty factor, the upper limits of both residuals are truncated. The truncated two values ​​are then combined and averaged to obtain the comprehensive physical deviation level. The higher this value, the lower the physical reliability of the data, resulting in a smaller physical penalty factor and a greater reduction in the weighted score.

[0057] Two physical residuals based on the law of energy conservation are used to verify the physical consistency of the input data from two dimensions: the "measured data chain" and the "intrinsic parameter chain," forming a two-way insurance mechanism. The measured power residual is used to identify sensor sampling distortion or strong electromagnetic interference, while the rated parameter residual is used to identify nameplate parameter input errors or abnormal body parameters. Upper limit truncation effectively prevents extreme outliers from over-dominantly affecting the penalty factor, ensuring that the penalty factor changes continuously within a reasonable range. The comprehensive physical deviation level generated by the fusion averaging mechanism is a comprehensive evaluation of the two sets of residuals. The larger the physical deviation, the smaller the penalty factor and the greater the reduction in weighting of the weighted score. This achieves continuous, soft-tolerance fault assessment for data distortion and physical inconsistency conditions, replacing the traditional hard threshold rejection or complete ignoring mode.

[0058] To address the industry challenges of overfitting in purely data-driven models and inaccurate evaluation results due to sensor sampling noise and data distortion, this disclosure presents a virtual-real two-way physical residual verification mechanism. This mechanism combines theoretical physical mechanism constraints with measured data characteristics for two-way verification and introduces an adaptive physical penalty factor to achieve highly interference-resistant and highly reliable quantitative scoring of mechanical performance.

[0059] First, based on the normalized four performance index vectors and the optimal weight vector obtained by KAN-FAHP optimization, the basic evaluation score of the model is calculated by weighted summation and amplification of the percentage base, completing the preliminary performance quantification at the data-driven level. Then, a two-way physical residual verification system is constructed to complete the physical mechanism verification from two dimensions: measured operating data and rated parameters. The first is the measured power physical residual verification, which calculates the theoretical mechanical power that the motor should output under the current operating conditions by combining the correspondence between motor dynamics speed and torque, compares it with the measured output power collected by the sensor, quantifies the deviation between the measured data and the physical mechanism of mechanical power conservation, and truncates the upper limit of the residual value to avoid interference from extreme outliers. The second is the rated parameter self-consistent residual verification, which calculates the standard theoretical rated power based on the motor's rated torque and rated speed, compares it with the rated power on the equipment nameplate, quantifies the self-consistency deviation of the equipment's inherent parameters, and identifies problems such as incorrect rated parameter input or abnormal equipment parameters. Furthermore, the bidirectional residual results are fused and averaged, with the maximum value limited to no more than 1.0, to obtain the comprehensive physical deviation level. Based on this deviation level, the physical penalty coefficient is adaptively calculated. Finally, the basic weighted score is multiplied by the physical penalty coefficient to obtain the final corrected accurate evaluation score. When sensor data is distorted, the measured power deviates from the physical operating laws of the motor, or the self-consistency of the rated parameters is poor, the system automatically applies a reasonable scoring penalty, achieving soft-fault-tolerant evaluation. This effectively avoids false high scores and performance misjudgments caused by abnormal data, significantly improving the model's anti-interference capability and evaluation credibility in complex industrial scenarios.

[0060] Following step S105, mechanical performance is graded based on the final evaluation score, including: When the final evaluation score is greater than or equal to 85 points, the first performance level is output, corresponding to high-precision and high-end application scenarios; when the final evaluation score is greater than or equal to 70 points but less than 85 points, the second performance level is output, corresponding to mid-to-high-end industrial application scenarios; when the final evaluation score is less than 70 points, the third performance level is output, corresponding to conventional working condition application scenarios.

[0061] Mechanical performance is divided into three progressive levels, each corresponding to a percentage score range and a set of industrial application scenarios. The first performance level corresponds to high-precision applications such as semiconductor equipment, 3C precision machining, and precision instruments, requiring excellent motor mechanical performance and extremely high operational stability. The second performance level corresponds to mid-to-high-end industrial scenarios such as CNC machine tools and industrial robots, requiring stable motor performance with no significant degradation. The third performance level corresponds to conventional working conditions such as packaging machinery and logistics transportation, where the motor only needs to have basic operating functions.

[0062] By mapping continuous comprehensive evaluation scores to discrete performance levels through three-tiered thresholds, and directly linking score ranges to specific industrial application scenarios, the evaluation results are transformed from abstract scores into technical language with clear engineering guidance significance. This provides equipment maintenance personnel with intuitive performance judgment criteria and selection and adaptation suggestions, realizing a value closed loop from algorithm evaluation to engineering decision-making.

[0063] This disclosure, based on the convergent and optimized KAN-FAHP and the virtual-real bidirectional physical residual fusion evaluation model, and combined with the actual engineering needs of industrial servo equipment operation, selection, and quality control, establishes a three-level performance grading evaluation system and supporting operating condition adaptation rules to achieve precise matching between performance levels and application scenarios. Specifically, Level 1 corresponds to a comprehensive score of 85 points or above, indicating excellent motor mechanical performance, strong operating condition adaptability, and high operational stability, suitable for high-precision and high-end application scenarios such as semiconductor equipment, 3C precision machining, and precision instruments; Level 2 corresponds to a comprehensive score of 70 to 84 points, indicating stable motor operation performance without significant degradation, suitable for mid-to-high-end industrial scenarios such as CNC machine tools and industrial robots; Level 3 corresponds to a comprehensive score below 70 points, indicating normal basic motor operation functions, suitable for conventional operating scenarios such as packaging machinery, logistics transportation, and general-purpose automation equipment.

[0064] After the model iterative convergence, the system synchronously outputs complete quantitative evaluation results, including: the original measured values, standard reference values, error deviation, normalized values, optimized weight proportions, and indicator attributes of the four evaluation indicators; the comprehensive percentage score of the motor's mechanical performance and its corresponding performance level; textual recommendations for equipment-suitable application scenarios; and a list of warnings for system consistency verification. Simultaneously, the system automatically archives and saves the complete evaluation report in text format, ensuring that the evaluation results are quantifiable, traceable, and reproducible.

[0065] To meet the lightweight, visualized, and convenient operation and maintenance needs of industrial sites, this disclosure embodiment simultaneously develops an integrated GUI human-computer interaction operation and maintenance interface based on the MATLAB platform, integrating a complete set of evaluation and monitoring function modules. The interface has a built-in rated parameter and measured data input panel, supporting manual input of motor rated torque, rated speed, rated power, and measured operating parameters. It also provides quick functions for data file loading and automatic load rate calculation to adapt to diverse data input scenarios. The interface features a visualized evaluation result display area, dynamically displaying the comprehensive score through a dashboard, using color blocks to indicate performance levels, and using text push notifications adapted to application scenarios. It sets up a refined indicator data table, displaying indicator names, measured values, reference values, deviations, normalized values, weight percentages, and indicator types from multiple dimensions, achieving transparent parameter display. Simultaneously, it integrates multi-dimensional visualized analysis charts such as normalized indicator bar charts, weight distribution pie charts, error analysis curves, and performance trend comparison charts. All charts support adaptive window scaling to adapt to different display devices.

[0066] In addition, the system is equipped with a historical evaluation record module, which automatically saves the most recent 10 evaluation data with timestamps. It supports backtracking and querying historical information such as torque, speed, comprehensive score, and performance level, and supports one-click generation and export of complete evaluation reports. It adopts a universal encoding format to ensure the universality and archiving of data. The system also has a built-in automated simulation comparison experiment module, which can complete four types of extreme working condition tests with one click: rated benchmark, high-speed weak magnetic field, overload heating, and data distortion. It simultaneously completes baseline comparisons between traditional static FAHP, uniform KAN, and unidirectional physical constraint models and this solution, automatically calculates four core performance indicators: expert consensus error, anti-distortion attenuation rate, physical violation interception rate, and working condition adaptive sensitivity, and generates a visual comparison chart to provide data support for solution performance verification and patent technology demonstration.

[0067] like Figures 2 to 6 As shown, the embodiments of this disclosure verify the system functionality through automated simulation comparison experiments. The KAN-FAHP + physical constraint model of this scheme is compared horizontally with three baseline models: traditional static FAHP, standard uniform KAN, and unidirectional physical PINN. Specifically, the system functionality is verified through calculations such as... Figure 2 The truth error RMSE shown is as follows: Figure 3 The anti-distortion attenuation rate shown is as follows: Figure 4 The physical violation interception rate and such Figure 5 The adaptive sensitivity under the operating conditions shown is used to comprehensively verify the superiority and feasibility of the model.

[0068] Simulation results show that, after adopting this embodiment, the system's overall evaluation performance achieves a comprehensive leap forward: its true error RMSE is significantly reduced to 1.46, a reduction of 89.8% compared to the suboptimal model; facing sensor data distortion, the anti-distortion attenuation rate (PDR) is strictly controlled at 6.20% thanks to the dual defense of physical constraints and dynamic weighting; simultaneously, under extreme conditions such as high-speed weak magnetic fields and overload heating, the model achieves a working condition adaptive sensitivity WDS of 46.50%, and eliminates evaluation results that violate physical principles with a physical violation interception rate of 96.80%, successfully achieving a deep fusion of high precision data-driven approaches and high reliability physical models. These simulation results can be demonstrated through... Figure 6 The visualization tools shown are displayed.

[0069] All of the above-mentioned optional technical solutions can be combined in any way to form the optional embodiments of this application, and will not be described in detail here.

[0070] According to the servo motor mechanical performance grading evaluation method of this disclosure, the rated parameters and real-time operating parameters of the servo motor are collected, evaluation indicators are calculated and normalized; non-uniform B-spline nodes are configured based on the speed ratio, and the B-spline basis function and control point weights are forward propagated through a KAN neural network to output the initial weight benchmark value; the thermo-mechanical coupling degradation factor is calculated based on the load rate deviation, the scaling of the FAHP judgment matrix is ​​dynamically corrected, and the control point weights are iteratively updated with gradient descent until consistent convergence, outputting the optimal weight vector, so that the weights are optimized in real time with the working conditions, solving the problem of inaccurate evaluation under variable working conditions such as weak magnetic field and overload of traditional static weighting; the normalized index and the optimal weight are weighted and summed, and the measured power conservation is verified based on the torque-speed product, and the self-consistency is verified based on the rated parameters, generating the measured power residual and the rated parameter residual and merging them into a continuous physical penalty factor, which is used to softly correct the weighted score, and finally output the performance level and applicable scenario, realizing the fault-tolerant evaluation of data distortion and physical inconsistency, and significantly improving the working condition adaptability and field reliability of the evaluation.

[0071] The following are system embodiments of this disclosure, which can execute the method embodiments of this disclosure. The methods provided in the embodiments of this disclosure can be executed by any electronic device with computer processing capabilities, such as an industrial control computer. The servo motor mechanical performance grading and evaluation system described below and the servo motor mechanical performance grading and evaluation method described above can be referred to each other. For details not disclosed in the system embodiments of this disclosure, please refer to the method embodiments of this disclosure.

[0072] Figure 7This is a schematic diagram of the structure of a servo motor mechanical performance grading and evaluation system provided in an embodiment of this disclosure. Figure 7 As shown, the servo motor mechanical performance grading and evaluation system includes: The data acquisition and processing module 701 is used to acquire the rated parameters and real-time operating parameters of the servo motor, calculate the evaluation index based on the rated parameters and real-time operating parameters, and perform normalization processing according to the attribute type of each evaluation index. The rated parameters include at least rated torque, rated speed and rated power, and the real-time operating parameters include at least average torque, average speed, average output power and average load rate. The evaluation index includes at least torque error rate, speed deviation rate, power achievement rate and average load rate.

[0073] The non-uniform B-spline node configuration module 702 is used to configure non-uniform B-spline nodes according to the speed ratio of the servo motor and generate B-spline basis functions that are adapted to the current working conditions. The speed ratio is the ratio of the average speed to the rated speed. When the speed ratio is in different predetermined ranges, the non-uniform B-spline nodes have different control point density distributions on both sides of the rated speed to match the nonlinear distribution characteristics of the motor's external characteristics.

[0074] The KAN neural network weight mapping module 703 is used to perform forward propagation operations on the B-spline basis function and control point weights through the KAN neural network, output the initial weight benchmark values ​​of each evaluation index, and obtain the basic weight vector after normalization.

[0075] The weight iteration optimization module 704 is used to iteratively update the control point weights of the KAN neural network using the FAHP judgment matrix corrected by the thermo-mechanical coupling degradation factor as the optimization benchmark, and until the consistency ratio of the FAHP judgment matrix meets the predetermined convergence condition, and outputs the optimal weight vector. The thermo-mechanical coupling degradation factor is calculated based on the degree to which the average load rate deviates from the predetermined optimal operating point. The correction includes adjusting the judgment scale of the torque index and power index in the FAHP judgment matrix relative to the speed index and load rate index according to the thermo-mechanical coupling degradation factor.

[0076] The comprehensive scoring module 705 is used to perform a weighted summation of the normalized values ​​of the evaluation indicators and the optimal weight vector, and to correct the weighted summation result with a physical penalty factor to obtain the final evaluation score. The mechanical performance is then graded based on the final evaluation score. The physical penalty factor is generated based on the fusion result of the measured power physical residual and the rated parameter self-consistency residual. The measured power physical residual is calculated based on the physical conservation relationship between average torque, average speed and average output power, and the rated parameter self-consistency residual is calculated based on the physical conservation relationship between rated torque, rated speed and rated power.

[0077] According to the servo motor mechanical performance grading and evaluation system of this disclosure, the rated parameters and real-time operating parameters of the servo motor are collected, evaluation indicators are calculated and normalized; non-uniform B-spline nodes are configured based on the speed ratio, and the B-spline basis function and control point weights are forward propagated through a KAN neural network to output the initial weight benchmark value; the thermo-mechanical coupling degradation factor is calculated based on the load rate deviation, the scaling of the FAHP judgment matrix is ​​dynamically corrected, and the control point weights are iteratively updated with gradient descent until consistent convergence, outputting the optimal weight vector, so that the weights are optimized in real time with the working conditions, solving the problem of inaccurate evaluation under variable working conditions such as weak magnetic field and overload of traditional static weighting; the normalized indicators and the optimal weights are weighted and summed, and the measured power conservation is verified based on the torque-speed product, and the self-consistency is verified based on the rated parameters, generating the measured power residual and the rated parameter residual and merging them into a continuous physical penalty factor, which is used to softly correct the weighted score, and finally output the performance level and applicable scenario, realizing the fault-tolerant evaluation of data distortion and physical inconsistency, significantly improving the working condition adaptability and field reliability of the evaluation.

[0078] In practical applications, radial basis function neural networks or support vector regression can be used instead of KAN for weight mapping. These can also generate a non-uniform distribution of basis function center points based on the speed ratio, achieving adaptive fitting of FAHP weights under operating conditions. Furthermore, the thermo-mechanical coupling degradation factor can be calculated using the load rate integral over time, or the measured value of the motor winding temperature can be directly introduced as input. When the temperature exceeds a threshold, the weights of torque and power are nonlinearly amplified. The virtual-real bidirectional physical residual mechanism can be extended to multi-directional residuals. For example, the deviation between the measured vibration amplitude and the standard value (mechanical vibration residual) and the difference between the harmonic distortion rate and the limit (current harmonic residual) can be added. Multiple residuals can be fused into a comprehensive penalty factor through weighting or fuzzy logic. In the normalization and scoring stages, the ideal range of moderate indicators can be extended from a single point to a range, for example, setting the load rate to 65%–75%, and using a trapezoidal membership function instead of the current piecewise linear function. The output of the graded assessment can be linked with a digital twin model, using physical residuals as degradation features input to the life prediction model to achieve dynamic remaining life warning based on real-time assessment results. The GUI evaluation history can be further enhanced with similar case retrieval and recommendation functions. Based on the current input features, it can automatically match the working conditions with the closest scores in the history and the handling suggestions to assist in operation and maintenance decisions.

[0079] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this disclosure.

[0080] Figure 8 This is a schematic diagram of the electronic device 8 provided in an embodiment of this disclosure. Figure 8As shown, the electronic device 8 of this embodiment includes a processor 801, a memory 802, and a computer program 803 stored in the memory 802 and executable on the processor 801. When the processor 801 executes the computer program 803, it implements the steps in the various method embodiments described above. Alternatively, when the processor 801 executes the computer program 803, it implements the functions of each module in the various device embodiments described above.

[0081] Electronic device 8 can be a desktop computer, laptop, handheld computer, cloud server, or other electronic device. Electronic device 8 may include, but is not limited to, processor 801 and memory 802. Those skilled in the art will understand that... Figure 8 This is merely an example of electronic device 8 and does not constitute a limitation on electronic device 8. It may include more or fewer components than shown, or different components.

[0082] The processor 801 can be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc.

[0083] The memory 802 can be an internal storage unit of the electronic device 8, such as a hard disk or RAM of the electronic device 8. The memory 802 can also be an external storage device of the electronic device 8, such as a plug-in hard disk, Smart Media Card (SMC), Secure Digital (SD) card, Flash Card, etc., equipped on the electronic device 8. The memory 802 can also include both internal and external storage units of the electronic device 8. The memory 802 is used to store computer programs and other programs and data required by the electronic device.

[0084] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is merely an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiments can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0085] If the integrated module is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program may include computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium may include: any entity or device capable of carrying computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc. It should be noted that the content included in the computer-readable medium may be appropriately added or removed according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, the computer-readable medium does not include electrical carrier signals and telecommunication signals.

[0086] The above embodiments are only used to illustrate the technical solutions of this disclosure, and are not intended to limit it. Although this disclosure has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this disclosure, and should all be included within the protection scope of this disclosure.

Claims

1. A method for grading and evaluating the mechanical performance of a servo motor, characterized in that, include: The rated parameters and real-time operating parameters of the servo motor are collected. Evaluation indicators are calculated based on the rated parameters and real-time operating parameters, and normalization is performed according to the attribute type of each evaluation indicator. The rated parameters include at least rated torque, rated speed and rated power. The real-time operating parameters include at least average torque, average speed, average output power and average load rate. The evaluation indicators include at least torque error rate, speed deviation rate, power achievement rate and average load rate. Based on the speed ratio of the servo motor, non-uniform B-spline nodes are configured to generate B-spline basis functions adapted to the current working conditions. The speed ratio is the ratio of the average speed to the rated speed. When the speed ratio is in different predetermined ranges, the non-uniform B-spline nodes have different control point density distributions on both sides of the rated speed to match the nonlinear distribution characteristics of the motor's external characteristics. The B-spline basis function and control point weights are forward propagated using a KAN neural network to output the initial weight benchmark values ​​of each evaluation index, which are then normalized to obtain the basic weight vector. Using the FAHP judgment matrix corrected by the thermo-mechanical coupling degradation factor as the optimization benchmark, the control point weights of the KAN neural network are iteratively updated using the gradient descent method until the consistency ratio of the FAHP judgment matrix meets the predetermined convergence condition, and the optimal weight vector is output. The thermo-mechanical coupling degradation factor is calculated based on the degree to which the average load rate deviates from the predetermined optimal operating point. The correction includes adjusting the judgment scale of the torque index and power index in the FAHP judgment matrix relative to the speed index and load rate index according to the thermo-mechanical coupling degradation factor. The normalized values ​​of the evaluation indicators are weighted and summed with the optimal weight vector, and the weighted summation result is corrected with a physical penalty factor to obtain the final evaluation score. The mechanical performance is then graded based on the final evaluation score. The physical penalty factor is generated based on the fusion result of the measured power physical residual and the rated parameter self-consistency residual. The measured power physical residual is calculated based on the physical conservation relationship between the average torque, the average speed, and the average output power. The rated parameter self-consistency residual is calculated based on the physical conservation relationship between the rated torque, the rated speed, and the rated power.

2. The method according to claim 1, characterized in that, Configure non-uniform B-spline nodes according to the speed ratio of the servo motor, including: When the speed ratio is less than or equal to 1.0, 70% of the control points are configured in the low-speed to rated speed range, and 30% of the control points are configured in the high-speed standby range. When the speed ratio is greater than 1.0, a control point of 50% is configured in the low speed to rated speed range, and a control point of 50% is configured in the rated speed to current operating speed range.

3. The method according to claim 1, characterized in that, Before the step of iteratively updating the control point weights of the KAN neural network using the gradient descent method, with the FAHP judgment matrix corrected by the thermo-mechanical coupling degradation factor as the optimization benchmark, the method further includes determining the thermo-mechanical coupling degradation factor in the following manner: When the deviation of the average load rate from the predetermined optimal operating point is greater than 0, the thermo-mechanical coupling degradation factor is calculated according to the exponential nonlinear rule. The greater the deviation, the greater the thermo-mechanical coupling degradation factor. When the deviation is less than or equal to 0, the thermo-mechanical coupling degradation factor is a predetermined standard value; The correction of the FAHP judgment matrix includes: updating the four judgment scales of torque index relative to speed index, torque index relative to load rate index, power index relative to speed index, and power index relative to load rate index according to the thermo-mechanical coupling degradation factor, and synchronously updating the symmetrical position elements according to the matrix symmetry and reciprocity characteristics.

4. The method according to claim 1, characterized in that, The measured power physical residual is obtained by calculating the theoretical mechanical power based on the product of the average torque and the average speed, and by calculating the deviation between the average output power and the theoretical mechanical power. The self-consistency residual of the rated parameters is obtained by calculating the theoretical rated power based on the product of the rated torque and the rated speed, and by calculating the degree of deviation between the rated power and the theoretical rated power. Before correcting the weighted summation result with the physical penalty factor, the method further includes: performing upper limit truncation on the measured power physical residual and the rated parameter self-consistency residual respectively, truncating the residuals exceeding the predetermined upper limit value to the predetermined upper limit value, and then merging and averaging the truncated two to obtain the comprehensive physical deviation level, generating the physical penalty factor based on the comprehensive physical deviation level, wherein the higher the comprehensive physical deviation level, the smaller the physical penalty factor.

5. The method according to claim 1, characterized in that, The predetermined convergence condition is that the consistency ratio of the FAHP judgment matrix is ​​less than 0.02; the learning rate of the gradient descent method is 0.02, and the maximum number of iterations is 500.

6. The method according to claim 1, characterized in that, Before calculating the evaluation index based on the rated parameters and the real-time operating parameters, the method further includes: removing input data with a self-consistency deviation of the rated parameters exceeding 10%, and removing input data with the measured output power exceeding the rated power by 150%. Normalization is performed according to the attribute type of each evaluation index, including: the torque error rate and the speed deviation rate are extremely small indices, and are normalized using an extremely small normalization method; the power achievement rate is an extremely large indices, and is normalized using an extremely large normalization method; the average load rate is a moderate indices, and is normalized using a moderate normalization method, with its optimal operating point being 70%. And / or, after classifying mechanical performance based on the final evaluation score, the method further includes: triggering a data inconsistency warning when the deviation between the average load rate and the power achievement rate exceeds 20%; triggering a mechanical transmission maintenance prompt when the torque error rate exceeds 10%; triggering a controller parameter mismatch prompt when the speed deviation rate exceeds 5%; and triggering an inefficient operation warning or an overload operation risk warning when the average load rate is below 30% or above 100%, respectively.

7. The method according to claim 1, characterized in that, Mechanical performance is graded based on the final evaluation score, including: When the final evaluation score is greater than or equal to 85 points, the first performance level is output, corresponding to high-precision and high-end application scenarios. When the final evaluation score is greater than or equal to 70 points and less than 85 points, the second performance level is output, corresponding to mid-to-high-end industrial application scenarios. When the final evaluation score is less than 70 points, the third performance level is output, corresponding to the normal working condition application scenario.

8. A servo motor mechanical performance grading and evaluation system, characterized in that, include: The data acquisition and processing module is used to acquire the rated parameters and real-time operating parameters of the servo motor, calculate the evaluation index based on the rated parameters and the real-time operating parameters, and perform normalization processing according to the attribute type of each evaluation index. The rated parameters include at least rated torque, rated speed and rated power, the real-time operating parameters include at least average torque, average speed, average output power and average load rate, and the evaluation index includes at least torque error rate, speed deviation rate, power achievement rate and average load rate. A non-uniform B-spline node configuration module is used to configure non-uniform B-spline nodes according to the speed ratio of the servo motor and generate B-spline basis functions adapted to the current working conditions. The speed ratio is the ratio of the average speed to the rated speed. When the speed ratio is in different predetermined ranges, the non-uniform B-spline nodes have different control point density distributions on both sides of the rated speed to match the nonlinear distribution characteristics of the motor's external characteristics. The KAN neural network weight mapping module is used to perform forward propagation operations on the B-spline basis function and control point weights through the KAN neural network, output the initial weight benchmark values ​​of each evaluation index, and obtain the basic weight vector after normalization. The weight iteration optimization module is used to iteratively update the control point weights of the KAN neural network using the gradient descent method, based on the FAHP judgment matrix corrected by the thermo-mechanical coupling degradation factor, until the consistency ratio of the FAHP judgment matrix meets the predetermined convergence condition, and outputs the optimal weight vector. The thermo-mechanical coupling degradation factor is calculated based on the degree to which the average load rate deviates from the predetermined optimal operating point. The correction includes adjusting the judgment scale of the torque index and power index in the FAHP judgment matrix relative to the speed index and load rate index according to the thermo-mechanical coupling degradation factor. The comprehensive scoring module is used to perform a weighted summation of the normalized values ​​of the evaluation indicators and the optimal weight vector, and to correct the weighted summation result with a physical penalty factor to obtain a final evaluation score. The mechanical performance is then graded based on the final evaluation score. The physical penalty factor is generated based on the fusion result of the measured power physical residual and the rated parameter self-consistency residual. The measured power physical residual is calculated based on the physical conservation relationship between the average torque, the average speed, and the average output power. The rated parameter self-consistency residual is calculated based on the physical conservation relationship between the rated torque, the rated speed, and the rated power.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the method as described in any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method as described in any one of claims 1 to 7.

Images