Fuzzy Calibration Method for Photoelectric Angle Sensors Based on ANSYS Simulation

Abstract

This invention discloses a fuzzy calibration method for photoelectric angle sensors based on ANSYS simulation, belonging to the field of fuzzy calibration technology for photoelectric angle sensors. The method includes the following steps: Based on the identification results, the nodal displacement data as a function of angle from the sensor simulation output are retrieved to construct a family of periodic displacement offset fitting functions. Curve fitting processing is then performed on one rotation data to determine the angular response skew caused by structural eccentric loading under conditions where no non-axisymmetric loading is set during the simulation. Based on the determined angular response skew results, a residual attention-enhanced Transformer structure is constructed to extract the error change trajectory feature set from the structural skew envelope, and a drift discrimination result reflecting the direction, level, and stability of structural error drift is output. This invention solves the problem of error modeling distortion in photoelectric angle sensors under non-axisymmetric loads and achieves error drift feature identification and dynamic accuracy control.

Description

Technical Field

[0001] This invention relates to the field of fuzzy calibration technology for photoelectric angle sensors, and specifically to a fuzzy calibration method for photoelectric angle sensors based on ANSYS simulation. Background Technology

[0002] Fuzzy calibration of photoelectric angle sensors based on ANSYS simulation is a novel calibration method that combines multiphysics simulation modeling with fuzzy logic algorithms. Its core idea is to utilize ANSYS to accurately model and simulate the structure, material properties, stress state, and photoelectric response process of the photoelectric angle sensor. This obtains error characteristic data between the sensor output and the actual angle under different environmental and operating conditions. Based on this high-dimensional simulation data, a fuzzy calibration model is constructed to achieve high-precision fitting and compensation for actual measurement errors. In existing technologies, a finite element model of the sensor is first established using ANSYS to simulate its response behavior under conditions such as pressure, torsion, and temperature changes. Simultaneously, key output parameters are extracted by simulating changes in the photoelectric signal. Subsequently, the simulation results are compared and analyzed with the actual sensor output to extract the error distribution pattern. A fuzzy logic-based calibration algorithm is then designed, including fuzzy rule base construction, membership function setting, and inference mechanism optimization, to achieve fuzzy modeling and nonlinear compensation of errors. Finally, through experimental verification, the calibration model is embedded into the sensor system to achieve real-time correction and adaptive optimization of the sensor output, thereby improving its measurement accuracy and environmental robustness. The entire calibration process includes four core components: sensor modeling and simulation, data acquisition and error analysis, fuzzy algorithm construction and training, and model verification and optimization. These components together form a closed-loop calibration system based on the integration of simulation-driven, data-driven, and intelligent algorithm-driven approaches.

[0003] The existing technology has the following shortcomings:

[0004] When photoelectric angle sensors are applied in environments with eccentric rotation or asymmetric structures, they are often subjected to non-axisymmetric loads (such as lateral forces or tilting torques) during operation. These loads cause periodic skewness in the structural response of the encoder disk or photoelectric receiving component within one rotation, resulting in asymmetric error fluctuations in the sensor's output angle signal. However, in existing fuzzy calibration techniques for photoelectric angle sensors based on ANSYS simulation, the simulation loading conditions are often idealized to an axisymmetric form to simplify the modeling process. This fails to realistically reproduce the actual effect of non-axisymmetric loads, causing the error simulation results to exhibit uniform amplitude symmetrical fluctuations, which cannot accurately reflect the error skewness characteristics under actual working conditions. Therefore, existing fuzzy calibration techniques for photoelectric angle sensors based on ANSYS simulation cannot accurately identify structural error drift characteristics based on the angle response skewness caused by structural eccentricity when non-axisymmetric loading conditions are not set during the simulation. This leads to errors in the extracted results deviating from the true distribution, resulting in distorted error learning during fuzzy modeling. Consequently, problems such as inconsistent zeroing, amplified error eccentricity, and decreased repeatability occur during sensor zeroing, repositioning, or repeated measurements, seriously affecting the reliability of angle measurement and the stability of system control.

[0005] The information disclosed in the background section is only intended to enhance the understanding of the background of this disclosure, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0006] The purpose of this invention is to provide a fuzzy calibration method for photoelectric angle sensors based on ANSYS simulation, so as to solve the problems in the background art mentioned above.

[0007] To achieve the above objectives, the present invention provides the following technical solution: a fuzzy calibration method for photoelectric angle sensors based on ANSYS simulation, specifically including the following steps:

[0008] S1. Extract all discrete load vector data from the ANSYS simulation settings, construct a load symmetry mapping rate function based on the directional distribution of each load vector about the principal rotation axis, and generate a mapping rate value by calculating the mirror fitting deviation, which is used to identify whether non-axisymmetric loading conditions are not set during the simulation process.

[0009] S2. Based on the recognition results, call the node displacement data with angle change in the sensor simulation output, construct a family of periodic displacement offset fitting functions, and perform curve fitting processing on one rotation data to determine the angular response skew caused by structural off-center loading when no non-axisymmetric loading conditions are set during the simulation.

[0010] S3. Based on the determined angle response skew results, construct a residual attention-enhanced Transformer structure, extract the error change trajectory feature set in the structural skew envelope, and output the drift discrimination results that reflect the drift direction, level and stability of structural error.

[0011] S4. Based on the drift discrimination results, construct a nonlinear entropy gate mapping network to map the error drift characteristics into fuzzy classification categories and divide them into a predictable response region, a dynamic boundary transition region, and an uncorrectable response region.

[0012] S5. Based on the fuzzy classification, generate a drift self-sensing classification factor sequence. Combined with the error evolution trend within the angle segment, adjust the convergence priority and learning rate curve of the training objective function in the fuzzy calibration modeling process of the photoelectric angle sensor to complete the dynamic accuracy control of the error modeling output.

[0013] Preferably, S1 is as follows:

[0014] Extract all discrete load vector data from the ANSYS simulation settings, project each load vector onto a cross-sectional plane perpendicular to the principal axis of rotation, extract the angle between each load vector and the reference radial reference direction within the cross-sectional plane, and generate an ordered distribution set containing all angle values.

[0015] The load symmetry mapping rate function is constructed based on the ordered distribution set. The polar angle symmetry mapping method is adopted to pair each included angle value with the included angle value that is symmetrically distributed with respect to the center line of the main rotation axis, and the numerical difference of each pair of load vectors is extracted.

[0016] All numerical differences are calculated into a set of mirror fit deviation vectors. Mean square fitting is performed according to the vector magnitude to obtain the mean square value of the mirror fit deviation as the mirror fit deviation metric.

[0017] A normalized proportional relationship is constructed based on the mirror fitting deviation metric, and a mapping rate value is generated. When the mapping rate value is greater than a set threshold, it is used to identify whether non-axisymmetric loading conditions were not set during the simulation process.

[0018] Preferably, S2 specifically includes the following steps:

[0019] S201. When the mapping rate value constructed based on the mirror fitting deviation metric exceeds the set threshold, the nodal displacement data with angle change in the ANSYS simulation output is called, and the nodal trajectory is organized according to the angle sequence within one rotation interval.

[0020] S202. Extract the displacement change sequence of each node in the rotation interval into a periodic response vector, extract the fundamental frequency component based on discrete Fourier transform and construct a family of periodic displacement offset fitting functions, and record the phase center position of the trajectory of each node in the function family.

[0021] S203. Perform multi-curve superposition on all functions in the family of periodic displacement offset fitting functions, and extract the main response envelope curve using the minimum deviation fitting strategy.

[0022] S204. Calculate the angular response skew value based on the relative phase difference between the master response envelope curve and the ideal axisymmetric fitting curve, and determine the angular response skew caused by structural off-center loading when no non-axisymmetric loading conditions are set during the simulation.

[0023] Preferably, S202 specifically refers to:

[0024] The displacement change sequence of each node within the complete rotation interval is converted into a periodic response vector according to the angle sampling order, and the amplitude range of the periodic response vector is normalized to unify the numerical scale of the response vector.

[0025] Perform a discrete Fourier transform on each normalized periodic response vector to extract the fundamental frequency component in the frequency domain features, which serves as the dominant feature parameter for the periodic displacement change.

[0026] A family of periodic displacement fitting functions is constructed based on the phase and amplitude parameters in the fundamental frequency component, and the phase center position of the periodic response of each node in the family of fitting functions is recorded as a relative position reference.

[0027] Preferably, S204 is as follows:

[0028] Based on the master response envelope curve, a periodic amplitude trajectory in the complete angular domain is constructed. An ideal axisymmetric fitting curve is constructed using Fourier cosine expansion. The ideal axisymmetric fitting curve and the master response envelope curve are then subjected to node resampling within a unified angular interval.

[0029] The response amplitudes of the main response envelope curve and the ideal axisymmetric fitting curve at each sampling angle point are respectively mapped to complex numbers, and the relative phase difference sequence between the two curves is calculated based on the complex plane phase relationship.

[0030] The position of the main peak and the angle index offset in the relative phase difference sequence are extracted as the angle response skew value. Based on the offset directionality and offset amplitude, it is determined whether the response skew caused by the structural off-center loading is valid. This is used to confirm the response offset characteristics when the non-axisymmetric loading condition is not set.

[0031] Preferably, S3 specifically includes the following steps:

[0032] S301. Based on the angle response skew results, construct a residual attention-enhanced Transformer structure, construct the angle response skew results into a two-dimensional input tensor according to the angle sequence, insert position index encoding and response residual feature channels in each encoding layer, and introduce a multi-head attention structure to extract feature dependencies in non-axisymmetric responses.

[0033] S302. The residual attention-enhanced Transformer structure is used to extract features from the skew envelope of the structure. The feature changes in the continuous response path are preserved through the residual jump structure, and the attention channel outputs an error change trajectory feature set containing the magnitude offset gradient, response period variation and phase characteristic trend.

[0034] S303. Perform a multi-dimensional drift discrimination process based on the error change trajectory feature set, and output the discrimination results reflecting the drift direction, drift level and drift stability of structural error. The drift direction is determined based on the peak offset direction of the response envelope, the drift level is determined according to the normalized offset amplitude, and the drift stability is determined based on the persistence of phase perturbation.

[0035] Preferably, S303 is as follows:

[0036] Extract the response envelope peak sequence within a continuous angular interval from the error change trajectory feature set, calculate the offset direction of the envelope peak relative to the reference angular position, and generate the discrimination result of the structural error drift direction based on the angular quadrant where the maximum offset direction is located.

[0037] Based on the normalized offset value of each response amplitude in the error change trajectory feature set, an offset classification vector is constructed. The offset classification vector is segmented into intervals. The number of high offset intervals is counted and an amplitude threshold is set. The offset level is compared with the threshold mapping table to generate the drift level discrimination result.

[0038] A perturbation persistence measure function is constructed by utilizing the phase perturbation change trend in the error change trajectory feature set. The length of the continuous perturbation interval is calculated based on the perturbation persistence measure function. The phase stability is evaluated by combining the average perturbation density and standard deviation, and the drift stability discrimination result is generated.

[0039] Preferably, S4 is as follows:

[0040] Based on the drift discrimination results, a nonlinear entropy gate mapping network is constructed. The drift direction, drift level and drift stability are used as input parameters. A three-dimensional drift feature tensor is constructed by concatenating the input features. An entropy gate control unit is introduced to adjust the contribution intensity of each drift dimension to the output mapping. A nonlinear mapping path is constructed using a two-layer fully connected network.

[0041] The drift discrimination result is mapped based on a nonlinear entropy gate mapping network. The multi-channel response value corresponding to the input vector is extracted. The normalization function is used to convert all channel responses into the attribution probability and a fuzzy attribution category label set is constructed. Each category label corresponds to the attribution probability channel of the predictable response region, the dynamic transition region of the boundary, and the uncorrectable response region.

[0042] The attribution result is determined based on the category channel corresponding to the maximum attribution probability value in the fuzzy attribution category label set. When the maximum value channel is a channel in the predictable response zone, it is determined to be in the predictable response zone. When the maximum value channel is a channel in the dynamic transition zone of the boundary, it is determined to be in the dynamic transition zone of the boundary. When the maximum value channel is a channel in the uncorrectable response zone, it is determined to be in the uncorrectable response zone.

[0043] Preferably, S5 is as follows:

[0044] Based on the probability value of each angle sampling point in the fuzzy classification, the predictable response area, the dynamic transition area of ​​the boundary, and the uncorrectable response area are mapped to three levels of classification weights: low, medium, and high. The drift self-aware classification factor sequence is constructed by arranging them according to the angle sequence to describe the error drift response characteristics corresponding to different angle segments.

[0045] The error evolution trend is extracted within each angle segment. An error evolution gradient sequence is constructed based on the slope of error change between continuous angle sampling points. A joint control weight matrix is ​​formed by combining the drift self-perception attribution factor sequence to reflect the modeling complexity and response uncertainty of each angle segment.

[0046] By introducing the joint control weight matrix into the fuzzy calibration modeling process of photoelectric angle sensor, the convergence priority parameter of the training objective function is dynamically adjusted, and the adjustment factor of the learning rate curve is set according to the changing trend of the error evolution gradient. During the model training process, the convergence efficiency is improved in the high response drift section, and the weight convergence step size is slowed down in the dynamic transition section, thereby completing the dynamic accuracy control of the error modeling output.

[0047] The technical effects and advantages provided by the present invention in the above technical solution are as follows:

[0048] 1. This invention solves the problem of distorted error extraction and model learning bias caused by the lack of non-axisymmetric loading conditions in simulation modeling of traditional photoelectric angle sensors by constructing a complete multi-level intelligent modeling process from load symmetry identification to error attribution control. The scheme first extracts load distribution characteristics by jointly using mirror fitting deviation and polar angle mapping rate function, enabling the identification of structural off-center loading states without explicitly setting non-axisymmetric loading conditions. Furthermore, it extracts the skew trend in the angle response based on Fourier frequency domain feature analysis and periodic fitting strategies, thus accurately establishing the basis for identifying non-axisymmetric responses. During error modeling, a residual attention-enhanced Transformer structure is used to achieve high-dimensional perception and analysis of amplitude, phase, and periodic error changes. Based on multi-dimensional discrimination of drift direction, level, and stability, an error evolution model that more closely resembles the real application state is constructed.

[0049] 2. This invention introduces a nonlinear entropy gate mapping mechanism to map multidimensional drift results into fuzzy classification categories. A joint control weight matrix, constructed from classification factors and error gradients, guides the parameter update path in the modeling process, enabling dynamically adjustable convergence priority and learning rate strategies. This differentiated optimization approach, targeting the error drift region, dynamic transition region, and high uncertainty region, effectively improves the fitting ability to asymmetric perturbation responses during fuzzy modeling, solving the problems of insufficient generalization and inconsistent zero-adjustment accuracy in existing technologies. Overall, this technical solution not only improves the adaptability of modeling results to actual non-ideal working conditions but also significantly enhances the robustness and accuracy consistency of sensor error models under simulation-driven conditions. Attached Figure Description

[0050] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0051] Figure 1 This is a flowchart illustrating the fuzzy calibration method for photoelectric angle sensors based on ANSYS simulation, as described in this invention. Detailed Implementation

[0052] Exemplary embodiments will now be described more fully with reference to the accompanying drawings. However, these exemplary embodiments can be implemented in many forms and should not be construed as limited to the examples set forth herein; rather, they are provided so that the description of this disclosure will be more complete and fully convey the concept of the exemplary embodiments to those skilled in the art.

[0053] This invention provides, for example Figure 1The fuzzy calibration method for photoelectric angle sensors based on ANSYS simulation shown includes the following steps:

[0054] S1. Extract all discrete load vector data from the ANSYS simulation settings, construct a load symmetry mapping rate function based on the directional distribution of each load vector about the principal rotation axis, and generate a mapping rate value by calculating the mirror fitting deviation, which is used to identify whether non-axisymmetric loading conditions are not set during the simulation process.

[0055] In this embodiment, S1 specifically refers to:

[0056] Extract all discrete load vector data from the ANSYS simulation settings, project each load vector onto a cross-sectional plane perpendicular to the principal axis of rotation, extract the angle between each load vector and the reference radial reference direction within the cross-sectional plane, and generate an ordered distribution set containing all angle values.

[0057] Before performing the ANSYS simulation, the simulation setup includes multiple discrete load vectors, which describe the direction and magnitude of the force applied at different locations on the sensor structure. Each load vector can be represented as a three-dimensional vector using three-dimensional coordinates. To analyze its distribution symmetry relative to the rotating structure, all load vectors are mapped to a plane perpendicular to the main rotation axis using vector projection. This plane is typically a cross-section passing through the center of the rotation axis. Subsequently, a unified two-dimensional polar coordinate system is established based on this cross-section, with a fixed radial direction set as the reference reference direction, for example, the direction from the center to the positive X-axis. In this two-dimensional coordinate system, the angle between each projected load vector and the reference radial reference direction is calculated. This can be accurately solved using the arctangent function (such as atan2) combined with the vector dot product and cross product formulas, ensuring that the angle range is uniformly between 0 and 360 degrees. All angle values ​​are sorted in ascending order, generating an ordered distribution set recording all direction angle information, providing the input basis for subsequently constructing the symmetry function.

[0058] Discrete load vector data represents the three-dimensional force situation at each loading point in the structure, and is the basic mechanical input information derived from the initial conditions of ANSYS simulation. The load vectors are projected onto a cross-sectional plane perpendicular to the principal axis of rotation. This transforms the three-dimensional distribution problem into a two-dimensional planar angle distribution problem, facilitating subsequent symmetry analysis. The principal axis of rotation typically refers to the central axis of the encoder disk during rotation, serving as the symmetry reference axis for the entire angle measurement system. The reference radial reference direction is the initial angular direction set in the two-dimensional polar coordinate system, facilitating a unified angle calculation benchmark. The included angle value reflects the absolute angular magnitude of each projected load vector direction deviating from the reference radial reference direction, and is a key characteristic parameter of the load direction distribution. The ordered distribution set is an ordered array composed of all included angle values ​​arranged in ascending order. It reflects the overall distribution trend of the load direction and is the basis for subsequently constructing the symmetry mapping rate function and identifying symmetry deviations.

[0059] The load symmetry mapping rate function is constructed based on the ordered distribution set. The polar angle symmetry mapping method is adopted to pair each included angle value with the included angle value that is symmetrically distributed with respect to the center line of the main rotation axis, and the numerical difference of each pair of load vectors is extracted.

[0060] The core of constructing a load symmetry mapping rate function based on an ordered distribution set lies in determining whether the load directions are symmetrically distributed around the principal rotation axis. First, using the previously generated set of included angle values, a polar coordinate system is established, cycling from 0° to 360°, uniformly mapping all included angle values ​​within this angular domain. For each included angle value 'a', the included angle value 'b' that is symmetrical about the principal rotation axis is found, calculated as b = (a + 180) mod 360, thus ensuring that 'a' and 'b' are symmetrically positioned on the rotation section. Subsequently, the load vector magnitudes corresponding to included angles 'a' and 'b' are extracted, and the absolute difference between them is calculated as the symmetry deviation value for that diagonal point. After processing all angle pairs sequentially, the deviation values ​​of all angle pairs are summarized and normalized to establish the load symmetry mapping rate function. Its expression can be the reciprocal of the sum of the mean squares of all differences or its variations; the closer the value is to zero, the more asymmetrical it is, and the closer it is to the limit value, the more symmetrical it is.

[0061] The ordered distribution set represents the ordered arrangement of all projected angle values ​​on the unit circle, providing a basis for discrete distribution in the angle domain. The load symmetry mapping rate function is used to quantify the degree of symmetry of the entire load distribution and is a key tool in this technique for determining whether non-axisymmetric loading exists in the simulation. The polar angle symmetry mapping method is a method that analyzes each angle in pairs with its 180-degree opposite angle, reflecting whether there are symmetrical structural loading characteristics when the principal rotation axis is the axis of rotation. The correspondence between the included angle value and its symmetry angle is maintained periodically through modulo transformation to avoid angle out-of-bounds problems. The numerical difference is the structural symmetry deviation represented by the difference in the magnitude of the load vectors of corresponding angle pairs, reflecting whether the loading direction intensity is geometrically balanced. The finally constructed load symmetry mapping rate function is an important quantitative input for subsequent identification of asymmetric loading simulation conditions, possessing continuity and sensitivity.

[0062] All numerical differences are calculated into a set of mirror fit deviation vectors. Mean square fitting is performed according to the vector magnitude to obtain the mean square value of the mirror fit deviation as the mirror fit deviation metric.

[0063] The mirror fitting deviation vector set is used to quantify the degree of load symmetry deviation. First, the load vector differences corresponding to each pair of axisymmetric angles extracted in the previous stage are constructed into a two-dimensional vector set. The magnitude of each vector represents the symmetry error magnitude of that angle pair, and the direction can be uniformly pointed to a preset offset direction for standardized calculation. Then, the entire vector set is squared, i.e., the magnitude of each vector is squared to amplify the impact of the deviation magnitude. All squared magnitude values ​​are averaged to form the mean square value, which serves as a metric for the overall degree of symmetry deviation. A higher value indicates a greater deviation in load symmetry, while a lower value indicates a load closer to axisymmetric conditions. This can be used to determine whether there are simulations where non-axisymmetric loading conditions were not set.

[0064] The specific process for constructing the mirror fitting deviation vector set is as follows: For each pair of included angles a and b, corresponding to two load vector magnitudes V1 and V2 respectively, calculate their difference ΔV = |V1-V2|. Map this difference to a two-dimensional vector of length ΔV and store it along a coordinate axis consistent with the direction of included angle a. Repeat the above process to obtain n sets of mirror fitting deviation vectors. Let n = 4 sets, with magnitudes of 2, 3, 4, and 1 respectively, then the squares are 4, 9, 16, and 1 respectively, the sum is 30, and the mean is 30 / 4 = 7.5. This mean square value is the mirror fitting deviation metric. If this value exceeds the preset symmetry threshold, such as 6.0, it indicates that the load distribution exhibits strong asymmetric characteristics, suggesting that the simulation may not include non-axisymmetric loading conditions.

[0065] The mirror fit deviation vector set is a high-dimensional numerical construction method for expressing the symmetry error distribution. Its core lies in transforming scalar differences into a structural expression usable for vector norm calculation. The vector magnitude squaring operation amplifies large errors while weakening the weight differences of small errors, helping to improve the sensitivity of identifying anomalous symmetry shifts. Mean square fitting is a commonly used statistical regression algorithm that extracts the overall deviation trend by weighted averaging of all magnitude squares, eliminating the interference of single-point anomalies and providing a stable index value reflecting the symmetry of the simulation settings. The mirror fit deviation metric is essentially a quantitative judgment of whether the entire load distribution conforms to the theoretical symmetry assumption, providing a reliable load condition screening basis for subsequent error drift identification.

[0066] A normalized proportional relationship is constructed based on the mirror fitting deviation metric, and a mapping rate value is generated. When the mapping rate value is greater than a set threshold, it is used to identify whether non-axisymmetric loading conditions were not set during the simulation process.

[0067] The purpose of constructing a normalized proportional relationship is to convert the mirror fit deviation metric into a mapping rate value under a unified dimension, so as to ensure comparability in different simulation scenarios. First, the minimum mirror fit deviation metric under perfect symmetry is set as the lower limit L, and the maximum mirror fit deviation metric under the theoretically maximum permissible error is set as the upper limit U. For any actual simulation result corresponding to the mirror fit deviation metric M, the mapping rate value R is constructed using the normalization formula R = (M - L) / (U - L), ensuring its range falls between 0 and 1. For example, when L=0.5, U=5.5, and a simulation result M=4.0, the mapping rate value R = (4.0 - 0.5) / (5.5 - 0.5) = 0.7. This value indicates a high degree of asymmetry. Subsequently, this mapping rate value is compared with a set recognition threshold T. If R > T, for example, T=0.6, it is determined that there is a possibility that non-axisymmetric loading conditions were not set during the simulation process, indicating that error drift feature processing should be performed in subsequent calibration.

[0068] The mapping rate value is a normalized index reflecting the degree of symmetry in the load direction. Its core function is to construct a symmetry discrimination reference with a unified numerical range. The mirror fit deviation metric is a raw characteristic value with physical dimensions that fluctuates with simulation conditions. To facilitate comparison across various structural types, material models, or load scenarios, numerical standardization is required through proportional mapping. The threshold is derived from experience extracted from a large amount of historical simulation data or actual test scenarios, typically based on the fitting performance of the simulation error and the actual error to infer a reasonable boundary point. The existence of the threshold gives the mapping rate value engineering-significant discriminative power, making fuzzy and uncertain simulation errors explicit and serving as a criterion for whether the structural loading symmetry meets the standard, thereby accurately identifying whether the simulation has omitted non-axisymmetric loading settings. This judgment directly affects the branch settings of subsequent fuzzy modeling paths and is a crucial starting point for the entire error compensation logic.

[0069] S2. Based on the recognition results, call the node displacement data with angle change in the sensor simulation output, construct a family of periodic displacement offset fitting functions, and perform curve fitting processing on one rotation data to determine the angular response skew caused by structural off-center loading when no non-axisymmetric loading conditions are set during the simulation.

[0070] In this embodiment, S2 specifically includes the following steps:

[0071] S201. When the mapping rate value constructed based on the mirror fitting deviation metric exceeds the set threshold, the nodal displacement data with angle change in the ANSYS simulation output is called, and the nodal trajectory is organized according to the angle sequence within one rotation interval.

[0072] When the mapping rate calculated by the mirror fitting deviation metric exceeds a set threshold, it can be considered that the current simulation model may lack non-axisymmetric loading conditions. Based on this identification result, the data sequence of nodal displacements varying with angle is extracted from the ANSYS simulation result file. When calling the data, the reference coordinate system of the rotating entity is used as the baseline, and all displacement records within a complete rotation angle range are selected to ensure that the data includes the structural response within the complete cycle. Subsequently, the rotation angle range is divided into fixed angle sampling intervals, and the nodal position at each time point or angle state is projected onto the rotation plane, establishing a one-dimensional ordered trajectory array in order of angle from zero to the maximum rotation. For each node, a set of three-dimensional displacement vector sequences corresponding to the angle is constructed as the basic dataset for subsequent periodic response feature extraction.

[0073] The mapping rate value is a numerical index generated by establishing a normalized ratio between the mirror fitting deviation metric and the minimum deviation standard value under theoretical symmetrical loading. It is used to measure the degree of symmetry in load distribution; when this index is greater than a threshold, it indicates a significant asymmetric trend in the load. Nodal displacement versus angle data refers to the response sequence of three-dimensional displacements as a function of rotation angle, recorded node-by-node in ANSYS simulations. This is often exported in batches via post-processing APIs such as ANSYS MAPDL or PyMAPDL. Angle sequence alignment involves resampling the nodal responses at a uniform angular resolution to align the angles of different nodal trajectories within the same rotation period. This operation not only ensures the periodic consistency of the nodal responses but also establishes a unified angular domain foundation for subsequent discrete Fourier analysis and migration function family fitting.

[0074] S202. Extract the displacement change sequence of each node in the rotation interval into a periodic response vector, extract the fundamental frequency component based on discrete Fourier transform and construct a family of periodic displacement offset fitting functions, and record the phase center position of the trajectory of each node in the function family.

[0075] S203. Perform multi-curve superposition on all functions in the family of periodic displacement offset fitting functions, and extract the main response envelope curve using the minimum deviation fitting strategy.

[0076] The purpose of performing multi-curve overlay processing on all functions in the family of periodic displacement offset fitting functions is to extract representative periodic response features from a large number of nodal response fitting curves. Specifically, this is achieved by first synchronously resampling all periodic displacement offset fitting functions under a unified angle sampling sequence, ensuring that each function curve has an equal-length vector representation within the same angle interval. Then, all curves are superimposed point-to-point at each sampling angle point, calculating the set of response amplitudes corresponding to each angle point. A minimum deviation fitting strategy is then applied to this set, selecting the response amplitude that minimizes the sum of squared deviations at each angle as the principal response value for that angle, thus constructing the principal response envelope curve. The minimum deviation fitting strategy is a statistical method based on the least squares approach, capable of selecting the median trajectory that best represents the overall trend among multiple curves, effectively avoiding interference from extreme anomalies, and giving the principal response envelope curve strong robustness and representativeness. This curve will serve as an important benchmark for judging structural response skewness in subsequent analyses, as it accurately preserves the most important trajectory features of the periodic response while eliminating high-frequency disturbances in local individual responses.

[0077] S204. Calculate the angular response skew value based on the relative phase difference between the master response envelope curve and the ideal axisymmetric fitting curve, and determine the angular response skew caused by structural off-center loading when no non-axisymmetric loading conditions are set during the simulation.

[0078] The purpose of calculating the angular response skew value based on the relative phase difference between the master response envelope curve and the ideal axisymmetric fitting curve is to quantitatively analyze the deviation between the periodic trajectory of the actual structural response under load and the ideal symmetric state, thereby revealing the structural response skew behavior caused by the lack of asymmetric loading conditions. This approach is based on the fact that the master response envelope curve is a global response trajectory formed by the superposition of the displacement behaviors of multiple nodes within a complete rotation period, truly reflecting the actual response state of the structure under the current load configuration; while the ideal axisymmetric fitting curve is a symmetric reference model constructed based on Fourier cosine expansion, representing the response behavior that should exist under ideal symmetric loading. The phase difference between the two is manifested in the complex plane as the relative rotation or drift of the angular response trajectory. By performing sequential calculations on the relative phase difference, the deviation trend induced by asymmetric loading can be accurately captured. Converting this phase difference into a specific angular response skew value helps to reveal the periodic disturbance effect of off-center loading on the response position of the encoder disk or sensitive element. Furthermore, determining whether the skewness is valid, i.e. whether it exceeds the set symmetry tolerance threshold, can serve as the logical basis for identifying undefined non-axisymmetric load conditions in the simulation settings. This technique utilizes the high sensitivity and directional characteristics of phase information, enabling it to overcome the error fitting limitations at the amplitude level and achieve accurate quantification and diagnosis of structural response imbalances.

[0079] In this embodiment, S202 specifically refers to:

[0080] The displacement change sequence of each node within the complete rotation interval is converted into a periodic response vector according to the angle sampling order, and the amplitude range of the periodic response vector is normalized to unify the numerical scale of the response vector.

[0081] The process of converting the displacement change sequence of each node within the complete rotation interval into a periodic response vector according to the angular sampling order first extracts the three-dimensional displacement values ​​corresponding to each angle point during the rotation process from the node data in the simulation output, and then constructs a strictly ordered one-dimensional vector sequence from the start to the end based on the angle. After this periodic response vector is constructed, amplitude range normalization processing needs to be performed to ensure that the responses of all nodes have a uniform comparison scale. Specifically, a maximum-minimum normalization method can be used, that is, subtract the minimum displacement value in each periodic response vector from all displacement values, and then divide by the amplitude range of the vector (maximum value minus minimum value), so that the value range of all normalized response vectors is uniformly in the interval of 0 to 1. This normalization process can be performed in the three-dimensional magnitude dimension, or it can be normalized separately according to the coordinate axis direction after component separation, depending on the needs of subsequent calculations. The core technical feature of the normalization processing is to eliminate the numerical differences of different nodes in the original displacement magnitude, so that in the subsequent Fourier frequency domain transformation and function fitting process, we can focus on the differences in periodic mode, phase structure and offset behavior, without being affected by the amplitude scale, thereby improving the accuracy and consistency of error drift feature modeling.

[0082] Perform a discrete Fourier transform on each normalized periodic response vector to extract the fundamental frequency component in the frequency domain features, which serves as the dominant feature parameter for the periodic displacement change.

[0083] The process of performing a Discrete Fourier Transform (DFT) on each normalized periodic response vector aims to transform the displacement change sequence in the time domain to the frequency domain, thereby revealing the spectral composition characteristics of the periodic signal. Specifically, the Fast Fourier Transform (FFT) algorithm can be used to perform a complex form DFT on the normalized displacement vector, mapping the real displacement value of each angle sampling point to a set of complex coefficients for frequency components. Each frequency component in the transformation result represents the amplitude and phase information of the signal at that frequency. The fundamental frequency component in the frequency domain corresponds to the frequency characteristics of the main periodic waveform of the signal and is the first non-zero frequency point in the spectrum. Its amplitude characterizes the intensity of the periodic change, while the angle represents the phase shift of the response. Extracting this fundamental frequency component as the dominant feature parameter can capture the overall pattern and directionality of the periodic shift, which is helpful for subsequent modeling and identification of structural off-center loading in the nodal response. The core technical feature of the DFT lies in decomposing the complex response signal induced by asymmetric structures into a linear superposition of sine and cosine functions, allowing the hidden shift trend to be concisely described by the principal components in the frequency domain, thereby improving the analytical capability and simulation accuracy of the identification process.

[0084] A family of periodic displacement fitting functions is constructed based on the phase and amplitude parameters in the fundamental frequency component, and the phase center position of the periodic response of each node in the family of fitting functions is recorded as a relative position reference.

[0085] The process of constructing a family of periodic displacement fitting functions based on the phase and amplitude parameters of the fundamental frequency component essentially reconstructs the response behavior of each node within the rotation period in the form of a standard sine function. Specifically, the fundamental frequency component extracted using the Discrete Fourier Transform (DFT) can be decomposed into amplitude and phase components. The amplitude parameter represents the amplitude of the node response, reflecting the response intensity and determining the peak and valley amplitudes of the fitting function. The phase parameter represents the time offset within the response period, determining the lateral offset position of the fitting curve within the period. These parameters are substituted into a standard sine expression, such as A·sin(ωt + φ), where A is the amplitude, φ is the phase, and ω is the known principal rotation angular frequency. This generates a unique fitting function for each node, forming a family of functions encompassing the periodic responses of all nodes. Based on this, the position of the peak or the angular index of the maximum phase in each fitting function is recorded, defined as the phase center position of the node's periodic response. This phase center position describes the offset centroid within the response period and serves as an important spatial reference for subsequently determining response skewness and structural symmetry imbalance. The introduction of the family of periodic displacement fitting functions not only simplifies the expression of the response signal, but also enhances the comparability between the responses of each node through consistency modeling, providing precise mathematical support for further extraction of the main response envelope curve and judgment of structural skewness.

[0086] In this embodiment, S204 specifically refers to:

[0087] Based on the master response envelope curve, a periodic amplitude trajectory in the complete angular domain is constructed. An ideal axisymmetric fitting curve is constructed using Fourier cosine expansion. The ideal axisymmetric fitting curve and the master response envelope curve are then subjected to node resampling within a unified angular interval.

[0088] When constructing the periodic amplitude trajectory across the complete angular domain, the master response envelope curve is used as a foundation, defined within a full 360-degree angular range. A continuous periodic response amplitude sequence is formed through high-density angular sampling. This amplitude trajectory reflects the periodic characteristics of the structural response at each angular point and serves as the initial input for subsequent symmetry analysis. To construct an ideal axisymmetric fitting curve, a Fourier cosine expansion is employed to represent the periodic amplitude trajectory as a linear combination of cosine basis functions. Only components with even-function symmetry are retained, while odd-function components are excluded, ensuring the generated fitting curve is symmetric about the principal rotation axis. This process uses the minimum mean square error criterion to solve for the cosine coefficients, ensuring the fitting curve best approximates the amplitude trend of the master response envelope curve overall. After fitting, the master response envelope curve and the ideal axisymmetric fitting curve are mapped to the same angular sampling sequence. Node resampling is then performed on both curves under this sequence, extracting amplitude points at the same angular position from both curves to form a set of angular correspondences. This provides a matching basis for subsequent complex mapping and phase difference calculations. The periodic amplitude trajectory is used to fully express the response fluctuation characteristics, while the ideal axisymmetric fitting curve serves as a reference curve for theoretical symmetry. Node resampling ensures the consistency of the two in the angular domain, forming the basis for comparison in structural response skew identification.

[0089] The response amplitudes of the main response envelope curve and the ideal axisymmetric fitting curve at each sampling angle point are respectively mapped to complex numbers, and the relative phase difference sequence between the two curves is calculated based on the complex plane phase relationship.

[0090] When mapping the response amplitude corresponding to each sampling angle point of the master response envelope curve and the ideal axisymmetric fitting curve to a complex number representation, it is necessary to use the unit complex plane as a basis, taking the angle value of each angle point as the argument of the complex number and its amplitude at that point as the modulus of the complex number, thereby converting the response amplitude of each point into a complex point represented in polar coordinates. For example, for a certain angle point θ, the amplitude of its master response envelope curve is A1, and the amplitude of its ideal axisymmetric fitting curve is A2, then the corresponding complex numbers are represented as A1·e^(jθ) and A2·e^(jθ), respectively, where j is the imaginary unit. Since the two complex numbers have the same argument but may have different moduli, their positions on the complex plane will form different vector lengths along the argument direction. Next, the vector difference between each pair of complex numbers is converted into a phase offset, i.e., the relative phase difference is calculated. This relative phase difference can be expressed in complex division form as (A1·e^(jθ)) ÷ (A2·e^(jθ)) = (A1÷A2). The phase difference is zero at this point. However, in the case of asymmetric loading leading to asymmetric response, the matching of the two complex numbers at the argument will shift, resulting in a non-zero phase difference value. By traversing all sampling points in the entire angular domain and performing phase difference calculations on each pair of complex numbers, a complete relative phase difference sequence can be constructed. This sequence reflects the deviation of the main response envelope curve from the ideal symmetric curve at various angular points. The complex plane phase relationship provides a quantitative means for analyzing response symmetry based on the rotational characteristics of complex vectors and is the core calculation method for identifying response skew caused by structural off-center loading.

[0091] The position of the main peak and the angle index offset in the relative phase difference sequence are extracted as the angle response skew value. Based on the offset directionality and offset amplitude, it is determined whether the response skew caused by the structural off-center loading is valid. This is used to confirm the response offset characteristics when the non-axisymmetric loading condition is not set.

[0092] When extracting the principal peak position and angle index offset from the relative phase difference sequence as the angle response skew value, it is necessary to first perform principal extremum analysis on the entire relative phase difference sequence. By finding the maximum point in the sequence, that is, identifying the position where there is the maximum phase offset between the principal response envelope curve and the ideal axisymmetric fitting curve within a complete rotation cycle, the angle index corresponding to this position is regarded as the principal peak position. The offset of this angle index relative to the theoretical symmetry point is the angle response skew value caused by the structural eccentric loading. By recording the positive or negative direction of this offset value, it can be determined whether the response skew occurs in a clockwise or counterclockwise direction, thus revealing the directional trend of structural force eccentricity. The magnitude of the offset value is used to measure the skew intensity caused by non-axisymmetric loading on the angle response; the larger the offset value, the more severe the degree of error symmetry destruction. The core of this analysis method is to use the relative phase difference sequence as a high-sensitivity indicator, and to intuitively reflect the non-uniformity of the structural response by using the spatial offset behavior of the principal phase peak point. The position of the main peak of the relative phase difference sequence represents the location where the relative imbalance is most concentrated, while the angle index offset provides accurate deviation quantification information and is a key criterion for confirming the response offset characteristics caused by the lack of non-axisymmetric loading conditions in the simulation model.

[0093] S3. Based on the determined angle response skew results, construct a residual attention-enhanced Transformer structure, extract the error change trajectory feature set in the structural skew envelope, and output the drift discrimination results that reflect the drift direction, level and stability of structural error.

[0094] In this embodiment, S3 specifically includes the following steps:

[0095] S301. Based on the angle response skew results, construct a residual attention-enhanced Transformer structure, construct the angle response skew results into a two-dimensional input tensor according to the angle sequence, insert position index encoding and response residual feature channels in each encoding layer, and introduce a multi-head attention structure to extract feature dependencies in non-axisymmetric responses.

[0096] When constructing the residual attention-enhanced Transformer structure, the angle response skew results must first be converted into a standardized two-dimensional input tensor. Specifically, using the angle increment as the row index, the corresponding angle response skew values ​​are arranged as column vectors to form a fixed-dimensional angle sequence input matrix. Subsequently, position index encoding is added to each encoding layer of the Transformer structure to embed angle information into the sequence representation, preserving sequential features. Furthermore, a response residual feature channel is introduced into the input channel to extract the error offset of each angle point relative to the ideal response model, used to model the differential characteristics of the skew trajectory. Next, a multi-head attention structure is used to capture feature dependencies at multiple scales in parallel, and the relative weights between different angles are calculated using the attention matrix, thereby achieving global modeling of the non-axisymmetric error characteristics.

[0097] The residual attention-enhanced Transformer architecture refers to an improved structure that integrates the response residual feature channel and residual connection path into the standard Transformer architecture. This structure preserves the original error sequence information by introducing a residual connection mechanism, avoiding the gradient decay problem in deep networks. Position index encoding is an embedding method that converts the position information of the angle sequence into vector form. It often uses sine and cosine functions to construct fixed position vectors to enhance the network's ability to perceive angle order. The response residual feature channel is a separate branch channel, whose input is the difference between the actual response value and the ideal model response value at each angle point, used to extract anomalous response trends of structural skew. The multi-head attention structure refers to using multiple independent attention heads in the Transformer, obtaining dependencies in multiple feature subspaces through parallel attention weight calculation, thereby improving the network's ability to recognize nonlinear drift patterns. These mechanisms work together to achieve deep modeling and feature capture of complex patterns in the angle response skew results of photoelectric angle sensors.

[0098] S302. The residual attention-enhanced Transformer structure is used to extract features from the skew envelope of the structure. The feature changes in the continuous response path are preserved through the residual jump structure, and the attention channel outputs an error change trajectory feature set containing the magnitude offset gradient, response period variation and phase characteristic trend.

[0099] When extracting features from the structural skew envelope using a residual attention-enhanced Transformer structure, the structural skew envelope is first input into the Transformer structure as a standardized two-dimensional sequence tensor. The structural skew envelope refers to the main response amplitude envelope extracted by fitting the angular trajectories of each node in the periodic response, reflecting the overall skew trend. In the Transformer structure, each encoding layer learns the interdependence patterns of response shifts between angles through a self-attention computation module. Simultaneously, residual skip structures are inserted between layers to fuse shallow angular response features with deep features, enhancing the trend preservation capability under continuous angular changes. Each attention head encodes the response data at different angular positions into an attention matrix, and the fused output sequence is generated through weighted summation of the matrices. Finally, the attention channel outputs a high-dimensional feature set containing three types of change patterns: the gradient trajectory of the response amplitude shift, the perturbation index of the periodic change trend, and the temporal path of phase characteristic changes.

[0100] The structural skew envelope represents the trajectory of the principal amplitude of the angular response as it changes with the angle during periodic rotation, reflecting the main trend of the skew response generated by the structure under non-axisymmetric loading. This trajectory is the principal component line in a family of periodic displacement fitting functions, exhibiting strong periodicity and overall offset characteristics. The residual jump structure refers to a connection method in the Transformer network where the output of each layer is directly weighted and superimposed with the original input, ensuring that low-level features are not excessively replaced by high-level abstractions. This is particularly suitable for capturing weak but continuous response change paths. Attention channels refer to the multi-head attention computation path in the Transformer. Each attention channel learns drift features at different scales in the angular response through different combinations of query, key, and value matrices, playing a role in focusing on abnormal perturbations and enhancing feature resolution within the overall structure. These feature extraction methods work together to systematically capture various error trajectory shapes in the structural skew envelope, supporting subsequent multi-dimensional drift discrimination calculations.

[0101] S303. Perform a multi-dimensional drift discrimination process based on the error change trajectory feature set, and output the discrimination results reflecting the drift direction, drift level and drift stability of structural error. The drift direction is determined based on the peak offset direction of the response envelope, the drift level is determined according to the normalized offset amplitude, and the drift stability is determined based on the persistence of phase perturbation.

[0102] The multi-dimensional drift discrimination process based on the feature set of error change trajectories aims to accurately reconstruct the internal error evolution mechanism of angular response skewness caused by structural off-center loading in a simulation environment where non-axisymmetric loading conditions are not set. Structural errors in angular responses often exhibit complex characteristics such as directional drift, abnormal offset amplitude, and weakened periodic stability; single-dimensional analysis methods cannot fully describe their complexity. Therefore, by determining the drift direction from the peak offset direction of the response envelope, extracting the drift level from the normalized offset amplitude, and assessing the drift stability by combining the continuity of phase perturbation, a complete three-dimensional discrimination system can be constructed. This multi-dimensional error drift discrimination strategy can accurately characterize the evolution characteristics of different types of errors while maintaining the original data structure, thus providing a solid theoretical basis and decision support for subsequent modeling optimization, adaptive parameter tuning, and response correction. This processing flow effectively overcomes the error identification distortion problem caused by traditional methods relying solely on amplitude fitting, enhancing the consistency between simulation results and real-world operating conditions.

[0103] In this embodiment, S303 specifically refers to:

[0104] Extract the response envelope peak sequence within a continuous angular interval from the error change trajectory feature set, calculate the offset direction of the envelope peak relative to the reference angular position, and generate the discrimination result of the structural error drift direction based on the angular quadrant where the maximum offset direction is located.

[0105] When extracting the peak sequence of the response envelope within a continuous angular interval from the feature set of error change trajectories, a sliding window strategy can be used to search for the maximum amplitude point on the structural skew envelope curve according to the angular step size, constructing a peak sequence containing pairs of angular positions and amplitudes. The difference between the angular position of each peak and a preset reference angular position is calculated to determine the pointing angle of the offset vector in the two-dimensional angular domain. Based on the weighted composite direction of all peak offset vectors, the position with the largest offset vector magnitude is identified, and the angular quadrant containing this maximum offset direction is determined. Using the relative relationship between this quadrant and the structural installation direction, the result of the structural error drift direction is output. For example, if the maximum peak offset direction appears within the range from zero degrees to ninety degrees, the drift direction can be determined to be towards the first quadrant, reflecting that the structural stress forms the maximum eccentric load response in this angular segment.

[0106] The response envelope peak sequence refers to the multiple local maximum amplitude points extracted from the structural skew envelope, exhibiting periodic distribution characteristics and concentrating on the top positions reflecting the error drift trend. The reference angular position is the zero-skew starting point set when constructing the analysis coordinate system, typically corresponding to the standard symmetry point in an ideal axisymmetric structure, used as the offset benchmark. The offset direction of the envelope peak relative to the reference angular position describes the angular region towards which the maximum drift of the actual structural response points, reflecting the concentrated trend of the abnormal response under non-axisymmetric loads. The maximum offset direction is the direction indicated by the largest magnitude among all offset vectors. Angular quadrant division is used to discretely express the orientation of the drift direction within the circular distribution, providing a structured output for drift direction classification. This classification result provides preliminary directional information for subsequent calculations of error level and stability.

[0107] Based on the normalized offset value of each response amplitude in the error change trajectory feature set, an offset classification vector is constructed. The offset classification vector is segmented into intervals. The number of high offset intervals is counted and an amplitude threshold is set. The offset level is compared with the threshold mapping table to generate the drift level discrimination result.

[0108] When constructing an offset grading vector based on the normalized offset value of each response amplitude in the error change trajectory feature set, all response amplitude offset values ​​can be mapped to a range of 0 to 1 according to their absolute value range, and normalization is completed through min-max normalization. This normalized offset sequence constitutes the offset grading vector, representing the relative offset degree of each angle point under a uniform numerical scale. The offset grading vector is then divided into several fixed-width intervals, such as low offset, medium offset, and high offset regions, and the number of angle nodes in the high offset region is counted. Combined with a pre-set amplitude threshold standard, it can be determined whether the offset exceeds the critical range. By mapping the number of high offsets to the amplitude threshold to a drift level discrimination table, a quantitative judgment result of the drift level can be output. For example, when the number of nodes in the high offset region exceeds one-third of the total number of nodes and the average offset value exceeds 0.7, it can be judged as a high-level structural drift risk.

[0109] Normalized offset values ​​represent the relative intensity of the response offset at each angle point within the overall distribution. Scaling ensures comparability of amplitudes across different nodes. The offset grading vector is a sequence of normalized offset values, essentially a mapping of error offset intensity at each point in the angular domain. The number of high-offset intervals refers to the number of nodes in the vector belonging to the high-intensity offset range, reflecting the concentrated characteristics of anomalous drift in local or overall structural responses. The amplitude threshold serves as the baseline for drift level classification, mapping continuous offset data to discrete drift level categories to ensure robustness and consistency in the classification process. These features work together to accurately determine whether the structure exhibits significant error drift levels during simulation, effectively serving the input reliability determination in the error modeling stage.

[0110] A perturbation persistence measure function is constructed by utilizing the phase perturbation change trend in the error change trajectory feature set. The length of the continuous perturbation interval is calculated based on the perturbation persistence measure function. The phase stability is evaluated by combining the average perturbation density and standard deviation, and the drift stability discrimination result is generated.

[0111] When constructing a perturbation persistence metric function using the phase perturbation change trend in the error change trajectory feature set, a perturbation index curve can be constructed by extracting the instantaneous phase value corresponding to each angle node and analyzing its curvature along the angle sequence. This curve reflects the coherence of phase perturbation changes with angle. When the phase change amplitude of multiple consecutive angle points exceeds a preset perturbation threshold and the change trend remains unchanged, it can be regarded as a perturbation interval. The angular span of each perturbation interval is taken as the length of the continuous perturbation interval, and the average length and number of all perturbation intervals are counted. For example, if there are five phase perturbation segments with a length exceeding 30 degrees within 360 degrees, it can be inferred that the perturbation exhibits a persistent trend. Subsequently, the average perturbation density of perturbation intensity within each perturbation interval is calculated, and the standard deviation is used to measure the fluctuation of perturbation intensity. Combining these two parameters, phase stability can be classified and judged, thereby generating a quantitative output result of drift stability.

[0112] The trend of phase perturbation variation refers to the differential variation characteristics of a periodic response signal in the phase dimension, typically manifested as the continuity, amplitude, and abrupt changes in the phase difference between adjacent angular nodes. The perturbation persistence measure is a functional index used to characterize the continuous temporal or spatial span of phase perturbations; essentially, it is a statistical evaluation standard for phase noise or unstable response segments in the response data. The length of a continuous perturbation interval represents the width of the segment where the perturbation phenomenon persists in the angular domain, and is a key parameter for determining whether the perturbation exhibits a systematic trend. The average perturbation density measures the concentration of phase perturbation intensity within a unit angular range, providing a quantitative description of the perturbation's "density." The standard deviation is used to assess the fluctuation range of perturbation amplitude within each perturbation interval, serving as a statistical tool characterizing the differences in perturbation stability. These parameters collectively establish the basis for discriminating phase perturbation patterns in the structural response, thereby enabling a refined distinction of error drift stability.

[0113] S4. Based on the drift discrimination results, construct a nonlinear entropy gate mapping network to map the error drift characteristics into fuzzy classification categories and divide them into a predictable response region, a dynamic boundary transition region, and an uncorrectable response region.

[0114] In this embodiment, S4 specifically refers to:

[0115] Based on the drift discrimination results, a nonlinear entropy gate mapping network is constructed. The drift direction, drift level and drift stability are used as input parameters. A three-dimensional drift feature tensor is constructed by concatenating the input features. An entropy gate control unit is introduced to adjust the contribution intensity of each drift dimension to the output mapping. A nonlinear mapping path is constructed using a two-layer fully connected network.

[0116] The process of constructing a nonlinear entropy-gated mapping network uses the drift direction, drift level, and drift stability extracted from the drift discrimination results as basic input variables. First, these three parameters are normalized and their dimensions expanded, concatenating them into a three-dimensional drift feature tensor with angular distribution, drift attribute, and response intensity dimensions. An entropy gate control unit is embedded in the network structure design. This control unit dynamically allocates weight coefficients based on the distribution pattern of drift attributes in different angular regions to adjust the influence of each drift dimension on the final category output. The entropy gate control unit can be composed of trainable gating functions, using a sigmoid-like activation function combined with a weight matrix to control the opening and closing states of the feature channels, thereby achieving information filtering between feature channels during network forward propagation. Then, the processed feature tensor is input into a mapping path containing two layers of nonlinear fully connected structures. The first layer is used for high-dimensional feature transformation, and the second layer performs output space projection. The output vector is used as the probability distribution vector of the assigned category to support subsequent fuzzy classification.

[0117] In one specific implementation, the discrimination results of structural error drift are first obtained, where the drift direction is represented by the angle quadrant number, the drift level by the grade label, and the drift stability by the proportion of continuous perturbation time. These values ​​are normalized and combined into a basic tensor of shape [N, 3], where N is the number of angle samples. This tensor is expanded to form a three-dimensional structure of [N, 3, 1] as input. The introduced entropy gate control unit adopts a channel-by-channel gating strategy, setting independent gating weights for each drift dimension and dynamically adjusting them. During training, the parameters of the gating function are optimized through backpropagation to improve the adaptability of the drift dimension to the predicted output. Next, the input features are mapped to a high-dimensional space through a first fully connected network for nonlinear activation processing. The output results are then dimensionality-reduced through a second fully connected network and mapped to three output nodes, each node corresponding to a probability value of a fuzzy classification. Finally, the category indicated by the maximum probability value is selected as the classification result of the current response state. This method can effectively integrate multiple drift features and dynamically adjust their importance, thereby enhancing the accuracy and robustness of drift detection.

[0118] The drift discrimination result is mapped based on a nonlinear entropy gate mapping network. The multi-channel response value corresponding to the input vector is extracted. The normalization function is used to convert all channel responses into the attribution probability and a fuzzy attribution category label set is constructed. Each category label corresponds to the attribution probability channel of the predictable response region, the dynamic transition region of the boundary, and the uncorrectable response region.

[0119] When performing mapping operations on drift discrimination results using a nonlinear entropy gate mapping network, the constructed 3D drift feature tensor is first input into the network structure. Through forward propagation, multiple channel response outputs are obtained, each corresponding to a structural state category. The multi-channel response values ​​refer to multiple activation values ​​generated in parallel at the network output layer, representing the response strength or confidence level of the structural state in different categories. After calculating the response values, a normalization function is used to standardize all channel response values, making their sum a constant value, thus transforming the response of each channel into a probabilistic attribution value. The normalization function can be softmax or max-min normalization, ensuring the comparability of outputs from different state categories and maintaining the uniformity of the sum of attribution probabilities. Finally, the normalized response results are divided into three independent category channels, representing the attribution probabilities corresponding to the predictable response region, the dynamic transition boundary region, and the uncorrectable response region, respectively, to construct a fuzzy attribution category label set.

[0120] In one implementation, after fusing features such as drift direction, drift level, and drift stability, the input tensor is fed into a constructed nonlinear entropy gate mapping network. This network consists of two fully connected mapping layers and a set of parallel output nodes, with the output dimension set to three channels. After nonlinear processing of the input features, the network outputs a two-dimensional matrix of the form [N,3], where each row corresponds to an angle sampling point and each column corresponds to a category channel. The three channel values ​​of each row are standardized using a softmax normalization function, transforming them into three probability values ​​representing the probability that the current angle point belongs to the predictable response region, the dynamic boundary transition region, or the uncorrectable response region. These probability values ​​are combined to form a fuzzy category label set, which supports a threshold filtering mechanism based on different classification probabilities. In practical applications, this set is used to identify the reliability state of the structural response or to implement compensation mechanisms, achieving uncertainty modeling and boundary risk prediction of structural behavior.

[0121] The attribution result is determined based on the category channel corresponding to the maximum attribution probability value in the fuzzy attribution category label set. When the maximum value channel is a channel in the predictable response zone, it is determined to be in the predictable response zone. When the maximum value channel is a channel in the dynamic transition zone of the boundary, it is determined to be in the dynamic transition zone of the boundary. When the maximum value channel is a channel in the uncorrectable response zone, it is determined to be in the uncorrectable response zone.

[0122] In the fuzzy classification label set, each angle sampling point corresponds to a set of three classification probability values. These three probability values ​​are mapped to channels in the predictable response region, the dynamic transition boundary region, and the uncorrectable response region, respectively. By comparing the channel corresponding to the maximum value in this set of classification probability values, the most likely classification state of the response behavior at the current angle position in the fuzzy classification can be determined. The core of this method is to use the maximum classification probability value as the basis for classification determination, thereby maintaining the characteristics of multi-class fuzzy inference while outputting a clear discrimination result to meet the needs of actual classification or strategy selection. This method can effectively avoid the uncertainty diffusion problem in fuzzy classification, while enabling the system to have a clear response strategy path and supporting dynamic updates of the structural state. The classification channel is determined based on the relative strength of the classification probability rather than the absolute value, thus exhibiting good robustness and adaptability.

[0123] In the specific implementation, for each input angle point, its three corresponding attribution probabilities are input into the classification and discrimination module. This module compares the response values ​​of the three channels and extracts the channel index corresponding to the maximum value. If the maximum value is located in the predictable channel, the system determines that the structural response corresponding to that angle point is highly predictable and belongs to a stable state. If the maximum value is located in the boundary dynamic transition channel, the system marks that point as a transition area with uncertain fluctuations, requiring the introduction of a dynamic evaluation mechanism or adjustment of monitoring parameters. If the maximum value is located in the uncorrectable channel, it indicates that the structural response corresponding to that angle point has deviated from the safe range and cannot be effectively corrected by the compensation mechanism, triggering an alarm mechanism or a shutdown strategy. This judgment mechanism based on the maximum attribution probability enables a rapid response to the structural behavior state and is suitable for online diagnostics and real-time control systems.

[0124] S5. Based on the fuzzy classification, generate a drift self-sensing classification factor sequence. Combined with the error evolution trend within the angle segment, adjust the convergence priority and learning rate curve of the training objective function in the fuzzy calibration modeling process of the photoelectric angle sensor to complete the dynamic accuracy control of the error modeling output.

[0125] In this embodiment, S5 specifically refers to:

[0126] Based on the probability value of each angle sampling point in the fuzzy classification, the predictable response area, the dynamic transition area of ​​the boundary, and the uncorrectable response area are mapped to three levels of classification weights: low, medium, and high. The drift self-aware classification factor sequence is constructed by arranging them according to the angle sequence to describe the error drift response characteristics corresponding to different angle segments.

[0127] In the error modeling task of photoelectric angle sensors, to achieve dynamic adaptive processing of different response regions, a drift-aware attribution factor sequence needs to be constructed based on the attribution probability value of each angle sampling point in the fuzzy attribution category. First, the attribution probability values ​​are mapped to three category channels: the predictable response region, the dynamic transition boundary region, and the uncorrectable response region, defined as low, medium, and high attribution weight levels, respectively. This hierarchical mapping method can transform the continuous characteristics of fuzzy attribution into numerical indicators with direct reference significance in error control, thereby applying differentiated adjustment strategies in training scheduling. The attribution factor sequence generated after arranging according to the angle sequence can be regarded as an information expression form that integrates drift discrimination conclusions and spatial distribution structure, providing an intuitive characterization of the error fluctuation state within the angle segment, and providing directional basis and granular control capability for subsequent convergence weight adjustment of the objective function.

[0128] In the specific implementation, the probability values ​​of each angle point in the three category channels are first extracted from the model's output probability matrix, and the value of each channel is selected as the weight basis for that point under different drift levels. For example, when the probability of an angle point in the uncorrectable response zone channel is higher than that in other channels, its assignment factor is marked as high-level. This factor is then encoded numerically as a high-weight value, and all angle points are arranged sequentially to generate a drift-aware assignment factor sequence. This sequence can be further processed by convolutional smoothing or sliding window weighting to remove noise influence and achieve continuous modeling of local error response trends. Finally, this assignment factor sequence will be used as a dynamic modeling control input to the subsequent training mechanism to determine the modeling resource allocation strategy for different angle segments, thereby enhancing the system's adaptability to non-uniform error drift phenomena.

[0129] The error evolution trend is extracted within each angle segment. An error evolution gradient sequence is constructed based on the slope of error change between continuous angle sampling points. A joint control weight matrix is ​​formed by combining the drift self-perception attribution factor sequence to reflect the modeling complexity and response uncertainty of each angle segment.

[0130] To achieve dynamic adaptation across different angle segments during the modeling of photoelectric angle sensors, it is necessary to extract the error evolution trend within each angle segment and construct a joint control mechanism based on drift characteristics. The error evolution trend is calculated by analyzing the slope of error change between consecutive angle sampling points and can be represented as an error evolution gradient sequence, describing the rate and directionality of error change during angle progression. This sequence reflects the dynamic process of error increasing or decreasing with angle. By performing positional matching and numerical fusion of the error evolution gradient sequence with the previously constructed drift self-perception attribution factor sequence, a joint control weight matrix can be generated. This matrix reflects the degree of attention that should be given to each angle segment in error modeling. Using this matrix can significantly enhance the model's adaptability to modeling complexity and response uncertainty, avoid overfitting the model to error stationary segments, and enhance the focus on error abrupt change segments.

[0131] The specific implementation involves first performing a difference operation on the error values ​​of two adjacent sampling points within each angular segment to calculate the slope of error change for each segment, and then forming an error evolution gradient sequence from all slope values. Next, the numerical characteristics of this sequence are used to determine the error change trend; for example, segments with drastic changes are marked as high-dynamic regions, and segments with gradual changes are marked as stable regions. Then, this sequence is multiplied term-by-term by the drift self-aware attribution factor sequence at each angular point to construct a joint control weight matrix reflecting both characteristics. This matrix contains both the directionality and gradient information of error changes, and also incorporates the drift identification results' judgment of regional uncertainty. During actual training, this matrix can be embedded as a weighting factor into the model's objective function to adjust the contribution of the error response of each angular segment to the loss function, thereby dynamically guiding model resources towards high-complexity or high-uncertainty regions, improving overall modeling accuracy and robustness.

[0132] By introducing the joint control weight matrix into the fuzzy calibration modeling process of photoelectric angle sensor, the convergence priority parameter of the training objective function is dynamically adjusted, and the adjustment factor of the learning rate curve is set according to the changing trend of the error evolution gradient. During the model training process, the convergence efficiency is improved in the high response drift section, and the weight convergence step size is slowed down in the dynamic transition section, thereby completing the dynamic accuracy control of the error modeling output.

[0133] Introducing a joint weighting matrix into the fuzzy calibration modeling process of photoelectric angle sensors aims to achieve dynamic accuracy control of the error modeling output. This control mechanism dynamically adjusts two key factors in model training: the convergence priority parameter of the training objective function and the adjustment factor of the learning rate curve. The convergence priority parameter controls the distribution of loss weights in different angle segments during training, allowing the model to optimize faster and converge in segments with large errors or significant drift. In transitional regions where error changes drastically but lacks clear directional characteristics, smaller weights are used to increase the step size, slowing down the training speed and preventing the model from falling into local fitting traps or over-adjusting. The adjustment factor of the learning rate curve flexibly adjusts the gradient descent step size based on the error evolution gradient trend, enabling the model to have strong tracking ability in high dynamic response segments and maintain training stability in low dynamic segments. Through this mechanism, the model can adaptively adjust the accuracy requirements of the sensor under different response states, thereby improving generalization ability and error control in the overall modeling process.

[0134] In its implementation, the joint control weight matrix is ​​first weighted and fused with the loss term in the original objective function. The calculation ratio of the corresponding loss is adjusted based on the weight value at each angle point, thereby dynamically allocating training priorities. Secondly, the rate of change information at each angle point in the error evolution gradient sequence is extracted. A threshold is set to determine the strength of the changing trend, and an adjustment factor for the learning rate curve is generated accordingly. This adjustment factor can be used to correct the base learning rate in real time, accelerating gradient descent in high error slope regions to improve optimization speed, and slowing down parameter update frequency in unstable or fluctuating error slope regions to enhance the stability and robustness of the training process. During training, this dynamic control mechanism can be implemented through policy functions or conditional mapping logic modules and integrated with the model optimizer's update process. This ensures that each round of parameter updates fully considers the error behavior characteristics of the angle segment, thus forming a high-precision modeling process that is sensitive to response partitions and has a controllable optimization rhythm.

[0135] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, as a computer program product. A computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, all or part of the processes or functions according to the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. Computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired or wireless means (e.g., infrared, wireless, microwave, etc.). A computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more sets of available media. Available media can be magnetic media (e.g., floppy disks, hard disks, magnetic tapes), optical media (e.g., DVDs), or semiconductor media. Semiconductor media can be solid-state drives.

[0136] It should be understood that in the various embodiments of this application, the order of the above-mentioned processes 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 application.

[0137] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0138] In the several embodiments provided in this application, it should be understood that the disclosed systems and methods can be implemented in other ways. For example, the embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be an indirect coupling or communication connection through some interfaces, devices, or units, and may be electrical, mechanical, or other forms.

[0139] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0140] In addition, the functional units in the various embodiments of this application 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.

[0141] The above are merely specific embodiments of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A fuzzy calibration method for photoelectric angle sensors based on ANSYS simulation, characterized in that, Specifically, the following steps are included: S1. Extract all discrete load vector data from the ANSYS simulation settings, construct a load symmetry mapping rate function based on the directional distribution of each load vector about the principal rotation axis, and generate a mapping rate value by calculating the mirror fitting deviation, which is used to identify whether non-axisymmetric loading conditions are not set during the simulation process. S2. Based on the recognition results, call the node displacement data with angle change in the sensor simulation output, construct a family of periodic displacement offset fitting functions, and perform curve fitting processing on one rotation data to determine the angular response skew caused by structural off-center loading when no non-axisymmetric loading conditions are set during the simulation. S3. Based on the determined angle response skew results, construct a residual attention-enhanced Transformer structure, extract the error change trajectory feature set in the structural skew envelope, and output the drift discrimination results that reflect the drift direction, level and stability of structural error. S4. Based on the drift discrimination results, construct a nonlinear entropy gate mapping network to map the error drift characteristics into fuzzy classification categories and divide them into a predictable response region, a dynamic boundary transition region, and an uncorrectable response region. S5. Based on the fuzzy classification, generate a drift self-sensing classification factor sequence. Combined with the error evolution trend within the angle segment, adjust the convergence priority and learning rate curve of the training objective function in the fuzzy calibration modeling process of the photoelectric angle sensor to complete the dynamic accuracy control of the error modeling output.

2. The fuzzy calibration method for photoelectric angle sensors based on ANSYS simulation according to claim 1, characterized in that, S1 specifically refers to: Extract all discrete load vector data from the ANSYS simulation settings, project each load vector onto a cross-sectional plane perpendicular to the principal axis of rotation, extract the angle between each load vector and the reference radial reference direction within the cross-sectional plane, and generate an ordered distribution set containing all angle values. The load symmetry mapping rate function is constructed based on the ordered distribution set. The polar angle symmetry mapping method is adopted to pair each included angle value with the included angle value that is symmetrically distributed with respect to the center line of the main rotation axis, and the numerical difference of each pair of load vectors is extracted. All numerical differences are calculated into a set of mirror fit deviation vectors. Mean square fitting is performed according to the vector magnitude to obtain the mean square value of the mirror fit deviation as the mirror fit deviation metric. A normalized proportional relationship is constructed based on the mirror fitting deviation metric, and a mapping rate value is generated. When the mapping rate value is greater than a set threshold, it is used to identify whether non-axisymmetric loading conditions were not set during the simulation process.

3. The fuzzy calibration method for photoelectric angle sensors based on ANSYS simulation according to claim 1, characterized in that, S2 specifically includes the following steps: S201. When the mapping rate value constructed based on the mirror fitting deviation metric exceeds the set threshold, the nodal displacement data with angle change in the ANSYS simulation output is called, and the nodal trajectory is organized according to the angle sequence within one rotation interval. S202. Extract the displacement change sequence of each node in the rotation interval into a periodic response vector, extract the fundamental frequency component based on discrete Fourier transform and construct a family of periodic displacement offset fitting functions, and record the phase center position of the trajectory of each node in the function family. S203. Perform multi-curve superposition on all functions in the family of periodic displacement offset fitting functions, and extract the main response envelope curve using the minimum deviation fitting strategy. S204. Calculate the angular response skew value based on the relative phase difference between the master response envelope curve and the ideal axisymmetric fitting curve, and determine the angular response skew caused by structural off-center loading when no non-axisymmetric loading conditions are set during the simulation.

4. The fuzzy calibration method for photoelectric angle sensors based on ANSYS simulation according to claim 3, characterized in that, S202 specifically refers to: The displacement change sequence of each node within the complete rotation interval is converted into a periodic response vector according to the angle sampling order, and the amplitude range of the periodic response vector is normalized to unify the numerical scale of the response vector. Perform a discrete Fourier transform on each normalized periodic response vector to extract the fundamental frequency component in the frequency domain features, which serves as the dominant feature parameter for the periodic displacement change. A family of periodic displacement fitting functions is constructed based on the phase and amplitude parameters in the fundamental frequency component, and the phase center position of the periodic response of each node in the family of fitting functions is recorded as a relative position reference.

5. The fuzzy calibration method for photoelectric angle sensors based on ANSYS simulation according to claim 3, characterized in that, S204 specifically refers to: Based on the master response envelope curve, a periodic amplitude trajectory in the complete angular domain is constructed. An ideal axisymmetric fitting curve is constructed using Fourier cosine expansion. The ideal axisymmetric fitting curve and the master response envelope curve are then subjected to node resampling within a unified angular interval. The response amplitudes of the main response envelope curve and the ideal axisymmetric fitting curve at each sampling angle point are respectively mapped to complex numbers, and the relative phase difference sequence between the two curves is calculated based on the complex plane phase relationship. The position of the main peak and the angle index offset in the relative phase difference sequence are extracted as the angle response skew value. Based on the offset directionality and offset amplitude, it is determined whether the response skew caused by the structural off-center loading is valid. This is used to confirm the response offset characteristics when the non-axisymmetric loading condition is not set.

6. The fuzzy calibration method for photoelectric angle sensors based on ANSYS simulation according to claim 1, characterized in that, S3 specifically includes the following steps: S301. Based on the angle response skew results, construct a residual attention-enhanced Transformer structure, construct the angle response skew results into a two-dimensional input tensor according to the angle sequence, insert position index encoding and response residual feature channels in each encoding layer, and introduce a multi-head attention structure to extract feature dependencies in non-axisymmetric responses. S302. The residual attention-enhanced Transformer structure is used to extract features from the skew envelope of the structure. The feature changes in the continuous response path are preserved through the residual jump structure, and the attention channel outputs an error change trajectory feature set containing the magnitude offset gradient, response period variation and phase characteristic trend. S303. Perform a multi-dimensional drift discrimination process based on the error change trajectory feature set, and output the discrimination results reflecting the drift direction, drift level and drift stability of structural error. The drift direction is determined based on the peak offset direction of the response envelope, the drift level is determined according to the normalized offset amplitude, and the drift stability is determined based on the persistence of phase perturbation.

7. The fuzzy calibration method for photoelectric angle sensors based on ANSYS simulation according to claim 6, characterized in that, S303 specifically refers to: Extract the response envelope peak sequence within a continuous angular interval from the error change trajectory feature set, calculate the offset direction of the envelope peak relative to the reference angular position, and generate the discrimination result of the structural error drift direction based on the angular quadrant where the maximum offset direction is located. Based on the normalized offset value of each response amplitude in the error change trajectory feature set, an offset classification vector is constructed. The offset classification vector is segmented into intervals. The number of high offset intervals is counted and an amplitude threshold is set. The offset level is compared with the threshold mapping table to generate the drift level discrimination result. A perturbation persistence measure function is constructed by utilizing the phase perturbation change trend in the error change trajectory feature set. The length of the continuous perturbation interval is calculated based on the perturbation persistence measure function. The phase stability is evaluated by combining the average perturbation density and standard deviation, and the drift stability discrimination result is generated.

8. The fuzzy calibration method for photoelectric angle sensors based on ANSYS simulation according to claim 1, characterized in that, S4 specifically refers to: Based on the drift discrimination results, a nonlinear entropy gate mapping network is constructed. The drift direction, drift level and drift stability are used as input parameters. A three-dimensional drift feature tensor is constructed by concatenating the input features. An entropy gate control unit is introduced to adjust the contribution intensity of each drift dimension to the output mapping. A nonlinear mapping path is constructed using a two-layer fully connected network. The drift discrimination result is mapped based on a nonlinear entropy gate mapping network. The multi-channel response value corresponding to the input vector is extracted. The normalization function is used to convert all channel responses into the attribution probability and a fuzzy attribution category label set is constructed. Each category label corresponds to the attribution probability channel of the predictable response region, the dynamic transition region of the boundary, and the uncorrectable response region. The attribution result is determined based on the category channel corresponding to the maximum attribution probability value in the fuzzy attribution category label set. When the maximum value channel is a channel in the predictable response zone, it is determined to be in the predictable response zone. When the maximum value channel is a channel in the dynamic transition zone of the boundary, it is determined to be in the dynamic transition zone of the boundary. When the maximum value channel is a channel in the uncorrectable response zone, it is determined to be in the uncorrectable response zone.

9. The fuzzy calibration method for photoelectric angle sensors based on ANSYS simulation according to claim 1, characterized in that, S5 specifically refers to: Based on the probability value of each angle sampling point in the fuzzy classification, the predictable response area, the dynamic transition area of ​​the boundary, and the uncorrectable response area are mapped to three levels of classification weights: low, medium, and high. The drift self-aware classification factor sequence is constructed by arranging them according to the angle sequence to describe the error drift response characteristics corresponding to different angle segments. The error evolution trend is extracted within each angle segment. An error evolution gradient sequence is constructed based on the slope of error change between continuous angle sampling points. A joint control weight matrix is ​​formed by combining the drift self-perception attribution factor sequence to reflect the modeling complexity and response uncertainty of each angle segment. By introducing the joint control weight matrix into the fuzzy calibration modeling process of photoelectric angle sensor, the convergence priority parameter of the training objective function is dynamically adjusted, and the adjustment factor of the learning rate curve is set according to the changing trend of the error evolution gradient. During the model training process, the convergence efficiency is improved in the high response drift section, and the weight convergence step size is slowed down in the dynamic transition section, thereby completing the dynamic accuracy control of the error modeling output.

Images