A method and device for integrated collaborative optimization design of a permanent magnet synchronous motor rotor magnetic circuit

Abstract

This invention relates to the field of permanent magnet synchronous motor (PMSM) design technology, and discloses a method and device for integrated collaborative optimization design of the rotor magnetic circuit of a PMSM. The method includes: collecting time-series data of multi-source vibration and electromagnetic load under heavy-load conditions of forklift operation, and constructing a dynamic excitation phase space mesh for the rotor interface; performing multi-field coupled response simulation based on the mesh to drive a parameterized rotor model, generating a structural eddy current coupled simplex complex; constructing a combined constraint and sub-mode gain model for robust magnetic circuit optimization based on the complex, generating a geometrically reinforced candidate solution set; performing multi-objective collaborative convergence and three-dimensional model generation under the constraints of the candidate solution set, and outputting an anti-disturbance optimized rotor structure. This invention solves the problem that existing technologies cannot cope with dynamic multi-field positive feedback under complex forklift operating conditions, leading to decreased efficiency and low reliability of the motor during continuous operation, thus enabling the motor to adapt to forklift operation requirements.

Description

Technical Field

[0001] This invention relates to the field of permanent magnet synchronous motor design technology, and discloses a method and device for integrated collaborative optimization design of the rotor magnetic circuit of a permanent magnet synchronous motor. Background Technology In warehouse forklift operation scenarios, permanent magnet synchronous motors are the core power source. The optimization of their rotor magnetic circuit needs to focus on static parameters such as the number of magnet segments, rotor yoke thickness, and contact gap between magnets and laminations. Existing designs mostly use finite element simulation tools, combined with static coupling analysis of dual physical fields such as electromagnetic and thermal, and structural and electromagnetic, to evaluate motor losses and strength characteristics in order to ensure operating performance under normal working conditions.

[0002] However, when forklifts operate in warehouse areas, they frequently encounter combined conditions of bumpy roads and heavy-load uphill climbing, and the existing rotor magnetic circuit design cannot adapt to the dynamic operating requirements under these conditions. When the forklift travels on bumpy roads, the rotor vibrates at high frequencies along with the vehicle body, causing micro-deformation and gradual widening of the contact gap between the magnets and laminations. This widening gap directly exacerbates magnetic flux distortion at the edges of the magnetic circuit, increasing local eddy current losses in the rotor. The heat generated by these eddy current losses then triggers thermal expansion of the rotor components, further widening the contact gap between the magnets and laminations, creating a cycle. This problem leads to a gradual decrease in motor efficiency after continuous operation, with the degradation trend intensifying over time. It also accelerates the aging of internal motor components, reducing reliability. Summary of the Invention

[0003] To address the aforementioned problems, the present invention provides the following technical solution: A method for integrated collaborative optimization design of rotor magnetic circuit of permanent magnet synchronous motor includes: S1 collects multi-source vibration and electromagnetic load time-series data of forklift under typical bumpy heavy-load conditions, and constructs a dynamic excitation phase space grid for rotor interface. S2, based on the dynamic excitation phase space grid of the rotor interface, drives the parameterized rotor model to perform multi-field coupled response simulation and generates a structural eddy current coupled simple complex. S3, based on the structural eddy current coupled simple complex, constructs a combined constraint and sub-mode gain model for magnetic circuit robustness optimization, and generates a set of candidate solutions for geometric reinforcement; S4, under the constraint of geometrically reinforced candidate solution set, performs multi-objective collaborative convergence and 3D model generation, and outputs a disturbance-resistant optimized rotor structure.

[0004] Furthermore, the steps for constructing the dynamic excitation phase space mesh of the rotor interface include: S11, collect multi-source vibration time-series data and electromagnetic load time-series data under typical bumpy heavy-load conditions of forklifts to obtain raw multi-source vibration time-series data and raw electromagnetic load time-series data. S12, based on the original multi-source vibration time series data and the original electromagnetic load time series data, perform time-frequency alignment processing to obtain time-frequency aligned multi-source vibration-electromagnetic load collaborative time series data; S13, Based on the time-frequency aligned multi-source vibration-electromagnetic load collaborative time series data, a high-dimensional mapping operation is performed using the sliding window embedding method to generate phase space trajectory points; S14, based on phase space trajectory points, performs phase space adaptive partitioning operation to construct a dynamic excitation phase space mesh for the rotor interface.

[0005] Furthermore, the steps for performing high-dimensional mapping operations using the sliding window embedding method include: S131, Determine the size of the sliding window, which is determined based on the sampling period of the time-frequency aligned multi-source vibration-electromagnetic load collaborative timing data and the characteristic duration of the forklift bumpy heavy load condition. S132, Determine the embedding dimension. Determining the embedding dimension includes: calculating the embedding dimension using the false nearest neighbor method, gradually increasing the dimension and statistically analyzing the proportion of false nearest neighbors, and the dimension when the proportion drops below a preset threshold is the final embedding dimension. S133 expands the time-frequency aligned multi-source vibration-electromagnetic load collaborative time series data within each sliding window according to the final embedding dimension to form a trajectory point in a high-dimensional space. All trajectory points together constitute the phase space trajectory point in the phase space.

[0006] Furthermore, the steps for generating a structural eddy current coupled simple complex through multi-field coupled response simulation of the driven parameterized rotor model include: S21, call the parameterized rotor model to provide a basic simulation carrier for subsequent multi-field coupling simulation; S22, based on the dynamic excitation phase space grid of the rotor interface and the parameterized rotor model, perform the excitation data loading operation to obtain the loaded parameterized rotor model; S23, based on the loaded parameterized rotor model, performs a structure-electromagnetic-thermal multi-field coupled transient simulation operation to generate the micro-deformation field, local eddy current density field and contact interface temperature field of the magnet-laminated lamination contact interface; S24. Based on the micro-deformation field, local eddy current density field and contact interface temperature field of the magnet-stamping interface, perform a triangulation operation on the contact interface to obtain a triangulation set of the magnet-stamping interface. S25, based on the triangulation set of the magnet-stamp contact interface, perform the structural eddy current coupling simple complex construction operation to obtain the structural eddy current coupling simple complex; S26. Based on the structural eddy current coupled simple complex, perform temporal complex sequence construction and continuous cohomology analysis to obtain a structural eddy current coupled simple complex containing time evolution characteristics.

[0007] Furthermore, the steps for performing a structure-electromagnetic-thermal multi-field coupled transient simulation include: S231, Initialize simulation parameters, set simulation time step, convergence threshold and boundary conditions; S232, based on the parameterized rotor model and vibration excitation after loading, uses an explicit dynamic algorithm to calculate the displacement field and stress field of each component of the rotor within each time step, extracts the displacement data of all discrete vertices of the magnet-laminated contact interface, and forms the micro-deformation field of the magnet-laminated contact interface. S233, the micro-deformation field of the magnet-laminated lamination contact interface is fed back to the electromagnetic field module to update the air gap size of the parameterized rotor model. At the same time, the initial temperature distribution of the rotor is calculated based on the frictional heat generated by structural vibration, and the material magnetic property parameters of the permanent magnet and lamination are corrected to obtain the updated parameterized rotor model. S234, based on the updated parameterized rotor model and electromagnetic load excitation, uses the time-step finite element method to calculate the magnetic flux density distribution and eddy current density distribution of the rotor magnetic circuit within each time step, extracts the eddy current data at the vertex of the magnet-laminated contact interface, and forms a local eddy current density field. S235, based on the local eddy current density field, calculates the heat generated by eddy current loss, superimposes the heat generated by structural vibration and friction, uses the finite volume method to calculate the temperature distribution of the rotor in each time step, extracts the temperature data of the vertex of the magnet-laminated contact interface, and forms the contact interface temperature field. S236 determines whether the simulation is complete. If the current simulation time has not reached the preset duration, return to S232 to repeat the multi-field coupling calculation. If the preset duration has been reached, output the micro-deformation field, local eddy current density field, and contact interface temperature field of the magnet-stamp contact interface.

[0008] Furthermore, the steps for constructing a combined constraint and sub-mode gain model for magnetic circuit robustness optimization and generating a set of geometrically reinforced candidate solutions include: S31, based on structural eddy current coupled simple complex, performs high-weight simplex identification and clustering operations to generate a set of vulnerable regions; S32, based on the vulnerable region set and magnetic circuit design rules, performs local geometry modification operation definition operations and generates a set of local geometry modification operations; S33, based on the set of local geometric modification operations and the performance requirements of the rotor magnetic circuit, performs a combined constraint construction operation to generate combined constraints for magnetic circuit robustness optimization; S34, based on the set of vulnerable regions, the set of local geometric modification operations, and combined constraints, performs the sub-mode gain model construction operation to generate a sub-mode gain model with optimized magnetic circuit robustness; S35, based on the sub-modulus gain model and combined constraints, adopts a greedy search strategy to perform candidate solution generation operations, generating a geometrically enhanced candidate solution set.

[0009] Furthermore, the steps for performing the sub-mode gain model construction operation include: S341 defines the input and output of the robustness gain function; specifically, the input is any subset of the set of local geometric modification operations, and the output is the robustness gain value corresponding to that subset. S342, Determine the method for calculating the gain value, which specifically includes: First, verify whether each operation in the subset satisfies the combination constraint. Operations that satisfy the constraint are "valid operations", and those that are not are "invalid operations". Extract the coverage area of ​​all valid operations and take the union of the coverage areas. Calculate the sum of the risk weights of all vulnerable areas in the union. This sum is the robustness gain value. S343, determine whether the function satisfies the submodularity property; S344, set the gain threshold; specifically including: taking a preset proportion of the total risk weight of the vulnerable region set as the gain threshold; combining the input and output of the robustness gain function, the gain value calculation method, the determination function sub-modulus and the gain threshold to obtain the sub-modulus gain model for magnetic circuit robustness optimization.

[0010] Furthermore, the steps for performing multi-objective collaborative convergence and 3D model generation include: S41, Based on the geometrically enhanced candidate solution set, perform parameterized feature tree instantiation operation to generate a candidate 3D model set; S42, based on the candidate 3D model set, construct a multiphysics closed-loop evaluation loop, perform model performance evaluation operations, and generate a model performance dataset; S43, based on the model performance dataset, uses a non-dominated ranking algorithm to perform multi-objective collaborative convergence operation and selects the optimal parameterized feature tree; S44, based on the optimal parameterized feature tree, performs the original rotor model structure update operation and outputs the disturbance-resistant optimized rotor structure.

[0011] Furthermore, the steps for performing multi-objective cooperative convergence operations using a non-dominated sorting algorithm include: S431, define the dominance relationship, specifically including: for any two candidate models M1 and M2 in the model performance dataset, if the efficiency decay rate of M1 is less than or equal to the efficiency decay rate of M2, the eddy current concentration suppression rate is greater than or equal to the eddy current concentration suppression rate of M2, the structural dynamic stiffness is greater than or equal to the structural dynamic stiffness of M2, and at least one of the indicators is strictly better, then it is determined that M1 dominates M2. S432, traverse the model performance dataset, use dominance relations to filter out all candidate models that are not dominated by other models, and form the first non-dominated layer; if the number of models in the first non-dominated layer does not exceed the preset limit, directly use their corresponding parameterized feature trees as the candidate set; if it exceeds the limit, proceed to S433. S433, for each model in the first non-dominated layer, calculate the sum of its distances to neighboring models in three performance index dimensions and record it as the crowding degree; sort the models in descending order of crowding degree, select a preset upper limit number of models, and the corresponding parameterized feature trees constitute the candidate set; S434, verifying optimality, specifically includes: comparing the performance data of the candidate models corresponding to the candidate set with the performance data of the original rotor model, and eliminating the parameterized feature trees that are not superior to the original model; if there are multiple remaining, the parameterized feature tree is finally obtained by screening through manufacturing complexity.

[0012] A device for integrated collaborative optimization design of rotor magnetic circuit of permanent magnet synchronous motor, used to implement the above-mentioned integrated collaborative optimization design method of rotor magnetic circuit of permanent magnet synchronous motor, the device comprising: Dynamic excitation phase space mesh construction module: used to collect multi-source vibration and electromagnetic load time series data under typical bumpy heavy load conditions of forklifts, and construct dynamic excitation phase space mesh of rotor interface; Multi-field coupling simulation and simple complex generation module: used to drive the parameterized rotor model to perform multi-field coupling response simulation based on the dynamic excitation phase space mesh of the rotor interface, and generate structural eddy current coupled simple complex; Combined constraint and candidate solution set generation module: used to construct a combined constraint and sub-mode gain model for magnetic circuit robustness optimization based on structural eddy current coupled simple complex, and generate a geometrically enhanced candidate solution set; Multi-objective convergence and optimized structure output module: used to perform multi-objective collaborative convergence and 3D model generation under geometric reinforcement candidate solution set constraints, and output disturbance-resistant optimized rotor structure.

[0013] Compared with the prior art, the beneficial effects of the present invention are as follows: The present invention effectively solves the problem of losing the spatiotemporal correlation information of dynamic excitation by collecting multi-source vibration and electromagnetic load time series data under the bumpy heavy load condition of forklift and constructing a dynamic excitation phase space grid of rotor interface. This provides a physically consistent input basis for subsequent multi-field coupling simulation and avoids the disconnect between simulation and actual working conditions due to incomplete excitation characterization.

[0014] Based on the phase space grid-driven parameterized rotor model, the structure-electromagnetic-thermal multi-field coupled transient simulation can realistically reproduce the dynamic multi-field positive feedback process of "vibration → gap deformation → increased eddy current loss → thermal expansion → further expansion of gap". The generated structural eddy current coupled simple complex can accurately characterize the spatial correlation and temporal evolution law of high-risk areas, breaking through the technical limitations of existing dual-field static coupling that cannot simulate dynamic feedback and traditional scalar fields that cannot reflect risk correlation.

[0015] By combining the robustness optimization constraints of the magnetic circuit constructed by simple complex shape with the sub-mode gain model, it is ensured that local geometric modification operations do not damage the magnetic circuit function and manufacturing feasibility. The generated geometrically enhanced candidate solutions and the final anti-disturbance optimized rotor structure can weaken the dynamic multi-field positive feedback effect from the root, enabling the motor to maintain stable performance during long-term bumpy heavy-load operation, reducing downtime maintenance caused by efficiency degradation, extending the service life of motor components, and meeting the core requirements of forklifts for reliability and endurance. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 This is a flowchart of an integrated collaborative optimization design method for the rotor magnetic circuit of a permanent magnet synchronous motor according to the present invention; Figure 2 This is a flowchart of the sliding window embedding method in an embodiment of the present invention; Figure 3 This is a schematic flowchart illustrating the transient simulation of structure-electromagnetic-thermal multi-field coupling in an embodiment of the present invention; Figure 4 This is a functional block diagram of a permanent magnet synchronous motor rotor magnetic circuit integrated collaborative optimization design device according to the present invention. Detailed Implementation

[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] Example 1: Please see Figure 1 As shown, this embodiment provides a method for integrated collaborative optimization design of the rotor magnetic circuit of a permanent magnet synchronous motor, including: S1: Collect multi-source vibration and electromagnetic load time-series data under typical bumpy and heavy-load conditions of forklifts to construct a dynamic excitation phase space mesh for the rotor interface. This step focuses on the dynamic excitation modeling requirements of the forklift motor rotor interface. First, it performs multi-source data acquisition and high-dimensional phase space construction. The core objective is to address the problem of traditional excitation models losing "spatiotemporal correlation information," providing a physically consistent input foundation for subsequent multi-field coupled simulations. Specifically, the steps for acquiring data and constructing the phase space mesh are as follows: S11: Collect multi-source vibration time-series data and electromagnetic load time-series data under typical bumpy and heavy-load conditions of forklifts to obtain raw multi-source vibration time-series data and raw electromagnetic load time-series data.

[0020] The multi-source vibration time-series data refers to the vibration acceleration data of key parts of the motor under bumpy and heavy-load conditions. Specifically, it includes the vibration acceleration at the bearings at both ends of the rotor, the area of ​​the stator housing near the rotor, and the motor output shaft. This data is collected by deploying triaxial vibration sensors at these locations. The collected data consists of continuous instantaneous vibration acceleration values, and the sampling frequency is determined by analyzing the dominant vibration frequency under bumpy conditions of the forklift to ensure complete capture of transient vibration peaks. The electromagnetic load time-series data refers to the instantaneous current and terminal voltage data of the motor during heavy-load climbing. This data is acquired through the signal acquisition interface of the motor controller. The collected data includes the instantaneous values ​​of the three-phase stator current, the instantaneous values ​​of the rotor-side excitation current, and the instantaneous values ​​of the stator terminal voltage. The acquisition frequency is consistent with that of the multi-source vibration time-series data to ensure the temporal correlation between the two types of data.

[0021] S12: Based on the original multi-source vibration time series data and the original electromagnetic load time series data generated in S11, perform time-frequency alignment processing to obtain time-frequency aligned multi-source vibration-electromagnetic load collaborative time series data.

[0022] Specifically, to eliminate the time difference caused by the acquisition delay of different sensors and ensure that the transient coupling relationship between vibration and electromagnetic load is not disrupted, the time-frequency alignment process is as follows: S121: Extract the timestamp information of the original multi-source vibration time series data and the original electromagnetic load time series data during acquisition, and select the timestamp with the highest accuracy between the two types of data as the reference timestamp.

[0023] S122: Compare the deviations between the non-reference timestamp and the reference timestamp, and use linear interpolation to calibrate the data segments with large time deviations so that the calibrated data points correspond one-to-one with the reference timestamps.

[0024] S123: Integrate the calibrated multi-source vibration time-series data with the electromagnetic load time-series data to form time-frequency aligned multi-source vibration-electromagnetic load collaborative time-series data.

[0025] S13: Based on the time-frequency aligned multi-source vibration-electromagnetic load collaborative time series data generated in S12, a high-dimensional mapping operation is performed using the sliding window embedding method to generate phase space trajectory points.

[0026] The sliding window embedding method is a method for mapping one-dimensional time series data into high-dimensional phase space trajectories. The sliding window embedding method process is described in [reference needed]. Figure 2 Specifically, the process of the high-dimensional mapping operation is as follows: S131: Determine the sliding window size. The window size is determined based on the sampling period of the time-frequency aligned multi-source vibration-electromagnetic load coordinated timing data and the characteristic duration of the forklift bumpy heavy load condition, ensuring that each window contains complete single transient excitation characteristics.

[0027] S132: Determine the embedding dimension. The false nearest neighbor (FNN) method is used to calculate the embedding dimension. By gradually increasing the dimension and statistically analyzing the proportion of false nearest neighbors, the dimension at which the proportion falls below a preset threshold is the final embedding dimension. The false nearest neighbor (FNN) method is a method used to detect whether the embedding dimension is appropriate, and it has wide applications in time series analysis, signal processing, chaos theory, and nonlinear system dynamics. The preset threshold is determined comprehensively based on the noise level and excitation characteristic stability of the multi-source vibration-electromagnetic load coordinated time series data under typical bumpy heavy-load conditions of forklifts, as well as the requirement that subsequent phase space trajectory points can accurately reflect the dynamic excitation evolution law of the rotor interface.

[0028] S133: Generate phase space trajectory points. Expand the time-frequency aligned multi-source vibration-electromagnetic load collaborative time series data in each sliding window according to the final embedding dimension to form a trajectory point in a high-dimensional space. All trajectory points together constitute the excitation trajectory in phase space, thus obtaining the phase space trajectory points.

[0029] S14: Based on the phase space trajectory points generated in S13, perform phase space adaptive subdivision operation to construct the dynamic excitation phase space mesh of the rotor interface; please refer to the sliding window embedding method process. Figure 2 .

[0030] The rotor interface dynamic excitation phase space grid refers to a non-uniform grid structure divided according to the correlation between excitation dynamic characteristics and rotor spatial orientation, using a high-dimensional phase space as a carrier. Its core components include a grid cell set, an excitation intensity statistics module, and a spatial mapping label module. The grid cell set is an independent region unit formed by partitioning the phase space according to local curvature and energy density thresholds. Each cell corresponds to a continuous excitation evolution region in the phase space. The excitation intensity statistics module stores the mean, variance, and peak value of the excitation intensity of all trajectory points within each grid cell, quantifying the strength and fluctuation characteristics of the excitation in that region. The spatial mapping label module assigns a unique rotor circumferential direction position identifier to each grid cell, establishing a geometric correspondence between phase space excitation and rotor mechanical structure (e.g., associating high-frequency vibration excitation regions with the vicinity of rotor bearings, and associating high-current load regions with the corresponding rotor magnetic pole positions), thus achieving precise binding of excitation information to the rotor interface. Specifically, the phase space adaptive partitioning operation process is as follows: S141: Calculate the local curvature and energy density of the phase space trajectory points. The local curvature is obtained by fitting a quadratic curve to adjacent phase space trajectory points and then calculating it according to the curve curvature formula. It is used to characterize the degree of curvature of the trajectory in this region. The higher the degree of curvature, the more intense the excitation change in this region. The energy density is obtained by counting the number of phase space trajectory points per unit volume in the phase space and combining it with the excitation intensity corresponding to each phase space trajectory point (which is calculated by weighting the vibration acceleration amplitude and electromagnetic load amplitude at that point). The higher the energy density, the stronger the excitation effect in this region.

[0031] S142: Determine the partitioning threshold. Based on the typical excitation intensity range and phase space trajectory distribution characteristics under the bumpy heavy load condition of the forklift, set the local curvature threshold and energy density threshold. When the local curvature or energy density of a certain area exceeds the corresponding threshold, the area needs to be densified.

[0032] S143: Perform meshing and record mesh information. Divide the phase space into several mesh units according to the meshing threshold. Record the corresponding excitation intensity statistics (mean, variance, peak value) and spatial mapping label (rotor circumferential direction position identifier) ​​for each mesh unit. Integrate all mesh units to obtain the dynamic excitation phase space mesh of the rotor interface.

[0033] This step addresses the core shortcomings of traditional excitation modeling methods by constructing a dynamic excitation phase space mesh for the rotor interface. Traditional methods only use the root mean square value of vibration or the peak value of the electromagnetic load spectrum as excitation input, losing the transient coupling relationship of the excitation in the time dimension and the orientation characteristics in the spatial dimension. This results in subsequent simulations being unable to reproduce the disturbance mode of the rotor interface under real working conditions. In contrast, the dynamic excitation phase space mesh for the rotor interface generated in this step fully encodes the dynamic complexity and spatial correlation of multi-source excitation in a topological form. The "excitation-orientation" coupling information carried by each mesh element ensures that the excitation loading is completely matched with the actual force position and force sequence of the rotor during subsequent multi-field coupled simulations. This provides a physically consistent input basis for accurately simulating the positive feedback process of "vibration-gap-loss-heat". At the same time, the adaptive meshing method avoids the problem of insufficient accuracy in key areas caused by uniform meshing and prevents the waste of computing power caused by excessive meshing. It ensures both modeling accuracy and the efficiency of subsequent simulations, providing the primary guarantee for the accuracy and feasibility of the entire optimization design process.

[0034] S2: Based on the dynamic excitation phase space grid of the rotor interface, drive the parameterized rotor model to perform multi-field coupled response simulation and generate a simple complex structure with eddy current coupling.

[0035] This step focuses on the needs of rotor dynamic multi-field coupling response analysis and topology risk characterization. First, it calls upon the rotor interface dynamic excitation phase space mesh generated by S1, and then performs multi-field coupling simulation and topology construction operations in conjunction with the parameterized rotor model. The core objective is to overcome the shortcomings of existing technologies, such as "dual-field static coupling, lack of topological association, and missing temporal evolution," and generate a structural eddy current coupled simple complex that can accurately characterize the dynamic risk state of the rotor magnetic circuit, providing a topological constraint basis for subsequent magnetic circuit robustness optimization. Specifically, the specific steps of the driving simulation and topology generation are as follows: S21: Call the parameterized rotor model to provide a basic simulation platform for subsequent multi-field coupled simulations.

[0036] The parametric rotor model refers to a three-dimensional parametric simulation model constructed using finite element modeling tools. This model uses the geometric parameters (magnet dimensions, magnetic bridge thickness, magnet-laminate contact gap, rotor yoke dimensions) and material parameters (permeability and remanence of permanent magnets, elastic modulus and permeability of laminations) of the forklift permanent magnet synchronous motor rotor as adjustable variables. The range of geometric parameter values ​​is determined based on the forklift motor's installation space constraints and power density requirements. The material parameter values ​​are obtained from material supplier testing reports and include characteristic data on material properties changing with temperature (such as the decay coefficient of permanent magnet remanence with temperature) to support multiphysics coupling calculations.

[0037] S22: Based on the rotor interface dynamic excitation phase space mesh generated by S1 and the parameterized rotor model generated by S21, perform the excitation data loading operation to obtain the loaded parameterized rotor model.

[0038] Specifically, to achieve accurate spatial and temporal matching between the excitation data and the rotor model, the excitation data loading process is as follows: S221: Establish the spatial mapping relationship between the dynamic excitation phase space grid of the rotor interface and the parameterized rotor model. Based on the spatial mapping label (corresponding to the rotor circumferential direction position) carried by each grid cell in the dynamic excitation phase space grid of the rotor interface, determine the orientation of the excitation data of the cell in the parameterized rotor model.

[0039] S222: Extract the excitation intensity statistics (mean, variance, and peak value) of each element in the dynamic excitation phase space grid of the rotor interface, and perform dimensional conversion in combination with the rated parameters of the forklift motor (rated voltage and rated speed) to convert the vibration excitation into an acceleration load and the electromagnetic load excitation into a current density load.

[0040] S223: Based on the temporal characteristics of the excitation data in the dynamic excitation phase space grid of the rotor interface, the converted load is applied to the corresponding orientation of the parameterized rotor model in a transient form to obtain the parameterized rotor model after loading.

[0041] S23: Based on the parameterized rotor model generated in S22, perform a structural-electromagnetic-thermal multi-field coupled transient simulation to generate the micro-deformation field, local eddy current density field, and contact interface temperature field at the magnet-laminate contact interface. For the structural-electromagnetic-thermal multi-field coupled transient simulation process, please refer to [link / reference]. Figure 3 .

[0042] The aforementioned structure-electromagnetic-thermal multi-field coupled transient simulation refers to a simulation method that uses a coupling algorithm to achieve real-time interaction of structural field, electromagnetic field, and thermal field data, simulating the coordinated response of a rotor under dynamic excitation across multiple physics fields. Specifically, to reproduce the positive feedback process of "vibration → gap deformation → increased eddy current loss → thermal expansion → further gap widening," the process of the multi-field coupled transient simulation is as follows: S231: Initialize simulation parameters, set simulation time step, convergence threshold, and boundary conditions. The simulation time step is determined based on the sampling frequency of the dynamic excitation phase space grid at the rotor interface to ensure the capture of transient excitation changes; the convergence threshold is determined through trial and error, and convergence is determined when the changes in structural displacement and electromagnetic field in adjacent iteration steps are both less than the threshold; boundary conditions include structural constraints (bearing constraints are set according to the actual installation method of the motor to limit the rotor's non-rotational displacement), electromagnetic constraints (stator winding current boundaries and air gap magnetic field transmission boundaries are set), and thermal constraints (rotor surface heat dissipation coefficient is set, determined based on the forklift motor cooling method).

[0043] S232: Calculate the structural field response. Based on the loaded parametric rotor model and vibration excitation, an explicit dynamic algorithm is used to calculate the displacement and stress fields of each rotor component within each time step. Displacement data of all discrete vertices at the magnet-laminated lamination contact interface are extracted to form the micro-deformation field of the magnet-laminated lamination contact interface. The micro-deformation field of the magnet-laminated lamination contact interface refers to a dynamic dataset containing spatial and temporal dimensions, recording the displacement change of each node along the interface normal and tangent directions, with each vertex of the magnet-laminated lamination contact interface as a data node. Its spatial resolution is consistent with the mesh accuracy of the parametric rotor model, and its temporal resolution is consistent with the simulation time step.

[0044] S233: Feedback the micro-deformation field of the magnet-laminated lamination contact interface to the electromagnetic field module, update the air gap size of the parameterized rotor model (the air gap size is adjusted in real time with the micro-deformation of the interface), and calculate the initial temperature distribution of the rotor based on the frictional heat generated by structural vibration, correct the material magnetic property parameters of the permanent magnet and lamination (such as the change of the magnetic permeability of the permanent magnet with temperature), and obtain the updated parameterized rotor model.

[0045] S234: Calculate the electromagnetic field response. Based on the updated parameterized rotor model and electromagnetic load excitation, the time-step finite element method is used to calculate the magnetic flux density distribution and eddy current density distribution of the rotor magnetic circuit within each time step. Eddy current data at the vertices of the magnet-laminated contact interface are extracted to form a local eddy current density field. The local eddy current density field refers to a dynamic dataset containing spatial and temporal information, with each vertices of the magnet-laminated contact interface as a data node, recording the magnitude and direction of the eddy current density at each node. Its resolution is consistent with the micro-deformation field of the magnet-laminated contact interface.

[0046] S235: Calculate the thermal field response. Based on the local eddy current density field, calculate the heat generated by eddy current losses (using the eddy current loss power density formula combined with material resistivity). Superimpose the heat generated by structural vibration and friction. Use the finite volume method to calculate the rotor temperature distribution within each time step. Extract the temperature data at the vertices of the magnet-laminated contact interface to form the contact interface temperature field. The contact interface temperature field refers to a dynamic dataset containing spatial and temporal information, with the vertices of the magnet-laminated contact interface as data nodes, recording the temperature change at each node. Its resolution is consistent with the local eddy current density field.

[0047] S236: Determine if the simulation is complete. If the current simulation time has not reached the preset duration (the preset duration is determined based on the typical continuous operation time of a forklift during bumpy, heavy-load, uphill driving), return to S232 to repeat the multi-field coupling calculation. If the preset duration has been reached, output the micro-deformation field, local eddy current density field, and contact interface temperature field of the magnet-laminated lamination contact interface. Please refer to the schematic flowchart of the structure-electromagnetic-thermal multi-field coupling transient simulation for details. Figure 3 .

[0048] S24: Based on the micro-deformation field, local eddy current density field, and contact interface temperature field generated in S23, perform a triangulation operation on the contact interface to obtain a triangulated set of the magnet-stamp contact interface.

[0049] The triangulation set of the magnet-stamping interface refers to discretizing the magnet-stamping interface into several non-overlapping triangular facets. Each triangular facet contains three vertices (corresponding to discrete vertices of the interface) and three edges (connecting adjacent vertices), forming a geometric dataset containing a vertex set, an edge set, and a triangular facet set. The side length of the triangular facet is determined according to the resolution of the interface physical field data to ensure that the physical field changes uniformly within each facet.

[0050] S25: Based on the triangulation set of the magnet-laminated contact interface generated by S24, perform the structural eddy current coupling simple complex construction operation to obtain the structural eddy current coupling simple complex.

[0051] Specifically, to characterize the spatial correlation and risk level of risk regions through topological structure, the process of constructing the eddy current coupled simple complex structure is as follows: S251: Construct 0-simpliforms and calculate weights. The vertex set in the triangulation set of the magnet-laminated contact interface is taken as the 0-simpliform. Each 0-simpliform is associated with the corresponding micro-deformation amplitude (the maximum displacement value of the vertex is extracted from the micro-deformation field of the magnet-laminated contact interface), eddy current concentration (the ratio of the eddy current density value of the vertex to the 90th quantile of the global eddy current density of the interface is extracted from the local eddy current density field), and temperature value (the temperature data of the vertex is extracted from the temperature field of the contact interface). Robust normalization is performed on the three physical quantities (each physical quantity value is divided by the global 95th quantile of the corresponding physical quantity). The weights are then summed according to preset weighting coefficients (determined based on forklift motor failure mode analysis) to obtain the weight value of each 0-simpliform.

[0052] S252: Construct a 1-simplicium and calculate its weights. For any two 0-simplicium, calculate their scalar similarity in terms of micro-deformation amplitude, eddy current concentration, and temperature (using the Gaussian kernel similarity formula, normalized by absolute difference and global standard deviation, and then input into the Gaussian function); take the average of the three similarities, and if it is greater than a preset similarity threshold, connect the two to form a 1-simplicium; the weight of the 1-simplicium is the maximum of the weights of the two associated 0-simplicium.

[0053] S253: Construct a 2-simplicium and calculate its weights. For any three 0-simplicium, if a 1-simplicium already exists between each pair of the three and the variance of the physical quantities within the resulting triangular facet is less than a preset variance threshold, then the triangular facet is considered a 2-simplicium; the weight of the 2-simplicium is the average of the weights of the three associated 0-simplicium.

[0054] S254: Integrate to form a structural eddy current coupled simplex complex. Integrate all 0-simulacra, 1-simulacra, and 2-simulacra according to topological relationships (such as the 0-simulacra identifier associated with 1-simulacra) to form a complete topological structure containing the geometric information, weight values, and topological relationships of each simulacra. Notably, 3-simulacra are not constructed because the magnet-laminated lamination contact interface is a two-dimensional curved surface without 3D volumetric mesh support and strictly satisfies the surface closure condition of a simplex complex.

[0055] S26: Based on the structural eddy current coupled simplex generated in S25, perform temporal complex sequence construction and continuous cohomology analysis to obtain a structural eddy current coupled simplex containing time evolution characteristics.

[0056] Specifically, to capture the temporal evolution patterns of high-risk areas, the time series analysis and integration process is as follows: S261: Divide the full duration of the multi-field coupling simulation into several time windows (the length is determined according to the rate of change of the physical field). Repeat the construction method of S25 for each time window to generate the structural eddy current coupling simple complex for the corresponding time window. Arrange all complexes in chronological order to form a temporal complex sequence.

[0057] S262: The continuous homology analysis method is used to track the birth time, death time and duration of high-weight simplexes to obtain the evolution law of high-risk areas.

[0058] S263: Embed the evolution law of high-risk areas into the topological correlation information of the structural eddy current coupled simple complex generated in S25 to obtain a structural eddy current coupled simple complex containing time evolution characteristics.

[0059] This step addresses the core shortcomings of existing technologies in dynamic multi-field analysis and topological characterization through multi-field coupled simulation and standardized simple complex construction. From the perspective of multi-field coupling, existing methods can only achieve static coupling between two physical fields, and the lack of thermal field feedback leads to a lack of basis for material parameter correction. However, the structure-electromagnetic-thermal closed-loop coupling constructed in this step can realistically reproduce the positive feedback process of "vibration → gap deformation → increased eddy current loss → thermal expansion → gap expansion", ensuring that the physical field data is consistent with the actual working conditions. From the perspective of topological representation, the scalar field or hotspot list output by traditional methods cannot reflect the spatial correlation of risk areas. However, this step constructs a simple complex "based on triangular mesh and physical constraints", and accurately quantifies the similarity of physical quantities through scalar Gaussian kernel similarity. Robust normalization maintains the weight discrimination, so that the complex can not only characterize the location of high-risk areas, but also reflect the degree of risk and spatial connectivity. From the perspective of time, existing methods ignore the temporal evolution of dynamic processes, while the temporal complex sequence and continuous cohomology analysis in this step can capture the changing pattern of high-risk areas over time, providing complete information on risk evolution for subsequent optimization. These improvements enable the simple complex structure with eddy current coupling to accurately characterize the dynamic risk state of the rotor magnetic circuit, while also possessing mathematical rigor and engineering practicality. This provides a reliable topological constraint basis for subsequent magnetic circuit robustness optimization and directly addresses the efficiency degradation and reliability issues caused by dynamic multi-field positive feedback in the background technology, providing an accurate risk identification tool.

[0060] For example, when a forklift motor experiences bumpy heavy loads, a certain edge region at the contact interface between the magnet and the lamination exhibits large deformation, high eddy currents, and high temperature. Simulation S23 extracts the three-field data of the vertices in this region. S24 divides the region into triangular facets. In S25, the high similarity of physical quantities of the vertices in this region forms a high-weight 1-simulacra, and the small variance of the triangular facets forms a high-weight 2-simulacra. Time series analysis in S26 reveals that these high-weight simplexes have a long duration and are marked as long-term high-risk areas. Finally, the eddy current coupling simple complex structure identifies this region as the focus of optimization, providing topological constraints for subsequent adjustments to the thickness of the magnetic bridge and optimization of the magnet gap.

[0061] S3: Based on the structural eddy current coupled simple complex, a combined constraint and sub-mode gain model for magnetic circuit robustness optimization is constructed, and a set of geometrically enhanced candidate solutions is generated. This step focuses on the constraint construction, model design, and candidate solution generation requirements for rotor magnetic circuit robustness optimization. First, it utilizes the eddy current coupled simplex structure generated by S2. Through high-risk region identification, constraint definition, model construction, and search algorithms, it generates a geometrically reinforced candidate solution set with global synergy and engineering feasibility. The core objective is to address the shortcomings of traditional optimization methods, such as "local optimization, lack of constraints, and low efficiency," providing a precise optimization direction for subsequent generation of disturbance-resistant rotor structures. Specifically, the steps for constructing constraints, modeling, and generating the candidate solution set are as follows: S31: Based on the structural eddy current coupled simple complex generated by S2, perform high-weight simplex identification and clustering operations to generate a set of vulnerable regions.

[0062] The vulnerable region set refers to a set of risk regions formed by clustering simplexes (including 0-similarity, 1-similarity, and 2-similarity) with weight values ​​exceeding a preset risk threshold from structural eddy current coupled simplexes, based on their spatial location relationships. Specifically, to accurately locate high-risk core regions, the process of high-weight simplex identification and clustering is as follows: S311: The weight values ​​of all simplexes in the statistical structure eddy current coupled simplex are used as the global 90th percentile of the weight values ​​as the preset risk threshold. The set of high-weight simplexes with weight values ​​exceeding the threshold is selected to ensure that the selection results are simplexes with a risk level significantly higher than the average level.

[0063] S312: Density clustering algorithm is used to cluster high-weight simplex sets. The spatial coordinates of the simplex are used as input. High-weight simplexes with a spatial distance less than the preset clustering radius (the clustering radius is determined according to the rotor magnetic circuit geometry and simplex dispersion) are grouped into the same risk region.

[0064] S313: Record the monomorphic identifier, spatial range, and average weight value (denoted as the risk weight of the region) for each risk region, and integrate all risk regions to obtain the vulnerable region set.

[0065] S32: Based on the vulnerable region set generated by S31 and the magnetic circuit design rules of the permanent magnet synchronous motor for forklifts, execute the local geometry modification operation definition operation to generate a set of local geometry modification operations.

[0066] The set of local geometric modification operations refers to a collection of geometric adjustment operations designed to enhance structural robustness and suppress multi-field coupling risks, tailored to the characteristics of different risk regions within the vulnerable region set. Specifically, to ensure the relevance and feasibility of the operations, the flow of the local geometric modification operations is defined as follows: S321: Analyze the associated components of each risk area within the vulnerable area set. For example, risk areas directly associated with magnets correspond to magnet size adjustment operations, risk areas associated with magnetic isolation bridges correspond to magnetic isolation bridge thickness adjustment operations, and risk areas associated with contact interfaces correspond to local support structure addition operations. Clarify the modification target of each operation.

[0067] S322: Combining magnetic circuit design rules (such as the minimum thickness of the magnetic bridge must meet structural strength requirements, and the maximum curvature of the magnetic isolation groove must avoid magnetic flux concentration) and manufacturing process limitations (such as the allowable dimensional adjustment range for processing accuracy), set the modification location (the spatial coordinate range of the corresponding risk area), modification type (such as thickening, thinning, bending, adding protrusions), and modification range (such as the magnetic bridge thickness adjustment range not exceeding 30% of the original thickness) for each operation, and at the same time clarify the vulnerable area subset that the operation can cover (denoted as the operation coverage area C(a)).

[0068] S323: Each operation is uniquely encoded, and the correspondence between the operation and the coverage area and the dependency relationship between operations are recorded (e.g., operation B must be executed after operation A is completed to avoid geometric conflicts). All operations are integrated to obtain a set of local geometric modification operations containing five core attributes: encoding, modified object, parameter range, coverage area C(a), and dependency relationship.

[0069] S33: Based on the set of local geometric modification operations generated by S32 and the performance requirements of the rotor magnetic circuit, perform a combined constraint construction operation to generate a combined constraint for magnetic circuit robustness optimization.

[0070] The combined constraints refer to a set of multi-dimensional constraints set to ensure that local geometric modifications do not compromise magnetic circuit function, structural strength, and manufacturing feasibility. These constraints include geometric constraints, electromagnetic constraints, and process constraints. Specifically, to construct a comprehensive and verifiable constraint system, the process for building the combined constraints is as follows: S331: Construct geometric constraints. Set constraint conditions for geometric conflicts that may be caused by local geometric modification operations, such as "the spatial influence range of any two modification operations does not overlap" and "the change in the contact gap between the magnet and the lamination after modification does not exceed 50% of the original gap". The constraint threshold is determined by rotor three-dimensional assembly space analysis and geometric interference detection.

[0071] S332: Construct electromagnetic constraints. Based on the core indicators of magnetic circuit design (such as the sinusoidal nature of air gap magnetic flux density and the harmonic content of back EMF), set constraint conditions, such as "the distortion rate of air gap magnetic flux density of the modified rotor magnetic circuit does not exceed 120% of that before the modification" and "the eddy current loss suppression rate is not lower than the preset target value". The constraint quantification is verified through electromagnetic finite element pre-simulation.

[0072] S333: Construct process constraints. Combine the technical limitations of the motor manufacturing process to set constraint conditions, such as "the minimum wall thickness of the modified lamination local structure shall not be less than the minimum value allowed by the manufacturing process" and "the modified dimensions of the magnet embedding groove shall match the magnet tolerance". The constraint parameters are determined by consulting the manufacturing department or referring to industry process standards.

[0073] S334: Geometric constraints, electromagnetic constraints, and process constraints are sorted in order of priority: geometric constraints first, electromagnetic constraints as the core, and process constraints as the bottom line, and integrated to form a combined constraint for magnetic circuit robustness optimization.

[0074] S34: Based on the vulnerable region set generated by S31, the local geometric modification operation set generated by S32, and the combined constraints generated by S33, perform the sub-mode gain model construction operation to generate a sub-mode gain model with optimized magnetic circuit robustness.

[0075] The submodulus gain model refers to a coverage function model that takes a subset of local geometric modification operations as input, outputs the total risk weight of the vulnerable regions covered by that subset, and satisfies the characteristic of diminishing marginal gain. Specifically, to construct a mathematically rigorous optimization model, the process for constructing the submodulus gain model is as follows: S341: Define the input and output of the robust gain function. The input is any subset (denoted as S) of the set of local geometric modification operations, and the output is the robust gain value (denoted as f(S)) corresponding to that subset.

[0076] S342: Determine the method for calculating the gain value. First, verify whether each operation in the subset S satisfies the combination constraint. Operations that satisfy the constraint are "valid operations" and those that do not are "invalid operations". Extract the coverage area C(a) of all valid operations and take the union of the coverage areas (denoted as U(S), which is the set of vulnerable areas covered by at least one valid operation). Calculate the sum of the risk weights of all vulnerable areas in U(S), and this sum is the robustness gain value f(S).

[0077] S343: Prove the submodularity of the function based on the mathematical property of covering functions: For any subset of operations A⊂B and any operation a not in B, if a is a valid operation and the region covered by A, U(A)⊂U(B), then the marginal gain of a added to A is ≥ the marginal gain of a added to B; if a is an invalid operation, the marginal gain is 0. Therefore, the function f(S) naturally satisfies the submodularity and does not require sampling verification.

[0078] S344: Set a gain threshold. Take a preset proportion (e.g., 70%) of the total risk weight of the vulnerable region set (the sum of the risk weights of all vulnerable regions) as the gain threshold. This is used to terminate the subsequent greedy search process, balance gain and computational efficiency, and finally obtain the sub-mode gain model with optimized magnetic circuit robustness.

[0079] S35: Based on the sub-mode gain model generated by S34 and the combined constraints generated by S33, a greedy search strategy is used to perform candidate solution generation operations to generate a geometrically enhanced candidate solution set.

[0080] The greedy search strategy refers to an optimized search method that iteratively constructs a subset of operations by using the core logic of "selecting the operation with the maximum current gain that satisfies the combinatorial constraints each time". Specifically, to efficiently generate feasible candidate solutions, the process for generating candidate solutions is as follows: S351: Initialize parameters. Set the initial operation subset S0 to be an empty set and the geometric reinforcement candidate solution set to be an empty set. Record the current iteration number (initially 0).

[0081] S352: Filter feasible operations, exclude operations already in S0 from the set of local geometric modification operations, verify whether the remaining operations satisfy the combination constraints, and obtain a list of feasible operations.

[0082] S353: Calculate the marginal gain. For each operation a in the list of feasible operations, calculate the marginal gain (f(S0∪{a})-f(S0)) after adding it to S0, and select the operation a_max with the largest marginal gain.

[0083] S354: Update the operation subset, add a_max to S0, and obtain a new operation subset S1=S0∪{a_max}, increment the iteration count by 1.

[0084] S355: Determine if the iteration terminates. If the robustness gain value f(S1) of S1 reaches the gain threshold, the feasible operation list is empty, or the number of iterations reaches the preset upper limit, then stop the iteration and encode S1 into a parameterized feature tree; otherwise, return to S352 and repeat the execution.

[0085] S356: Generate a candidate solution set. Repeat S351 to S355, adjusting the initial parameters (such as the gain threshold ratio and the order of feasible operations) in each iteration, and generating multiple parameterized feature trees (containing the encoding, position, type, magnitude and dependency of the modification operation). All parameterized feature trees together constitute the geometric enhancement candidate solution set.

[0086] This step addresses the core shortcomings of traditional optimization methods in improving magnetic circuit robustness through the collaborative design of combined constraints and sub-module gain models. From a global optimization perspective, traditional methods often employ local gradient optimization or stochastic search, which can easily lead to "fixing one area and damaging another" by neglecting the constraints between operations (e.g., thickening the magnetic bridge in a certain area may reduce the risk of deformation, but it may cause magnetic flux concentration and eddy current surge in adjacent areas). This step, however, constructs a "safety boundary" from geometric, electromagnetic, and technological perspectives based on combined constraints, ensuring that each modification does not compromise the core performance of the rotor. From the perspective of optimization efficiency and effectiveness, traditional stochastic heuristic methods (such as genetic algorithms) require numerous iterations to approach the optimal solution and struggle to guarantee global optimality. This step, based on the covering function model of the natural sub-module, theoretically achieves optimal performance through greedy search. The 1-1 / e approximation ratio of the global optimal solution (the classic greedy approximation ratio of the covering function) significantly reduces computational cost, adapting to the engineering efficiency requirements of forklift motor design. From a risk-specific perspective, traditional methods often target generalized objectives such as "minimum loss" and "maximum stiffness," neglecting the risks caused by dynamic multi-field positive feedback. This step, however, starts from the set of vulnerable regions extracted from the simple complex structure with eddy current coupling. All modifications directly target the key links in the "vibration-gap-eddy current-heat" positive feedback (such as strengthening contact interface support to suppress gap deformation and optimizing the magnetic bridge to reduce eddy current concentration), ensuring that the optimized structure weakens the positive feedback effect at its source. These improvements ensure that each solution in the geometrically strengthened candidate solution set possesses "global synergy, performance safety, and engineering feasibility," providing a precise and reliable optimization direction for the subsequent generation of disturbance-resistant rotor structures. This directly addresses the problem of "design schemes being out of sync with actual working conditions" in the background technology, laying the foundation for improving the reliability and endurance of forklift motors.

[0087] S4: Under the constraints of the geometrically reinforced candidate solution set, perform multi-objective collaborative convergence and 3D model generation, and output the disturbance-resistant optimized rotor structure. This step focuses on the instantiation and optimization screening requirements of the disturbance-resistant rotor structure. First, it calls upon the geometrically reinforced candidate solution set generated by S3. Through parametric modeling, multiphysics evaluation, multi-objective convergence, and structural mapping, it generates a disturbance-resistant optimized rotor structure that meets the operating conditions. The core objective is to overcome the shortcomings of traditional optimization methods, such as subjective manual screening and a disconnect between design and manufacturing, ensuring that the final structure possesses both performance advantages and engineering feasibility. Specifically, the steps for performing multi-objective collaborative convergence and generating the 3D model are as follows: S41: Based on the geometric enhancement candidate solution set generated by S3, perform parameterized feature tree instantiation operation to generate a candidate 3D model set.

[0088] Specifically, the parameterized feature tree instantiation refers to the process of converting the modification operation attributes (modification position coordinates, geometry type, size range, operation dependencies) recorded in the parameterized feature tree of the geometry enhancement candidate solution set into feature parameters of a 3D solid model through the parameterized interface of the 3D modeling software. Specifically, to convert abstract operation parameters into a solid model, the parameterized feature tree instantiation process is as follows: S411: Establish the coordinate mapping relationship between the parametric feature tree and the 3D modeling software, and convert the relative coordinates of the modified position in the parametric feature tree (local coordinate system with the rotor center as the origin) into the global coordinate system coordinates of the 3D modeling software to ensure that the modified position is accurately aligned with the spatial position of the rotor model.

[0089] S412: Analyze the operation dependencies of the parametric feature tree, and call the solid editing functions of the modeling software (such as extrusion, rotation, and chamfer) in sequence according to the order of "parent node operation is executed first and child node operation is executed later". Convert the modification type and size range in the parametric feature tree into the corresponding geometric operation parameters (such as the thickness parameter of the extrusion operation corresponding to "magnetic bridge thickening" and the curvature radius parameter of the rotation operation corresponding to "magnetic groove curvature adjustment").

[0090] S413: Verify the geometric integrity of the instantiated model, check whether all modification operations have successfully generated solid features (such as whether the support protrusion is seamlessly connected with the original structure, and whether the size of the magnet embedding slot matches the magnet parameters); if there are geometric conflicts (such as feature overlap, dimensional deviation), return to the corresponding parametric feature tree to adjust the modification range until a conflict-free 3D CAD model is generated; integrate all conflict-free models to obtain a candidate 3D model set.

[0091] S42: Based on the candidate 3D model set generated in S41, construct a multiphysics closed-loop evaluation loop, perform model performance evaluation operations, and generate a model performance dataset.

[0092] The multiphysics closed-loop evaluation circuit refers to an automated evaluation process that simulates the bumpy, heavy-load, and hill-climbing conditions of a forklift, performs structural-electromagnetic-thermal multi-field coupled simulation on candidate 3D models, and quantitatively evaluates the key performance indicators of the models. Specifically, to ensure that the evaluation is consistent with real-world conditions, the model performance evaluation process is as follows: S421: Load the turbulence excitation, call the rotor interface dynamic excitation phase space mesh generated in S1, and use the load mapping function of the simulation software to load the vibration and electromagnetic load excitation in the mesh to the corresponding position of the candidate three-dimensional model (consistent with the excitation loading position of the parameterized rotor model in S2).

[0093] S422: Calculate the eddy current-thermal-structural response using the S2 structural-electromagnetic-thermal multi-field coupled transient simulation method. Calculate the efficiency decay rate, eddy current concentration suppression rate, and structural dynamic stiffness of the candidate 3D model over the entire evaluation period (the evaluation period is consistent with the S2 simulation period). The efficiency decay rate refers to the ratio of the difference between the initial and final efficiency of the model within the evaluation period to the initial efficiency, used to characterize efficiency stability. The eddy current concentration suppression rate refers to the ratio of the difference between the maximum eddy current density of the model and the maximum eddy current density of the original rotor model before optimization to the maximum value of the original model, used to characterize the eddy current suppression effect. The structural dynamic stiffness refers to the ratio of the maximum deformation of key parts of the model (magnet-laminate contact interface, magnetic bridge) to the excitation load, used to characterize deformation resistance.

[0094] S423: Record performance data, associate the efficiency decay rate, eddy current concentration suppression rate, and structural dynamic stiffness numerical values ​​of each candidate 3D model with the model identifier, and integrate them to form a model performance dataset.

[0095] S43: Based on the model performance dataset generated by S42, a non-dominated sorting algorithm is used to perform multi-objective collaborative convergence operations to select the optimal parameterized feature tree.

[0096] The non-dominated ranking algorithm refers to a multi-objective optimization algorithm that extracts Pareto optimal solutions by hierarchically ranking candidate 3D models in the model performance dataset based on the dominance relationship of solutions. Specifically, to objectively select the optimal equilibrium model, the multi-objective collaborative convergence process is as follows: S431: Define the dominance relationship. For any two candidate models M1 and M2 in the model performance dataset, if the efficiency decay rate of M1 is less than or equal to the efficiency decay rate of M2, the eddy current concentration suppression rate is greater than or equal to the eddy current concentration suppression rate of M2, and the structural dynamic stiffness is greater than or equal to the structural dynamic stiffness of M2, and at least one of these indicators is strictly better, then it is determined that M1 dominates M2.

[0097] S432: Perform the first-level sorting, traverse the model performance dataset, and select all candidate models that are not dominated by other models to form the first non-dominated layer (Pareto optimal layer); if the number of models in the first non-dominated layer does not exceed the preset upper limit (determined according to engineering requirements, such as no more than 10), then directly use their corresponding parameterized feature trees as the candidate set; if it exceeds the upper limit, then proceed to S433.

[0098] S433: Calculate the crowding degree. For each model in the first non-dominated layer, calculate the sum of its distances to neighboring models in three performance index dimensions (the larger the distance, the more sparse the model is in and the better the diversity). Sort the models by crowding degree from largest to smallest, select the first preset number of models, and their corresponding parameterized feature trees form the candidate set.

[0099] S434: Verify optimality by comparing the performance data of the candidate models corresponding to the candidate set with the performance data of the original rotor model, and eliminating the parameterized feature trees that are not superior to the original model; if there are multiple remaining, filter by manufacturing complexity (select the one with fewer modification operations) to finally obtain one comprehensive optimal parameterized feature tree.

[0100] S44: Based on the optimal parameterized feature tree generated by S43, perform the original rotor model structure update operation and output the disturbance-resistant optimized rotor structure.

[0101] The original rotor model structure update refers to the process of precisely overlaying the modifications in the optimal parametric feature tree onto the original rotor model to update the geometric parameters and generate the final 3D model. Specifically, to ensure accurate and complete structure updates, the structure update process is as follows: S441: Import the original rotor model. Import the original 3D rotor model (unoptimized initial design model) of the permanent magnet synchronous motor for forklift into the 3D modeling software and coordinate it with the coordinate system of the optimal parameterized feature tree (with the rotor rotation axis as the Z-axis and the rotor end face center as the origin).

[0102] S442: Match the modified position. Based on the position coordinates of each modification operation in the optimal parametric feature tree, locate the corresponding modification area (such as the magnetic bridge position, the magnetic isolation slot position, and the contact interface between the magnet and the lamination) on the original rotor model. Ensure that the positioning deviation is less than the allowable range of modeling accuracy through coordinate comparison.

[0103] S443: Perform structural update, and make geometric adjustments to the corresponding areas of the original rotor model according to the modification operation type and magnitude of the optimal parameterized feature tree: for the magnetic bridge outline, modify the size of the magnet solid according to the thickness parameter; for the curvature of the magnetic isolation slot, adjust the arc outline according to the curvature parameter; for the local support structure, add support solids according to the protrusion size and position parameters.

[0104] S444: Verify model integrity. Use the geometric inspection tool of the modeling software to check whether there are geometric defects (such as missing surfaces or overlapping edges) in the updated original rotor model. Repair minor defects to ensure that the model can be used for subsequent engineering analysis (such as finite element simulation and mold design).

[0105] S445: Output disturbance-resistant optimized rotor structure. Save the updated original rotor model in a general three-dimensional format (such as "STEP format"). This model is the three-dimensional geometric model of the permanent magnet synchronous motor rotor with disturbance-resistant capability, denoted as the disturbance-resistant optimized rotor structure.

[0106] This step addresses the core shortcomings of traditional optimization processes—namely, reliance on manual performance selection, disconnect between design and operating conditions, and lack of assurance regarding engineering manufacturability—through multi-objective collaborative convergence and automated model generation. From an optimization closed-loop perspective, traditional methods require manual comparison of performance data after generating candidate solutions, easily leading to the omission of optimal solutions due to subjective experience. This step, however, utilizes a non-dominated sorting algorithm to achieve automated multi-objective convergence, objectively selecting solutions with optimal efficiency, stability, eddy current suppression, and structural stiffness balance. Furthermore, the selection process forms a logical closed loop with the preceding sub-mode gain model and multi-field coupled simulation, ensuring that the optimization direction always points to "suppressing dynamic multi-field positive feedback." From an engineering adaptability perspective, traditional optimization often suffers from the problem of "excellent simulation performance but unmanufacturable," while this step's optimization process uses a geometrically reinforced candidate solution set as a constraint (the candidate set already satisfies the engineering requirements). (Industry constraints), and when the optimal solution is mapped back to the original model, the geometric integrity needs to be verified to ensure that the final structure meets manufacturing requirements; from the perspective of problem solving, traditional methods cannot continuously optimize for the dynamic working conditions of forklifts under bumpy and heavy loads, while the multiphysics closed-loop evaluation in this step adopts the bumpy excitation consistent with the real working conditions, and the evaluation index is directly related to core pain points such as "efficiency decay" and "structural deformation resistance". The resulting anti-disturbance optimized rotor structure can fundamentally weaken the positive feedback loop of "vibration → gap deformation → eddy current increase → thermal expansion", so that the motor maintains stable performance during continuous operation and directly meets the reliability and endurance requirements of forklift-specific motors.

[0107] Example 2: This embodiment, based on Embodiment 1, provides an integrated collaborative optimization design device for the rotor magnetic circuit of a permanent magnet synchronous motor, such as... Figure 4 As shown, it includes: A device for integrated collaborative optimization design of rotor magnetic circuit of permanent magnet synchronous motor, used to implement the above-mentioned integrated collaborative optimization design method of rotor magnetic circuit of permanent magnet synchronous motor, the device comprising: Dynamic excitation phase space mesh construction module: used to collect multi-source vibration and electromagnetic load time series data under typical bumpy heavy load conditions of forklifts, and construct dynamic excitation phase space mesh of rotor interface; Multi-field coupling simulation and simple complex generation module: used to drive the parameterized rotor model to perform multi-field coupling response simulation based on the dynamic excitation phase space mesh of the rotor interface, and generate structural eddy current coupled simple complex; Combined constraint and candidate solution set generation module: used to construct a combined constraint and sub-mode gain model for magnetic circuit robustness optimization based on structural eddy current coupled simple complex, and generate a geometrically enhanced candidate solution set; Multi-objective convergence and optimized structure output module: used to perform multi-objective collaborative convergence and 3D model generation under geometric reinforcement candidate solution set constraints, and output disturbance-resistant optimized rotor structure.

Claims

1. A method for integrated collaborative optimization design of the rotor magnetic circuit of a permanent magnet synchronous motor, characterized in that, The method includes: S1: Collect multi-source vibration and electromagnetic load time-series data of forklift under typical bumpy heavy-load conditions, and construct a dynamic excitation phase space grid of rotor interface. S2: Based on the dynamic excitation phase space grid of the rotor interface, drive the parameterized rotor model to perform multi-field coupled response simulation and generate a simple complex structure with eddy current coupling. S3: Based on the simple complex of structural eddy current coupling, construct a combined constraint and sub-mode gain model for magnetic circuit robustness optimization, and generate a set of candidate solutions for geometric reinforcement; S4: Under the constraint of geometrically reinforced candidate solution set, perform multi-objective collaborative convergence and 3D model generation, and output disturbance-resistant optimized rotor structure.

2. The integrated collaborative optimization design method for the rotor magnetic circuit of a permanent magnet synchronous motor according to claim 1, characterized in that, The steps for constructing the dynamic excitation phase space mesh of the rotor interface include: S11: Collect multi-source vibration time-series data and electromagnetic load time-series data under typical bumpy and heavy-load conditions of forklifts to obtain raw multi-source vibration time-series data and raw electromagnetic load time-series data. S12: Based on the original multi-source vibration time series data and the original electromagnetic load time series data, perform time-frequency alignment processing to obtain time-frequency aligned multi-source vibration-electromagnetic load collaborative time series data; S13: Based on the time-frequency aligned multi-source vibration-electromagnetic load collaborative time series data, a high-dimensional mapping operation is performed using the sliding window embedding method to generate phase space trajectory points; S14: Based on the phase space trajectory points, perform phase space adaptive partitioning operation to construct a dynamic excitation phase space mesh for the rotor interface.

3. The integrated collaborative optimization design method for the rotor magnetic circuit of a permanent magnet synchronous motor according to claim 2, characterized in that, The steps for performing high-dimensional mapping operations using the sliding window embedding method include: S131: Determine the size of the sliding window, which is determined based on the sampling period of the time-frequency aligned multi-source vibration-electromagnetic load collaborative timing data and the characteristic duration of the forklift bumpy heavy load condition. S132: Determine the embedding dimension, which includes: calculating the embedding dimension using the false nearest neighbor method, gradually increasing the dimension and statistically analyzing the proportion of false nearest neighbors, and the dimension when the proportion drops below a preset threshold is the final embedding dimension. S133: Expand the time-frequency aligned multi-source vibration-electromagnetic load collaborative time series data within each sliding window according to the final embedding dimension to form a trajectory point in a high-dimensional space. All trajectory points together constitute the phase space trajectory point in the phase space.

4. The integrated collaborative optimization design method for the rotor magnetic circuit of a permanent magnet synchronous motor according to claim 1, characterized in that, The steps for generating a structural eddy current coupled simple complex by performing multi-field coupled response simulation on a parameterized rotor model include: S21: Call the parameterized rotor model to provide a basic simulation platform for subsequent multi-field coupling simulation; S22: Based on the dynamic excitation phase space grid and parameterized rotor model of the rotor interface, perform excitation data loading operation to obtain the loaded parameterized rotor model; S23: Based on the loaded parametric rotor model, perform a structural-electromagnetic-thermal multi-field coupled transient simulation operation to generate the micro-deformation field, local eddy current density field, and contact interface temperature field of the magnet-laminated lamination contact interface. S24: Based on the micro-deformation field, local eddy current density field and contact interface temperature field of the magnet-stamping interface, perform a triangulation operation on the contact interface to obtain a triangulation set of the magnet-stamping interface; S25: Based on the triangulation set of the magnet-stamping interface, perform the structural eddy current coupling simple complex construction operation to obtain the structural eddy current coupling simple complex; S26: Based on the structural eddy current coupled simple complex, perform temporal complex sequence construction and continuous cohomology analysis to obtain a structural eddy current coupled simple complex containing time evolution characteristics.

5. The integrated collaborative optimization design method for the rotor magnetic circuit of a permanent magnet synchronous motor according to claim 4, characterized in that, The steps for performing a transient simulation of a structure-electromagnetic-thermal multi-field coupling include: S231: Initialize simulation parameters, set simulation time step, convergence threshold and boundary conditions; S232: Based on the parameterized rotor model and vibration excitation after loading, an explicit dynamic algorithm is used to calculate the displacement field and stress field of each component of the rotor within each time step, extract the displacement data of all discrete vertices of the magnet-laminated contact interface, and form the micro-deformation field of the magnet-laminated contact interface. S233: Feedback the micro-deformation field of the magnet-laminated lamination contact interface to the electromagnetic field module, update the air gap size of the parameterized rotor model, and calculate the initial temperature distribution of the rotor based on the frictional heat generated by structural vibration, correct the material magnetic property parameters of the permanent magnet and lamination, and obtain the updated parameterized rotor model. S234: Based on the updated parametric rotor model and electromagnetic load excitation, the magnetic flux density distribution and eddy current density distribution of the rotor magnetic circuit within each time step are calculated using the time-step finite element method. Eddy current data at the apex of the magnet-laminated contact interface are extracted to form a local eddy current density field. S235: Based on the local eddy current density field, the heat generated by eddy current loss is calculated, and the heat generated by structural vibration and friction is superimposed. The temperature distribution of the rotor within each time step is calculated using the finite volume method. The temperature data of the vertex of the magnet-laminated contact interface is extracted to form the contact interface temperature field. S236: Determine whether the simulation is complete. If the current simulation time has not reached the preset duration, return to S232 to repeat the multi-field coupling calculation. If the preset duration has been reached, output the micro-deformation field of the magnet-laminated contact interface, the local eddy current density field, and the contact interface temperature field.

6. The integrated collaborative optimization design method for the rotor magnetic circuit of a permanent magnet synchronous motor according to claim 1, characterized in that, The steps for constructing a combined constraint and submode gain model for magnetic circuit robustness optimization and generating a set of geometrically reinforced candidate solutions include: S31: Based on structural eddy current coupling simple complex, perform high-weight simplex identification and clustering operations to generate a set of vulnerable regions; S32: Based on the vulnerable region set and magnetic circuit design rules, execute the local geometry modification operation definition operation and generate a set of local geometry modification operations; S33: Based on the set of local geometric modification operations and the performance requirements of the rotor magnetic circuit, perform a combined constraint construction operation to generate combined constraints for magnetic circuit robustness optimization; S34: Based on the set of vulnerable regions, the set of local geometric modification operations, and combined constraints, perform sub-mode gain model construction operations to generate a sub-mode gain model with optimized magnetic circuit robustness; S35: Based on the sub-modulus gain model and combined constraints, a greedy search strategy is used to perform candidate solution generation operations to generate a geometrically enhanced candidate solution set.

7. The integrated collaborative optimization design method for the rotor magnetic circuit of a permanent magnet synchronous motor according to claim 6, characterized in that, The steps for performing the sub-mode gain model construction operation include: S341: Define the input and output of the robustness gain function; specifically, the input is any subset of the set of local geometric modification operations, and the output is the robustness gain value corresponding to that subset; S342: Determine the method for calculating the gain value, which specifically includes: First, verify whether each operation in the subset satisfies the combination constraint. Operations that satisfy the constraint are "valid operations", and those that do not are "invalid operations". Extract the coverage area of ​​all valid operations and take the union of the coverage areas. Calculate the sum of the risk weights of all vulnerable areas in the union. This sum is the robustness gain value. S343: Determine if the function satisfies the submodularity property; S344: Set a gain threshold; specifically including: taking a preset proportion of the total risk weight of the vulnerable region set as the gain threshold; combining the input and output of the robustness gain function, the gain value calculation method, the determination function sub-modulus and the gain threshold to obtain a sub-modulus gain model for magnetic circuit robustness optimization.

8. The integrated collaborative optimization design method for the rotor magnetic circuit of a permanent magnet synchronous motor according to claim 1, characterized in that, The steps for performing multi-objective collaborative convergence and 3D model generation include: S41: Based on the geometrically enhanced candidate solution set, perform parameterized feature tree instantiation to generate a candidate 3D model set; S42: Based on the candidate 3D model set, construct a multiphysics closed-loop evaluation loop, perform model performance evaluation operations, and generate a model performance dataset; S43: Based on the model performance dataset, a non-dominated ranking algorithm is used to perform multi-objective collaborative convergence operations and select the optimal parameterized feature tree; S44: Based on the optimal parameterized feature tree, perform the original rotor model structure update operation and output the disturbance-resistant optimized rotor structure.

9. The integrated collaborative optimization design method for the rotor magnetic circuit of a permanent magnet synchronous motor according to claim 8, characterized in that, The steps for performing multi-objective collaborative convergence operations using a non-dominated sorting algorithm include: S431: Define the dominance relationship, which specifically includes: for any two candidate models M1 and M2 in the model performance dataset, if the efficiency decay rate of M1 is less than or equal to the efficiency decay rate of M2, the eddy current concentration suppression rate is greater than or equal to the eddy current concentration suppression rate of M2, the structural dynamic stiffness is greater than or equal to the structural dynamic stiffness of M2, and at least one of the indicators is strictly better, then it is determined that M1 dominates M2. S432: Traverse the model performance dataset and use the dominance relation to filter out all candidate models that are not dominated by other models to form the first non-dominated layer; if the number of models in the first non-dominated layer does not exceed the preset limit, then directly use their corresponding parameterized feature trees as the candidate set; if it exceeds the limit, then proceed to S433. S433: For each model in the first non-dominated layer, calculate the sum of its distances to neighboring models in three performance index dimensions and record it as the crowding degree; sort the models in descending order of crowding degree, select a preset upper limit number of models, and the corresponding parameterized feature trees constitute the candidate set; S434: Verify optimality, specifically including: comparing the performance data of the candidate models corresponding to the candidate set with the performance data of the original rotor model, and eliminating the parameterized feature trees that are not better than the original model; if there are multiple remaining, filter them by manufacturing complexity, and finally obtain one parameterized feature tree that is optimal in all aspects.

10. A device for integrated collaborative optimization design of rotor magnetic circuit of a permanent magnet synchronous motor, used to implement the integrated collaborative optimization design method of rotor magnetic circuit of a permanent magnet synchronous motor according to any one of claims 1-9, characterized in that, The device includes: Dynamic excitation phase space mesh construction module: used to collect multi-source vibration and electromagnetic load time series data under typical bumpy heavy load conditions of forklifts, and construct dynamic excitation phase space mesh of rotor interface; Multi-field coupling simulation and simple complex generation module: used to drive the parameterized rotor model to perform multi-field coupling response simulation based on the dynamic excitation phase space mesh of the rotor interface, and generate structural eddy current coupled simple complex; Combined constraint and candidate solution set generation module: used to construct a combined constraint and sub-mode gain model for magnetic circuit robustness optimization based on structural eddy current coupled simple complex, and generate a geometrically enhanced candidate solution set; Multi-objective convergence and optimized structure output module: used to perform multi-objective collaborative convergence and 3D model generation under geometric reinforcement candidate solution set constraints, and output disturbance-resistant optimized rotor structure.

Images