Vehicle permanent magnet motor cogging torque and torque ripple optimization method

Abstract

The invention relates to the field of motor structure design and optimization, and discloses a cogging torque and torque ripple optimization method for a permanent magnet motor for a vehicle. The method comprises the steps that based on an electromagnetic scheme, a geometric boundary and a term dictionary, a parameterized geometric prototype containing four symmetric semi-arc auxiliary grooves is constructed, and a design variable set structure is generated; converting the structure into a target evaluation interface structure containing a target item and a constraint item; an improved adaptive particle swarm algorithm is combined with finite element rapid check and a new iteration function to carry out optimization solution, and an updated candidate group structure is generated; and finally, optimal parameters are screened through Pareto sorting and robustness recheck, and an engineering configuration structure is generated. According to the method, through systematic parameter modeling, an efficient collaborative optimization algorithm and simulation evaluation, the optimization efficiency and reliability are remarkably improved, the cogging torque and the torque ripple of the motor are effectively reduced, and meanwhile, the process feasibility and the performance robustness of a design scheme are ensured.

Description

Technical Field

[0001] This invention relates to the field of motor structure design and optimization, and in particular to a method for optimizing the cogging torque and torque ripple of a permanent magnet motor for automobiles. Background Technology

[0002] New energy vehicles have high requirements for the noise of permanent magnet motors, and cogging torque and torque ripple are important indicators for judging whether the motor noise is serious. Therefore, achieving low cogging torque and torque ripple is of great significance for improving the competitiveness of automotive permanent magnet motors.

[0003] Cogging torque and torque ripple can be optimized by altering the combination of slot and pole numbers and changing the geometry of the stator and rotor. However, this method is based on changing the topology of the motor stator and rotor. Although it can optimize cogging torque and torque ripple, it will also affect other motor performance. Optimizing cogging torque and torque ripple through auxiliary slots can effectively avoid this problem. However, this method requires optimization of the auxiliary slot size based on a scanning method. This method relies on expert experience and is prone to getting trapped in local optima, making it impossible to find the optimal size.

[0004] Furthermore, in the field of motor structure design and optimization, existing solutions for optimizing the cogging torque and torque ripple of automotive permanent magnet motors typically rely on offline simulation and empirical scanning. These solutions consist of a fragmented pipeline of parameter tables, evaluation scripts, and manual rules, leading to limitations such as inconsistent parameter definitions, disconnect between constraints and evaluation, and decoupling between search and simulation. Existing methods often employ rule-based scanning or simple heuristic search to drive offline finite element verification, relying on manual selection and multiple rounds of trial and error. Under structural and electrical constraints, as well as manufacturing and assembly constraints, this approach is prone to task chain fragmentation and non-reproducible iterations, making it difficult to ensure the stable generation of updated candidate structures. Regarding the joint processing of target evaluation interface structure, new iterative function updates, and rapid finite element verification, existing technologies generally suffer from insufficient integration, asynchronous timing, and difficulty in refactoring error feedback in target item mapping, constraint injection, and interface transcribing. This makes it difficult to establish a continuous process of data acquisition, alignment, judgment, control, and recording in search and evaluation collaborative application scenarios, resulting in mutual obstruction between candidate parameter generation and evaluation, and difficulty in accumulating engineering configurations into reusable assets. Summary of the Invention

[0005] To address the aforementioned technical problems, this invention provides a method for optimizing the cogging torque and torque ripple of a permanent magnet motor for automobiles, comprising: The process involves acquiring electromagnetic schemes, geometric boundaries, and a terminology dictionary; performing input validation and conflict detection; constructing a parametric geometric prototype including symmetrical four-semi-circular arc auxiliary slots based on field naming and definitions from the terminology dictionary; performing hierarchical discretization and feasibility pre-judgment; performing variable interval transcription and linkage constraint configuration operations; and handling structured encapsulation and protocol declaration to generate a design variable set structure. The electromagnetic scheme includes the range of pole pairs, lamination thickness, core material grade, winding form and connection method, permanent magnet magnetization direction and magnetic circuit partitioning, stator and rotor assembly constraints, and thermal environment grade. The geometric boundaries include tooth width, tooth tip chamfer range, air gap range, rotor outer diameter, magnetic pole pitch, magnetic bridge thickness, and end fillet radius. The terminology dictionary includes symmetrical four-semi-circular arc auxiliary slots, a parametric geometric prototype, a parametric geometric model, a feature parameter discretization table, a variable interval library, linkage constraints, mesh-friendly markers, and a design variable set structure. Obtain the design variable set structure, perform target item mapping of cogging torque and torque ripple items, and inject and interface transcribing of constraint items such as average torque constraint, manufacturing constraint, assembly constraint and mesh stability constraint to generate the target evaluation interface structure; Obtain the target evaluation interface structure, perform improved adaptive particle swarm initialization and finite element fast verification with adaptive inertia weight and adaptive learning factor, and update the new iterative function with global guiding term, local mining term and nonlinear compression term to generate updated candidate population structure; The updated candidate population structure is obtained, and the optimal individual selection, consistency comparison, and parameter solidification are performed based on Pareto sorting and neighborhood robustness verification to generate the engineering configuration structure.

[0006] Furthermore, the symmetrical four-semi-circular auxiliary groove includes: The symmetrical four-semi-circular arc auxiliary groove is a groove formed by four continuously curvature circular arc segments, arranged symmetrically around the center line of the magnetic pole, and has a smooth transition at the end of the groove and the magnetic bridge area; its parameter set includes the arc radius group, groove depth, groove opening width, relative angle group and groove end transition fillet.

[0007] Furthermore, the process of generating the design variable set structure includes: The geometry generator is invoked to construct a parametric geometric prototype. The geometry generator adopts a parametric script mechanism driven by the computer-aided design CAD kernel. First, stator and rotor basic elements are formed by combining slot sketches, revolution bodies and extrusion bodies. Then, symmetrical four semi-circular arc auxiliary slots are arranged on the rotor surface. Mesh-friendly markers are generated. Corner smoothing, minimum transition radius constraints and minimum channel width constraints reduce the risk of element distortion caused by extremely fine features.

[0008] Furthermore, the process of generating the design variable set structure also includes: The core fields are read one by one from the parametric geometric prototype to construct a discrete table of feature parameters. The discrete strategy adopts a collaborative mechanism of hierarchical sampling and boundary expansion: hierarchical sampling establishes multi-level sampling steps for each field and covers a large range of combination space through inter-layer staggering; boundary expansion inserts high-density sampling points near the lower limit of process and assembly.

[0009] Furthermore, the process of generating the design variable set structure also includes: The geometric boundaries and process constraint templates are transcribed into interval expressions, and a variable interval library is established. To reduce the risk of implicit violations caused by cross-field linkage, a linkage constraint is introduced. The linkage constraint gives the coordination range based on the geometric relationship between the magnetic bridge and the slot depth, the slot width and the relative angle group, and the arc radius group and the transition fillet.

[0010] Furthermore, the improved adaptive particle swarm initialization and rapid finite element verification process, which performs adaptive inertia weights and adaptive learning factors, also includes: The system reads the upper and lower boundaries, step size, and linkage constraint summaries of each field from the input template, and generates an initial parameter set within the feasible interval using stratified uniform sampling. The adaptive inertia weight is driven by the population statistical structure, which calculates the diversity measure, the ratio of stagnant rounds to improved rounds in real time. Based on the statistical results, the system updates the inertia weight in segments between the relaxed and convergent intervals. The adaptive learning factor is triggered by the progress rate at the individual level.

[0011] Furthermore, the new iteration function includes: The new iterative function consists of three parts: a global guiding term, a local mining term, and a nonlinear compression term. The weights of the three are jointly adjusted by the adaptive inertial weight and the adaptive learning factor. The global guiding term uses the group's historically optimal parameter set for directional guidance, the local mining term uses the individual's historically optimal parameter set for neighborhood search, and the nonlinear compression term gradually reduces the step size as the group's diversity decreases.

[0012] Furthermore, the process of rapid verification using the finite element method also includes: For each individual, its parameter set is read, the geometry generator is called to construct the symmetrical four semi-circular auxiliary groove geometry on the rotor surface, and the mesh strategy is selected based on the mesh-friendly marker; the solver is driven to generate the magnetic field and electromagnetic force sequence, the index mapping table in the target evaluation interface structure is called to perform item-level statistics on the cogging torque and torque pulsation; and parameter fingerprints are jointly generated from the geometric fields and the mesh-friendly markers, and the evaluation cache is queried.

[0013] Furthermore, after generating and updating the candidate population structure, the optimal individual is selected based on Pareto ranking and neighborhood robustness verification.

[0014] Furthermore, the structured encapsulation generates a set of design variables that uses a serialized organization method that combines key-value pairs and hierarchical structures. The upper layer stores a list of field names and field attributes, while the lower layer stores feasible combinations, diagnostic label references, linkage constraint summaries, and mesh-friendly tag mappings, along with version numbers, source time, and engineering baseline fingerprints for traceability.

[0015] The key innovations of this invention include: (1) Construct a target evaluation interface structure, organize the target item mapping, constraint item injection and interface transcribing into a callable object, including input template, output template, call sequence, tracking relationship and task description sheet, and connect the design variable set structure and evaluation collaboration in the loop call.

[0016] (2) In the improved adaptive particle swarm, the co-rotational mutation perturbation trigger, the new iterative function update and elite retention are introduced, and the constraint receipt and mesh-friendly marker are introduced to participate in the update decision, forming a search and evaluation closed loop with the finite element fast verification.

[0017] (3) The parameter fingerprint is generated by combining geometric fields and grid-friendly markers, and is used in conjunction with evaluation cache, error channel and tracking relationship to support evaluation result recovery, retry and source tracing, and generate structured storage of updated candidate population structure according to convergence criteria.

[0018] The following are its main beneficial effects: (1) It applies to the target evaluation interface structure and evaluation collaboration link, unifies the input template, output template and calling sequence into the same object, reduces the discrepancy between target item mapping, constraint item injection and task generation, and maintains field alignment and process coherence in the search-driven batch evaluation scenario.

[0019] (2) The in-loop process of search and evaluation is driven by the new iterative function update and elite retention to steadily promote the evolution of the population under the judgment of the constraint list and mesh-friendly marker. Combined with the finite element fast verification feedback, invalid search is reduced. It is suitable for optimization scenarios with structural and electrical constraints.

[0020] (3) It is used to evaluate task and result management, reduce repeated solutions through parameter fingerprints and evaluation cache, locate the source of anomalies through error channels and tracking relationships and reuse existing receipts, so as to facilitate the accumulation of reusable records in the engineering configuration structure after generating and updating the candidate group structure. Attached Figure Description

[0021] Figure 1 This application provides an overall flowchart of a method for optimizing the low noise of a permanent magnet motor for vehicles. Figure 2 A flowchart illustrating a method for optimizing the cogging torque and torque ripple of a permanent magnet motor for vehicles, provided in an embodiment of this application; Figure 3 A schematic diagram illustrating the parameterized definition of a symmetrical four-semi-circular arc auxiliary groove provided in an embodiment of this application; Figure 4 A flowchart of an improved adaptive particle swarm optimization algorithm provided in this application embodiment; Figure 5 A schematic diagram of a test function and iteration curve provided for an embodiment of this application; Figure 6 A comparison chart showing the performance of an improved adaptive particle swarm optimization algorithm provided in this application in practical applications. Detailed Implementation

[0022] Figure 1 A flowchart illustrating a low-noise optimization method for automotive permanent magnet motors provided in this application is shown below. Figure 1 As shown, the process begins by constructing a parametric finite element model of the permanent magnet motor. This step corresponds to S100 in this patent, which involves constructing a parametric geometric prototype based on a terminology dictionary and geometric boundaries, laying the model foundation for subsequent optimization.

[0023] The process then moves to the stage of determining input variables and optimization objectives: based on the constructed finite element model, initial performance parameters (such as cogging torque, torque pulsation, etc.) without auxiliary slots are calculated; simultaneously, the value range of each input variable (such as auxiliary slot size parameters) is confirmed. This stage corresponds to the connection between S100 and S200 in this patent, completing the generation of the design variable set structure and the mapping and injection of objective and constraint terms.

[0024] The core of the process is to construct an augmented objective function that includes an objective function and a penalty function, and then optimize this augmented objective function based on an improved adaptive particle swarm optimization algorithm to find the optimal value. This core step completely corresponds to S300 of this patent, namely, performing improved adaptive particle swarm initialization, rapid finite element verification, and new iterative function update operations, and efficiently obtaining the optimized solution through the algorithm's adaptive mechanism and rapid simulation verification.

[0025] Based on the optimal value obtained through optimization, the optimal size parameters of the auxiliary slot can be derived. Finally, the finite element model is updated based on this optimal size to verify whether the various performance indicators (such as cogging torque, torque ripple, etc.) of the motor with the optimized auxiliary slot meet the expected targets. This final verification and output link corresponds to S400 of this patent, namely, the optimal individual selection and parameter solidification to generate the final engineering configuration structure.

[0026] This concludes the process. Figure 1 The patent clearly demonstrates the complete technical path from model construction, parameter initialization, core optimization to result verification, showcasing the advantages of the method's systematicity, automation, and efficiency.

[0027] Example 1, refer to Figure 2 This is a flowchart illustrating a method for optimizing the cogging torque and torque ripple of a permanent magnet motor for vehicles, provided by an embodiment of the present invention. The process may include at least steps S100-S400: S100: Obtain electromagnetic scheme, geometric boundary and terminology dictionary, perform input verification and conflict detection processing, construct parameterized geometric prototype including symmetrical four semi-circular arc auxiliary groove based on field naming and caliber of terminology dictionary, perform hierarchical discretization and feasibility pre-judgment, variable interval transcribing and linkage constraint configuration operation, structured encapsulation and protocol declaration processing, and generate design variable set structure. S200: Obtain the design variable set structure, perform target item mapping of cogging torque and torque pulsation items, and inject and interface-transfer processing of constraint items such as average torque constraints, manufacturing constraints, assembly constraints and mesh stability constraints to generate the target evaluation interface structure. S300: Obtain the target evaluation interface structure, perform improved adaptive particle swarm initialization and finite element fast verification with adaptive inertial weight and adaptive learning factor, and update the new iterative function containing global guiding terms, local mining terms and nonlinear compression terms to generate updated candidate population structure; S400: Obtain the updated candidate population structure, perform optimal individual screening, consistency comparison, and parameter solidification based on Pareto sorting and neighborhood robustness verification, and generate the engineering configuration structure.

[0028] S100: Obtain electromagnetic scheme, geometric boundary and terminology dictionary, perform input verification and conflict detection processing, construct parametric geometric prototype, perform hierarchical discretization and feasibility pre-judgment, variable interval transcription and linkage constraint configuration operation, structured encapsulation and protocol declaration processing, and generate design variable set structure; The electromagnetic scheme refers to the structural and power-on coordination configuration for automotive permanent magnet motors, including items such as pole pair range, lamination thickness range, core material grade, winding form and connection method, permanent magnet magnetization direction and magnetic circuit partitioning, stator and rotor assembly constraints and thermal environment level; the geometric boundary refers to the set of physical constraints on stator slot shape, tooth width, tooth tip chamfer range, air gap range, rotor outer diameter, magnetic pole pitch, magnetic bridge thickness and end fillet, and simultaneously records process items such as machining toolpath restrictions and minimum chamfer radius; the terminology dictionary refers to the specification list used to unify field naming and scope within the scope of this invention, covering terms such as symmetrical four-half-circular auxiliary slots, parametric geometric prototypes, parametric geometric models, feature parameter discrete tables, variable interval libraries, linkage constraint devices, mesh-friendly markers, and design variable set structures, and provides definitions, constituent fields and value constraints for each term. Specifically, the system first reads the electromagnetic scheme and geometric boundaries and injects them into the environment configuration layer of the geometry generator. Simultaneously, it attaches the terminology dictionary to the global namespace, triggering field name conflict detection and difference report generation. If homonyms or unregistered terms are found, they are marked for review and written to the version log, preventing further processing. During the input validation phase, scope compliance checks and assembly logic checks are performed concurrently. If geometric entries contradicting the electromagnetic scheme are found, they are sent back to the engineering end via a prompt channel to generate external correction records. If process entries conflicting with geometric boundaries are found, a boundary rollback strategy is triggered, and the triggering reason and rollback level are recorded until the input set reaches a usable state.

[0029] like Figure 3 As shown, Figure 3 This embodiment of the application provides a parameterized definition diagram of a symmetrical four-semi-circular auxiliary slot. This geometric prototype is the core object for constructing the parameterized geometric model in step S100. As shown in the figure, the four sets of auxiliary slots (1-4) are symmetrically distributed about the central axis L, and are explicitly defined in the terminology dictionary as two sets of symmetrical parameter pairs: auxiliary slot 1 and auxiliary slot 4 are symmetrical about the L-axis and share the same set of parameters, including angle x1, depth x2, and the fillet radius (x2+x3) derived from the depth and offset; auxiliary slot 3 and auxiliary slot 2 are symmetrical about the L-axis and share another set of parameters, including angle x4, depth x5, and the fillet radius (x5+x6). This parameterized definition directly corresponds to the design variable set generated in step S100, and the linkage constraint ensures the parameter linkage of the symmetrical slots, thereby maintaining the symmetry and mesh friendliness of the model. The figure also clearly states that minimizing the cogging torque Tc and torque ripple Tpk are the optimization objectives, and maintaining the rated torque T is the constraint condition. These are the specific physical quantities on which the target item mapping and constraint item injection operations in step S200 are based. Figure 3It intuitively demonstrates the mapping relationship from geometric parameters (x1-x6) to electromagnetic performance targets (Tc, Tpk, T), providing clear input, output and constraint definitions for the subsequent optimization process based on the improved adaptive particle swarm algorithm.

[0030] Furthermore, during the geometric prototype generation process, the system calls the geometry generator to construct a parametric geometric prototype. The geometry generator adopts a parametric script mechanism driven by the Computer-Aided Design (CAD) kernel. It first forms the stator and rotor basic elements by combining slot sketches, revolution bodies, and extruded bodies, and then arranges symmetrical four semi-circular arc auxiliary slots on the rotor surface. The symmetrical four semi-circular arc auxiliary slots are grooves formed by four continuously curving circular arc segments, arranged symmetrically around the magnetic pole centerline, and have smooth transitions at the slot ends and magnetic bridge areas; its parameter set includes fields such as the arc radius group, slot depth, slot opening width, relative angle group, and slot end transition fillet. The system creates parameter nodes according to the field names in the terminology dictionary, and establishes units, precision, value step size, and upper and lower boundaries for each node. During the generation process, the geometry kernel performs geometric intersection detection, closed curve detection, and machinability detection for chamfers and fillets. When a cut body is detected to penetrate a magnetic bridge or the minimum chamfer radius is lower than the process lower limit, a backtracking mechanism is automatically triggered. The system backtracks to the nearest feasible position within a predetermined step size and generates diagnostic labels. The diagnostic labels contain the violation field, violation location, and backtracking level, and are written into the structural diagnostic set for subsequent tracing. To balance the modeling efficiency of subsequent finite element (FE) evaluation, mesh-friendly markers are generated simultaneously during the geometric prototype stage. Corner smoothing, minimum transition radius constraints, and minimum channel width constraints reduce the risk of element distortion caused by extremely fine features. The mesh-friendly markers are written into the model tree in the form of geometric attributes for the evaluation collaboration stage to read.

[0031] During the discretization of feature parameters, the system reads core fields such as the arc radius group, groove depth, groove width, and relative angle group from the parameterized geometric prototype to construct a feature parameter discretization table. The discretization strategy adopts a collaborative mechanism of hierarchical sampling and boundary expansion: hierarchical sampling establishes multi-level sampling steps for each field and covers a large range of combination space through inter-layer staggering; boundary expansion inserts high-density sampling points near the lower limits of the process and assembly, covering sensitive areas near magnetic bridges and air gaps. A geometric feasibility pre-judgment unit is pre-processed before combination generation. The pre-judgment unit eliminates combinations of potential penetration, weak magnetic bridges, and excessively narrow transition zones based on fast Boolean calculations and distance field estimation of the geometric kernel, and outputs feasible and conflict markers; combinations that hit process no-go zones trigger an alternative step size strategy, replacing the original step size with a smaller step size and re-evaluating; if it is still not feasible, the corresponding field is frozen at the boundary safety value and a freezing description is generated. After discretization, the system generates a feature parameter discretization table based on the field order and naming conventions of the terminology dictionary. It records the list of field discrete values, adjacent step distances, boundary sources, and diagnostic label references. The sensitivity to magnetic bridges, air gaps, and groove-shaped transition zones is marked as an independent labeling domain for item-level mapping in the subsequent target construction stage.

[0032] During variable range setting, the system transcribes geometric boundaries and process constraint templates into range expressions, compares them with the boundary sources recorded in the feature parameter discrete table, and sends discrepancies to the engineering department for review and generation of discrepancy receipts. Subsequently, a variable range library is established, setting start values, end values, step sizes, and out-of-bounds handling methods for each field. Out-of-bounds handling methods include three strategies: rollback, return, and rejection. To reduce the hidden violation risks caused by cross-field linkage, a linkage constraint is introduced. The linkage constraint provides a coordinated range based on the geometric relationships between the magnetic bridge and slot depth, slot width and relative angle groups, and arc radius groups and transition fillets. When an update of any field violates the coordinated range, a small-scale callback is executed on the relevant fields in priority order, and a re-verification is performed. Variable range setting simultaneously generates a range event stream, which records the triggering reasons and timestamps for range creation, rollback, return, and rejection for auditing and reproduction. The system then merges the feature parameter discrete table with the variable interval library to obtain a parameterized geometric model. The parameterized geometric model includes a set of fields, a list of feasible combinations, linkage constraint rules, and a diagnostic label index, which serves as the entry point for subsequent variable combination generation and solution.

[0033] During the output construction and cross-step integration process, the system performs structured encapsulation on the parametric geometric model, generating a design variable set structure for cross-step invocation. The design variable set structure adopts a serialized organization method that combines key-value pairs and hierarchical structure. The upper layer stores a list of field names and field attributes, while the lower layer stores feasible combinations, diagnostic label references, linkage constraint summaries, and mesh-friendly tag mappings, along with version numbers, source times, and engineering baseline fingerprints for traceability. At the same time, a call protocol list is created, declaring field names, units, step sizes, reading order, and null value handling methods. This list is attached as metadata segments to the design variable set structure and submitted to the project library. Upon submission, the system registers the design variable set structure as the sole external output of this main step at the process level. The process description explicitly states that this output serves as the input for the design variable set structure in subsequent steps (S210), which is used by the target construction and executable main step to perform target item mapping and constraint item injection. In closed-loop scenarios, after the engineering configuration structure generated by the optimal write-back and engineering solidification main steps is stored in the database, it can be used as an incremental reference to feed back into the input set of this main step during the next process initiation, driving continuous updates to parametric geometry and boundary settings. To support the modeling efficiency of the collaborative evaluation main step, the design variable set structure also retains mesh-friendly marker mappings, enabling subsequent steps to select mesh strategies as needed and shortening modeling preparation time.

[0034] In summary, the technical effects of this step are as follows: a stable and traceable set of design variables is formed through the linkage mechanism of geometric prototype generation, feature parameter discretization, and variable range setting; cross-step calling ambiguity and communication costs are reduced through unified field naming and calling protocols; and the recording method of linkage constraint and interval event flow supports the reproduction and auditing of subsequent steps.

[0035] S200: Obtain the design variable set structure, perform target item mapping and constraint item injection and interface transcription processing, and generate the target evaluation interface structure.

[0036] The input sources for S200 are the design variable set structure, terminology dictionary, structural diagnostic set, and mesh-friendly tag mapping registered in the preceding S100. These objects are on the same version line in the engineering library and have source time and baseline fingerprints. The design variable set structure is a field container for collaborative search and evaluation, containing a list of field names, field attributes, feasible combinations, linked constraint summaries, and diagnostic tag references. The terminology dictionary is a standardized naming list, providing terms and field sequences. The structural diagnostic set records geometric rollback and freeze descriptions. The mesh-friendly tag mapping is used to select mesh strategies during the evaluation phase. Specifically, the system reads these objects from the engineering library and performs consistency alignment. If a field is missing or a naming conflict is found, input interception and discrepancy report generation are triggered. The discrepancy report is written to the version log, suspending subsequent actions of S200 until it is approved by the engineering side.

[0037] In the target item mapping phase, the system establishes an item-level mapping between the fields in the design variable set structure and the evaluation indicators. Target item mapping maps the parameter combinations of candidate geometries to a set of rules for the indicator calculation interface, corely covering two indicator entries: cogging torque and torque pulsation, while retaining auxiliary statistical entries directly related to these two entries. Specifically, the system constructs an indicator mapping table based on the field sequence of the terminology dictionary. Table entries include indicator name, data source, sampling window, statistical method, alignment rules, and missing data handling strategy. The data source refers to the time-domain and angular-domain sequences generated during the evaluation collaboration phase; the sampling window describes the alignment method between the angular step and the mechanical cycle; the statistical method covers implementation steps such as peak selection, mean statistics, and envelope estimation; the alignment rules handle the index relationship between electrical and mechanical angles; and the missing data handling strategy provides placeholder values ​​and recalculation flags when mesh generation fails or the solution is interrupted. When constructing the mapping table, the system synchronously reads the grid-friendly marker mapping. If the marker indicates that there is a risk of grid instability in a local area with excessively narrow channels or sharp corners, an early warning marker is added to the corresponding combination. The early warning marker enters the subdomain of the target item mapping and triggers the densification grid or corner step reduction action when the evaluation task is generated.

[0038] During the constraint injection phase, the system establishes a constraint list for the average torque constraint and integrates geometric manufacturing constraints, assembly constraints, material constraints, and mesh stability constraints into the same constraint framework. The constraint list is a structured object composed of decision rules, range thresholds, and violation handling strategies, providing boundary conditions around the target value ranges of the cogging torque and torque pulsation terms. Decision rules reference field names from the terminology dictionary, avoiding temporary naming; range thresholds are read from the vehicle platform file recorded in the engineering baseline and are not directly modified in this step; violation handling strategies employ a penalty structure, recording the violation type, severity level, weight adjustment method, and recalculation limit. During constraint injection, the system cross-checks the linked constraint summary. If a field update is found to disrupt the coordination range between the magnetic bridge and slot depth or slot width and relative angle group, a linked callback is triggered, and a constraint conflict entry is written; if a material entry and temperature level combination contradicts an assembly clearance entry, the combination is frozen, and the freezing reason is recorded in the list. After the constraint framework is constructed, the system generates a constraint decision maker. During the evaluation and collaboration phase, the constraint decision maker is triggered item by item according to the input order of the mapping table. It returns a penalty record for illegal combinations and a pass record for feasible combinations. All records are entered into the evaluation log.

[0039] During the interface transcribing phase, the system transcribs target item mappings and constraint items into structured objects that can be directly invoked during the evaluation and collaboration phase. Interface transcribing consists of three parts: an input template, an output template, and a call sequence. The input template describes the order of geometric field readings, units, step sizes, and null value handling methods, and binds mesh-friendly and warning tags. The output template describes the organization order of indicator names, statistical fields, constraint judgment receipts, and penalty records. The call sequence describes the execution order, retry strategy, and error channels for the evaluation and collaboration phase. The interface objects are described using the Application Programming Interface (API) style, the serialization format uses JavaScript Object Notation (JSON), and the remote invocation method uses Representational State Transfer (REST). After generating the interface objects, the system performs a validity check, including field completeness, naming consistency, and template structure integrity. If any are missing, the system reverts to the target item mapping phase to regenerate the mapping table entries; if the naming is inconsistent, the system reads the standard naming from the terminology dictionary, automatically rewrites the entries, and marks the source of the rewriting. To reduce coupling, the system creates an evaluation task description sheet. The task description sheet binds the input template slice of a single candidate combination to the corresponding output template slice, and declares the sequence number in the call sequence, the maximum number of retry attempts, and the error channel strategy. The description sheet serves as the smallest granular unit of the task in the evaluation collaboration phase.

[0040] Furthermore, during the objective term mapping and constraint term injection process in step S200, the system constructs an augmented objective function for subsequent optimization. Specifically, a design variable vector is defined. The parameter combination (such as angle, depth, etc.) representing the symmetrical four-semi-circular auxiliary groove is given by the design variable set structure. The objective function consists of the weighted normalized values ​​of the cogging torque term and the torque ripple term, and a penalty function is introduced to handle the violation of the average torque constraint. Augmented objective function. The definition is as follows: in: : Design variable vector, representing the combination of geometric parameters (such as arc radius, groove depth, relative angle, etc.) of the symmetrical four semi-circular auxiliary groove. Its field names and ranges are derived from the design variable set structure generated by S100. : Normalized weighted objective function, used to comprehensively evaluate cogging torque and torque ripple performance; : Penalty function, used to handle violations of average torque constraints; the penalty is activated when the optimized average torque value is lower than the standard value; : The augmented objective function serves as the fitness function for the improved adaptive particle swarm optimization algorithm in the S300 step; and : Weighting coefficient, given by the engineering baseline, used to adjust the relative importance of the cogging torque term and torque ripple term in the objective function (default value is usually 1).

[0041] The optimized value of cogging torque is a design variable. The function is obtained through the collaborative stage of finite element evaluation.

[0042] The standard value of cogging torque is derived from vehicle platform files or engineering specifications and is used for normalization.

[0043] The optimized value for torque ripple is a design variable. The function is obtained through the collaborative stage of finite element evaluation.

[0044] Torque pulsation standard value, derived from vehicle platform files or engineering specifications, is used for normalization.

[0045] : Penalty weight coefficient, given by the violation handling strategy in the constraint list, is used to control the strength of the penalty item.

[0046] The average torque optimization value is a design variable. The function is obtained through the collaborative stage of finite element evaluation.

[0047] The standard value of average torque (i.e., rated torque) is derived from the threshold value of the average torque constraint term in the engineering baseline. Furthermore, the system establishes a tracking relationship for target item mapping and constraint item injection. The tracking relationship records the position indexes of mapping table entries and constraint list entries in the interface object, forming a bidirectional reference with the indicator and penalty records in the evaluation log, facilitating rapid entry location when anomalies occur during the evaluation collaboration phase. During the tracking relationship generation process, the system generates a tag for each entry, containing the field name source, entry creation time, and triggerer identity. If the number of retries reported during the evaluation collaboration phase exceeds the limit, the system will locate the corresponding entry based on the tag and mark it as unavailable, while simultaneously registering the exemption reason in the project library to avoid repeatedly calling the same unavailable entry within the same iteration. This strategy ensures the stable operation of the interface object across multiple iterations without altering the indicator calculation logic and constraint judgment rules.

[0048] Understandably, there is a two-way information channel between S200 and S100. The current step reads the design variable set structure to complete the target item mapping, constraint item injection, and interface transcribing. At the same time, in the diagnostic scenario, it writes back the violation combination and freezing reason to the structural diagnostic set, providing a reference for the next round of parameter interval convergence. S200 and S300 are directly related. The interface object output by the current step is read by the initialization action of the evaluation collaboration phase, calls the input template and call sequence to create batch tasks, and writes back the index record and constraint receipt according to the output template after the finite element evaluation is completed. In the closed-loop scenario, after the engineering configuration structure generated by S400 enters the engineering library, the current step can read the size and index fields in it when the new round of process begins, update the default priority of the target item mapping and the range entries of the constraint list, so that the interface object keeps the platform baseline in a synchronous state.

[0049] During the output construction and destination declaration phase, the system performs version encapsulation and fingerprint generation on the aforementioned interface objects, and writes them to the evaluation interface partition of the project library. The encapsulation action assigns a version number, source time, and baseline fingerprint to the interface object, which are registered together with the task description sheet; fingerprint generation is used to quickly compare whether structural changes have occurred to the interface object. After encapsulation, the system explicitly names the output field as the target evaluation interface structure in the process text, and the next input location is the target evaluation interface structure of S310. The evaluation collaboration phase uses this to create candidate parameter groups and batch evaluation tasks. To support playback and auditing, the system retains tracking relationships and log anchors within the interface object, does not expose internal implementation details, and only returns indicator records and constraint receipts through the output template.

[0050] In summary, the technical effects of this step are as follows: Through the collaborative processing of target item mapping, constraint item injection, and interface transcribing, a unified interface object that can directly drive the evaluation and collaboration phase is formed; by tracking relationships and organizing task description sheets, the ability to align data across steps and locate anomalies is improved; and through version encapsulation and fingerprint generation, the structural stability and reproducibility in multiple iterations are enhanced.

[0051] S300: Obtain the target evaluation interface structure, perform improved adaptive particle swarm initialization, finite element fast verification, and new iterative function update operations, and generate updated candidate population structure.

[0052] Figure 4 A flowchart of an improved adaptive particle swarm optimization algorithm provided in this application embodiment is shown below, and its core algorithm flow is as follows: Figure 4 As shown. This process, based on the standard particle swarm optimization algorithm, integrates improvements such as adaptive inertia weight, adaptive learning factor, new iteration function, and mutation strategy. The specific steps are as follows: After starting, the population size N and the maximum number of iterations T are determined first. The population here corresponds to a candidate population, and each particle represents a set of design parameter combinations for the permanent magnet motor auxiliary slot.

[0053] Initialize the velocity and position of each particle. The initial position value is randomly or strategically initialized based on the variable range defined by the design variable set structure generated by S100, ensuring that the starting point is within the feasible region.

[0054] Calculate the fitness value for each particle. This step calls the target evaluation interface structure generated by S200, performs a fast finite element method check, and calculates the augmented objective function value F(x) corresponding to the current particle (i.e., a set of parameters).

[0055] Calculate the individual best value (pbest) for each particle and the global best value (gbest) for the entire swarm, and record the historical best position of each particle and the best position found so far for the entire swarm, respectively.

[0056] The algorithm then enters an iterative loop, performing the following improvements in each iteration: Inertia weight and learning factor update: Based on the population distribution and convergence status of this iteration, the inertia weight w and social learning factors c1 and c2 are adaptively adjusted to balance global exploration and local mining capabilities.

[0057] The velocity and position are updated according to the new formula: the updated w, c1, c2 are used, and the particle's velocity and position are updated based on a new iterative function that includes a global guiding term, a local mining term, and a nonlinear compression term.

[0058] Mutation strategy to update position: The particle position is mutated and perturbed with a specific probability to inject randomness, which effectively avoids premature convergence of the algorithm and escape from local optima.

[0059] Perform boundary condition processing: ensure that the updated particle position does not exceed the geometric and physical boundaries defined by the linkage constraint in S100.

[0060] Determine if the termination condition is met (e.g., reaching the maximum number of iterations T, or the quality of the solution showing no significant improvement in consecutive iterations). If not met, return to step 3 to continue iteration; if met, output the result, i.e., the optimal combination of motor parameters represented by the current global best position (gbest), and finally terminate the process. This output is the optimal solution in the updated candidate population structure.

[0061] The input source for S300 is the target evaluation interface structure generated by the preceding S200 and stored in the evaluation interface partition of the engineering library. This structure includes an input template, an output template, a call sequence, a tracing relationship, and a task description sheet. The input template specifies the order of geometric field reading, units, step size, null value handling method, and mesh-friendly tag binding relationship; the output template specifies the organization order of indicator names, statistical fields, and constraint judgment receipts; the call sequence specifies the execution order, retry strategy, and error channel of the evaluation collaboration phase; the tracing relationship records the position index of the mapping table entries and constraint list entries in the interface object; the task description sheet binds the input template slice and output template slice of a single candidate combination and declares the sequence number and the upper limit of the number of retries. Specifically, when the system starts this step, it first reads the target evaluation interface structure and performs a validity check. If it finds missing fields, inconsistent naming, or incomplete template structure, it triggers input interception and difference report generation. The difference report is written to the version log and subsequent actions of this step are suspended. When the verification passes, the system creates an evaluation task queue according to the call sequence and parses the geometric fields, step size, and boundary caliber required for each task based on the input template to generate a list of candidate parameter placeholders that can be assembled. The placeholder list is used to receive the set of candidate individual parameters produced by the improved adaptive particle swarm.

[0062] In the improved adaptive particle swarm initialization stage, the system constructs the population structure, individual structure, and population statistical structure of the particle swarm. The English name for the particle swarm algorithm is Particle Swarm Optimization, abbreviated as PSO. The improved adaptive particle swarm in this invention employs collaborative design in terms of adaptive inertia weights, adaptive learning factors, mutation perturbation triggering, and new iteration function updates. Specifically, the system reads the upper and lower boundaries, step size, and linkage constraint summaries of each field from the input template. It uses stratified uniform sampling to generate an initial parameter set within the feasible interval. During parameter set creation, a rapid check is performed with the linkage constraint mechanism. Any combination that disrupts the coordinated range of the magnetic bridge and slot depth or slot width and relative angle group is marked as invalid and replaced with a neighboring feasible combination. For values ​​recorded as high-risk by the structure diagnosis set, the system performs offset sampling within the same field to generate several safe neighborhood points. These safe neighborhood points, together with the origin, constitute the initial candidate set for individuals. Given the population size and initialization rounds as determined by the engineering baseline, the system sequentially fills several candidate sets into a placeholder list, forming a candidate parameter population. Simultaneously, it creates a status record entry for each individual, including the current position parameter set, the historical best parameter set, a feasibility flag, the cumulative constraint penalty value, and an error counter. Inertia weight adaptation is driven by the population statistical structure, which continuously calculates diversity metrics, the ratio of stagnant rounds to improvement rounds. Based on the statistical results, the system updates the inertia weight in segments between two intervals: lenient and convergent. Learning factor adaptation at the individual level is triggered by the progress rate, which is described by the sign and magnitude of the fitness change between nearest neighbor iterations. The system increases the weight of global terms when there is no continuous improvement and increases the weight of local terms when there is continuous improvement. None of the above adaptive strategies introduce new fields, and the upper and lower bounds of parameter updates are limited by the interval constraints given by the input template.

[0063] In one embodiment, in the improved adaptive particle swarm initialization step S300, the system is based on the standard Particle Swarm Optimization (PSO) algorithm and introduces several improved strategies to optimize the auxiliary slot size of the automotive permanent magnet motor rotor. The standard PSO process includes: confirming the population size. Maximum number of iterations Dimensions Initialize position and velocity, where the first... During the nth iteration The first particle The positional components of the dimension are The velocity component is ; Calculate the individual optimal value and global optimal value Update speed in, :No. During the nth iteration The first particle The velocity component controls the step size and direction of position updates; : Iteration count index; Inertia weight: controls the influence of the particle's current velocity on the velocity of the next generation; it is an adaptive variable in the improvement strategy. Learning factors control the weights of individual experience and social experience on speed updates, respectively. :No. The random number generated during the next iteration, with a value range of [0,1], is used to introduce randomness; :No. During the nth iteration The first particle The individual optimal position component of the dimension records the best position in the history of the particle; :No. During the nth iteration The global optimal position component of the dimension records the best position in the entire population's history; :No. During the nth iteration The first particle The positional component of the dimension represents the current value of the design variable; : Particle index, indicating the number of particles in the population. One particle; : Dimension index, indicating the first dimension of a variable. dimension; Update location: Repeat the iterations until the maximum number of iterations is reached.

[0064] Improvement strategies include: 1. Inertia Weight Improvement Strategy: Inertia Weight The adaptive update formula is: in, Inertia weight: controls the influence of the particle's current velocity on the velocity of the next generation; it is an adaptive variable in the improvement strategy. It is a constant. and These are the maximum and minimum values ​​of the inertia weight. and These represent the maximum and minimum values ​​among the individual optimal values. This strategy balances global search and local exploitation by dynamically adjusting weights.

[0065] 2. Learning Factor Adaptive Strategy: Learning Factor and It updates adaptively based on the iteration progress.

[0066] 3. New Iteration Function: Introducing a nonlinear function to enhance search diversity.

[0067] 4. Mutation strategy: based on probability Trigger mutation (where, Mutation probability (representing the probability threshold for triggering a mutation operation), the formula is: in, A random number in the range [0,1]. and These represent the maximum and minimum values ​​of the input variables. The mutation strategy injects random perturbations when population diversity decreases, preventing premature convergence.

[0068] The improved algorithm integrates the above strategies, and its effectiveness is verified through a test function. During testing, the population size is set... Number of iterations A function was used as the baseline. The results show that the improved algorithm can effectively overcome the local optimum problem and improve convergence performance. Figure 5 A schematic diagram of a test function and iteration curve provided for an embodiment of this application, such as... Figure 5 As shown, the algorithm improvement effect can be intuitively demonstrated. To address the problem that particle swarm optimization (PSO) is prone to getting trapped in local optima, an improved adaptive PSO algorithm is adopted. This is achieved by adding a new inertia weight function, an adaptive learning factor, a mutation strategy, and a new iteration function to improve the PSO algorithm's ability to avoid getting trapped in local optima. The results are tested using a test function with a population size of 30 and 1000 iterations. The results show that the improved adaptive PSO algorithm can effectively overcome the local optima problem.

[0069] Furthermore, in the finite element rapid verification stage, the system generates simulation tasks one by one according to the evaluation task queue and calls the finite element solver to evaluate the indicators. The English name for finite element is Finite Element, abbreviated as FE. Specifically, the system reads the parameter set and corresponding input template slices for each individual, calls the geometry generator to construct the symmetrical four-half-circular auxiliary groove geometry on the rotor surface, selects a mesh strategy based on mesh-friendly markings, and performs corner smoothing and local refinement on sharp corners and excessively narrow channels. Then, it drives the solver to generate magnetic field and electromagnetic force sequences, calls the indicator mapping table in the target evaluation interface structure to perform item-level statistics on cogging torque and torque pulsation, and simultaneously calls the constraint decision-maker to determine the average torque constraint, manufacturing constraint, assembly constraint, and mesh stability constraint. If the solver returns an error or mesh generation fails, it performs one or more recalculations according to the retry strategy in the call sequence and records the error path; if it still fails, it establishes a penalty record for that individual and increments the error counter. To avoid redundant evaluations, the system jointly generates parameter fingerprints using geometric fields and mesh-friendly tags, queries the evaluation cache, and if a cache hit occurs, directly reclaims the indicator record and constraint receipt, registering the cache source in the tracing relationship. If a cache miss occurs, a new round of solving is executed, and the results are written to the cache after evaluation. All individual indicator records and penalty records are uniformly entered into the evaluation log, which is bidirectionally referenced by the tracing relationship, facilitating subsequent fault location and replay.

[0070] In the new iteration function update phase, the system uniformly updates the particle swarm based on the fitness sequence and the population statistical structure. The fitness sequence consists of a weighted index of the cogging torque term and the torque ripple term, along with constraint penalties. The system first updates the historical best parameter set and the priority of the current position at the individual level, and then executes the new iteration function. The new iteration function includes three parts: a global guiding term, a local mining term, and a nonlinear compression term. The weights of these three parts are jointly adjusted by adaptive inertia weight and adaptive learning factor. The global guiding term uses the historical best parameter set of the population for directional guidance, the local mining term uses the historical best parameter set of the individual for neighborhood search, and the nonlinear compression term gradually reduces the step size as the population diversity decreases, thereby controlling the search radius. After the parameter update, the system performs boundary processing. Out-of-bounds entries enter the boundary bounce or reversal strategy, and the choice between bounce and reversal is determined based on the out-of-bounds processing mode in the input template. Subsequently, a deduplication process is executed, retaining only individuals with better fitness or lower error counts when the parameter fingerprints are consistent. Then, an elite retention strategy is executed, directly writing several optimal individuals into the next generation candidate set to avoid them being covered by random perturbations in the next generation. If the population diversity falls below a threshold or improvement stagnates to the threshold, the system triggers a mutation perturbation mechanism. Within the safe range of the variable interval library, a small random perturbation is applied to some fields. The perturbation amplitude is jointly pruned by the field step size and the linkage constraint summary. The perturbed individuals then re-enter the linkage constraint check. Those that fail the check are removed or recalled. After the above update is completed, the system forms the next generation of candidate parameter population and regenerates the evaluation task queue according to the call sequence to enter the next round of finite element rapid verification.

[0071] In the convergence and output phase, the system terminates or continues iteration based on convergence criteria. These criteria consist of a population diversity threshold, the number of improvement stagnation rounds, and the evaluation task budget. When the criteria are met, the system stops creating a new task and summarizes candidate individuals, fitness sequences, constraint receipts, and penalty records from the current generation and several previous generations to form a snapshot of the candidate population. Subsequently, several optimal individuals are selected according to the elite retention rule, and the elite entries retained from the previous generation are merged with the newly generated superior entries to form a structured updated candidate population structure. The updated candidate population structure includes a set of candidate individuals, fitness sequences, a summary of violation penalty records, tracking tags, and cached fingerprint summaries, and is stored in a temporary partition of the project repository. After encapsulation, the system explicitly states in the process text that the output field name is "Update Candidate Group Structure," and the next input position is the "Update Candidate Group Structure" in S410. After receiving the structure, S400 performs optimal individual selection, geometric write-back, and consistency comparison. At the same time, for error channels, penalty records, and freeze descriptions that occur during the evaluation and collaboration phase, the system writes back to the mapping table and constraint list entries in S200 by tracing relationships, providing a basis for interface optimization in subsequent rounds. For stability issues caused by mesh-friendly marking, the system writes the hit statistics of the markings back to the mesh-friendly marking mapping in S100 as an incremental reference for corner smoothing and minimum transition radius constraints in the geometric prototype phase.

[0072] The technical effects of this step can be summarized as follows: By driving the improved adaptive particle swarm initialization, rapid finite element verification, and new iterative function update through the target evaluation interface structure, a structured candidate population structure for updating is formed; by using a combination of adaptive, perturbation, and elite retention strategies, the stable progress of the population search is maintained; and by relying on the organization of tracking relationships and cached fingerprints, the closed-loop backtracking of evaluation records and parameter sources and cross-step docking are completed.

[0073] S400: Obtain the updated candidate group structure, perform optimal individual screening, consistency comparison and parameter solidification processing, and generate the engineering configuration structure.

[0074] The input to S400 comes from the update candidate population structure, evaluation log, and tracking relationships encapsulated in the preceding S300. It also references the target evaluation interface structure and constraint list established in S200, and reads the geometric boundaries and terminology dictionary registered in S100 as a baseline. The update candidate population structure includes a set of candidate individuals, a fitness sequence, a summary of violation penalty records, tracking tags, and cached fingerprint summaries. The evaluation log carries statistical records of cogging torque and torque ripple terms, as well as constraint decision receipts. The tracking relationship records the position indices of the mapping table entries and constraint entries within the interface object. Specifically, the system first performs input validity verification, aligning the field completeness, naming conventions, version fingerprints, and source time of the candidate individual set. If any candidate individual is found to have missing fields, inconsistent naming, or version drift, the input interception strategy is triggered, marking the individual as suspended and locating the corresponding mapping table entry or constraint entry based on the tracking relationship, and submitting a re-evaluation request through the error channel. If the evaluation log detects that the number of retries exceeds the threshold and still returns invalid records, the corresponding individual is removed from the current round of screening list and the reason for removal is written to the difference receipt partition of the engineering library.

[0075] In the optimal individual selection process, the selection of candidate individuals is a process of sorting and selecting according to the priority of multiple indicators and the feasibility of constraints. Specifically, based on the aligned indicator records in the evaluation log, the system constructs two main indicator tracks, corresponding to the cogging torque item and the torque pulsation item, respectively, and uses the feedback of average torque constraints, manufacturing constraints, assembly constraints, and mesh stability constraints as feasibility barriers. Under the premise that feasibility is passed, the system uses Pareto sorting to form a non-dominated front, and introduces the penalty accumulation and error count as parallel resolution conditions within the same level front, so that individuals with smaller penalties and lower error counts are selected first. Further, the system performs neighborhood robustness verification on the selected individuals according to field sensitivity: it reads the field step size and cooperative range related to the individual from the variable interval library, generates a small number of local alternative combinations around the selection point, calls the target evaluation interface structure to complete the rapid evaluation, and compares and records the indicator fluctuations of the selection point and the alternative combinations; if the alternative combination has a constraint violation or evaluation failure, only the diagnostic label is written back, without changing the current ranking of the selection point, but the verification priority of the selection point is increased in the subsequent consistency comparison stage. After the screening is completed, the system obtains a candidate optimization list sorted by multiple indicators and verified by the neighborhood, and retains evaluation log references and tracking tags in the list items for use in the next stage.

[0076] During the consistency comparison process, the consistency comparison is a process of checking the candidate preference list against geometric boundaries, process templates, assembly rules, and interface specifications item by item. Specifically, the system first performs a static comparison between the geometric fields of the selected individuals and the geometric boundaries registered in S100, checking whether the minimum thickness of the magnetic bridge, the air gap range, the groove turning radius, and the end transition fall within the boundary library. If there is a critical hit, the boundary backoff strategy is triggered, and the system is pulled back from the variable interval library to the safe boundary value according to the field priority, and the backoff record is written to the consistency comparison log. Further, the system calls the constraint determiner based on the constraint list in S200 to review the average torque constraint, manufacturing constraint, assembly constraint, and mesh stability constraint. The manufacturing constraint review covers tool accessibility, minimum channel width, and chamfer machinability; the assembly constraint review covers stator and rotor clearance, polar symmetry, and assembly tolerance accumulation; and the mesh stability review covers sharp corner density and local scale ratio. To avoid duplicate evaluations due to discrepancies in terminology, the system uses the input and output templates in the target evaluation interface structure as the basis for terminology alignment, reconfirming field units, reading order, and null value handling. If a template structure change is detected, the system traces back to the corresponding mapping table entry based on the tracking relationship and updates the terminology imprint in the consistency comparison log. Understandably, during the consistency comparison phase, the system only verifies terminology consistency and boundary compliance, without evaluating or judging indicator values; all judgments are expressed through receipts and labels.

[0077] During parameter solidification, the process involves converting the dimension fields, linkage constraint summaries, mesh-friendly tag mappings, and call protocol lists of selected individuals that have passed the consistency comparison into deliverable configuration assets and applying version freeze. Specifically, the system generates a parameter solidification sheet for each selected individual that passes the verification. The parameter solidification sheet includes field names, field values, step sizes, boundary sources, linkage constraint summaries, mesh-friendly tag mappings, and referenced evaluation log anchors. Subsequently, the parameter solidification sheet undergoes two levels of freeze: the first level is field freeze, which locks the field values ​​and their source versions, disallowing modifications within the current project baseline; the second level is relationship freeze, which writes the collaborative scope and order constraints between fields into a relationship table and prohibits cross-table rearrangement. To support subsequent project releases, the system binds the parameter solidification sheet to the configuration fragments of the geometry generator, generating geometric fragment assets that can be directly called in the design environment. Simultaneously, it generates process suggestion entries to indicate tool selection, minimum fillet radius, and machining priority for locally reinforced areas. These process suggestion entries are derived from boundary hit records and mesh-friendly tag mappings from the consistency comparison phase. If inconsistencies are found between the naming conventions and those in the terminology dictionary during the freezing process, the system will automatically rewrite the data and record the source and timestamp of the rewrite in the remarks section of the parameter freezing sheet. If a field is found to be missing in the input template of the target evaluation interface structure, the system will return to the consistency comparison stage to complete the missing field before entering the freezing process.

[0078] After the parameters in step S400 are solidified, the system verifies and analyzes the optimization results. The optimization results for the auxiliary slot size of the automotive permanent magnet motor rotor based on the Improved Adaptive Particle Swarm Optimization (IAPSO) algorithm are as follows: Algorithm convergence verification: The iteration curves show that the improved adaptive particle swarm optimization algorithm converges stably during the optimization process, and the objective function value gradually decreases with the increase of the number of iterations, indicating that the algorithm can effectively approximate the optimal solution.

[0079] Rotor structure optimization comparison: A comparison of the rotor structures before and after optimization shows that the auxiliary slot dimensions, adjusted by the algorithm, better meet the electromagnetic performance requirements. The rotor structure before optimization was the baseline design, and the optimized rotor structure achieved magnetic circuit optimization through parameter adjustments of the symmetrical four semi-circular arc auxiliary slots (such as slot depth and arc radius).

[0080] Specifically, the optimized auxiliary slots 3 and 4 show significant changes in size, reflecting the algorithm's fine adjustment of local geometric features.

[0081] Performance comparison: After optimization, the average torque fluctuation is guaranteed to not exceed [a certain value]. Under the premise of, cogging torque ( ) and torque ripple ( All decreased significantly. The specific performance improvement is quantified by the following formula: in, and The standard values ​​for cogging torque and torque ripple before optimization. and The corresponding value is the optimized value. The peak torque comparison further verifies that the optimization scheme reduces noise and vibration while maintaining output capability.

[0082] After completing the parameter solidification process in step S400, to verify the effectiveness of the improved adaptive particle swarm optimization (IAPSO) algorithm proposed in this invention in practical applications, a comparative analysis of the motor performance before and after optimization was conducted. The results are as follows: Figure 6 As shown, Figure 6 A comparison chart showing the performance of an improved adaptive particle swarm optimization algorithm provided in this application in practical applications.

[0083] Figure 6The comparison of cogging torque after IAPSO optimization in the figure above shows that, within the same time scale (0-1.4ms), the fluctuation amplitude of the optimized cogging torque (blue curve) is significantly reduced. Specifically, the cogging torque before optimization (orange curve) fluctuates drastically in the range of -80mNm to +80mNm, while after optimization by the IAPSO algorithm, the fluctuation amplitude of the cogging torque is greatly reduced, and the waveform becomes smoother. This indicates that the optimization of the symmetrical four semi-circular arc auxiliary slot parameters in this invention effectively suppresses the torque pulsation caused by the interaction between the stator slots and the permanent magnet.

[0084] Figure 6 The figure below compares the peak torque after IAPSO optimization, showing the peak torque variation within the 0-3.5ms time range. The peak torque before optimization (orange curve) fluctuated significantly between 163Nm and 172Nm, while the optimized peak torque (blue curve) exhibited better stability, with a significantly narrowed fluctuation range. This result verifies that the present invention can maintain the stability of the motor's output torque while suppressing torque ripple, ensuring that the motor's power performance is not affected.

[0085] While maintaining a torque ripple of 1 Nm, the cogging torque decreased by 84.26%, and the torque pulsation decreased by 4.24%.

[0086] These two sets of comparative data fully demonstrate the effectiveness of the permanent magnet motor optimization method based on the improved adaptive particle swarm optimization algorithm proposed in this invention. Through the complete optimization process from S100 to S400, the dual objectives of suppressing cogging torque and improving torque stability are achieved, providing a reliable technical solution for the low-noise, high-performance design of permanent magnet motors. The above results demonstrate that the present invention achieves efficient optimization of the rotor auxiliary slot size through the collaborative process of S100-S400, and the optimized engineering configuration structure (such as the output of S400) has practical application value.

[0087] Furthermore, during the generation of the engineering configuration structure, the structure is a hierarchical configuration object that carries dimension fields, linkage constraints, mesh-friendly tag mappings, a list of calling protocols, version fingerprints, and source times. This structure is used for transfer and reuse between the engineering library, design environment, and evaluation collaboration. Specifically, the system merges all parameter fixed sheets, removes field conflicts, and generates a master engineering configuration object. A hierarchical directory is set within this object: the upper layer contains a list of field names and field attributes; the middle layer contains feasible combinations, a summary of collaborative constraints, and mesh-friendly tag mappings; and the lower layer contains evaluation log anchors, tracking tag references, and a list of calling protocols. The system performs integrity verification and fingerprint generation on the master engineering configuration object, writing the version number, source time, and engineering baseline fingerprint into the object header. Subsequently, the object is submitted to the configuration partition of the engineering library, generating an external reference identifier for cross-system access between the design environment and evaluation collaboration. As a preliminary step in the closed-loop process, the system explicitly names the output field "Engineering Configuration Structure" in the process text and declares it as the input source for the parameterized geometric prototype of the next process S110, driving the regeneration of the parameterized geometric prototype and the discretization of feature parameters and the setting of variable ranges. Simultaneously, on the interface side, it declares that this Engineering Configuration Structure can be read by S200 to update the default priority of the target item mapping and the range entries of the constraint list, and can also be read by S300 to improve the initialization distribution and retry strategy of the adaptive particle swarm optimization. After storage, the system establishes a bidirectional link between the consistency comparison log, the parameter fixation sheet, and the external reference identifier, facilitating the tracing of evaluation records and boundary sources from the Engineering Configuration Structure at any time.

[0088] In summary, the technical effects of this step are as follows: Through the sequential processing of optimal individual selection, consistency comparison, and parameter solidification, the candidate results of the search phase are transformed into a reusable, auditable, and replayable engineering configuration structure; the organization of log anchors and tracking tags ensures traceability of sources and alignment of standards; and through cross-main step destination declarations, a stable closed loop is constructed from configuration entry to the next round of modeling and evaluation.

Claims

1. A method for optimizing cogging torque and torque ripple of a permanent magnet motor for vehicle, characterized in that, The method comprises the following steps of: obtaining an electromagnetic scheme, a geometric boundary and a term dictionary, performing input verification and conflict detection processing, field naming and caliber construction based on the term dictionary, parameterized geometric prototype containing symmetric four semi-circular arc auxiliary slot, hierarchical discrete and feasibility pre-judgment, variable interval transcription and linkage constraint configuration operation, structured packaging and protocol declaration processing, and generating a design variable set structure; the electromagnetic scheme contains the pole pair number range, the lamination thickness range, the core material grade, the winding form and the connection mode, the permanent magnet magnetization direction and the magnetic circuit partition, the stator and rotor assembly constraint and the thermal environment grade; the geometric boundary contains the tooth width, the tooth tip chamfer range, the air gap range, the rotor outer diameter, the magnetic pole pitch, the magnetic bridge thickness and the end roundness; the term dictionary contains the symmetric four semi-circular arc auxiliary slot, the parameterized geometric prototype, the parameterized geometric model, the feature parameter discrete table, the variable interval library, the linkage constraint device, the grid-friendly marker and the design variable set structure; obtaining the design variable set structure, performing target item mapping and average torque constraint of the cogging torque item and the torque ripple item, constraint item injection and interface transcription processing of the manufacturing constraint, the assembly constraint and the grid stability constraint, and generating a target evaluation interface structure; obtaining the target evaluation interface structure, performing improved adaptive particle swarm initialization of inertia weight self-adaptation and learning factor self-adaptation and finite element quick checking, and generating an updated candidate group structure through the new iteration function update operation containing the global guide item, the local exploitation item and the non-linear compression item; obtaining the updated candidate group structure, performing optimal individual screening and consistency comparison based on the Pareto ordering and the neighborhood robustness review, and generating an engineering configuration structure through the parameter solidification processing.

2. The method of claim 1, wherein, The symmetric four semi-circular arc auxiliary slot comprises: The symmetric four semi-circular arc auxiliary slot is a groove formed by four curvature-continuous circular arc segments, which are symmetrically arranged around the magnetic pole center line and smoothly transitioned at the slot end and the magnetic bridge area; the parameter set contains a circular arc radius group, a slot depth, a slot opening width, a relative angle group and a slot end transition roundness.

3. The method of claim 1, wherein, The process of generating the design variable set structure comprises: calling a geometric generator to construct a parameterized geometric prototype, the geometric generator adopts a parameter script mechanism driven by a computer-aided design (CAD) kernel, first forms a stator and rotor base element through a slot sketch, a rotation body and a stretching body combination, and then arranges symmetric four semi-circular arc auxiliary slots on the rotor surface; and generating a grid-friendly marker to reduce the risk of cell distortion caused by extremely small features through corner smoothing, minimum transition radius constraint and channel minimum width constraint.

4. The method of claim 1, wherein, The process of generating the design variable set structure further comprises: reading core fields from the parameterized geometric prototype one by one to construct a feature parameter discrete table; the discrete strategy adopts a hierarchical sampling and boundary expansion cooperative mechanism: hierarchical sampling establishes multiple sampling step distances for each field and covers a large range of combined space through interlayer interlacing; boundary expansion inserts high-density sampling points near the lower limit of process and assembly.

5. The method of claim 1, wherein, The process of generating the design variable set structure further comprises: The geometric boundary and the process constraint template are transcribed into interval expressions to establish a variable interval library; to reduce the risk of implicit violation caused by cross-field linkage, a linkage constraint generator is introduced, which gives the collaborative range according to the geometric relationship between the magnetic bridge and the slot depth, the slot opening width and the relative angle group, and the transition fillet group and the transition fillet.

6. The method of claim 1, wherein, The process of improved adaptive particle swarm initialization and finite element rapid checking with inertia weight adaptation and learning factor adaptation also includes: The upper and lower boundaries, step distance and linkage constraint summary of each field are read from the input template, and the initial parameter set is generated in the feasible interval by hierarchical uniform sampling; inertia weight adaptation is driven by group statistical structure, which real-time statistics diversity measure, stagnation rounds and improvement rounds ratio, and the system implements segmented update of inertia weight between the two intervals of relaxation and convergence according to the statistical results; learning factor adaptation is triggered by progress rate at individual level.

7. The method of claim 1, wherein, The new iteration function includes: The new iteration function includes global guidance item, local exploitation item and nonlinear compression item, and the weights of the three are jointly adjusted by inertia weight adaptation and learning factor adaptation; the global guidance item uses the group historical optimal parameter set for directional guidance, the local exploitation item uses the individual historical optimal parameter set for neighborhood search, and the nonlinear compression item gradually reduces the step when the group diversity decreases.

8. The method of claim 1, wherein, The process of finite element rapid checking also includes: For each individual, read its parameter set, call the geometry generator to build the symmetric four semi-circular arc auxiliary slot geometry of the rotor surface, and select the grid strategy according to the grid-friendly marker; drive the solver to generate the magnetic field and electromagnetic force sequence, and call the index mapping table in the target evaluation interface structure to perform item-level statistics on the cogging torque and torque ripple; and generate parameter fingerprints jointly with geometric fields and grid-friendly markers, and query the evaluation cache.

9. The method of claim 1, wherein, After generating the updated candidate population structure, the optimal individual screening based on Pareto sorting and neighborhood robustness review is performed.

10. The method of claim 1, wherein, The design variable set structure generated by structured packaging adopts a serialized organization method with key-value and hierarchical coexistence, the upper layer saves the field name list and field attributes, the lower layer saves the feasible combination, diagnostic label reference, linkage constraint summary and grid-friendly marker mapping, and is attached with version number, source time and engineering baseline fingerprint for tracing.

Images