Multi-physical field optimization method for high-voltage synchronous motor

Abstract

The invention relates to the technical field of motor design and optimization, and discloses a multi-physical field optimization method of a high-voltage synchronous motor. The method comprises the following steps: determining a multi-physics field optimization target and selecting a structure parameter as an optimization factor; performing finite element calculation on each experiment scheme based on orthogonal experiment design to obtain a simulation response value of each optimization target; aggregating the multi-target response of each scheme into a single comprehensive score by using grey correlation analysis; calculating average scores and ranges of the optimization factors under different levels based on the comprehensive scores, and dividing the optimization factors into an iteration layer and an output layer according to the ranges; and for the iteration layer factors, by taking the current optimal level as a center, adjusting the level setting of the iteration layer factors according to a preset dynamic space reconstruction rule to carry out iteration until convergence, and outputting a final structure parameter combination. According to the method, collaborative optimization of multi-physical field performance can be realized while the calculated amount is reduced, and the efficiency and stability of optimization design of the high-voltage synchronous motor are improved.

Description

Technical Field

[0001] This application relates to the field of motor design and optimization technology, and in particular to a multiphysics optimization method for high-voltage synchronous motors. Background Technology

[0002] With the development of power systems and new energy equipment towards larger capacity and higher voltage, high-voltage synchronous motors are widely used in fields such as wind power generation. Compared with medium and low voltage level motors, high-voltage synchronous motors can effectively reduce stator winding current, reduce cable cross-sectional area and auxiliary equipment size under the same power conditions, which is beneficial to improving the overall system economy. However, changes in motor structural parameters under high voltage conditions will simultaneously affect the responses of multiple physical fields such as electromagnetic field, temperature field and structural field, and its multi-physics coupling characteristics are significantly enhanced, leading to a significant increase in design and optimization difficulty.

[0003] Existing optimization design methods for high-voltage synchronous motors mainly include methods based on empirical parameter adjustment, multi-objective optimization methods based on surrogate models, and direct optimization methods based on finite element simulation. Among them, empirical parameter adjustment methods rely on engineering experience and are difficult to consider multi-physics performance indicators. Optimization methods based on surrogate models usually require the construction of multiple sub-models such as electromagnetic field, temperature field, and structural field. Under high-voltage conditions, due to the increased thickness of the insulation system and the complexity of the physical field response mechanism, the accuracy of the surrogate model is difficult to guarantee, and the reliability of the optimization results is limited. While multi-physics optimization methods based on finite element simulation have higher accuracy, they require a large number of simulation calculations in a large parameter space, resulting in high computational costs and long cycles, which is not conducive to quickly obtaining a reasonable optimization scheme in the early stages of engineering design.

[0004] Furthermore, existing methods often use weighted summation to transform multiple objectives into a single objective when dealing with multi-physics and multi-performance index optimization problems. However, the selection of weights is highly subjective and cannot objectively reflect the comprehensive influence relationship between the performance indices of each physics field. It is also not conducive to the targeted reduction and iterative optimization of the parameter space in the future.

[0005] Therefore, there is an urgent need to propose a multiphysics optimization method suitable for high-voltage synchronous motors. This method should ensure the accuracy of multiphysics analysis, effectively reduce the computational scale, objectively evaluate the comprehensive optimization degree of multiple performance indicators, and improve the efficiency and stability of high-voltage synchronous motor optimization design through reasonable parameter hierarchical and iterative strategies. Summary of the Invention

[0006] To address the problems of high computational cost, large parameter space, difficulty in comprehensive evaluation of multiple performance indicators, and low optimization efficiency in the optimization design of high-voltage synchronous motors under multi-physics coupling conditions, this invention proposes a multi-physics optimization method for high-voltage synchronous motors. This method, while ensuring the accuracy of multi-physics field analysis (electromagnetic field, temperature field, and structural field), constructs a unified multi-objective comprehensive evaluation index and combines parameter hierarchical and dynamic iterative optimization strategies to achieve efficient and stable optimization of the structural parameters of high-voltage synchronous motors.

[0007] In a first aspect, this application provides a multiphysics optimization method for a high-voltage synchronous motor, the method comprising: S1. Determine multiple physical field performance indicators to be optimized as optimization targets; S2. Select multiple structural parameters to be optimized as optimization factors, and set the initial value range of each optimization factor; S3. Discretize the initial value range into several levels, and conduct orthogonal experimental design based on the number of optimization factors and level settings. Perform finite element calculations on each experimental scheme to obtain the simulation response values ​​of all optimization objectives under each experimental scheme. S4. Normalize the simulation response values, then construct a reference sequence with the ideal optimal values ​​of each optimization objective, calculate the grey relational degree of each experimental scheme relative to the reference sequence using the grey relational coefficient method, and use the grey relational degree as the comprehensive score of the corresponding experimental scheme. S5. Based on the comprehensive score, calculate the average score of the same optimization factor at different levels, and then obtain the range of each optimization factor. Optimization factors with a range greater than a preset threshold are assigned to the iteration layer, and the rest are assigned to the output layer. S6. For the optimization factor of the iteration layer, take its current optimal level as the center, adjust the level setting of the optimization factor according to the preset dynamic space reconstruction rule, and keep the optimal level of the optimization factor of the output layer unchanged to form a new round of orthogonal experimental scheme. S7. Based on the new round of orthogonal experimental scheme, return to S3 for iterative calculation until the preset convergence condition is met, and output the final combination of structural parameters.

[0008] In conjunction with the first aspect, in the first implementation of the first aspect of this application, the optimization objectives include at least the air gap magnetic flux density of the electromagnetic field, the winding temperature rise of the temperature field, and the vibration acceleration amplitude of the structural field.

[0009] In conjunction with the first aspect, in the second implementation of the first aspect of this application, the optimization factors include at least the insulation thickness of the synchronous motor, the stator slot width, the stator slot depth, and / or the air gap length.

[0010] In conjunction with the first aspect, in the third implementation of the first aspect of this application, the method for calculating the grey relational degree is as follows: The grey relational coefficients for each optimization objective are calculated using the following formula: , in, Let Δ be the grey relational coefficient of the j-th experimental scheme on the i-th optimization objective. ij Let be the absolute difference between the reference sequence and the normalized response sequence of the j-th experimental scheme at the i-th optimization objective, and let minΔ and maxΔ be the values ​​of all Δ... ij The minimum and maximum values ​​in the range, where ρ is the resolution coefficient; The formula for calculating grey relational degree is: , Among them, R j Let represent the grey relational degree corresponding to the j-th experimental scheme, and N be the total number of optimization objectives.

[0011] In conjunction with the first aspect, in the fourth implementation of the first aspect of this application, S5 includes: For any optimization factor, all experimental schemes are grouped according to the level of that optimization factor; The arithmetic mean of the comprehensive scores of each experimental scheme at the same level is taken as the average score of any optimization factor at the corresponding level. Calculate the difference between the maximum and minimum average scores, and use it as the range of any optimization factor.

[0012] In conjunction with the first aspect, in the fifth implementation of the first aspect of this application, for any optimization factor, the level corresponding to the maximum average score is determined as the current optimal level of any optimization factor.

[0013] In conjunction with the first aspect, in the sixth implementation of the first aspect of this application, the dynamic space reconstruction rule is as follows: for any iteration layer optimization factor, taking the current optimal level of any iteration layer optimization factor as the center, the value range of the current optimal level is narrowed and re-discrete into several new levels.

[0014] In conjunction with the first aspect, in the seventh implementation of the first aspect of this application, the dynamic space reconstruction rules include: For any iteration layer optimization factor, calculate the difference between the current best level and the current second best level of any iteration layer optimization factor, and determine whether the difference is less than half of the range of values ​​of any iteration layer optimization factor. If so, during reconstruction, maintain or increase the number of levels, and reduce the step size between adjacent level values ​​with the current best level as the center; If not, during reconstruction, increase the number of levels and expand the step size between adjacent level values, using the current best level as the center.

[0015] In conjunction with the first aspect, in the eighth implementation of the first aspect of this application, the preset convergence condition is that the range of the optimization factors of all iterative layers is less than the preset convergence threshold.

[0016] In conjunction with the first aspect, in the ninth implementation of the first aspect of this application, the resolution coefficient in the grey relational coefficient method is taken as 0.5.

[0017] Compared with the prior art, the beneficial effects of the technical solution of this application are at least as follows: 1. Improve the objectivity and consistency of multiphysics optimization evaluation. This invention uses the grey relational coefficient method to unify the performance indicators of multiple physical fields such as electromagnetic field, temperature field and structural field into a comprehensive score of experimental scheme, avoiding the subjective problem of weight selection in traditional multi-objective optimization, and making the comprehensive performance evaluation of multiphysics more objective and consistent.

[0018] 2. Significantly reduce computational costs while maintaining analytical accuracy. This invention replaces the full-parameter space traversal search with orthogonal experimental design combined with finite element simulation, and gradually narrows the optimization range through parameter hierarchies and dynamic space reconstruction strategies. While ensuring the accuracy of multiphysics simulation, it effectively reduces the number of finite element calculations required during the optimization process, thereby improving overall computational efficiency.

[0019] 3. Improve the convergence efficiency and stability of high-voltage synchronous motor optimization design. By adopting a hierarchical optimization factor strategy based on range analysis, iterative reconstruction is performed only on optimization factors that have a significant impact on overall performance, while the optimal level of optimization factors with a smaller impact is directly fixed. This reduces the dispersion of the parameter space, which helps the optimization process converge quickly and obtain stable optimization results.

[0020] 4. The method of the present invention can effectively address the problems of increased insulation system thickness and enhanced multi-physics coupling effect under high voltage conditions. It is suitable for rapid optimization of high-voltage synchronous motors in the early stage of engineering design and scheme comparison stage, and provides an efficient and reliable multi-physics optimization method for the engineering design of high-voltage synchronous motors. Attached Figure Description

[0021] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0022] Figure 1 This is a flowchart of a multiphysics optimization method for a high-voltage synchronous motor according to this application. Detailed Implementation

[0023] The terms “first,” “second,” “third,” “fourth,” etc. (if present) in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a particular order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments described herein can be implemented in a sequence other than that illustrated or described herein. Furthermore, the terms “comprising” or “having,” and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0024] For ease of understanding, the specific process of the embodiments of this application is described below. Figure 1 The diagram shows a flowchart of a multiphysics optimization method for a high-voltage synchronous motor provided by the present invention. The flowchart specifically includes the following steps: S1. Determine multiple physical field performance indicators to be optimized as optimization targets.

[0025] In one specific embodiment, the optimization objectives include at least the air gap magnetic flux density of the electromagnetic field, the winding temperature rise of the temperature field, and the vibration acceleration amplitude of the structural field.

[0026] Specifically, in optimization design projects for high-voltage synchronous motors (rated operating voltage greater than 20kV), defining the optimization objective is the starting point for implementing the methodology. Based on the motor's performance requirements, operational reliability, and design specifications, designers identify the key physical field output parameters that need to be improved or constrained simultaneously.

[0027] The reason for identifying the air gap magnetic flux density of the electromagnetic field, the winding temperature rise of the temperature field, and the vibration acceleration amplitude of the structural field as optimization targets is that in high-voltage synchronous motors, the main impact path of structural parameter changes on performance is manifested through these three types of physical fields (electromagnetic field, temperature field, and structural field). The air gap magnetic flux density directly reflects the coupling relationship between the stator and rotor magnetic circuits and structural parameters such as insulation thickness and air gap length. Its distribution changes determine the electromagnetic torque capacity and electromagnetic force amplitude, and are a direct quantitative indicator for assessing whether electromagnetic performance has deteriorated under high voltage conditions. The winding temperature rise is jointly determined by the current density, insulation system thermal resistance, and heat dissipation path. In high-voltage synchronous motors, thickening the insulation layer significantly changes the heat conduction conditions, and the winding temperature rise can comprehensively characterize the impact of structural parameters on thermal safety margin. The vibration acceleration amplitude is formed by the electromagnetic force generated by the spatial harmonics of the air gap magnetic flux density through structural transmission. It is the direct response of the structural field to changes in the electromagnetic field, and its magnitude reflects the dynamic stability of the stator structure under high-voltage electromagnetic excitation.

[0028] By selecting the above three types of indicators simultaneously, the responses of the same set of structural parameters in the electromagnetic, thermal, and structural dimensions can be synchronously quantified and entered into a unified data processing flow, thereby solving the problem of fragmented multi-physics responses and difficulty in coordinating evaluation.

[0029] S2. Select multiple structural parameters to be optimized as optimization factors, and set the initial value range of each optimization factor.

[0030] In one specific embodiment, the optimization factors include at least the insulation thickness of the synchronous motor, the stator slot width, the stator slot depth, and / or the air gap length.

[0031] Specifically, the structural parameters of the high-voltage synchronous motor are initially selected. The influence of various structural parameters, including insulation thickness, on the physical field is calculated using the finite element method. The range and number of optimization factors are then selected. The selection of the optimization factor range is based on the chosen optimization objective. The number of optimization factors can be selected using the commonly used Taguchi method.

[0032] In specific implementations, the structural parameters most sensitive to electromagnetic-thermal-structural response coupling and engineering-adjustable are extracted as optimization factors. A one-to-one correspondence is established between each factor and the geometric input fields of the finite element model, ensuring that each experimental scheme is driven by a set of factor values ​​throughout the same mesh generation and boundary assignment process. The insulation thickness is determined by the lower limit of manufacturing feasibility and the upper limit required for voltage withstand, for example, set to 0.10–0.19 and used to update the geometry of the insulation layer between the winding and the core. The stator slot width is determined by the constraints of winding inlay, leakage flux, and heat dissipation channels, for example, 2–5, and used to update the slot opening. The stator slot depth is determined by the slot fill factor and tooth magnetic saturation constraints, for example, 45–48, and used to update the slot depth dimension. The air gap length is determined by the constraints of mechanical tolerances, magnetic flux density, and vibration sensitivity, for example, 2.0–3.5, and used to update the stator-rotor gap. The interval data serves as the input for subsequent discrete and orthogonal combinations, solving the problem of difficulty in determining the optimization search boundary when the multiphysics response mechanism is unclear due to the thickening of the insulation system under high voltage conditions, and enabling the simulation data to have a comparable and consistent geometric benchmark among different schemes.

[0033] S3. Discretize the initial value range into several levels, and conduct orthogonal experimental design based on the number of optimization factors and level settings. Perform finite element calculations on each experimental scheme to obtain the simulation response values ​​of all optimization objectives under each experimental scheme.

[0034] Specifically, an orthogonal experimental design was used to select the number of optimization factors and the level range to obtain the optimized results of the calculation scheme. The experiment here uses an orthogonal experimental table to test the finite element calculation values ​​of various physical fields under different combinations of parameter values.

[0035] In a specific implementation, the initial value range of each optimization factor is discretized into several levels at equal intervals or according to manufacturing distinguishable step sizes, and a level table is formed. For example, the insulation thickness is 0.10, 0.13, 0.16, and 0.19, the stator slot width is 2, 3, 4, and 5, the stator slot depth is 45, 46, 47, and 48, and the air gap length is 2.0, 2.5, 3.0, and 3.5. Thus, each factor generates a discrete value set, and the elements of the set correspond one-to-one with the finite element geometric parameter fields.

[0036] Based on the number of factors and levels, an orthogonal array is selected to generate an experimental scheme matrix. In this embodiment, there are 4 factors and 4 levels per factor. An L16 orthogonal array is used to obtain 16 rows of schemes. Each row corresponds to a set of parameter values. The factor column and level number are mapped to specific values, so that the scheme number and the geometric parameter set form a definite correspondence. The same finite element modeling process is called for each row of schemes. The insulation layer thickness, slot geometry, slot depth and air gap are updated according to the parameters of that row, and the mesh is reconstructed. The electromagnetic field solution outputs the magnetic flux density distribution in the air gap region, and the air gap magnetic flux density index is obtained according to the specified sampling rules. The electromagnetic loss is used as a heat source to input the thermal field solution, and the winding temperature rise index is obtained from the difference between the highest winding temperature and the ambient temperature. The electromagnetic force or equivalent excitation load is mapped to the structural model and the dynamic response is solved. The vibration acceleration index is obtained from the acceleration amplitude of the monitoring point. Therefore, each scheme forms a set of three response values, which correspond one-to-one with the parameter set of the scheme and are summarized into a response data table. By using orthogonal arrays to perform balanced sampling of parameter combinations, the large parameter space that is difficult to exhaustively search under high voltage conditions is compressed into a finite number of schemes. The response data is generated under the same solution settings and can be used for subsequent unified evaluation and iterative shrinkage, reducing the computational burden caused by a large number of global simulation searches and avoiding insufficient parameter coverage caused by relying solely on experience.

[0037] S4. Normalize the simulation response values, then construct a reference sequence using the ideal optimal values ​​of each optimization objective, calculate the grey relational degree of each experimental scheme relative to the reference sequence using the grey relational coefficient method, and use the grey relational degree as the comprehensive score of the corresponding experimental scheme.

[0038] In one specific embodiment, the gray relational degree is calculated as follows: The grey relational coefficients for each optimization objective are calculated using the following formula: , in, Let Δ be the grey relational coefficient of the j-th experimental scheme on the i-th optimization objective. ij Let be the absolute difference between the reference sequence and the normalized response sequence of the j-th experimental scheme at the i-th optimization objective, and let minΔ and maxΔ be the values ​​of all Δ... ij The minimum and maximum values ​​in the range, where ρ is the resolution coefficient; The formula for calculating grey relational degree is: , Among them, R j Let represent the grey relational degree corresponding to the j-th experimental scheme, and N be the total number of optimization objectives.

[0039] Specifically, a response matrix is ​​constructed for the three types of simulation response values ​​obtained for each experimental scheme. The rows correspond to the scheme number, and the columns correspond to the three indicators: air gap magnetic flux density, winding temperature rise, and vibration acceleration amplitude. Since the dimensions and magnitudes are different, direct comparison cannot reflect the comprehensive deviation under multi-physics coupling. Therefore, the maximum and minimum values ​​of each column are taken and linear normalization is performed. After normalization, each scheme still maintains the data correspondence of one row and three columns, and the values ​​of each column fall into the same scale.

[0040] A reference sequence is constructed using the ideal optimal values ​​of each optimization objective, and set to 1 under a normalized scale. The difference matrix Δ is obtained by calculating the absolute difference between the normalized response sequence of each scheme and the reference sequence at the same index position. ij Subsequently, in all Δ ij The minimum value minΔ and the maximum value maxΔ are taken as the global scale, and the grey relational coefficient of each scheme on each index is calculated in combination with the resolution coefficient ρ. This makes the index position with a smaller difference correspond to a larger value. And maintain the correspondence between "scheme and indicator" unchanged. The grey relational degree R is obtained by arithmetically averaging the N grey relational coefficients within the same scheme row. j This is used as the overall score for the scheme. The response matrix is ​​then normalized to obtain a normalized matrix, and the difference is calculated to obtain Δ. ij Matrix and mapping The matrix is ​​finally compressed into a comprehensive score vector R. j The greater the grey relational degree, the closer the optimization objective is to the ideal value.

[0041] The comprehensive score corresponds one-to-one with the scheme number and can be directly used as input for subsequent range analysis, forming a unified multi-indicator measurement link that does not rely on subjective weights. This helps alleviate the difficulty in scheme selection caused by the incomparability of multiple objectives and inconsistent evaluation standards.

[0042] In a preferred embodiment, the resolution coefficient in the grey relational coefficient method is set to 0.5.

[0043] Specifically, in this embodiment, the resolution coefficient is set to 0.5, which is used to balance the difference amplification capability and anti-interference capability between different experimental schemes in the calculation of grey relational coefficients, so that the impact of difference changes on the correlation coefficient remains moderate, avoiding excessive compression or amplification of differences, and meeting the evaluation requirements when there are fluctuations in multi-physics response.

[0044] In addition, the resolution coefficient values ​​in the range of 0.3 to 0.7 are all within the conventional selectable range of the grey relational coefficient method, with 0.5 being a commonly used neutral value in engineering applications.

[0045] S5. Based on the comprehensive score, calculate the average score of the same optimization factor at different levels, and then obtain the range of each optimization factor. Optimization factors with a range greater than a preset threshold are assigned to the iteration layer, and the rest are assigned to the output layer.

[0046] In one specific embodiment, the process of performing step S5 may specifically include the following steps: For any optimization factor, all experimental schemes are grouped according to the level of that optimization factor; The arithmetic mean of the comprehensive scores of each experimental scheme at the same level is taken as the average score of any optimization factor at the corresponding level. Calculate the difference between the maximum and minimum average scores, and use it as the range of any optimization factor.

[0047] Specifically, the range of the comprehensive scores of experimental schemes corresponding to a single optimization factor at different levels is determined, and this range is used as the stratification criterion. The comprehensive score reflects the overall closeness of the experimental scheme to the ideal state under multiple optimization objectives, while the dispersion of the comprehensive scores of the same optimization factor at different levels reflects the sensitivity of the factor to the comprehensive performance of multiphysics. Optimization factors with large ranges indicate that their different levels have significantly different impacts on comprehensive performance, and the corresponding parameter ranges still need to be further refined; optimization factors with ranges below a set threshold indicate that their optimal level has tended to stabilize, and the optimal level obtained in the current experiment is used as the output layer parameter; optimization factors with ranges above the set threshold are assigned to the iteration layer and enter subsequent level reconstruction.

[0048] In the specific implementation, the comprehensive score obtained by each experimental scheme in step S4 corresponds one-to-one with the scheme number, and the scheme number corresponds one-to-one with the factor level combination in the orthogonal table. Therefore, the comprehensive score is used as the response quantity and backfilled into the orthogonal table to form a mapping table of "level combination - comprehensive score".

[0049] For a given optimization factor, the level number in the column containing the factor is used as the grouping key. All schemes are divided into several groups according to the level of the factor. For example, if the factor has four levels, there are four groups, and each group contains several scheme numbers. The other factor levels of the schemes within a group are kept orthogonally balanced to reduce the bias of the interaction term on the statistic. For each group, the comprehensive score of all schemes within the group is extracted, and the arithmetic mean is calculated to obtain the average score of the factor at that level. Each level of the factor corresponds to an average score, thus forming a correspondence of "factor-level-average score". The maximum and minimum values ​​of multiple average scores for the same factor are taken, and the difference is calculated to obtain the range of the factor. The range reflects the sensitivity of the comprehensive score to changes in the factor's different levels.

[0050] The range of each factor is compared with a preset threshold. Factors with a range greater than the threshold are marked as iterative layers and retained for subsequent dynamic space reconstruction to further refine the search. Factors with a range not greater than the threshold are marked as output layers and fixed at the current optimal level corresponding to the maximum average score. This layering is used to explicitly screen out the "key parameters" that are difficult to define under multi-physics coupling by the statistical results driven by the comprehensive score, thereby avoiding the problem of excessive computation caused by relying on experience to select optimization focus or repeated simulation in the full parameter space.

[0051] The preset threshold is a criterion parameter used to determine the sensitivity of optimization factors, and its setting can be adjusted according to the requirements of optimization accuracy.

[0052] S6. For the optimization factor of the iterative layer, take its current optimal level as the center, adjust the level setting of the optimization factor according to the preset dynamic space reconstruction rules, and keep the optimal level of the optimization factor of the output layer unchanged to form a new round of orthogonal experimental scheme.

[0053] In one specific embodiment, for any optimization factor, the level corresponding to the maximum average score is determined as the current optimal level of any optimization factor.

[0054] In one specific embodiment, the dynamic space reconstruction rule is as follows: for any iteration layer optimization factor, taking the current optimal level of any iteration layer optimization factor as the center, the value range of the current optimal level is narrowed and re-discrete into several new levels.

[0055] Specifically, step S5 obtains a mapping table of "optimization factor - level - average score" and simultaneously obtains the range and stratification label of each optimization factor. The output layer optimization factor takes the level with the highest average score in this mapping table and locks it as a fixed level. This locking action is equivalent to hardcoding the values ​​of each experimental scheme of the factor to the same level number in the new round of level table and keeping the corresponding geometric parameter values ​​unchanged, thus ensuring that the new round of orthogonal experiments revolves around only a small number of parameter changes. For the iterative layer optimization factor, the level with the highest average score is located according to the mapping table as the current optimal level, and this level number is mapped back to the actual parameter value of the factor as the center point. Then, according to the dynamic space reconstruction rule, the current value range span and adjacent level step size of the factor are read and a new, reduced range is generated. The new range is then re-discrete into several new levels to form a new level set. The new levels corresponding to the center point maintain the same order or value as the current optimal level to ensure inheritance. Other new levels are symmetrically distributed on both sides of the center point or expanded according to the step size to form a new discrete value sequence.

[0056] Taking insulation thickness as an example, if the previous round's four levels were 0.10, 0.13, 0.16, and 0.19, and the highest average score corresponded to 0.16, then the level corresponding to 0.16 is determined as the current optimal level and used as the center point. The interval is then shrunk from 0.10 to 0.19 to 0.13 to 0.19 and re-discrete into 0.13, 0.15, 0.17, and 0.19, or the number of levels and step size are adjusted according to the rules. The air gap length, slot width, and slot depth also generate new levels according to their respective center points and new ranges.

[0057] By combining the fixed level of the output layer with the new level set of the iteration layer to form a new level table, and then generating a new scheme matrix according to the same orthogonal table while maintaining the correspondence between "scheme number - factor level combination - geometric parameter value" and passing it to the finite element modeling input, the problem of high computational cost and difficulty in fast convergence caused by repeated simulation in a large parameter space can be solved.

[0058] Preferably, the principle for reselecting the optimization factor levels relies on a dynamic spatial reconstruction strategy centered on the optimization parameters obtained from the experiment. This involves reconstructing the range of design parameters in the optimization space, updating the design factor levels, and redesigning the hybrid orthogonal experimental scheme. Here, the main purpose of spatial reconstruction is to reduce the discreteness of the optimization scheme; therefore, it is assumed that the multiple level values ​​of the optimization factors are monotonically increasing or decreasing. Assuming the step size between levels in the first Taguchi experiment is d, and based on the fact that the difference between the optimal and suboptimal schemes obtained in the first Taguchi experiment is less than or equal to half of the optimization factor interval, the number of configured levels is equal to the number of levels in the first experiment. The configured step size is the first experiment step size divided by the number of iterations. The optimization center is the optimal value x. When the difference between the optimal and suboptimal schemes is greater than half of the optimization factor interval, the number of configured levels is 4 / 3 times the number of levels in the first experiment, rounded up. The configured step size is 3 / 2 times the original step size, and a hybrid orthogonal array is constructed.

[0059] In a preferred embodiment, the dynamic space reconstruction rules include: For any iteration layer optimization factor, calculate the difference between the current best level and the current second best level of any iteration layer optimization factor, and determine whether the difference is less than half of the range of values ​​of any iteration layer optimization factor. If so, during reconstruction, maintain or increase the number of levels, and reduce the step size between adjacent level values ​​with the current best level as the center; If not, during reconstruction, increase the number of levels and expand the step size between adjacent level values, using the current best level as the center.

[0060] Specifically, the optimization factors in the iterative layer have formed a "level-average score" correspondence table in the previous round of orthogonal experiments, and based on this, the current optimal level corresponding to the maximum average score and the current second-best level corresponding to the second-largest average score are obtained. The factor level number is also mapped to the actual parameter value of the factor. Therefore, the difference between the two parameter values ​​is obtained and compared with half of the current value range of the factor. The range is given by the difference between the maximum and minimum values ​​of the factor in the level table of this round and corresponds one-to-one with the factors.

[0061] When the difference is no greater than half the span, the optimal and second-best values ​​fall relatively close within the interval. During reconstruction, the number of levels remains unchanged or is slightly increased according to the rules, and the current optimal value is used as the center point. The new interval is shrunk to a narrower range with the center point as the axis of symmetry. At the same time, the step size between adjacent levels is set to a fraction of the previous step size to increase sampling density, making the new level set denser around the center point and maintaining a monotonic sequence for easy orthogonal array encoding. When the difference is greater than half the span, the optimal and second-best values ​​are scattered on both sides of the interval, reflecting insufficient coverage of the current discrete levels. During reconstruction, the number of levels is increased, and the value points are expanded to both sides with the center point as the reference. At the same time, the step size of adjacent levels is set to a multiple of the previous step size to widen the coverage range and re-discrete into a new level sequence, so that the new level includes values ​​near the center point and covers a wider search boundary.

[0062] This branch rule transforms the "two sets of horizontal information obtained from the score ranking" into a reconstruction decision of "range shrinkage or range expansion and horizontal densification or sparsity". Under the premise of controlling the number of finite element simulations, it can refine or complete the coverage of the sensitive parameters of multi-physics coupling of high-voltage synchronous motors in a targeted manner, and alleviate the problem of excessive computation caused by full parameter space traversal and the problem of rough parameter intervals caused by empirical point selection.

[0063] S7. Based on the new round of orthogonal experimental scheme, return to S3 for iterative calculation until the preset convergence condition is met, and output the final combination of structural parameters.

[0064] In one specific embodiment, the preset convergence condition is that the range of all optimization factors in the iterative layers is less than a preset convergence threshold.

[0065] Specifically, the new round of orthogonal experimental schemes is generated by the fixed level of the output layer factor and the new level table after the reconstruction of the iterative layer factor. The scheme number and the combination of factor level are kept in one-to-one correspondence and mapped to the synchronous motor geometric parameter input. The finite element solution outputs three types of simulation response values ​​for each scheme: air gap magnetic flux density, winding temperature rise and vibration acceleration amplitude, and backfills to form a new round of response matrix. After the response matrix is ​​normalized according to the index column, the reference sequence is constructed with the ideal optimal value to calculate the difference matrix and obtain the grey relational coefficient matrix. The average value is calculated by row to obtain the comprehensive score vector of each scheme and backfills to the orthogonal table. For each iterative layer factor, the comprehensive score of the schemes within the group is extracted according to its level, and the arithmetic mean is taken to form a new round of "level-average score" mapping for the factor. The difference between the maximum and minimum average scores in this mapping is taken to obtain the range of the factor and compared with the convergence threshold. If the range of any iterative layer factor is not less than the threshold, the level reconstruction is triggered again and the orthogonal experiment generation and finite element calculation link is returned. If the ranges of all iterative layer factors are less than the threshold, it is regarded as level sensitivity convergence, and the combination of structural parameter values ​​corresponding to the fixed level of the output layer and the current optimal level of the iterative layer is output as the structural parameter combination. In this way, range convergence replaces exhaustive simulation of the whole space to reduce the amount of computation and alleviate the problem of difficulty in determining the parameter search boundary caused by multi-physics coupling under high voltage conditions.

[0066] The preset convergence threshold is a positive control parameter used to determine whether the iteration has converged. It is preset by the user according to the required optimization accuracy of the multiphysics synthesis performance. In one specific embodiment, this threshold can be set to 0.1. Typically, its value can be selected between 0.05 and 0.15. The setting of this threshold can also be related to the fluctuation range of the overall score of the initial experimental scheme. By adjusting this threshold, a trade-off can be made between the accuracy of the optimization results and the computational cost.

[0067] Finally, ablation experiments can be designed to verify the optimization effect. The baseline group uses the traditional Taguchi method without grey relational analysis or iteration; Experiment 1 obtains results through grey relational analysis without sequential iteration; Experiment 2 removes grey relational analysis, uses entropy weighting for weight evaluation, and performs experimental iteration. Experiment 4 uses grey relational analysis + hierarchical optimization iteration adopted in this invention. The iteration strategy for Experiment 3 is as follows: 1. Treat the signal-to-noise ratio (SNR) of each performance indicator as a column of a decision matrix, with each experimental scheme as a row; 2. Standardize the decision matrix (the standardization method may need to be adjusted depending on the SNR's tendency to increase or decrease); 3. Calculate the entropy value of each indicator and then calculate the weight. The weighted SNR (weighted sum) of each experimental scheme is calculated. The influence of each factor on the overall SNR is analyzed to determine the optimal factor level combination. Based on the optimal factor level combination of this iteration, the level range is narrowed down, centered on the current optimal level.

[0068] To facilitate understanding of the implementation process of the method of the present invention and the data correspondence between each step, specific embodiments are described below with reference to exemplary parameters. Note: The following data are exemplary and not actual motor or experimental values. The data are for reference only, and parameters may be unreasonable.

[0069] I. Set the initial parameter values ​​for the motor. Here, we take a 20kV electrically excited synchronous motor as an example.

[0070] II. The air gap magnetic flux density B of the motor electromagnetic field, the winding temperature rise T of the temperature field, and the vibration acceleration amplitude A of the structural field are selected as optimization targets.

[0071] The initial structural parameters of the high-voltage synchronous motor were selected and calculated using the finite element method. Here, the motor insulation thickness d is selected. i The four parameters—stator slot width bs, stator slot depth hs, and air gap length δ—are used as optimization factors, with each variable having four horizontal values ​​within the preliminary design range. The parameter settings are shown in Table 1.

[0072] Table 1 III. Design an orthogonal experimental scheme and perform normalization. The parameters are shown in Table 2.

[0073] Table 2 IV. Calculate the grey relational coefficient C and its average grey relational degree R(k) using the formula. The parameters are shown in Table 3.

[0074] Table 3 5. Determine the range of grey relational degree at different levels of the optimization factor, and divide the iterative layer and output layer. The parameters are shown in Table 4.

[0075] Table 4 VI. Implementing a strategy for reconstructing the dynamic space of optimization factors Optimize and reselect di and δ. The maximum grey relational degree of di is 0.74, corresponding to a level value of 0.13; the second largest relational degree is 0.64, corresponding to a level value of 0.16; the difference in level values ​​is 0.03, which is less than half of the optimization range of hs (0.09). The maximum grey relational degree of bs is 0.68, corresponding to a level value of 2; the second largest relational degree is 0.65, corresponding to a level value of 5; the difference in level values ​​is 3, which is greater than half of the optimization range of hs.

[0076] Therefore, the new optimization parameters are shown in Table 5.

[0077] Table 5 Redesign the orthogonal experiment, here for L6. 4 Orthogonal experimental table. Through the second iteration, it was found that the grey relational degree range was less than 0.1, and the iteration ended.

[0078] VII. Ablation Experiment Design No ablation experiments were performed here, but the overall approach to the ablation experiments is as described above. It can be inferred that, as shown in Table 6, the accuracy of experimental group 4 is significantly higher than that of experimental groups 1 and 2, and similar to that of experimental group 3, but the number of calculations is significantly lower than that of experimental group 3.

[0079] Table 6 The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A multiphysics optimization method for a high-voltage synchronous motor, characterized in that, The method includes: S1. Determine multiple physical field performance indicators to be optimized as optimization targets; S2. Select multiple structural parameters to be optimized as optimization factors, and set the initial value range of each optimization factor; S3. Discretize the initial value range into several levels, and conduct orthogonal experimental design based on the number of optimization factors and level settings. Perform finite element calculations on each experimental scheme to obtain the simulation response values ​​of all optimization objectives under each experimental scheme. S4. Normalize the simulation response value, then construct a reference sequence with the ideal optimal value of each optimization objective, calculate the grey relational degree of each experimental scheme relative to the reference sequence using the grey relational coefficient method, and use the grey relational degree as the comprehensive score of the corresponding experimental scheme. S5. Based on the comprehensive score, calculate the average score of the same optimization factor at different levels, and then obtain the range of each optimization factor. Optimization factors with a range greater than a preset threshold are assigned to the iteration layer, and the rest are assigned to the output layer. S6. For the optimization factor of the iterative layer, take its current optimal level as the center, and adjust the level setting of the optimization factor according to the preset dynamic space reconstruction rule. The optimization factor of the output layer remains at its optimal level, forming a new round of orthogonal experimental scheme. S7. Based on the new round of orthogonal experimental scheme, return to S3 for iterative calculation until the preset convergence condition is met, and output the final combination of structural parameters.

2. The method according to claim 1, characterized in that, The optimization objectives include at least the air gap magnetic flux density of the electromagnetic field, the winding temperature rise of the temperature field, and the vibration acceleration amplitude of the structural field.

3. The method according to claim 1, characterized in that, The optimization factors include at least the insulation thickness, stator slot width, stator slot depth, and / or air gap length of the synchronous motor.

4. The method according to claim 1, characterized in that, The method for calculating grey relational degree is as follows: The grey relational coefficients for each optimization objective are calculated using the following formula: ， in, Let Δ be the grey relational coefficient of the j-th experimental scheme on the i-th optimization objective. ij Let be the absolute difference between the reference sequence and the normalized response sequence of the j-th experimental scheme at the i-th optimization objective, and let minΔ and maxΔ be the values ​​of all Δ... ij The minimum and maximum values ​​in the range, where ρ is the resolution coefficient; The formula for calculating grey relational degree is: ， Among them, R j Let represent the grey relational degree corresponding to the j-th experimental scheme, and N be the total number of optimization objectives.

5. The method according to claim 1, characterized in that, S5 include: For any optimization factor, all experimental schemes are grouped according to the level of any optimization factor. The arithmetic mean of the comprehensive scores of each experimental scheme at the same level is taken as the average score of any of the optimization factors at the corresponding level. The difference between the maximum and minimum average scores is calculated and used as the range of any of the optimization factors.

6. The method according to claim 1, characterized in that, For any optimization factor, the level corresponding to the maximum average score is determined as the current optimal level of any optimization factor.

7. The method according to claim 1, characterized in that, The dynamic space reconstruction rule is as follows: for any iteration layer optimization factor, taking the current optimal level of any iteration layer optimization factor as the center, the value range of the current optimal level is narrowed and re-discrete into several new levels.

8. The method according to claim 7, characterized in that, The dynamic space reconstruction rules include: For any iteration layer optimization factor, calculate the difference between the current optimal level and the current suboptimal level of any iteration layer optimization factor, and determine whether the difference is less than half of the value range span of any iteration layer optimization factor. If so, during reconstruction, maintain or increase the number of levels, and reduce the step size between adjacent level values ​​with the current best level as the center; If not, during reconstruction, increase the number of levels and expand the step size between adjacent level values, using the current best level as the center.

9. The method according to claim 1, characterized in that, The preset convergence condition is that the range of the optimization factors of all iterative layers is less than the preset convergence threshold.

10. The method according to claim 4, characterized in that, In the grey relational coefficient method, the resolution coefficient is set to 0.5.

Images