Optimization method for multi-working-condition motor

Abstract

The invention discloses an optimization method for a multi-working-condition motor. The optimization method specifically comprises the following steps: S1, obtaining performance index requirements of the motor under multiple working conditions; s2, performing topological design of a motor structure according to performance index requirements; and S3, performing optimization design on the motor performance according to the performance index requirement. Through collaborative optimization of electromagnetism, heat and structure, the barrier of discrete design in the traditional motor design process is broken through, and integrated multi-physical field and multi-objective optimization is realized, so that the whole joint motor is optimal, and the strict requirement of global temperature rise control and the requirement of structural light weight are met at the same time.

Description

Technical Field

[0001] This invention relates to the field of motor design technology, and in particular to an optimization method for multi-condition motors. Background Technology

[0002] Motor optimization aims to achieve the best balance between design goals and constraints, covering aspects such as power density, torque density, efficiency, size, weight, and cost.

[0003] Traditional strategies seek optimal values ​​for parametric geometric variables within a predetermined range. In design, the rated operating point is often used to optimize the motor design to achieve the highest efficiency and performance at the operating point.

[0004] like Figure 1 As shown, the traditional design optimization process follows a linear progression: first, the motor output performance indicators are evaluated; second, the motor electromagnetic scheme is designed and parameters are optimized; third, the motor thermal scheme is designed and temperature rise is verified; and finally, the mechanical structure strength is verified. Currently, most robot joint drive motor development still follows this method. This method is based on finite element simulation, repeatedly performing simulation calculations, which consumes significant time and computing resources. While the designed motor may meet the design specifications, the electromagnetic performance and temperature verification are based on a single equivalent working condition, which cannot fully meet the specific needs of the robot joint, thus limiting its motion performance and endurance. Furthermore, the design process fails to fully integrate the coupling effects of electromagnetic, thermal, and mechanical disciplines, resulting in redundancy and performance imbalances, failing to reach the extreme balance boundaries of motor electromagnetic performance, thermal performance, and lightweight design. Summary of the Invention

[0005] To address the problem of performance imbalance in motor optimization under multiple operating conditions in existing technologies, this invention proposes an optimization method for motors under multiple operating conditions. By synergistically optimizing electromagnetic, thermal, and structural aspects, it breaks through the barriers of discrete design in traditional motor design, achieving integrated multi-physics field and multi-objective optimization. This enables the entire joint motor to reach its optimal state while meeting the stringent requirements of global temperature rise control and lightweight structure.

[0006] To achieve the above objectives, the present invention provides the following technical solution:

[0007] An optimization method for multi-condition motors specifically includes the following steps:

[0008] S1: Obtain the performance requirements of the motor under multiple operating conditions;

[0009] S2: Design the motor structure topology according to performance requirements;

[0010] S3: Optimize the design of motor performance based on performance requirements.

[0011] Preferably, in S1, the multi-function includes static operating conditions, dynamic motion operating conditions, load change operating conditions, and environmental adaptability operating conditions; the performance index requirements include efficiency, dynamic response, temperature rise and thermal stability, overload capacity, power density, and reliability.

[0012] Preferably, S2 includes:

[0013] S2-1: First, select the basic motor type, which includes permanent magnet synchronous motor, asynchronous motor and switched reluctance motor;

[0014] S2-2: Design the structure of the motor;

[0015] S2-3: Perform secondary optimization of the motor structure based on the operating conditions.

[0016] Preferably, in step S2-2, the structural design method of the motor is as follows:

[0017] The motor's winding design employs short-pitch, distributed windings, and multi-branch windings; the magnetic circuit design uses a salient pole / non-salient pole structure; the permanent magnets are built-in; the cooling structure uses water-cooled / oil-cooled channels; and the stator and rotor cogging torque is reduced through skewed slots and unequal slot number design.

[0018] Preferably, in steps S2-3, the secondary optimization method includes:

[0019] When the motor is in a wide speed range, the salient pole ratio is weakened to reduce iron loss at high speed.

[0020] When the motor is in a frequent start-stop scenario, the rotor stiffness is enhanced to reduce mechanical shock;

[0021] When the motor is under variable load, the air gap length is optimized to balance low load efficiency and high load output capability.

[0022] Preferably, S3 includes:

[0023] S3-1: First, construct a first training set for training the electromagnetic field proxy model, a second training set for training the temperature field proxy model, and a third training set for training the structural field proxy model;

[0024] S3-2: Deep reinforcement learning is used to learn from the first training set, the second training set, and the third training set to construct electromagnetic field proxy models, temperature field proxy models, and structural field proxy models;

[0025] S3-3: The performance parameters of the motor are calculated based on the electromagnetic field proxy model, temperature field proxy model, and structural field proxy model to obtain the comprehensive optimization target;

[0026] S3-4: Combine the non-dominated sorting genetic algorithm with deep reinforcement learning methods, and then iteratively optimize the comprehensive optimization objective;

[0027] S3-5: Output motor design scheme, including electromagnetic design scheme, thermal design scheme and structural design scheme.

[0028] Preferably, in step S3-1, the electromagnetic field distribution under different structural parameters and excitation conditions is calculated using the finite element method, and the electric field strength, magnetic field strength, and magnetic flux density are obtained as the first training set.

[0029] A mathematical model for calculating the temperature field under different heat source layout conditions is constructed. Then, random sampling of sequential layout and random sampling of Gibbs layout are used to obtain samples of various heat source layouts. The temperature field distribution of different samples is then calculated through numerical simulation to form a second training set.

[0030] Based on finite element analysis, different loads and boundary conditions are applied to the structure to calculate the stress, strain, displacement and other responses of the structure, and these data are used as the third training set.

[0031] Preferably, in S3-2, the key constraint condition of the electromagnetic field proxy model is Maxwell's equations:

[0032]

[0033] In formula (1), R err1 This indicates an emphasis on peak field strength; E represents the predicted peak electric field value. peak λ represents the true peak value of the electric field; λ represents the variance weight, which penalizes the overall distribution bias; Var() represents the statistical function for calculating the variance of the sample data; The value of the predicted electric field distribution is represented by ; E represents the actual electric field distribution.

[0034] The key constraint of the temperature field proxy model is the heat conduction equation:

[0035]

[0036] In formula (2), R err2 This represents the boundary temperature error; N represents the total number of discrete points in the temperature field. T represents the predicted temperature value at the i-th discrete point; i This represents the true temperature value at the i-th discrete point; Indicates the predicted temperature at the boundary; T bound This represents the true temperature value at the boundary; μ is the boundary weight, emphasizing the accuracy of the boundary conditions.

[0037] The key constraints of the structural field proxy model are the equilibrium equations and the material yield conditions:

[0038]

[0039] In formula (3), R err3 Indicates the stress at the critical point; σ represents the predicted stress value at the i-th critical point; i σ represents the true stress value at the i-th critical point; yield I represents the yield strength of the material; I() represents the indicator function; ν represents the penalty coefficient, which deducts points if the predicted stress exceeds the yield limit.

[0040] Preferably, in step S3-3, the performance parameters of the motor include output power, efficiency, and torque ripple;

[0041] The formula for calculating output power is as follows:

[0042] P out =T×n / 9550 (4)

[0043] In formula (4), P out Indicates output power; T represents output torque; n represents rotational speed;

[0044] The formula for calculating efficiency is as follows:

[0045] η = P out / P in ×100% (5)

[0046] In formula (5), η represents efficiency; P out Indicates output power; P in Indicates input power;

[0047] The formula for calculating torque ripple is as follows:

[0048] ΔT=max(T i )-min(T i (6)

[0049] In formula (6), ΔT represents torque ripple; T i This represents the instantaneous torque at speed i.

[0050] Finally, the various performance parameters of the motor are weighted and combined to form a comprehensive optimization objective:

[0051]

[0052] In formula (7), F(x) is the comprehensive optimization objective; x is the optimization variable, including the rated torque, speed, efficiency, volume, and mass of the motor; w is the normalized value of the i-th performance metric; i Let i be the weight of the i-th indicator. n represents the total number of indicators.

[0053] Preferably, in steps S3-5, the output parameters of the electromagnetic design scheme include geometric parameters, electrical parameters, and performance parameters; the output parameters of the thermal design scheme include heat dissipation structure parameters and thermal performance parameters; and the output parameters of the structural design scheme include geometric and material parameters and mechanical performance parameters.

[0054] In summary, by adopting the above technical solution, the present invention has at least the following beneficial effects compared with the prior art:

[0055] This invention breaks through the barriers of discrete design in traditional motor design by synergistically optimizing electromagnetic, thermal, and structural aspects, achieving integrated multi-physics field and multi-objective optimization. This enables the entire joint motor to reach its optimal state while meeting the stringent requirements of global temperature rise control and lightweight structure. Attached image description:

[0056] Figure 1 This is a schematic diagram of the existing technology for motor optimization.

[0057] Figure 2 This is a schematic diagram of an optimization method for a multi-condition motor according to an exemplary embodiment of the present invention.

[0058] Figure 3 This is a schematic diagram illustrating the process of sampling the parameter space using the Latin hypercube sampling method according to an exemplary embodiment of the present invention. Detailed Implementation

[0059] The present invention will be further described in detail below with reference to embodiments and specific implementation methods. However, this should not be construed as limiting the scope of the above-described subject matter of the present invention to the following embodiments; all technologies implemented based on the content of the present invention fall within the scope of the present invention.

[0060] In the description of this invention, it should be understood that the terms "longitudinal", "lateral", "up", "down", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this invention.

[0061] This invention provides an optimization method for multi-condition motors, specifically including the following steps:

[0062] S1: Obtain the performance requirements of the motor under multiple operating conditions.

[0063] In this embodiment, the multi-worker includes static working conditions, dynamic motion working conditions, load change working conditions, and environmental adaptability working conditions.

[0064] In this embodiment, the performance requirements include efficiency, dynamic response, temperature rise and thermal stability, overload capacity, power density, and reliability.

[0065] S2: Design the motor structure topology according to performance requirements.

[0066] S2-1: First, select the basic motor type.

[0067] In this embodiment, the motor types include permanent magnet synchronous motors, asynchronous motors, and switched reluctance motors.

[0068] Among them, permanent magnet synchronous motors are suitable for high-precision and high-efficiency scenarios (such as servo systems), and require optimized magnet layout to suppress torque ripple; asynchronous motors are suitable for strong overload and harsh operating conditions (such as industrial drives), with no permanent magnets in the rotor and high reliability; switched reluctance motors are suitable for high-speed and wide-range speed regulation scenarios, with a simple structure, but require suppression of torque pulsation. Performance under different operating conditions can be verified through simulation (such as finite element analysis) to determine the motor type.

[0069] S2-2: Design the structure of the motor.

[0070] In this embodiment, the motor winding design employs short-pitch, distributed windings to reduce harmonic losses; multiple branch windings adapt to current distribution under varying loads; the magnetic circuit design uses a salient pole / non-salient pole structure to accommodate different speed ranges (non-salient poles are more suitable for high speeds); the permanent magnets are internally mounted (to enhance demagnetization resistance and adapt to impact loads); the cooling structure (due to significant temperature differences under various operating conditions) can utilize water-cooled / oil-cooled channels, or optimize the heat dissipation fin layout to enhance heat dissipation under high loads; skewed slots and unequal slot counts reduce stator and rotor cogging torque, improving low-speed stability. Performance under different operating conditions can be verified through simulation (such as finite element analysis) to determine the motor's structure.

[0071] S2-3: Perform secondary optimization of the motor structure based on the operating conditions.

[0072] In this embodiment, the motor needs to operate under different environmental conditions, therefore, the motor structure needs to be optimized again. For example, when the motor is in a wide speed range scenario, the saliency ratio should be weakened to reduce iron loss at high speeds;

[0073] When the motor is in a frequent start-stop scenario, the rotor stiffness should be strengthened to reduce mechanical shock; when the motor is in a variable load scenario, the air gap length should be optimized to balance low load efficiency and high load output capability.

[0074] S3: Optimize the performance of the motor according to the performance requirements.

[0075] S3-1: First, construct a first training set for training the electromagnetic field proxy model, a second training set for training the temperature field proxy model, and a third training set for training the structural field proxy model.

[0076] In this embodiment, the first training set typically uses numerical simulation software such as the finite element method to calculate the electromagnetic field distribution under different structural parameters and excitation conditions, and obtains data such as electric field strength, magnetic field strength, and magnetic flux density as a dataset. Alternatively, actual electromagnetic field data can be obtained through experimental measurement, such as using an electromagnetic field probe to measure the electromagnetic field value in a specific area, but this is costly and inefficient. Existing datasets can also be used, such as electromagnetic field datasets constructed for devices such as microwave filters in some studies; there are also electromagnetic field datasets for wireless charging of electric vehicles, which contain electric and magnetic field data under different charging distances.

[0077] In this embodiment, the second training set can first construct mathematical models for temperature field calculation under different heat source layout conditions, and then use methods such as sequential layout random sampling and Gibbs layout random sampling to obtain various heat source layout samples. Finally, numerical simulation is used to calculate the temperature field distribution of different samples, forming a dataset. For example, the IDRL team at the National Defense Science and Technology Innovation Research Institute has constructed a standard dataset for thermal layout temperature field prediction research, providing 2000 training samples and 40000 test samples for layout problems under three different boundary conditions. In addition, NVIDIA has a dataset based on OpenFOAM simulation data for data center airflow and temperature field prediction.

[0078] In this embodiment, the third training set can be based on finite element analysis, applying different loads (such as tension, pressure, torque, etc.) and boundary conditions to the structure, calculating the stress, strain, displacement, and other responses of the structure, and using these data as a dataset; it can also be combined with experimental testing, such as measuring the surface strain of the structure through strain gauges; or in the design of the hull structure, the potential flow method (PFM) and viscous flow method (VFM) can be used to obtain hydrodynamic load data and construct a dataset for predicting the hydrodynamic drag of the hull structure.

[0079] S3-2: Deep reinforcement learning is used to learn from the first training set, the second training set, and the third training set to construct electromagnetic field proxy models, temperature field proxy models, and structural field proxy models.

[0080] In this embodiment, the goal of the surrogate model is to replace traditional numerical simulations (such as finite element method and finite difference method) with DRL (Deep Reinforcement Learning) to quickly predict the field distribution (such as electromagnetic field strength, temperature distribution, and structural stress) under given input parameters. The input consists of the design parameters of the physical system, including geometric dimensions, material properties, boundary conditions, etc.; the output consists of key indicators of the field distribution or the overall field distribution, including peak value, mean value, gradient, etc.

[0081] The DRL framework is as follows: the agent selects a combination of input parameters to obtain the field distribution of the environment (real physical field or high-precision simulation), optimizes the strategy through a reward mechanism, and finally achieves efficient mapping from input to output.

[0082] In this embodiment, the key constraint of the electromagnetic field proxy model is Maxwell's equations, such as... The curl represents the vortex characteristic of a vector field; H represents the magnetic field strength vector (unit: A / m), an auxiliary quantity introduced when calculating the magnetic field; J represents the conduction current density (unit: A / m). 2 The electric current distribution is described by the directional movement of free charges; D represents the electric displacement vector (electric flux density, unit: C / m). 2 ); Displacement current density (unit: A / m) 2 The electric field is a "virtual current" introduced by Maxwell, which describes a changing electric field as equivalent to an electric current, and reflects the law that changes in the electric field induce a magnetic field.

[0083] R err1 Key indicators such as peak field strength and energy density can be emphasized:

[0084]

[0085] In formula (1), R err1 This indicates a reward for prediction error, which focuses on peak field strength and energy density. E represents the predicted peak electric field value, in V / m. peak λ represents the true peak value of the electric field, in V / m; λ represents the variance weight, penalizing the overall distribution bias; Var() represents the statistical function for calculating the variance of the sample data. V represents the predicted value of the electric field distribution across the entire field, in V / m; E represents the actual value of the electric field distribution across the entire field, in V / m.

[0086] In this embodiment, the key constraint of the temperature field proxy model is the heat conduction equation: ρ represents the density of a substance (unit: kg / m³). 3), where represents the mass of a substance per unit volume, reflecting the mass distribution characteristics of the substance; c represents the specific heat capacity of the substance (unit: J / (kg·K)), representing the amount of heat required to raise the temperature of a unit mass of the substance by 1 Kelvin, measuring the substance's ability to store heat; T represents temperature (unit: K or ℃), describing the degree of hotness or coldness of the substance; t represents time (unit: s), representing the time dimension of the process, used to describe the change of the unsteady temperature field over time; k represents thermal conductivity (unit: W / (m·K)), a physical quantity characterizing the thermal conductivity of a substance; the higher the thermal conductivity, the better the thermal conductivity of the substance. Divergence representing heat flux density (unit: W / m³) 3 This describes the degree of heat convergence or dissipation per unit volume at a point in space, reflecting the rate of energy change caused by heat conduction. The temperature gradient (unit: K / m) represents the rate of change of temperature in space. Heat flux density vector (unit: W / m) 2 (This describes the direction and intensity of heat transfer.) q represents divergence (describing the "source / sink" characteristics of a vector field); q represents the intensity of the internal heat source (unit: W / m). 3 (), represents the rate at which a substance generates or consumes heat per unit volume.

[0087] R err2 Boundary temperature error can be included:

[0088]

[0089] In formula (2), R err2 This represents the boundary temperature error; N represents the total number of discrete points in the temperature field, such as the number of grid nodes. T represents the predicted temperature value at the i-th discrete point; i This represents the true temperature value at the i-th discrete point, derived from experimental measurements or high-precision simulations, such as finite element results. Indicates the predicted temperature at the boundary; T bound This represents the actual temperature value at the boundary, such as the set or measured value of an isothermal boundary or an adiabatic boundary; μ is the boundary weight, emphasizing the accuracy of the boundary conditions.

[0090] In this embodiment, the key constraints of the structural field proxy model are the equilibrium equations and the material yield conditions: σ represents the divergence operator, used to describe the divergence characteristics of the stress tensor in space (measuring the degree of "convergence / divergence" of stress at a point); σ represents the stress tensor (unit: Pa or MPa), a second-order tensor describing the force state inside an object, characterizing the force per unit area at a point within the object; f represents the volume force vector (unit: N / m).3 or kg / (m 2 ·s 2 Force refers to the force acting on a unit volume of an object.

[0091] R err3 Focus can be placed on stresses at critical points, such as near the yield limit:

[0092]

[0093] In formula (3), R err3 Indicates the stress at the critical point; σ represents the predicted stress value at the i-th critical point, in Pa or MPa; i σ represents the true stress value at the i-th critical point, in Pa or MPa, derived from experiments or high-precision simulations, such as finite element results; yield I represents the yield strength of the material; I() represents the indicator function; ν represents the penalty coefficient, which deducts points if the predicted stress exceeds the yield limit.

[0094] S3-3: The performance parameters of the motor are calculated based on the electromagnetic field proxy model, temperature field proxy model, and structural field proxy model to obtain a comprehensive optimization target that meets the optimal torque-speed characteristics, thereby improving efficiency.

[0095] In this embodiment, the motor's performance parameters include output power, efficiency (reflecting energy conversion efficiency, a key performance indicator), and torque fluctuation (assessing operational stability).

[0096] The formula for calculating output power is as follows:

[0097] P out =T×n / 9550 (4)

[0098] In formula (4), P out T represents output power; T represents output torque (unit: N·m), corresponding to the torque value on the torque-speed curve; n represents speed (unit: r / min), the speed operating point on the curve.

[0099] The formula for calculating efficiency is as follows:

[0100] η = P out / P in ×100% (5)

[0101] In formula (5), η represents efficiency; P out P represents the output power (unit: kW), the effective power output by the system. in This represents the input power (unit: kW), which is the total power consumed by the system.

[0102] The formula for calculating torque ripple is as follows:

[0103] ΔT=max(T i )-min(T i (6)

[0104] In formula (6), ΔT represents torque ripple (unit: N·m), the maximum range of torque ripple at a certain speed; T i This represents the instantaneous torque at speed i.

[0105] In this embodiment, the various performance parameters of the motor are weighted and combined to form a comprehensive optimization target:

[0106]

[0107] In formula (7), F(x) is the comprehensive optimization objective, which needs to be minimized. The smaller the value, the better the performance. x is the optimization variable, such as the rated torque, speed, efficiency, volume, mass, number of winding turns, wire diameter, magnet size and other design parameters of the motor. The normalized value of the i-th performance index is used to map the original index to the [0,1] interval, eliminating the influence of dimensions; w i Let i be the weight of the i-th indicator. The allocation is based on the working conditions of the joint motor. For example, dynamic response is given priority and has a higher weight. n represents the total number of indicators.

[0108] S3-4: Combine the Non-Dominated Sorting Genetic Algorithm (NSGA-II) with deep reinforcement learning methods, and then iteratively optimize the comprehensive optimization objective.

[0109] S3-4-1: Combine the comprehensive optimization objective with temperature constraints to select the non-dominated solution set.

[0110] In this embodiment, the specific expression for the temperature constraint is:

[0111] T max (x,g)≤T lim (8)

[0112] In formula (8), T max (x,g) represents the highest operating temperature of the motor under the optimized variables x and operating condition g. The peak temperature under all key operating conditions needs to be calculated; T lim The upper limit of the motor's temperature is determined by the insulation class, such as 155℃ for Class F insulation and 180℃ for Class H insulation; 'x' represents optimization variables, such as design parameters like rated torque, speed, efficiency, volume, mass, number of winding turns, wire diameter, and magnet dimensions; 'g' represents operating parameters, such as load torque T. L The parameters, such as rotational speed (n) and operating time (t), need to cover typical operating conditions including static, dynamic, and overload conditions.

[0113] The method for selecting non-dominated solutions using the Non-Dominated Sorting Genetic Algorithm (NSGA-II) is as follows, specifically reflected in the "non-dominated sorting" and "population evolution" stages, with the aim of gradually approaching and ultimately obtaining the Pareto optimal solution set:

[0114] Initialize the population: Randomly generate an initial solution set for the design variable x (satisfying boundary constraints);

[0115] Multi-field simulation evaluation: Call coupled simulation for each x and calculate the objective function value f. j (x) and constraint value g i (x), eliminate solutions that violate the constraints;

[0116] Multi-field coupled models can be established using coupled simulation tools (such as COMSOL Multiphysics); electromagnetic-temperature coupling relationship: electromagnetic loss (Joule heating) serves as the heat source of the temperature field; temperature-structure coupling relationship: temperature changes lead to thermal stress, which affects the stress distribution of the structural field.

[0117] Non-dominated sorting: The solutions are divided into different levels (the Pareto front is the first level), and the solutions within the same level are sorted according to their crowding (to ensure the diversity of solutions);

[0118] Selection and Evolution: Excellent solutions are selected from different levels through roulette wheel selection, and the next generation of population is generated through crossover and mutation.

[0119] Termination conditions: The iteration reaches the maximum number of iterations (e.g., 100 generations) or the Pareto front converges (the change in the solution is <1e-3);

[0120] Pareto optimal solution decision-making: Select the final solution from the Pareto front (such as using the TOPSIS method or fuzzy decision-making method), weigh the priority of the objectives (such as choosing the solution with minimum stress if safety is the priority), and obtain the non-dominated solution set.

[0121] S3-4-2: Local optimization of non-dominated solution sets using deep reinforcement learning methods:

[0122] Using the NSGA-II high-quality solution (i.e., the non-dominated solution set, in electromagnetic-thermal-structural coupling optimization, if a solution cannot be surpassed by other solutions in terms of efficiency, temperature and stress, and satisfies all constraints, it is a non-dominated solution and belongs to the high-quality solution) as a seed, we define the parameter fine-tuning range (state space) and clarify the optimization objectives (such as efficiency and temperature).

[0123] A deep reinforcement learning method is used to iteratively update the optimization target while checking the temperature constraint to avoid infeasible solutions; the iterative update of the deep reinforcement learning method is an existing technology, so it will not be repeated here.

[0124] After the iteration converges, the local optimal solution is output and merged with the original solution set (the non-dominated solution set, i.e. the Pareto front solution set obtained by NSGA-II optimization) to filter the final result and make up for the shortcomings of NSGA-II in local optimization.

[0125] Among them, the merging and filtering rules are as follows:

[0126] (1) Merge the local optimal solution with the original non-dominated solution set, and uniformly evaluate the objective value and constraints of all solutions;

[0127] (2) Eliminate redundant solutions that violate constraints and are dominated, and retain non-dominated solutions;

[0128] (3) If there are too many solutions, retain the sparsely distributed solutions according to the degree of crowding, and eventually form a better Pareto front.

[0129] In this embodiment, a collaborative optimization method using genetic algorithms and deep learning not only considers the real-time state of the motor under multiple operating conditions but also overcomes the limitations of the traditional separation of electromagnetic, thermal, and structural design in motors. This enables motor optimization under conditions of conflicting design objectives, complex operating conditions, and strong constraints. Artificial intelligence technology empowers the motor in the intelligent collaborative optimization design process involving multiple physics fields and multiple objectives, achieving high torque density, high efficiency, stable thermal management, and ultra-lightweight design. This method provides solid technical support for designing high torque density motors for robots.

[0130] S3-5: Output motor design scheme, including electromagnetic design scheme, thermal design scheme and structural design scheme.

[0131] In this embodiment, the output parameters of the electromagnetic design scheme include geometric parameters, electrical parameters, and performance parameters.

[0132] The geometric parameters include coil parameters and core / conductor structure. Coil parameters include: number of turns (N), wire diameter (d). wire ), winding method (such as number of layers, spacing); core / conductor structure includes: core material (μ r Relative permeability), cross-sectional area (A) c ), length (l) c ); tooth pitch (t) and slot dimensions (width w) of conductors (such as motor stators and rotors). s 、deep h s ).

[0133] Among them, electrical parameters include operating current (I rated Rated current), voltage (U) rated ), resistance (R), inductance (L), electromagnetic loss (P) em For example, copper loss P cu =I 2 R, iron loss P fe ).

[0134] Among them, the performance parameters include magnetic field strength (B max (maximum magnetic flux density), electromagnetic force (F) em (e.g., electromagnet attraction force), energy conversion efficiency (η) em = Output electromagnetic power / Input electrical power).

[0135] In this embodiment, the output parameters of the thermal design scheme include heat dissipation structure parameters and thermal performance parameters.

[0136] Among them, the heat dissipation structure parameters include the heat sink size (thickness d) fin Spacing s fin Height h fin ), cooling channel diameter (d) ch ), flow channel layout (e.g., parallel / series), material thermal conductivity (k, e.g., aluminum heat dissipation material k = 202 W / (m·K)), material specific heat capacity (c) p ).

[0137] Among them, the thermal performance parameters include the maximum temperature (T) max ), hotspot locations (e.g., coordinates of a point on the surface of a coil / chip), temperature gradient ( (Temperature change rate along the heat dissipation direction), heat flux density (q, heat dissipation power per unit area, W / m²) 2 Total heat dissipation (Q) total ), thermal resistance (R) th K / W reflects the resistance to heat transfer.

[0138] In this embodiment, the output parameters of the structural design scheme include geometric and material parameters and mechanical performance parameters.

[0139] Among them, the geometric and material parameters include the thickness (t) of the shell and the supporting beam. s ), length (l) s ), cross-sectional shape (e.g., rectangular, circular), material elastic modulus (E), material Poisson's ratio (v), material yield strength (σ) s ), material density (ρ), etc.

[0140] Among them, the mechanical performance parameters include the maximum stress (σ) max ), maximum strain (ε max ), stress concentration factor (K) t Reflects local stress amplification factor and maximum deflection (w) max Such as the bending deformation of a beam), natural frequency (f n To avoid resonance), and the safety factor (S = σ) s / σ max (Requires S≥1.2~2.0).

[0141] This application adopts a strategy that combines NSGA-II with deep reinforcement learning, giving full play to the advantages of traditional multi-objective genetic algorithms in global search and the ability of deep reinforcement learning to finely tune parameters in a high-dimensional continuous parameter space.

[0142] like Figure 3 As shown, in the application of artificial intelligence to motor design and optimization, the entire process based on Latin hypercube sampling parameter space has been realized in terms of datasets: given the geometric design space of the motor, Latin hypercube sampling is used to cover the parameter space to generate samples, FEA simulation is called to obtain key performance indicators, and geometric parameters and key performance indicators are stored.

[0143] Therefore, this application provides strong support in areas such as motor dataset construction, surrogate model training, and optimization algorithms that combine NSGA-II with deep reinforcement learning, and the overall optimization scheme has high technical feasibility.

[0144] Those skilled in the art will understand that the above embodiments are specific examples of implementing the present invention, and in practical applications, various changes in form and detail may be made without departing from the spirit and scope of the present invention.

Claims

1. An optimization method for a multi-operating condition electric machine, characterized in that, Specifically comprising the following steps: S1: obtaining performance index requirements of the motor under multiple working conditions; S2: performing motor structure topology design according to the performance index requirements; S3: performing optimized design of the motor performance according to the performance index requirements.

2. An optimization method for a multi-operating condition electric machine as claimed in claim 1, characterized in that, In the S1, the multiple working conditions include static working condition, dynamic motion working condition, load change working condition and environmental adaptability working condition; the performance index requirements include efficiency, dynamic response, temperature rise and thermal stability, overload capacity, power density and reliability.

3. An optimization method for a multi-duty motor as claimed in claim 1, wherein, The S2 includes: S2-1: first selecting a basic motor type, the motor type including permanent magnet synchronous motor, asynchronous motor and switched reluctance motor; S2-2: designing the structure of the motor; S2-3: performing secondary optimization of the motor structure according to the working condition scene.

4. An optimization method for a multi-duty motor as claimed in claim 3, characterized in that, In the S2-2, the structure design method of the motor is as follows: The winding design in the motor adopts short-pitch, distributed winding and multi-branch winding; the magnetic circuit design adopts salient pole / hidden pole structure; the permanent magnet adopts built-in type; the cooling structure adopts water / oil cooling channel; the tooth slot torque of the stator and the rotor is reduced through skew slot and unequal slot number design.

5. An optimization method for a multi-duty motor as claimed in claim 3, wherein, In the S2-3, the secondary optimization method includes: When the motor is in a wide speed regulation scene, the salient pole rate is weakened, and the iron loss at high speed is reduced; When the motor is in a frequent start-stop scene, the rotor stiffness is strengthened, and mechanical impact is reduced; When the motor is in a variable load scene, the air gap length is optimized, and the low load efficiency and high load output capacity are balanced.

6. An optimization method for a multi-duty motor as recited in claim 1, wherein, The S3 includes: S3-1: first constructing a first training set for training an electromagnetic field proxy model, a second training set for training a temperature field proxy model and a third training set for training a structure field proxy model; S3-2: learning the first training set, the second training set and the third training set by using a deep reinforcement learning method, and constructing the electromagnetic field proxy model, the temperature field proxy model and the structure field proxy model; S3-3: calculating the performance parameters of the motor according to the electromagnetic field proxy model, the temperature field proxy model and the structure field proxy model, and obtaining a comprehensive optimization target; S3-4: combining the non-dominated sorting genetic algorithm and the deep reinforcement learning method, and then iteratively optimizing the comprehensive optimization target; S3-5: outputting a motor design scheme, including an electromagnetic design scheme, a thermal design scheme and a structure design scheme.

7. An optimization method for a multi-duty motor as claimed in claim 6, characterized in that, In the S3-1, the electromagnetic field distribution under different structure parameters and excitation conditions is calculated by using the finite element method, and the electric field intensity, the magnetic field intensity and the magnetic flux density are obtained as the first training set; A mathematical model for calculating the temperature field under different heat source layout conditions is constructed, then a plurality of heat source layout samples are obtained by using sequential layout random sampling and Gibbs layout random sampling method, and the temperature field distribution of different samples is calculated through numerical simulation to form the second training set; Based on the finite element analysis, different loads and boundary conditions are applied to the structure to calculate the stress, strain and displacement responses of the structure, and these data are taken as the third training set.

8. An optimization method for a multi-duty motor as claimed in claim 6, characterized in that, In the S3-2, the key constraint condition of the electromagnetic field proxy model is the Maxwell equation group: In Equation (1), R err1 represents the peak value of the electric field; represents the peak value of the electric field; E peak represents the peak value of the electric field; λ represents the variance weight, which penalizes the overall distribution deviation; Var() represents a statistical function for calculating the variance of sample data; represents the peak value of the electric field; E represents the peak value of the electric field; The key constraint of the temperature field proxy model is the heat conduction equation: In formula (2), R err2 represents the boundary temperature error; N represents the total number of temperature field discrete points; represents the temperature prediction value of the i-th discrete point; T i represents the temperature true value of the i-th discrete point; represents the temperature prediction value at the boundary; T bound represents the temperature true value at the boundary; μ is the boundary weight, emphasizing the accuracy of the boundary condition; The key constraint of the structure field proxy model is the balance equation and the material yield condition: In Equation (3), R err3 represents the stress at the dangerous point; represents the predicted stress value at the i-th dangerous point; σ i represents the true stress value at the i-th dangerous point; σ yield represents the yield strength of the material; I() represents an indicator function; and v represents a penalty coefficient, which is deducted if the predicted stress exceeds the yield limit.

9. An optimization method for a multi-duty motor as claimed in claim 6, characterized in that, In the S3-3, the performance parameters of the motor include output power, efficiency and torque fluctuation; The calculation formula of the output power is: P out = T x n / 9550 (4) In Equation (4), P out represents output power; T represents output torque; and n represents rotational speed. The calculation formula of the efficiency is: η = P out / P in x 100% (5) In Equation (5), η represents efficiency; P out represents output power; P in represents input power; The calculation formula of the torque fluctuation is: ΔT = max(T i )-min(T i ) (6) In Equation (6), ΔT represents torque fluctuation; T i represents the instantaneous torque at i rotational speed; Finally, the various performance parameters of the motor are combined by weighting to form a comprehensive optimization target: In formula (7), F(x) is a comprehensive optimization target; x is an optimization variable, including rated torque, speed, efficiency, volume, and mass of the motor; is a normalized value of the i-th performance index; w i is a weight of the i-th index, n represents the total number of indexes.

10. An optimization method for a multi-duty motor as claimed in claim 6, wherein, In the S3-5, the output parameters of the electromagnetic design scheme include geometric parameters, electrical parameters and performance parameters; the output parameters of the thermal design scheme include heat dissipation structure parameters and thermal performance parameters; and the output parameters of the structural design scheme include geometric and material parameters and mechanical performance parameters.

Images