Model optimization method based on dynamic parameter identification of brushless DC motor products

Abstract

The present invention provides a model optimization method based on the dynamic parameter identification of brushless DC motor products, which belongs to the field of electronic product reliability simulation and includes the following steps: S1, collecting basic information of the brushless DC motor; S2, constructing a simulation model; S3, constructing a proxy model; S4, obtaining reliability enhancement test data; S5, using a reverse optimization algorithm to perform dynamic parameter identification to obtain a dynamic parameter combination; S6, optimizing and verifying the simulation model using the dynamic parameter combination. The simulation model dynamic parameter identification and calibration method proposed in the present invention can greatly improve the authenticity of the simulation model, and provide strong support for product performance evaluation and reliability prediction under complex working conditions. This method effectively improves the response accuracy and model stability of the brushless DC motor under complex working conditions, and provides a solid foundation for the reliability evaluation of avionics products.

Description

Technical Field

[0001] The present invention belongs to the field of electronic product reliability simulation, and in particular relates to a model optimization method based on dynamic parameter identification of brushless DC motor products. Background Art

[0002] Brushless DC motors are widely used in avionics products, providing efficient motor drive support for the system. However, in complex and changing environmental conditions, the key parameters of such motors (such as resistance, inductance, and motor constant) will vary significantly with the environmental conditions, thus affecting their overall performance and reliability. Some key parameters of avionics products will vary significantly with environmental conditions. Because avionics equipment typically operates in extreme and variable environments such as temperature, humidity, air pressure, and electromagnetic interference, these external conditions not only affect product performance but also accelerate component aging and performance degradation. Therefore, establishing a high-fidelity simulation model that can realistically represent the performance changes of the product under complex environmental conditions is crucial for reliability design and performance evaluation.

[0003] Currently, in the process of building a high-fidelity model, key parameters need to be effectively identified and calibrated to ensure the accuracy of the model under different working conditions. This process faces the following three major challenges:

[0004] First, parameters are highly dependent on and diverse operating conditions, and traditional fixed parameter assumptions cannot adapt to the actual operating conditions of avionics products. For example, when a certain type of flight computer operates in a high-temperature environment, the temperature of the core electronic components will rise significantly, resulting in a decrease in data processing speed and an increase in storage unit leakage current. In low-pressure and high-humidity environments, signal transmission speed and electromagnetic compatibility may also change, directly affecting the overall reliability of the system. To this end, simulation models need to be dynamically adjusted according to different operating conditions to truly reflect the performance changes of the product in specific environments. For brushless DC motors, operating conditions such as ambient temperature, humidity, and air pressure will affect their key parameters such as motor constants, resistance, and inductance, and existing fixed parameter assumptions cannot accurately describe these dynamic changes.

[0005] Secondly, the identification and calibration of operating parameters are difficult, and there is currently a lack of multi-operating condition identification technology suitable for complex environments. The operating parameters of avionics products are complex and diverse, including temperature, pressure, humidity, electromagnetic interference, vibration, etc. These variables do not exist independently, but are interrelated. For example, in a high-frequency vibration environment, the effects of temperature and humidity on circuit boards will be superimposed, resulting in a significant increase in the computational complexity of parameter identification. Existing calibration methods usually rely on experimental data, and the variation patterns of experimentally obtained parameters under various operating conditions are often difficult to accurately establish, making the operating condition identification and calibration accuracy unable to meet the requirements of high-fidelity models.

[0006] Third, high-fidelity model calibration experiments are costly and time-consuming. Avionics products typically have high reliability requirements. Traditional model calibration processes require a large amount of real-world test data to support parameter identification, resulting in lengthy testing times and high costs. Multivariable calibration experiments are particularly difficult to implement in complex environments. This often forces parameter adjustments to be made based on small sample sizes or simplified operating conditions, resulting in insufficient adaptability of the model to specific operating conditions.

[0007] To solve the above problems, it is necessary to further develop efficient and accurate multi-condition identification and calibration technologies, and combine simulation data and experimental data to support the construction of high-fidelity models of avionics products in complex environments, and achieve accurate prediction and effective optimization of product performance. Summary of the Invention

[0008] To address the shortcomings of existing technologies, the present invention provides a model optimization method based on dynamic parameter identification of brushless DC motor products. This method samples the product's response under different environmental conditions, combines a proxy model with an optimization algorithm, and accurately calibrates the dynamic parameters to obtain real-time and accurate dynamic parameters. The obtained dynamic parameters are then used to optimize the model, thereby constructing a simulation model that accurately reflects the product's performance. The present invention's dynamic parameter identification method is specifically designed for brushless DC motors in avionics systems. The dynamic parameters of such motor systems vary significantly in complex environments, and existing models have poor fidelity. Therefore, special calibration techniques are required to identify the dynamic parameters and calibrate and optimize the simulation model.

[0009] To achieve the above objectives, the present invention discloses the following technical solutions:

[0010] A model optimization method based on dynamic parameter identification of brushless DC motor products includes the following steps:

[0011] S1. Collect basic information about the brushless DC motor: The basic information about the brushless DC motor includes the motor's CAD model, structural parameters, key performance parameters, power and control parameters, and typical operating environment.

[0012] S2. Build a simulation model: Based on the CAD model of the brushless DC motor, combined with the motor's structural characteristics and electromagnetic dynamics characteristics, build a preliminary simulation model of the motor;

[0013] S3. Build a surrogate model: Based on the response data of the initial simulation model, a surrogate model is built through data fitting technology. The input of the surrogate model is a dynamic parameter combination, and the output is a key performance indicator;

[0014] S4. Obtain reliability enhancement test data: In a laboratory environment, reliability enhancement tests are performed to measure the actual response data of the motor under different operating conditions. The actual response data includes performance response and motor health status data;

[0015] S5. Dynamic parameter identification using a reverse optimization algorithm: Based on the surrogate model, a reverse optimization algorithm is applied to find the dynamic parameter combination that best matches the simulation model response with the actual data using experimental data under different working conditions. This step specifically includes the following sub-steps:

[0016] S51, determine the optimization variables and optimization targets of the reverse optimization algorithm, the optimization variables are the resistance value, the motor constant and the sensitivity coefficient K of the inductance value R , K Ce , K L ,The optimization goal is to minimize the error between the motor response output by the surrogate model and the measured data under all temperature conditions;

[0017] S52. Import the maximum torque and no-load speed data measured under different temperature conditions, construct an objective function, and use the particle swarm algorithm to collaboratively search the solution space with multiple particles, continuously iteratively update, and finally converge to the globally optimal dynamic parameter combination; the dynamic parameter combination includes the resistance value R, motor constant Ce, and inductance value L of the brushless DC motor; the objective function is specifically:

[0018] The objective function is as follows:

[0019]

[0020] Where x = [K R ,K Ce ,K L ] is the sensitivity coefficient to be optimized, T model,j and n model,j are the maximum torque and no-load speed calculated by the proxy model, T meas,j and n meas,j is the corresponding measured data, M is the number of temperature conditions;

[0021] S6. Use dynamic parameter combination to optimize and verify the simulation model.

[0022] Preferably, in step S1, the structural parameters of the motor include winding structure, permanent magnet material, stator and rotor structure dimensions, and air gap length; the key performance parameters of the motor include torque output, speed, efficiency, and temperature change; the power and control parameters of the motor include voltage, current, controller setting value, and excitation mode; typical working environment conditions include temperature, humidity, vibration, and load conditions.

[0023] Preferably, in step S4, the performance response is temperature, no-load speed and maximum torque.

[0024] Preferably, in step S5, the relationship between the resistance value R, the motor constant Ce and the inductance value L of the brushless DC motor and the temperature is as follows:

[0025] R=R0+K R (T-T0)

[0026] Ce=Ce0+K Ce (T-T0)

[0027] L=L0+k L (T-T0)

[0028] Among them, R0, Ce0, and L0 are dynamic parameter values ​​under the condition of T0, and T and T0 are time.

[0029] Preferably, in step S52, each particle represents a possible sensitivity coefficient combination, and the position and velocity of the particle are dynamically adjusted according to the individual optimal solution and the global optimal solution. The velocity update formula is:

[0030] v i,d (t+1)=ωv i,d (t)+c1r1(p i,d -x i,d (t))+c2r2(g d -x i,d (t));

[0031] The position update formula is:

[0032] x i,d (t+1)=x i,d (t)+v i,d (t+1)

[0033] Among them, ω is the inertia weight, c1 and c2 are learning factors, r1 and r2 are random numbers, and p i is the individual optimal position of the particle, g is the global optimal position, and the algorithm gradually converges through multiple iterations and finally finds the optimal sensitivity coefficient combination at the global optimal point.

[0034] Preferably, step S6 is specifically to embed the relationship between the corresponding dynamic parameters and temperature changes into the simulation model after determining the dynamic parameters of the simulation model, and apply it to the motor's resistance value R, motor constant Ce and inductance value L, so that it can be adaptively adjusted with the change of temperature L, so as to simulate the motor response under different temperature conditions.

[0035] Preferably, the simulation model in step S2 is constructed using MATLAB Simulink simulation software, and the simulation model includes an input signal, a load signal, feedback control, a motor principle model, and a reduction device.

[0036] Preferably, the proxy model in step S3 is a neural network architecture model.

[0037] Compared with the prior art, the present invention has the following beneficial effects:

[0038] (1) The proposed method for dynamic parameter identification and calibration of simulation models can significantly improve the realism of simulation models, providing strong support for product performance evaluation and reliability prediction under complex operating conditions. This method effectively improves the response accuracy and model stability of brushless DC motors under complex operating conditions, providing a solid foundation for reliability evaluation of avionics products.

[0039] (2) This paper proposes a brushless DC motor simulation model calibration method based on dynamic parameter identification under environmental working conditions. Aiming at the performance and reliability evaluation requirements of aviation airborne brushless DC motors, a technical process for dynamic parameter identification and calibration of simulation models is established. The study proposes key technical implementation methods such as working condition parameter combination sampling, proxy model construction, reliability enhancement test data acquisition, and dynamic parameter calibration of the reverse optimization algorithm. A dynamic parameter calibration experiment of the simulation model is carried out using a certain type of aviation brushless DC motor as the object, verifying the accuracy and practicality of the model calibration based on this method. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 It is a schematic diagram of the overall process of the present invention;

[0041] Figure 2 This is a functional block diagram of the brushless DC motor of the present invention;

[0042] Figure 3 It is a simulation model diagram of the present invention;

[0043] Figure 4 Schematic diagram of the brushless DC motor working circuit of the present invention;

[0044] Figure 5 This is the no-load input step signal output signal curve of the present invention;

[0045] Figure 6 The motor speed curve of the no-load input step signal of the present invention;

[0046] Figure 7 is the input torque variation curve of the present invention;

[0047] Figure 8a 、 Figure 8b and Figure 8c The resistance, motor constant and inductance of the present invention are sampled and distributed in each interval respectively;

[0048] Figure 9a and Figure 9b They are the idling speed and maximum torque distribution characteristics of the present invention respectively;

[0049] Figure 10a and Figure 10b The proxy model prediction data of the present invention are compared with the measured data respectively;

[0050] Figure 11a and Figure 11b The motor responses under different temperature conditions are measured for the step test of the present invention respectively;

[0051] Figure 12 This is a flow chart of reverse optimization of the coefficients of the dynamic parameter relationship equation of the present invention;

[0052] Figure 13 Schematic diagram for comparing the measured results and simulation results of the present invention. DETAILED DESCRIPTION

[0053] The exemplary embodiments, features, and aspects of the present invention will be described in detail below with reference to the accompanying drawings. The same reference numerals in the accompanying drawings represent elements with the same or similar functions. Although various aspects of the embodiments are shown in the accompanying drawings, the drawings are not necessarily drawn to scale unless otherwise indicated.

[0054] The overall method of the present application is based on a simulation model, which changes the model dynamic parameters respectively, compares the simulation response parameters under different model dynamic parameters, establishes an association relationship between the model dynamic parameters and the simulation response parameters that have an impact relationship, and provides constraint conditions for the establishment of an overall mapping relationship between the two. Secondly, since the product performance model has a large number of model parameters and a wide range of parameter values, parameter identification requires a lot of time and data. A method of establishing a mapping relationship between model dynamic parameters and simulation response parameters using an artificial neural network to construct a proxy model is used to quickly obtain simulation response parameters corresponding to different model dynamic parameter values. Based on the value constraints of the model dynamic parameters, multiple groups of dynamic parameter values ​​are generated, and the parameter values ​​are input into the simulation model respectively to obtain simulation response parameter values. The dynamic parameter values ​​are used as the input of the proxy model neural network, and the simulation response parameter values ​​are used as the output of the proxy model neural network. Based on multiple groups of dynamic parameters and corresponding simulation response parameters, the neural network is trained. After the model training is completed, the various stress conditions of the strengthening test and the model dynamic parameters generated according to the constraint conditions are input into the neural network model to obtain multiple groups of simulation outputs. Finally, a suitable optimization objective function is determined to represent the model output error under multiple stress conditions. A suitable optimization algorithm is used to find a set of model dynamic parameters that minimize the objective function, complete parameter identification, and use the identified dynamic parameters to optimize the simulation model to improve the accuracy of the model.

[0055] Specifically, the present invention provides a model optimization method based on dynamic parameter identification of brushless DC motor products, such as Figure 1 As shown, it includes the following steps:

[0056] S1. Collect basic information of the brushless DC motor: The basic information of the brushless DC motor includes the motor's CAD model, motor structural parameters, key performance parameters, motor power and control parameters, and typical working environment.

[0057] In step S1, the structural parameters of the motor include the winding structure, permanent magnet material, stator and rotor structural dimensions, and air gap length; the key performance parameters of the motor include torque output, speed, efficiency, and temperature change; the power and control parameters of the motor include voltage, current, controller setting value, and excitation mode; typical working environment conditions include temperature, humidity, vibration, and load conditions.

[0058] S2. Build a simulation model: Based on the CAD model of the brushless DC motor, combined with the motor's structural characteristics and electromagnetic dynamics characteristics, build a preliminary simulation model of the motor.

[0059] S3. Constructing a proxy model: Based on the response data of the initial simulation model, a proxy model is constructed through data fitting technology. The input of the proxy model is a dynamic parameter combination, and the output is a key performance indicator.

[0060] The proxy model is a data-driven approximate model that constructs a sample data set by combining dynamic parameters and simulation model output responses into 'input-output' array pairs. Next, the proxy model is used to construct the mapping relationship between data groups, thereby replacing the simulation model based on physical mechanisms. The proxy model establishes an approximate mathematical model with high computational efficiency and accuracy by acquiring the characteristic information in the sample. In the specific embodiment of the present application, an artificial neural network method is selected to construct the proxy model. The structure of the artificial neural network can capture the complex nonlinear relationships in the data, and by adjusting the network architecture to increase the complexity of the model, the adaptability and generalization ability of the model are improved. The construction of the proxy model of this application mainly includes the following steps:

[0061] Determine the neural network structure:

[0062] First, determine the architecture of the neural network. The number of nodes in the input layer of a neural network is typically the same as the number of features, and the number of nodes in the output layer is the same as the number of features. Through multiple trials or hyperparameter search strategies such as grid search, random search, or Bayesian optimization, find the optimal number of layers and nodes within a defined range.

[0063] Network weight initialization: After determining the structure of the neural network, initialize the weights of the neural network.

[0064] Data input and forward propagation process:

[0065] Data first enters the network's input layer. This layer typically contains a number of neurons equal to the number of features. Consider a dataset of shape m, n, where m is the number of samples and n is the number of features. Each sample is a 1×n vector that is fed into the n neurons in the input layer.

[0066] Before the data enters the hidden layer or output layer from the input layer, it will be linearly transformed by the weight matrix and bias vector. Assuming the weight matrix is ​​w and the bias vector is b, the net input z of each neuron j is j It can be calculated as:

[0067]

[0068] Among them, x i is the i-th element of the input vector, w ij is the (i,j) element of the weight matrix, b j is the bias term of the jth neuron. Then, the net input z j It is passed to an activation function (ReLU, sigmoid or tanh) to calculate the activation value a of the jth neuron j :

[0069] a j=f(z j )

[0070] Here, f is the activation function, and this activation value will be used as the input of the next layer to undergo similar linear transformation and nonlinear activation. After being processed by the above weighting, bias, and activation function, the activation value is passed to the next hidden layer. Then, the data is passed to the output layer through the last hidden layer. Secondly, the backpropagation method is used to calculate the gradient of the loss function with respect to the network weights. Using the chain rule, the gradient is calculated layer by layer from the output layer back to the input layer. An optimization algorithm is used to update the weights based on the calculated gradient. The weights are usually adjusted to reduce the prediction error. The above steps are repeated iteratively until the network performance meets the predetermined criteria or the pre-set maximum number of iterations is reached.

[0071] Finally, the coefficient of determination R 2 , root mean square error RMSE, mean square error MRAE, and maximum absolute error MAE are used to evaluate the performance of the proxy model.

[0072] S4. Obtain reliability enhancement test data: In a laboratory environment, reliability enhancement tests are performed to test the actual response data of the motor under different operating conditions. The actual response data includes performance response and motor health status data; performance response refers to temperature, no-load speed, and maximum torque.

[0073] S5. Dynamic parameter identification using a reverse optimization algorithm: Based on the surrogate model, a reverse optimization algorithm is applied to find the dynamic parameter combination that best matches the simulation model response with the actual data using experimental data under different working conditions. This step specifically includes the following sub-steps:

[0074] S51, determine the optimization variables and optimization targets of the reverse optimization algorithm, the optimization variables are the resistance value, the motor constant and the sensitivity coefficient K of the inductance value R , K Ce , K L ,The optimization goal is to minimize the error between the motor response output by the surrogate model and the measured data under all temperature conditions.

[0075] S52. Import the maximum torque and no-load speed data measured under different temperature conditions, construct an objective function, and use the particle swarm algorithm to collaboratively search the solution space with multiple particles, continuously iteratively update, and finally converge to the globally optimal dynamic parameter combination; the dynamic parameter combination includes the resistance value R, motor constant Ce, and inductance value L of the brushless DC motor; the objective function is specifically:

[0076] The objective function is as follows:

[0077]

[0078] Where x = [KR ,K Ce ,K L ] is the sensitivity coefficient to be optimized, T model,j and n model,j are the maximum torque and no-load speed calculated by the proxy model, T meas,j and n meas,j is the corresponding measured data, and M is the number of temperature conditions.

[0079] In step S5, the relationship between the resistance value R, the motor constant Ce, and the inductance value L of the brushless DC motor and the temperature is as follows:

[0080] R=R0+K R (T-T0

[0081] Ce=Ce0+K Ce (T-T0)

[0082] L=L0+k L (T-T0)

[0083] Among them, R0, Ce0, and L0 are dynamic parameter values ​​under the condition of T0, and T and T0 are time.

[0084] Preferably, in step S52, each particle represents a possible sensitivity coefficient combination, and the position and velocity of the particle are dynamically adjusted according to the individual optimal solution and the global optimal solution. The velocity update formula is:

[0085] v i,d (t+1)=ωv i,d (t)+c1r1(p i,d -x i,d (t))+c2r2(g d -x i,d (t)).

[0086] The position update formula is:

[0087] x i,d (t+1)=x i,d (t)+v i,d (t+1)

[0088] Among them, ω is the inertia weight, c1 and c2 are learning factors, r1 and r2 are random numbers, and p i is the individual optimal position of the particle, g is the global optimal position, and the algorithm gradually converges through multiple iterations and finally finds the optimal sensitivity coefficient combination at the global optimal point.

[0089] S6. Optimize and verify the simulation model using a combination of dynamic parameters. Specifically, after determining the dynamic parameters of the simulation model, embed the corresponding temperature-dependent relationship between the dynamic parameters into the simulation model. This relationship is then applied to the motor's resistance R, motor constant Ce, and inductance L, enabling adaptive adjustment with temperature L to simulate the motor's response under different temperature conditions. This optimizes the simulation model and improves its fidelity. Verification is then performed to ensure the effectiveness of this method.

[0090] The present invention will be described in detail below through examples.

[0091] Taking a certain type of aircraft-mounted brushless DC motor as the research object, this study identified and calibrated dynamic parameters, constructed a simulation model to improve the performance prediction accuracy of the motor under different operating conditions, carried out reliability simulation analysis, and established an integrated performance and reliability model of the motor to truly reflect the performance changes and reliability evolution of the product under different operating conditions. To further improve the accuracy of the simulation model, this study identified and calibrated the key dynamic parameters of the simulation model based on measurement data obtained from reliability hardening tests (HALT). By comparing the experimentally measured response data with the simulation results, and using a reverse optimization algorithm to optimize the key dynamic parameters, the simulation model can more accurately represent the performance and degradation patterns of the motor under different operating conditions, thereby significantly improving the model's realism and applicability, providing effective support for the reliability assessment and life prediction of aircraft-mounted motors.

[0092] Key performance characteristics of brushless DC motors for aircraft include no-load speed and output torque, with the output torque being characterized by selecting the maximum output torque. Specifically, the method of this embodiment includes the following steps:

[0093] S1. Product information collection.

[0094] A certain type of aircraft-mounted brushless DC motor consists of three major components: a stator, a rotor, and a rear cover. This DC motor is used to receive position commands from the controller and drive the output shaft of the servo. The composition and functions of the brushless DC motor are shown in Table 1, basic parameters are shown in Table 2, motor parameters are shown in Table 3, and actual performance data of the brushless DC motor is shown in Table 4. The functional block diagram of the brushless DC motor is shown in Figure 2 .

[0095] Table 1 Composition and function of brushless DC motor

[0096]

[0097] Table 2 Basic parameters

[0098]

[0099]

[0100] Table 3 Motor parameters

[0101] Serial number Parameter name unit Reference value tolerance 1 Stator coil resistance Ω 1.582 ±0.15 2 Stator coil inductance mH 0.479 ±0.045 3 Motor torque coefficient Nm*s 0.01979 ±0.0018

[0102] Table 4 Measured performance data of brushless DC motor

[0103] Serial number No-load speed (rpm) Maximum torque (Nm) 1 13080 0.2058 2 12978 0.2205

[0104] S2. Preliminary construction of simulation model.

[0105] According to the motor structure and working principle, a certain type of aviation brushless DC motor model is established using MATLAB Simulink simulation software. Figure 3 The basic parameters are shown in Table 2. The model consists of: (1) input signal, (2) load signal, (3) feedback control, (4) motor principle model, and (5) reduction device. This model provides data support for subsequent dynamic parameter identification and calibration, ensuring that the simulation model can accurately predict the dynamic response of the motor under different working conditions.

[0106] According to the principle of brushless DC motor, its working circuit diagram is shown in Figure 4 The six drive elements in the drive circuit operate according to the control algorithm, sequentially connecting the corresponding stator windings. This generates a rotating magnetic field on the stator, which in turn drives the rotor embedded with permanent magnets. As the rotor rotates, the stator windings intersect the magnetic flux lines, generating a back electromotive force.

[0107] For a brushless DC permanent magnet synchronous motor, the voltage balance equation of the stator three-phase winding is:

[0108]

[0109] Where e a 、e b 、e c is the stator back electromotive force of each phase,

[0110] R a 、R b 、R c , is the stator phase winding resistance,

[0111] L a 、L b 、L c is the inductance of each phase winding of the stator,

[0112] L ab 、L ac 、L ba 、L bc 、L ca 、L cb is the stator winding phase-to-phase inductance,

[0113] i a 、i b 、i c is the stator current of each phase.

[0114] Assuming that the three-phase windings of the brushless DC permanent magnet synchronous motor are symmetrical and the influence between the magnetic groups is ignored, the mutual inductance between the stator phase windings is considered to be constant. That is:

[0115] L a =L b =L c =L s , R a =R b =R c =R,L ab =L ba =L ac =L ca =L bc =L cb =M.

[0116] Since the three-phase winding is symmetrical, i a +i b +i c =0,Mi a +Mi b +Mi c =0, we can get:

[0117]

[0118] Where L = L i -M.

[0119] For Y-type windings without a neutral point, the phase voltage is difficult to measure, and the line voltage equation is more practical. When the power tube is turned on, the line voltage is approximately equal to the DC side voltage of the inverter bridge.

[0120]

[0121] For the Y-connected winding, the transfer function of the line voltage to the output speed is similar to that of the DC brush motor.

[0122] When a brushless DC motor is operating, phases A, B, and C are connected in pairs. When any two phases are conducting, the currents are equal in magnitude but opposite in direction. For example, phases A and B are connected.

[0123] i A =-i B =i

[0124]

[0125] When phases A and B are connected, the relationship between the line voltage and current is:

[0126]

[0127] Where, Ω is the motor speed, unit is rad / s, C e It is the torque coefficient of the motor, that is, the ratio of the back electromotive force of the motor winding to the speed. It is also the ratio of the motor electromagnetic torque (Nm) to the current (A).

[0128] At this time, the motor mechanical model is:

[0129]

[0130] Where, T L is the load torque, J is the moment of inertia of the motor shaft and rotor as a whole, and T is L =0, the relationship between line current and speed is obtained:

[0131]

[0132] Substituting the relationship between line voltage and current, we get the relationship between motor input voltage and speed:

[0133]

[0134] When the load torque is not 0, the load torque can be used as the system input, and its transfer function is:

[0135]

[0136] According to the linear system superposition principle, the output response is equal to U when there is load torque. d (s) and T L (s) The sum of their respective actions.

[0137] Based on the relationship between motor input voltage and output speed, a motor model was established in MATLAB Simulink. The overall input signal of the model is a position signal, which is converted into motor input voltage. The motor parameters are marked in the model. The motor parameters are shown in Table 3:

[0138] The measured data of the brushless DC motor are shown in Table 4. The no-load speed and maximum torque represent the two characteristic points on the motor characteristic curve when the load is zero and the rotor is locked, respectively. The relationship between the remaining speed and torque can be represented by the straight line between these two points.

[0139] The no-load speed and maximum torque of the motor model are simulated. The load torque is set to 0Nm, and a step signal is input. The time domain diagrams of the input signal and output signal and the no-load speed change curve are shown in Fig. Figure 5 and Figure 6 .

[0140] When the motor is unloaded, the speed increases rapidly after the input signal changes. When the motor approaches the required servo position, the speed gradually decreases and eventually becomes 0. The maximum speed during the process is 130004. When the model load torque is increased to make the motor stall, and the drive signal is still input in the first second, the output torque change curve is shown in Fig. Figure 7 After the motor is started, due to stalling, its output electromagnetic torque increases rapidly, with the maximum torque being 0.2127 Nm.

[0141] S3. Build an agent model.

[0142] Before establishing the proxy model, the data needs to be preprocessed. In this study, the input and output data only contain numerical data, so data preprocessing only includes data normalization and data segmentation. Through normalization, all input parameters are scaled to the range of [0,1]. Then, the data is segmented. The obtained sample data is divided into a training set and a test set in an 80 / 20 ratio. The training set is used to train the neural network, and the test set is used to evaluate the performance of the trained neural network. After obtaining the preprocessed data, the neural network proxy model is established. The number of neurons in the input layer is consistent with the dimension of the dynamic parameters used as input. There are three hidden layers with 128, 64, and 32 neurons respectively. The number of neurons in the output layer is the same as the output response dimension. The activation function is the ReLU function.

[0143] As environmental stresses change, the dynamic parameter values ​​in the simulation model will vary. When selecting dynamic parameters, we generally choose those that vary significantly with environmental stresses (temperature, load). Furthermore, fluctuations in the selected dynamic parameter values ​​can significantly impact the motor's output. Based on the aforementioned selection criteria, we chose resistance R, motor constant Ce, and inductance L as the dynamic parameters to be studied. The motor's no-load speed at zero load and maximum torque at stalled conditions were used as responses.

[0144] After determining the input and output data of the surrogate model, the dynamic parameter combination is generated by Latin hypercube. First, the dynamic parameter space is defined and the output parameters and their ranges of the dynamic parameters are determined. Then, Latin hypercube is used to generate samples to ensure that the samples cover the entire range of the parameters. The frequency of each parameter in the dynamic parameter group [R, Ce, L] in the sampling space is as follows: Figure 8a-8c As shown, the frequencies in each slice are basically the same in the entire value range, ensuring the randomness and uniformity of the distribution of data samples used for training.

[0145] After obtaining the dynamic parameters obtained by Latin hypercube sampling, the dynamic parameter combination is used as the parameter setting of the simulation model. The output response of the corresponding simulation model, namely the idling speed and maximum torque data, is obtained by calculation. The distribution of the two dynamic response data is as follows: Figure 9a and Figure 9b The idling speed distribution is basically uniform within the value range, and the maximum torque shows a quasi-normal distribution feature.

[0146] After obtaining simulation output data for different dynamic parameter combinations, a simplified model was constructed using surrogate modeling technology to efficiently simulate the motor's dynamic response. First, an appropriate fitting method was selected based on the distribution characteristics of no-load speed and maximum torque. In this case, a neural network was chosen to construct the surrogate model. Training and cross-validation on simulation data ensured the model's prediction accuracy across the entire parameter space.

[0147] To improve the reliability of the model, error analysis is performed and model parameters are adjusted when necessary to reduce the deviation, ultimately enabling the proxy model to achieve high accuracy in the prediction of no-load speed and maximum torque. In addition, the verification results of the proxy model are as follows Figure 10a As shown in the figure, the measured data are Figure 10b As shown in the figure, the comparison shows that the proxy model can stably predict the response characteristics of the motor under different dynamic parameter combinations, providing an accurate and fast-responding proxy model for subsequent parameter optimization and reliability evaluation.

[0148] S4. Acquisition of reliability enhancement test (HALT) data.

[0149] In the embodiment, in order to obtain the response data of the brushless DC motor under different temperature conditions, a reliability enhancement test (HALT) is carried out. This test mainly uses a temperature step test to gradually increase or decrease the temperature to simulate the operating state of the motor under extreme temperatures, so as to obtain the key response characteristics of the motor under these conditions. Since the resistance value, motor constant and inductance value are temperature-sensitive dynamic parameters, they will affect the performance of the motor with temperature changes. Therefore, the response data under these temperature conditions are particularly important for model calibration. In this case, the motor response signal in the range of 5℃-65℃ is mainly measured, with a step size of 5℃.

[0150] Through temperature testing, HALT test effectively captures the response characteristics of the motor under different temperature conditions. Figure 11a and Figure 11b As shown in the figure, the no-load speed and maximum torque values ​​at various temperatures are recorded. This data reflects the direct impact of temperature on motor performance and provides a basis for parameter identification of the simulation model, making the model's response predictions under different temperature conditions more realistic.

[0151] S5. Dynamic parameter identification of reverse optimization algorithm.

[0152] First, determine the optimization variables and optimization objectives for the inverse optimization algorithm. Dynamic parameters are usually adjusted as environmental conditions change. For example, the relationship between the resistance value R, motor constant Ce, and inductance value L of a brushless DC motor and temperature is shown below:

[0153] R=R0+K R (T-T0)

[0154] Ce=Ce0+K Ce (T-T0)

[0155] L=L0+k L (T-T0)

[0156] Among them, R0, Ce0, L0 values ​​are known, which are the dynamic parameter values ​​under T0 conditions, and the coefficients represent the sensitivity coefficients of each parameter with temperature changes. In order to ensure the physical rationality of the optimization results and avoid non-physical dynamic parameter combinations, these sensitivity coefficients K R , K Ce , K L These are set as optimization variables and used as adjustments in the inverse optimization algorithm. Ultimately, by adjusting these coefficients, the error between the motor response output by the surrogate model (such as no-load speed and maximum torque) and the measured data is minimized under all temperature conditions, achieving accurate identification and calibration of dynamic parameters.

[0157] In the dynamic parameter calibration, the particle swarm optimization algorithm (PSO) is used to perform reverse optimization to obtain the optimal dynamic parameter combination. First, the maximum torque and no-load speed data measured under different temperature conditions are imported to define the optimization objective function and the sensitivity coefficient K to be optimized. R , K Ce , K L The optimization goal is to minimize the error between the surrogate model's output response and the measured data. The particle swarm algorithm uses multiple particles to collaboratively search within the solution space, continuously iterating and updating, ultimately converging to the globally optimal dynamic parameter combination. This ensures that dynamic parameter adjustments accurately reflect the motor's performance response under different temperature conditions. This process ensures the accuracy of calibration results under multiple operating conditions and enables efficient optimization of dynamic parameters. Figure 12 This is a flowchart of reverse optimization, which illustrates the steps and iterative process of the optimization algorithm. The specific steps are as follows:

[0158] 1. Based on the relationship between dynamic parameters and stress, the position of each particle is determined as [p1, p2, ..., p n ], where p i =[p i1 ,pi2 ,…,p ij ].

[0159] 2. Set the sample size of the particle swarm to N, and use the Latin hypercube sampling technique to determine the position of the particles in the initial particle swarm sample within the value range of the coefficient of the dynamic parameter relationship.

[0160] 3. Through x i =f(C) and the position of each particle [p1,p2,…,p n ]Determine the dynamic parameter values ​​under various stress conditions.

[0161] 4. Call the aforementioned neural network proxy model, input the proxy model with the dynamic parameter combination under all stress conditions of each particle, and obtain the output response under the corresponding stress conditions.

[0162] 5. The fitness function is calculated based on the sum of the differences between the response outputs produced by the surrogate model under all stress conditions and the measured responses of the step test

[0163] 6. Update the historical optimal position of each particle and the global optimal position of the group.

[0164] 7. The algorithm terminates based on whether the number of iterations or the fitness function meets the predetermined criteria. If the termination criteria are met, the algorithm terminates; otherwise, it returns to step 3 and continues iterating.

[0165] S6. Optimization and verification of simulation model.

[0166] After determining the dynamic parameters of the simulation model using a reverse optimization algorithm, the corresponding temperature-dependent relationships are embedded in the simulation model to simulate the motor's response under different temperature conditions. Specifically, the updated relationships are applied to the motor's resistance R, motor constant Ce, and inductance L, enabling them to adapt adaptively to changes in temperature L.

[0167] Next, the optimized and calibrated simulation model was simulated and verified under multiple temperature conditions, and response data such as no-load speed and maximum torque at different temperatures were calculated. These simulation results were then compared with the experimental measured data, and the accuracy of the calibration results was further verified through comparative analysis. In this embodiment, the comparison results showed that the response values ​​of the simulation model under various temperature conditions were highly consistent with the measured data, with the root mean square error of no-load speed and maximum torque being 3.03×10 -7 and 1.16×10 -15 The measured response and the response obtained by the simulation model are as follows Figure 13As shown, through this verification step, the effectiveness and reliability of the calibrated simulation model under multiple working conditions are proved, and the performance response of the motor under different temperature conditions can be accurately predicted. Compared with the existing methods, the accuracy of the model is greatly improved. This high-precision model provides data support for the dynamic performance analysis and reliability evaluation of the brushless DC motor. At the same time, the overall method of the proxy model in the present invention is also more reliable. The construction of the proxy model makes the overall method have higher execution degree, and can quickly and accurately identify dynamic parameters. The optimized high-precision model can better perform dynamic performance analysis and reliability evaluation of the brushless DC motor.

[0168] Specifically, the application effect of this technology in aviation brushless DC motor simulation and reliability analysis is mainly reflected in the following three aspects:

[0169] (1) In the initial design stage of the brushless DC motor, by sampling the working condition combination and predicting the rapid response of the surrogate model, the performance changes of the brushless DC motor under different environmental stress conditions can be simulated, and the key dynamic parameters can be initially optimized, which helps to identify the potential risks of the motor under extreme working conditions during the design.

[0170] (2) During the verification phase of the brushless DC motor, dynamic parameter calibration and reverse optimization algorithms based on reliability enhancement test data can achieve high-precision calibration of temperature-sensitive parameters, effectively solving the error accumulation problem of traditional static parameter models in a changing environment, and ensuring that the response of the simulation model during the verification phase is consistent with the actual working conditions.

[0171] (3) During the use phase of the brushless DC motor, this method can support field fault cause analysis and rapid verification and optimization. By using simulation methods to verify the deviation between the dynamic response measured in the field and the response predicted by the proxy model, the impact of different environmental factors on the motor performance can be quantified, thereby providing reliable simulation data support for design improvement and maintenance, ensuring the dynamic adaptability of the model under changing environmental and load conditions, and realizing real-time monitoring and rapid optimization of motor performance.

[0172] The embodiments described above are merely descriptions of preferred implementations of the present invention and are not intended to limit the scope of the present invention. Without departing from the spirit of the present invention, various modifications and improvements made to the technical solutions of the present invention by ordinary technicians in this field should fall within the scope of protection determined by the claims of the present invention.

Claims

1. A model optimization method based on dynamic parameter identification of brushless DC motor products, characterized by: It includes the following steps: S1. Collect basic information about the brushless DC motor: The basic information about the brushless DC motor includes the motor's CAD model, structural parameters, key performance parameters, power and control parameters, and typical operating environment. S2. Build a simulation model: Based on the CAD model of the brushless DC motor, combined with the motor's structural characteristics and electromagnetic dynamics characteristics, build a preliminary simulation model of the motor; S3. Build a surrogate model: Based on the response data of the initial simulation model, a surrogate model is built through data fitting technology. The input of the surrogate model is a dynamic parameter combination, and the output is a key performance indicator; S4. Obtain reliability enhancement test data: In a laboratory environment, reliability enhancement tests are performed to measure the actual response data of the motor under different operating conditions. The actual response data includes performance response and motor health status data; S5. Dynamic parameter identification using a reverse optimization algorithm: Based on the surrogate model, a reverse optimization algorithm is applied to find the dynamic parameter combination that best matches the simulation model response with the actual data using experimental data under different working conditions. This step specifically includes the following sub-steps: S51, determine the optimization variables and optimization targets of the reverse optimization algorithm, the optimization variables are the resistance value, the motor constant and the sensitivity coefficient K of the inductance value R , K Ce , K L ,The optimization goal is to minimize the error between the motor response output by the surrogate model and the measured data under all temperature conditions; S52. Import the maximum torque and no-load speed data measured under different temperature conditions, construct an objective function, and use the particle swarm algorithm to collaboratively search the solution space with multiple particles, continuously iteratively update, and finally converge to the globally optimal dynamic parameter combination; the dynamic parameter combination includes the resistance value R, motor constant Ce, and inductance value L of the brushless DC motor; the objective function is specifically: The objective function is as follows: Where x = [K R ,K Ce ,K L ] is the sensitivity coefficient to be optimized, T model,j and n model,j are the maximum torque and no-load speed calculated by the proxy model, T meas,j and n meas,j is the measured data of maximum torque and no-load speed, M is the number of temperature conditions; S6. Use dynamic parameter combination to optimize and verify the simulation model.

2. The model optimization method based on dynamic parameter identification of brushless DC motor products according to claim 1, characterized in that: In step S1, the structural parameters of the motor include the winding structure, permanent magnet material, stator and rotor structural dimensions, and air gap length; the key performance parameters of the motor include torque output, speed, efficiency, and temperature change; the power and control parameters of the motor include voltage, current, controller setting value, and excitation mode; typical working environment conditions include temperature, humidity, vibration, and load conditions.

3. The model optimization method based on dynamic parameter identification of brushless DC motor products according to claim 1, characterized in that: In step S4 , the performance responses are temperature, no-load speed, and maximum torque.

4. The model optimization method based on dynamic parameter identification of brushless DC motor products according to claim 1, characterized in that: In step S5, the relationship between the resistance value R, the motor constant Ce, and the inductance value L of the brushless DC motor and the temperature is as follows: R=R0+K R (T-T0) Ce=Ce0+K Ce (T-T0) L=L0+k L (T-T0) Among them, R0, Ce0, and L0 are dynamic parameter values ​​under the condition of T0, and T and T0 are time.

5. The model optimization method based on dynamic parameter identification of brushless DC motor products according to claim 4, characterized in that: In step S52, each particle represents a possible sensitivity coefficient combination. The position and velocity of the particle are dynamically adjusted according to the individual optimal solution and the global optimal solution. The velocity update formula is: v i,d (t+1)=ωv i,d (t)+c1r1(p i,d -x i,d (t))+c2r2(g d -x i,d (t)); The position update formula is: x i,d (t+1)=x i,d (t)+v i,d (t+1) Among them, ω is the inertia weight, c1 and c2 are learning factors, r1 and r2 are random numbers, and p i is the individual optimal position of the particle, g is the global optimal position, and the algorithm gradually converges through multiple iterations and finally finds the optimal sensitivity coefficient combination at the global optimal point.

6. The model optimization method based on dynamic parameter identification of brushless DC motor products according to claim 5, characterized in that: Step S6 specifically involves, after determining the dynamic parameter combination of the simulation model, embedding the relationship between the dynamic parameter combination and temperature change into the simulation model, and applying it to the motor's resistance value R, motor constant Ce and inductance value L, so that it can be adaptively adjusted with changes in temperature L, so as to simulate the motor response under different temperature conditions.

7. The model optimization method based on dynamic parameter identification of brushless DC motor products according to claim 1, characterized in that: In step S2, the simulation model is constructed using MATLAB Simulink simulation software. The simulation model includes input signals, load signals, feedback control, motor principle model, and reduction device.

8. The model optimization method based on dynamic parameter identification of brushless DC motor products according to claim 1, characterized in that: The proxy model in step S3 is a neural network architecture model.

Images