Motor dynamic characteristic compensation test method and platform based on multi-modal data

Abstract

The invention relates to the technical field of motor testing, and discloses a motor dynamic characteristic compensation test method and platform based on multi-modal data, the method collects motor data through a multi-modal synchronous test system, and a processing module executes the following steps: constructing a state observer model to estimate winding temperature and loss torque; in the online test, the accuracy of the model is monitored by calculating the residual error between the model prediction and the measured value; and when the residual dynamic characteristic exceeds a threshold value, generating and applying an adaptive excitation sequence according to the residual characteristic, collecting high-resolution data to update model parameters on line, forming a self-optimization closed loop, after the test is finished, adopting a fixed interval smoothing algorithm to be combined with a final model to obtain a global optimal state track, and calculating a dynamic characteristic compensation result according to the global optimal state track. According to the invention, through adaptive excitation of residual error driving and online model updating, accurate identification and compensation of dynamic characteristics of the motor are realized, and the test automation level and accuracy are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of motor testing, in particular to a motor dynamic characteristic compensation testing method and platform based on multi-modal data. BACKGROUND

[0002] As a core power component, the dynamic characteristics of the motor directly affect the performance of the entire system during operation. The electromagnetic, thermal, mechanical and other physical processes of the motor are coupled with each other, especially under dynamic working conditions such as load mutation or environmental temperature fluctuation, the internal parameters of the motor will change significantly, resulting in deviation of performance from the design expectation. Therefore, it is crucial to accurately compensate the dynamic characteristics of the motor for improving its control accuracy and operation efficiency.

[0003] The existing motor testing technology has certain limitations in dealing with such complex dynamic coupling characteristics. In the establishment of the motor model, the traditional testing method often simplifies the thermal effect and mechanical loss, or analyzes the characteristics of different physical domains in isolation, which is difficult to accurately capture the dynamic response under the coupling action of multiple physical fields, which leads to insufficient compensation accuracy under complex working conditions, affecting the physical fidelity of the model.

[0004] In addition, in the design of the testing process, the existing technology usually adopts fixed testing specifications or preset working condition cycles to test the motor. This method lacks real-time feedback and adjustment of the testing process. When the motor shows model mismatch at a specific unexpected working condition point, the fixed testing process cannot apply effective excitation to deeply investigate the problem, thereby limiting the adaptability and diagnostic ability of the testing method.

[0005] In the processing and application of testing data, the existing method usually processes the testing data in segments or in real time, lacks a global optimization model to uniformly and consistently compensate the data of the entire testing process after completing the entire testing, and also fails to provide a complete data set recording the evolution trajectory of the key dynamic parameters of the motor in the entire testing process, which brings obstacles to the subsequent research and development improvement and in-depth analysis of the motor. SUMMARY

[0006] The purpose of the present application is to provide a motor dynamic characteristic compensation testing method and platform based on multi-modal data, which solves the problem that the existing motor testing method is difficult to obtain and compensate the dynamic characteristic changes caused by non-electromagnetic factors such as thermal effect and mechanical loss online and adaptively, resulting in deviation between the test results and the actual working conditions.

[0007] To achieve the above purpose, the present application is implemented by the following technical solutions: The first aspect of the present application provides a motor dynamic characteristic compensation testing method based on multi-modal data, comprising the following steps: S1. System and Model Construction Phase: Construct a multimodal synchronous test system for synchronously acquiring electrical parameter data, temperature data, vibration data, and speed data of the motor under test, and construct a state observer model with initial model parameters; S2. Online Testing and Self-Optimization Phase: The multimodal synchronous testing system is started to continuously collect data, and the following operations are performed cyclically during data collection: The state observer model is run to predict temperature and vibration data based on the collected electrical parameter data and rotational speed data, generating a prediction output. The residual between the prediction output and the temperature and vibration data collected during the data acquisition period is calculated. The residual e(k) is calculated as follows: Where k is a discrete time point; y(k) is the observation vector at discrete time point k, and its elements are the specific values ​​of the temperature data and the vibration data collected during the data acquisition period; The predicted output vector generated by the state observer model at discrete time point k, whose elements are the predicted values ​​of the temperature data and the vibration data.

[0008] Analyze the dynamic characteristics of the residuals, and trigger a test event when the dynamic characteristics of the residuals exceed a preset dynamic threshold; Based on the triggered test event, an adaptive excitation sequence corresponding to the dynamic characteristics of the residual is generated and applied to the motor under test. At the same time, high-resolution data generated under the adaptive excitation sequence is collected, and the model parameters of the state observer model are updated using the high-resolution data. The updated model parameters are used to construct the state observer model in the next cycle. S3. Global Processing and Archiving Stage: After step S2, the state observer model, after completing the cyclic update, is used to perform global compensation calculation on all collected data, generate dynamic characteristic compensation results, and archive the dynamic characteristic compensation results and the cyclic update values ​​of the model parameters completed in step S2.

[0009] Preferably, the state observer model is a state-space model. This state-space model is described by the following set of discrete-time linear time-invariant system equations: State equation: x(k+1)=A*x(k)+B*u(k); Observation equation: y(k)=C*x(k)+D*u(k); wherein, k is a discrete time point; x(k) is a state vector at the discrete time point k, which is used to describe the internal physical state of the motor under test at the time, which cannot be directly measured, for example, its elements can include real-time winding temperature and loss torque; x(k+1) is a state vector at the next discrete time point k+1; u(k) is a control input vector at the discrete time point k, whose elements are measurable control inputs applied to the motor under test, for example, specific values in the electrical parameter data and the speed data; y(k) is an observation vector at the discrete time point k, whose elements are directly measurable system outputs, for example, specific values in the temperature data and the vibration data; A is a state matrix; B is an input matrix; C is an output matrix; D is a feedforward matrix; the internal elements of the matrices A, B, C and D constitute the model parameters of the state observer model.

[0010] Preferably, in the step S2, the way of analyzing the dynamic characteristics of the residual error is to calculate the Mahalanobis distance M(e) of the residual error. When the Mahalanobis distance is greater than a preset threshold γ, it is determined that the dynamic characteristics of the residual error exceed the preset dynamic threshold. The calculation formula of the Mahalanobis distance is: wherein, e is a residual error vector at the current time; μ e is a mean vector composed of residual error vector data at historical time points collected under normal working conditions; T represents the transpose operation of a vector or a matrix; the sqrt function represents the square root operation; is the inverse matrix of the covariance matrix Σ e ; (e-μ e ) is a residual error deviation vector, which quantifies the position and direction of the current residual error deviating from its historical statistical center in numerical value; (e-μ e ) T is the transposed residual error deviation vector.

[0011] Preferably, in the step S2, the specific way of generating an adaptive excitation sequence corresponding to the dynamic characteristics of the residual error includes: specificating the dynamic characteristics of the residual error corresponding to the triggered test event into frequency domain and time domain features; and generating an excitation sequence for high-resolution scanning at a specific frequency and a specific load point according to the frequency domain and time domain features; the excitation sequence is selected from a group consisting of voltage excitation sequence and load torque excitation sequence.

[0012] Preferably, in the step S2, the specific way of updating the model parameters by using the high-resolution data is: performing parameter identification on the high-resolution data by selecting an algorithm from a group consisting of recursive least squares method and gradient descent method, solving and updating the model parameters, and minimizing the prediction error of the state observer model on the high-resolution data.

[0013] Preferably, in the step S3, before the global compensation calculation, a fixed interval smoothing algorithm is adopted to perform state re-estimation on all data by using the state observer model updated in the last cycle to obtain an optimal smoothed state trajectory, and the global compensation calculation is performed based on the optimal smoothed state trajectory.

[0014] Preferably, on the basis of the state observer model being a state space model, the state vector includes real-time winding temperature and loss torque. On this basis, the global compensation calculation in the step S3 includes: performing dynamic compensation on the copper loss of the measured motor by using the real-time winding temperature estimated by the state observer model updated in the last cycle; and performing compensation on the electromagnetic torque of the measured motor by using the loss torque estimated by the state observer model updated in the last cycle to obtain an effective load torque.

[0015] Preferably, the step S1 further includes: performing a preliminary offline parameter identification experiment, and taking the parameters identified through the offline parameter identification experiment as the initial model parameters.

[0016] The second aspect of the present application provides a motor dynamic characteristic compensation test platform based on multi-modal data, which is used to execute any of the above motor dynamic characteristic compensation test methods based on multi-modal data, and includes: A multi-modal synchronous test system configured to synchronously collect electrical parameter data, temperature data, vibration data and speed data of a measured motor.

[0017] A processing module connected with the multi-modal synchronous test system and configured to perform the following operations: Construct a state observer model with initial model parameters; During continuous data collection by the multi-modal synchronous test system, the following operations are cyclically performed: Run the state observer model to predict temperature data and vibration data according to the collected electrical parameter data and speed data, generate a predicted output, and calculate a residual error between the predicted output and the temperature data and vibration data collected during the data collection; Analyze the dynamic characteristic of the residual error, and trigger a test event when the dynamic characteristic exceeds a preset dynamic threshold; According to the test event, generate an adaptive excitation sequence corresponding to the dynamic characteristic of the residual error and apply it to the measured motor, collect high-resolution data generated under the excitation sequence, and update the model parameters of the state observer model by using the high-resolution data, wherein the updated model parameters are used to constitute the state observer model in the next cycle; And configure for after completing the operation of the cycle execution, using the state observer model updated with the completion cycle, performing global compensation calculation on all data collected, generating dynamic characteristic compensation results, and archiving the compensation results and the cycle update value of the model parameter completed in the cycle execution operation.

[0018] The application provides a motor dynamic characteristic compensation test method and platform based on multi-modal data. The application has the following beneficial effects: 1. The application fuses multi-modal data such as electrical parameters, temperature, and vibration by constructing a state observer model, and realizes accurate modeling and tracking of electromagnetic, thermal, and mechanical coupling dynamic characteristics of the motor by using online self-optimization continuous update model, which fundamentally improves the physical fidelity and accuracy of dynamic compensation.

[0019] 2. The application generates an adaptive excitation sequence corresponding to the specific dynamic characteristics of the residual error, so that the test process can specifically explore the specific working condition of the model that currently performs poorly, realizes the transformation from fixed test paradigm to dynamic adaptive test, and enhances the adaptability and problem diagnosis capability of the test method.

[0020] 3. The application uses the final optimization model to perform global compensation on all data after the test is completed, and archives the compensation results and the model parameter update value, which not only ensures the global consistency of the final compensation results, but also provides a complete data set support containing the dynamic evolution process for the subsequent research and optimization of the motor. BRIEF DESCRIPTION OF DRAWINGS

[0021] Figure 1 The figure is a structural schematic diagram of the test platform of an embodiment of the application. Figure 2 The figure is a flowchart of the test method of an embodiment of the application. Figure 3 The figure is a detailed flowchart of the online test and self-optimization stage of an embodiment of the application. DETAILED DESCRIPTION

[0022] The technical solutions in the embodiments of the application will be described clearly and completely below with reference to the drawings in the specification of the application. Obviously, the described embodiments are only part of the embodiments of the application, not all the embodiments. Based on the embodiments in the application, all other embodiments obtained by those skilled in the art without creative labor are within the protection scope of the application.

[0023] Refer to the drawings Figure 1 , Figure 1Fig. 1 is a structural schematic diagram of a test platform according to an embodiment of the present application. The present application provides a motor dynamic characteristic compensation test platform based on multi-modal data. The test platform is connected with a motor 1 to be tested and is used to perform dynamic characteristic compensation test on the motor 1, and includes a multi-modal synchronous test system 10 and a processing module 2.

[0024] The multi-modal synchronous test system 10 is configured to be physically and electrically connected with the motor 1 to be tested. The function of the multi-modal synchronous test system 10 is to synchronously collect data of multiple groups of different physical quantities generated by the motor 1 to be tested during operation, which specifically includes electric parameter data, temperature data, vibration data and rotation speed data.

[0025] The processing module 2 can be physically an industrial computer, an embedded computing system or a special hardware platform composed of a digital signal processor (DSP) and a field programmable gate array (FPGA).

[0026] The processing module 2 is connected with the multi-modal synchronous test system 10 through a data communication interface, and is used to receive a synchronous data stream collected and transmitted by the multi-modal synchronous test system 10.

[0027] The processing module 2 is also connected with the multi-modal synchronous test system 10 and a drive controller of the motor 1 to be tested through a control interface. The control interface is used to apply control instructions, such as a voltage excitation sequence, from the processing module 2 to the motor 1 to be tested, and to apply a load torque excitation sequence to a programmable load unit 15 in the multi-modal synchronous test system 10.

[0028] The processing module 2 is configured to perform the method steps described in subsequent embodiments of the present application, specifically including: constructing and storing a state observer model; receiving and processing the synchronous data stream; generating and outputting an adaptive excitation sequence based on the data processing result; online updating parameters of the state observer model; and performing global compensation calculation and archiving data after the end of the test process.

[0029] The multi-modal synchronous test system 10 includes an electric parameter measurement unit 11, a temperature sensor group 12, a vibration sensor group 13, a rotation speed sensor 14, a programmable load unit 15 and a synchronization and collection unit 16.

[0030] The electric parameter measurement unit 11 is connected with an electric input end of the motor 1 to be tested, and is used to measure phase voltage and phase current of the motor 1 to be tested and output raw electric parameter data.

[0031] The temperature sensor group 12 is composed of multiple thermocouples or platinum resistance sensors, which are arranged at predetermined temperature measurement points of the motor 1 to be tested, such as the shell and the winding end, and are used to measure temperature of the motor at different positions and output raw temperature data.

[0032] The vibration sensor group 13 is composed of one or more acceleration sensors, which are installed on the housing or bearing seat of the motor 1 under test, for measuring the vibration acceleration of the motor during operation, and outputting raw vibration data.

[0033] The rotational speed sensor 14 can be an optical encoder or a resolver installed on the shaft end of the motor, for measuring the angular speed or angular position of the rotor of the motor 1 under test, and outputting raw rotational speed data.

[0034] The programmable load unit 15 can be a dynamometer or a magnetic powder brake connected coaxially with the motor 1 under test, for receiving control instructions from the processing module 2, and applying precise load torque to the motor 1 under test.

[0035] The synchronization and acquisition unit 16 includes a unified clock source and a multi-channel synchronous data acquisition card. The synchronization and acquisition unit 16 receives all raw data from the electrical parameter measurement unit 11, the temperature sensor group 12, the vibration sensor group 13, and the rotational speed sensor 14, and attaches a precise time stamp to each frame of data according to the unified clock source, generates a synchronous data stream, and sends it to the processing module 2 through a data communication interface.

[0036] The processing module 2 is functionally configured to include a data interface and preprocessing unit, a model operation unit, an analysis and decision unit, an excitation generation unit, a parameter update unit, and a global processing and archiving unit.

[0037] The data interface and preprocessing unit is used to receive the synchronous data stream from the synchronization and acquisition unit 16, and to parse, align, and format the data for calling by other units inside the processing module 2.

[0038] The model operation unit is used to store and execute a state observer model. During the execution of the method, the model operation unit receives electrical parameter data and rotational speed data from the data interface and preprocessing unit as model inputs, runs the model, and outputs predicted temperature data and vibration data.

[0039] The analysis and decision unit is used to receive predicted outputs from the model operation unit, and to receive measured temperature data and vibration data from the data interface and preprocessing unit. The analysis and decision unit calculates the residual between the two, analyzes the dynamic characteristics of the residual according to a preset algorithm (such as calculating the Mahalanobis distance), and then compares the analysis result with a preset dynamic threshold value. When the threshold value is exceeded, the analysis and decision unit outputs a test event trigger signal.

[0040] The excitation generation unit is configured to generate digital instructions of the adaptive excitation sequence according to the residual dynamic characteristics associated with the triggering event after receiving the test event triggering signal from the analysis and decision unit. The digital instructions are sent to the drive controller or the programmable load unit 15 of the motor under test 1 through the control interface.

[0041] The parameter updating unit is configured to receive the high-resolution data from the data interface and the preprocessing unit during the application of the adaptive excitation sequence. The parameter updating unit executes a preset parameter identification algorithm (for example, the recursive least square method) and calculates and updates the model parameters of the state observer model stored in the model operation unit by using the high-resolution data.

[0042] The global processing and archiving unit is configured to execute global compensation calculation on all the data collected during the test by using the state observer model that has completed the final update in the model operation unit after the completion of the entire online test and self-optimization phase. After the calculation is completed, the global processing and archiving unit stores and archives the final dynamic characteristic compensation result and the cyclic update values of the model parameters in the test process.

[0043] The specific steps of the motor dynamic characteristic compensation test method executed by the processing module 2 in an embodiment of the present application will be described in detail below.

[0044] Referring to the accompanying Figure 1 and the accompanying Figure 2 , Figure 2 The flowchart of the test method of an embodiment of the present application is shown in FIG. 1. The present application provides a motor dynamic characteristic compensation test method based on multi-modal data. The first stage of the method is the system and model construction stage S1. The goal of this stage is to complete the hardware deployment of the test platform and construct a state observer model with initial model parameters in the processing module 2, so as to prepare for the subsequent online test and self-optimization stage S2.

[0045] In the present embodiment, the state observer model is specifically implemented as a discrete-time linear state space model. The state observer model is constructed and stored in the model operation unit of the processing module 2, and its mathematical structure is jointly defined by the following state equation and observation equation: State equation: x(k+1) = A*x(k) + B*u(k); Observation equation: y(k) = C*x(k) + D*u(k); Wherein, k is a discrete time point; x(k) is a state vector at the discrete time point k, and the elements of the state vector are physical quantities inside the motor that cannot be directly measured. In the present embodiment, the state vector is defined as x(k) = [θ w (k), τ loss (k)], wherein θ w(k) is the average equivalent temperature of the motor winding in a sampling period, τ loss (k) is the aggregated loss torque of the motor at the discrete time point k, which includes mechanical friction loss and stray loss; u(k) is the control input vector at the discrete time point k, the elements of which are directly measurable physical quantities for driving the model state evolution, in this embodiment, the control input vector is composed of electrical parameter data and speed data, and is specifically defined as u(k) = [u d( k), u q( k), ω r( k)] T , where u d (k) and u q (k) are stator voltage components in d, q rotating coordinate system, which are collected by the electrical parameter measurement unit 11 and obtained through coordinate transformation, ω r (k) is the mechanical angular velocity of the motor rotor, which is collected by the speed sensor 14; y(k) is the observation vector at the discrete time point k, the elements of which are directly measurable system outputs of the model; matrices A, B, C, D are parameter matrices of the state space model, and the internal elements thereof jointly constitute a model parameter vector of the state observer model. The initial values of these parameters, i.e. A0, B0, C0, D0, are obtained by performing a preliminary offline parameter identification experiment.

[0046] The specific process of the offline parameter identification experiment is as follows: first, the processing module 2 controls the programmable load unit 15 and the driver of the motor to be tested 1, so that the motor to be tested 1 runs at a group of preset representative steady state working points (for example, different combinations of speed and load torque). At these working points, the multi-modal synchronous test system 10 collects a complete time series data set containing all inputs (voltage, current, speed) and outputs (surface temperature, vibration), forming an offline identification data set.

[0047] Subsequently, in the processing module 2, a system identification algorithm (for example, the subspace identification method of N4SID or the prediction error method) is called, and the offline identification data set is taken as input. The algorithm solves the optimal state space matrices A0, B0, C0, D0 which can describe the dynamic characteristics of the data set by analyzing the input and output data. Finally, this set of solved initial parameter matrices is loaded into the model operation unit as the initial model parameters of the state observer model at the beginning of the online test and self-optimization stage S2.

[0048] After completing the system and model construction stage S1, the test method enters the online test and self-optimization stage S2.

[0049] Referring to the accompanying drawings, Figure 3 , Figure 3Detailed flow chart of the online testing and self-optimization phase for one embodiment of the present application. This phase is a continuous closed-loop process, which is repeated during the whole period of time when the motor 1 is running and continuous data acquisition is performed by the multi-modal synchronous testing system 10.

[0050] At each discrete time point k, the model operation unit within the processing module 2 first acquires the control input vector u(k) at the current time instant, which is provided by the data interface and pre-processing unit. Subsequently, the model operation unit performs a prediction calculation using its internally stored state observer model with the current model parameters, generating a predicted output vector

[0051] Meanwhile, the analysis and decision unit within the processing module 2 acquires the predicted output vector and the actual observation vector y(k) at the current time instant, which are provided by the data interface and pre-processing unit. The actual observation vector y(k) is composed of real-time measurements from the temperature sensor group 12 and the vibration sensor group 13, specifically: s (k), V a (k) T ; where T s (k) is the measured temperature on the surface of the motor; and V a (k) is the measured vibration amplitude of the motor.

[0052] After receiving the predicted output vector and the actual observation vector y(k), the analysis and decision unit calculates the difference between them, obtaining the residual vector e(k) at the current time instant k. This calculation is performed by the following formula:

[0053] After obtaining the residual vector e(k), the analysis and decision unit analyzes its dynamic characteristics to determine whether the accuracy of the current prediction by the model has decreased. In this embodiment, the analysis method is to calculate the Mahalanobis distance M(e) of the residual vector e(k). The calculation formula of the Mahalanobis distance is: where e is the residual vector at the current time instant; μ e is the mean vector of the residual vectors at historical time instants, which is a reference center obtained by statistically averaging the residual vectors over a period of time when the motor is in a stable running state and the model is considered to be accurate; and Σ e is the covariance matrix of the residual vectors at historical time instants collected at the same period as μ e , which describes the statistical dispersion of each component of the residual vector and the correlation between the components; is the determinant of the covariance matrix Σ ethe inverse matrix of A; (e - μ e ) is the residual deviation vector, which quantifies numerically the position and direction of the current residual deviation from its historical statistical center; (e - μ e ) T is the transposed residual deviation vector; T denotes the transposition operation of a vector or matrix.

[0054] The analysis and decision unit compares the calculated Mahalanobis distance M(e) with a pre-set dynamic threshold γ. If M(e)≤γ, it is determined that the current state of the model is accurate, and the multi-modal synchronous test system 10 does not perform any operation and continues to enter the loop of the next time point k+1. If M(e) > γ, it is determined that there is a significant mismatch between the model and the actual dynamic characteristics of the motor 1 under test, and the analysis and decision unit triggers a test event and outputs a test event trigger signal to the excitation generation unit to start the subsequent model parameter updating process.

[0055] When the analysis and decision unit outputs the test event trigger signal, the excitation generation unit in the processing module 2 is activated to perform the generation and application of the adaptive excitation sequence. This process aims to actively apply excitation with information to the motor 1 under test according to the specific characteristics of the current model mismatch, and collect high-resolution data that can effectively correct the model parameters.

[0056] Firstly, the excitation generation unit obtains the residual vector or set of residual vectors e(k) that caused the test event to be triggered. It analyzes the residual vector and specifies its dynamic characteristics as time domain features and frequency domain features. The analysis of the time domain features includes determining which component (e.g., temperature residual component or vibration residual component) in the residual vector has the largest contribution to the exceedance of the Mahalanobis distance.

[0057] The analysis of the frequency domain features includes performing a Fast Fourier Transform (FFT) on the time series of residual vectors (especially the vibration residual component) in a small time window before the trigger event to obtain a frequency spectrum. By analyzing the frequency spectrum, it can be determined whether there is a concentration of energy at a specific frequency point that indicates that the model is inaccurate in describing the vibration characteristics at the corresponding speed.

[0058] After completing the specification of the residual dynamic characteristics, the excitation generation unit generates an adaptive excitation sequence corresponding to the analyzed time domain and frequency domain features according to the pre-set mapping rules.

[0059] For example, if the time domain feature analysis shows that the temperature residual is the main reason for the model mismatch, the excitation generation unit generates a load torque excitation sequence. This sequence can be one or more step load torque commands, and the amplitude is set in a range that can cause significant changes in the thermal state of the motor.

[0060] For example, if the frequency domain feature analysis indicates that there is a significant residual vibration at a certain frequency, the excitation generation unit generates a voltage excitation sequence. The sequence can be a frequency sweep signal (e.g. chirp signal) with its center frequency set to the motor speed associated with the certain frequency and its sweep bandwidth covering a small range around the speed point.

[0061] After the excitation sequence is generated as a series of digital instructions, the excitation generation unit applies the excitation sequence to the motor under test 1 through the control interface. If the generated sequence is a load torque excitation sequence, the instructions are sent to the programmable load unit 15 in the multi-modal synchronous test system 10 for execution. If the generated sequence is a voltage excitation sequence, the instructions are sent to the external drive controller of the motor under test 1 for execution. In synchronization with the application of the excitation sequence, the processing module 2 instructs the synchronization and acquisition unit 16 to perform data acquisition at a higher data sampling rate than the regular test, to obtain high resolution data.

[0062] After the application of the adaptive excitation sequence is completed and the corresponding high resolution data acquisition is finished, the parameter update unit in the processing module 2 is activated to perform online update of the model parameters of the state observer model using the newly acquired data.

[0063] The parameter update unit first receives the high resolution time series data acquired during the excitation from the data interface and the pre-processing unit. The data includes a series of control input vectors u(k) = {u(l), u(2), …, u(N)} and the corresponding observation vectors y(k) = {y(l), y(2), …, y(N)}, where N is the length of the high resolution data sequence, which is a time-continuous sequence. Each point in the sequence includes a control input vector u(k) (e.g. voltage, speed) and the actual observation vector y(k) (e.g. measured temperature, vibration) synchronized with it.

[0064] In this embodiment, the parameter update unit employs a parameter identification algorithm to process the high resolution data set, to solve and update the model parameters of the state observer model. The goal of the algorithm is to minimize the cumulative error between the predicted output calculated by the model based on the input data and the actual observation output recorded in the high resolution data set, by adjusting the model parameters.

[0065] The parameter identification algorithm is a recursive process. It processes the data points in the high resolution data set one by one in time sequence. For each data point, the algorithm performs the following operations: First, the algorithm calculates a predicted output value using the model parameters updated at the previous time step, combined with the input data at the current time step.

[0066] Then, the algorithm compares this predicted output value with the actual measured value at the current time instant, and thus calculates a prediction error vector. The prediction error vector quantifies the inaccuracy of the current model parameters at this point.

[0067] Next, the algorithm calculates a gain based on an uncertainty assessment it maintains internally (the uncertainty assessment reflects the confidence in the current parameter estimates). The gain is used to determine how much the current prediction error can be used to correct the model parameters.

[0068] Finally, the algorithm multiplies the prediction error vector with the calculated gain to obtain a parameter correction, and applies the correction to the existing model parameter vector to generate an updated model parameter vector. At the same time, the algorithm also updates its internal uncertainty assessment in preparation for the next data point.

[0069] During this recursive process, the algorithm also applies a forgetting factor mechanism. The mechanism serves to reduce the weight of historical data on parameter updates and correspondingly increase the weight of current new data points in each iteration. This enables the model update process to give priority to the latest motor dynamic characteristics detected by the adaptive excitation, and thus to track changes in motor parameters over time or operating conditions.

[0070] After all data points in the high-resolution data set have been processed, the recursive process ends. The final model parameter vector obtained at this point is considered as the final result of the current online update. The parameter update unit converts the updated parameter vector into a new state space parameter matrix A and passes it to the model operation unit to replace the original model parameters.

[0071] At this point, a complete online model parameter update process is completed. The updated state observer model will be immediately applied to the residual calculation at subsequent time points, thus completing one closed-loop cycle of the online test and self-optimization phase S2.

[0072] After the execution of the online test and self-optimization phase S2 ends, for example when the preset total test duration is reached or the motor 1 under test is stopped, the test method enters the final global processing and archiving phase S3. The task of this phase is to use the final state observer model obtained after the cyclic update in the S2 phase to perform a one-time, globally optimized processing of all data collected during the test, to ensure the global consistency and accuracy of the final compensation result.

[0073] First, the global processing and archiving unit within the processing module 2 invokes the fixed-interval smoothing algorithm. The fixed-interval smoothing algorithm takes as input the entire time series of data collected during the online testing phase (S2) and processes these data using the model parameters that have been updated up to the final update. The execution of the fixed-interval smoothing algorithm involves a forward pass and a backward pass. The forward pass starts from the beginning of the data and processes in time sequence to the end. The backward pass starts from the end of the data and processes in reverse to the beginning. By combining the results of the two passes, the fixed-interval smoothing algorithm is able to compute for each time point a globally optimal state vector estimate that utilizes all the historical and future information. The output of this computation is a complete and smooth state trajectory that contains the best estimate sequence of the real-time winding temperature and the real-time loss torque of the motor 1 during the entire testing process.

[0074] Next, the global processing and archiving unit performs global compensation calculations based on the smoothed state trajectory, which is divided into two parts: The first part is the copper loss compensation of the motor. For each time point during the testing process, the global processing and archiving unit obtains the real-time winding temperature estimate output by the smoothing algorithm. Based on this real-time temperature value, a pre-determined reference temperature value of the winding resistance, and the known temperature coefficient of resistance of the motor winding material, the real-time winding resistance at that time point is calculated. Subsequently, using this real-time resistance value and the motor phase current measured at that time point, the accurate copper loss value after temperature compensation is calculated.

[0075] The second part is the output torque compensation of the motor. For each time point during the testing process, the global processing and archiving unit obtains the loss torque estimate output by the smoothing algorithm. At the same time, it obtains the electromagnetic torque calculated by the motor controller or derived from the measured values. By subtracting the loss torque estimate from the electromagnetic torque, the unit calculates a compensated shaft end torque value that better represents the actual output capability of the motor.

[0076] Finally, the global processing and archiving unit performs archiving operations. The contents of the archiving include two types of data: the first type is the time series of dynamic characteristic compensation results calculated above, which cover the entire testing process, i.e., the temperature-compensated copper loss sequence and the loss-compensated output torque sequence that vary with time. The second type is the set of model parameter values obtained after each successful update of the model parameters during the S2 phase, which completely records the entire evolution history of the model parameters with the testing process.

Claims

1. A motor dynamic characteristic compensation test method based on multi-modal data, characterized in that, The method comprises the following steps: S1, system and model construction stage: constructing a multi-modal synchronous test system for synchronously collecting electrical parameter data, temperature data, vibration data and rotating speed data of a measured motor, and constructing a state observer model with initial model parameters; S2, online test and self-optimization stage: starting the multi-modal synchronous test system to continuously collect data, and during the data collection, cyclically performing the following operations: running the state observer model, predicting temperature data and vibration data according to the collected electrical parameter data and rotating speed data, generating a predicted output, and calculating a residual error between the predicted output and the temperature data and vibration data collected during the data collection; analyzing the dynamic characteristics of the residual error, and triggering a test event when the dynamic characteristics of the residual error exceed a preset dynamic threshold; generating an adaptive excitation sequence corresponding to the dynamic characteristics of the residual error according to the triggered test event, and applying the adaptive excitation sequence to the measured motor, while collecting high-resolution data generated under the adaptive excitation sequence, and updating model parameters of the state observer model by using the high-resolution data, wherein the updated model parameters are used to constitute the state observer model in the next cycle; S3, global processing and archiving stage: after the S2 step, performing global compensation calculation on all collected data by using the state observer model updated in the completed cycle, generating dynamic characteristic compensation results, and archiving the dynamic characteristic compensation results and the model parameters updated in the S2 step.

2. The method of claim 1, wherein, The state observer model is a state space model, comprising: a state vector for internal states of the measured motor; a control input vector for control inputs applied to the measured motor; an observation vector for sensor outputs of the measured motor; a model parameter vector for physical characteristics of the measured motor.

3. The method of claim 1, wherein, The step S1 further comprises: performing a preliminary offline parameter identification experiment, and using parameters identified through the offline parameter identification experiment as the initial model parameters.

4. The method of claim 2, wherein, The state vector comprises real-time winding temperature and loss torque.

5. The method of claim 1, wherein, In the step S2, the judgment manner that the dynamic characteristics of the residual error exceed the preset dynamic threshold is: calculating Mahalanobis distance of the residual error, and determining that the dynamic characteristics of the residual error exceed the preset dynamic threshold when the Mahalanobis distance is greater than a preset threshold.

6. The method of claim 1, wherein, In the step S2, the adaptive excitation sequence corresponding to the dynamic characteristics of the residual error is generated by specifically comprising the following steps: specificating the residual error dynamic characteristics corresponding to the triggered test event into frequency domain and time domain characteristics; and generating an excitation sequence for high-resolution scanning at a specific frequency and a specific load point according to the frequency domain and time domain characteristics; The excitation sequence is selected from a group consisting of voltage excitation sequence and load torque excitation sequence.

7. The method of claim 1, wherein, In the step S2, the model parameters of the state observer model are updated by using the high-resolution data, and the specific steps are as follows: An algorithm is selected from the group consisting of recursive least square method and gradient descent method, parameter identification is performed on the high-resolution data, the model parameters are solved and updated, and the prediction error of the state observer model on the high-resolution data is minimized.

8. The method of claim 4, wherein, The global compensation calculation in step S3 includes: The copper loss of the measured motor is dynamically compensated by using the winding real-time temperature estimated by the state observer model updated after completing the cycle; The electromagnetic torque of the measured motor is compensated by using the loss torque estimated by the state observer model updated after completing the cycle, to obtain the effective load torque.

9. The method of claim 8, wherein, Before the global compensation calculation in step S3, the following steps are further included: A fixed interval smoothing algorithm is used to perform state re-estimation on all data by using the state observer model updated after completing the cycle, to obtain an optimal smoothed state trajectory, and the global compensation calculation is performed based on the optimal smoothed state trajectory.

10. A motor dynamic characteristic compensation test platform based on multi-modal data, characterized in that, The method for compensating the dynamic characteristics of a motor based on multi-modal data according to any one of claims 1-9, comprising: A multi-modal synchronous test system is used to synchronously collect electrical parameter data, temperature data, vibration data, and speed data of a measured motor. A processing module is configured to: Construct a state observer model with initial model parameters; During continuous data collection by the multi-modal synchronous test system, the following operations are cyclically performed: Run the state observer model to predict temperature data and vibration data according to the real-time collected electrical parameter data and speed data, generate a predicted output, and calculate the residual error between the predicted output and the temperature data and vibration data collected during the data collection; Analyze the dynamic characteristics of the residual error, and trigger a test event when the dynamic characteristics exceed a preset dynamic threshold; According to the test event, generate an adaptive excitation sequence corresponding to the dynamic characteristics of the residual error, and apply it to the measured motor, collect high-resolution data generated under the excitation sequence, and update the model parameters of the state observer model by using the high-resolution data, wherein the updated model parameters are used to construct the state observer model in the next cycle; And configure to use the state observer model updated after completing the cycle to perform global compensation calculation on all collected data after completing the cyclic operation, generate a dynamic characteristic compensation result, and archive the compensation result and the cyclic update value of the model parameters completed in the cyclic operation.

Images