Motor test system and test method

Abstract

The invention relates to the technical field of motor performance testing, in particular to a motor testing system and method, and the method comprises the steps: respectively calculating a voltage zero offset, a current range coefficient and a rotating speed difference value in collected no-load sampling data, and outputting a correction coefficient; an isolated forest algorithm is introduced to identify path length abnormal points in a multi-channel difference value set, abnormal data caused by noise interference, drifting and transient impact are accurately positioned, and a signal sequence is kept stable and consistent before entering an analysis link through weight weighting gap filling and continuous fluctuation correction. The method comprises the following steps: segmenting a stable operation sequence into time frame segments according to equal-length intervals, calculating amplitude and phase distribution, extracting multi-scale features by adopting a convolutional neural network, realizing correlation identification of a local energy peak value and a phase weight, enhancing the expression ability of the features and the mode discrimination, and performing similarity comparison with an existing mode to complete working condition judgment. And the classification result under the complex working condition is more reliable.

Description

Technical Field

[0001] This invention relates to the field of motor performance testing technology, and in particular to a motor testing system and testing method. Background Technology

[0002] The field of motor performance testing technology involves measuring, analyzing, and evaluating the electrical and mechanical parameters of motors to determine their operating status, performance indicators, and compliance with design or usage requirements. It encompasses testing equipment, measurement sensors, signal processing devices, control systems, and data analysis software, and its applications cover factory testing during motor manufacturing, performance verification during motor R&D, and maintenance and diagnostics during operation.

[0003] A motor testing system is designed to accurately acquire key operating parameters of a motor, assess whether it meets design specifications and usage requirements, and complete the testing and analysis of multiple performance indicators in a short time. This results in high precision and repeatability of test results, thereby improving testing efficiency and quality control.

[0004] Existing technologies lack multi-channel unified benchmark calibration of initial data during acquisition and analysis. This makes them prone to the accumulation of subsequent calculation errors due to sensor zero-point drift or range deviation. In operational status monitoring, they cannot effectively identify short-term interference or complex fluctuations hidden in multi-dimensional parameters, leading to misjudgments or omissions. Signal feature extraction often focuses on the amplitude or frequency statistics of a single parameter, lacking in-depth analysis of the joint distribution of amplitude and phase, resulting in insufficient differentiation in operating condition identification. Performance prediction mainly relies on extrapolation of single-parameter trends, failing to integrate the coupled changes of related parameters such as temperature rise, power, speed, and torque, thus limiting the accuracy and applicability of prediction results. Safety protection relies solely on single-threshold alarms, lacking early intervention of predictive information, which may lead to failure to respond promptly to anomalies such as overload and leakage, thereby causing equipment performance degradation or increased operational risks. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of the existing technology and to propose a motor testing system and testing method.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: a motor testing method comprising: S1: Based on the no-load sampling data of the voltage channel, current channel and speed detector of the stator winding test unit, calculate the voltage zero point offset, current range coefficient and speed difference and output the correction coefficient. Combine the parameter matrices of each channel and write them into the parameter register table of the console to generate the calibration parameter matrix. S2: Based on the calibration parameter matrix, the power load stage temperature rise, speed, vibration output and reference value are segmented. The isolated forest algorithm is used to calculate the difference and mark the anomaly. The gap is filled by multi-channel weighted filling, the continuous fluctuation segment is smoothed and the correction value is written back to generate a stable operation sequence. S3: Based on the stable operation sequence, a convolutional neural network is used to segment the fixed frequency voltage sequence of the rotor drive unit and calculate the amplitude and phase distribution by weighting. The peak value of the amplitude energy is accumulated and the feature vector is generated by weighting the peak value and phase. The existing patterns are compared and the item with the highest similarity is selected to generate a set of operating conditions. S4: Based on the aforementioned operating condition mode set, solve for the temperature rise coefficient, power coefficient, speed and torque coefficient, combine multiple coefficients to deduce the future temperature rise, power, speed and torque curves, and generate an operation prediction curve; S5: Based on the predicted operation curve, compare the peak power with the current limit value and write it into the current limit register table. Compare the casing to ground current with the leakage threshold and trigger the power-off protection. Load the customer model parameters into the console parameter register table. Archive the test records and judgment tags and generate the factory test file.

[0007] As a further aspect of the present invention, the calibration parameter matrix includes voltage offset, current range coefficient, and speed correction coefficient; the stable operation sequence includes temperature rise correction value, speed correction value, and vibration correction value; the operating condition mode set includes normal operation mode, light load mode, and fault mode; the operation prediction curve includes temperature rise change curve, power change curve, speed change curve, and torque change curve; and the factory test file includes batch number, test data record, and operating condition judgment label.

[0008] As a further aspect of the present invention, the specific steps for generating the calibration parameter matrix are as follows: Based on the no-load sampling data of the voltage channel, current channel and speed detector of the stator winding test unit, the initial output of the voltage channel under zero load is measured and compared with the standard reference value to obtain the zero offset. The output of the current channel under different loads is measured and compared with the range calibration value to obtain the range coefficient. The output of the speed detector at multiple speed points is collected and compared with the standard speed difference to generate a multi-channel initial parameter set. Based on the multi-channel initial parameter set, the three parameters, including voltage zero-point offset, current range coefficient, and speed difference, are arranged in channel order to form a parameter matrix. This matrix is ​​written into the console parameter register table as a reference for subsequent test calls, completing the matrix data storage and index update, and generating the calibration parameter matrix.

[0009] As a further aspect of the present invention, the specific steps for generating the stable operating sequence are as follows: Based on the calibration parameter matrix, the temperature rise, speed, and vibration output values ​​of the power load stage are segmented with the reference values ​​in chronological order. The difference between the temperature rise value and the reference temperature rise for each segment is calculated and recorded, the difference between the speed value and the reference speed for each segment is calculated and recorded, and the difference between the vibration value and the reference vibration for each segment is calculated and recorded, thus generating a multi-channel difference set. Based on the multi-channel difference set, the isolated forest algorithm is used to extract adjacent power, temperature rise, rotation speed and vibration values ​​at the marked anomaly locations. The weighted average replacement gap value is calculated according to the channel weight, the continuity of the replacement sequence is checked, and the fluctuation section is smoothed to generate the corrected channel sequence. Based on the corrected channel sequence, the corrected power, temperature rise, speed, and vibration data are recombined into a multi-channel sequence according to the time index, maintaining the time order for output, and generating a stable operating sequence.

[0010] As a further aspect of the present invention, the isolated forest, after acquiring the multi-channel difference set, constructs multiple random splitting trees for each difference sample, randomly selects parameter items and randomly sets a splitting threshold in each tree, continuously divides the data space where the sample is located until the sample is completely isolated, records the average path length of the sample in each tree, and compares the path length with a threshold set according to the normal data distribution, determines that samples below the threshold are outliers, marks the outlier locations and extracts adjacent power, temperature rise, rotation speed, and vibration values, calculates the weighted average replacement gap value according to the channel preset weight, performs a continuity check on the replaced sequence, and performs smoothing processing on continuous fluctuation segments.

[0011] As a further aspect of the present invention, the specific steps for generating the stable operating sequence are as follows: Based on the stable operation sequence, the time series of the rotor drive unit under a fixed frequency voltage is divided into frames with equal intervals. A convolutional neural network is used to extract features from the amplitude and phase distribution of the frames. The amplitude distribution of each frame is calculated and recorded, and the phase distribution of each frame is calculated and recorded to generate an amplitude and phase distribution set. Based on the amplitude and phase distribution set, the energy of adjacent intervals is accumulated sequentially in the amplitude distribution of each frame segment to determine and mark the energy peak position. The peak position is recalculated and recorded according to the set weight in combination with the phase distribution to generate a feature vector set. Based on the feature vector set, each feature vector is compared with the existing working condition mode features for similarity item by item. The item with the highest similarity is selected and the corresponding mode label is recorded. The mode labels are then summarized to form a working condition mode set.

[0012] As a further aspect of the present invention, the convolutional neural network, after obtaining a stable operating sequence, acquires the time sequence of the rotor drive unit under a fixed frequency voltage condition, and divides it into frames at equal intervals. For each frame, the amplitude distribution and phase distribution are extracted. In the convolution operation, multiple sets of convolution kernels slide along the time and parameter dimensions to calculate the local feature response. After the response result is nonlinearly mapped, sensitive information on temperature rise, power, speed, and vibration characteristic patterns is retained. Then, through multi-layer convolution and pooling operations, the data dimension is gradually compressed to extract a high-dimensional feature vector that can reflect the combination change law of amplitude and phase.

[0013] As a further aspect of the present invention, the specific steps for generating the running prediction curve are as follows: Based on the aforementioned operating condition mode set, the original values ​​of temperature rise, power, speed, and torque within a set time period are extracted. The temperature rise change is calculated according to the time index and divided by the corresponding time difference to obtain the temperature rise coefficient. The power change is calculated and divided by the corresponding time difference to obtain the power coefficient. The product of speed and torque is calculated and divided by a set base value to obtain the coupling coefficient. A multi-parameter coefficient set is generated. The temperature rise coefficient, power coefficient, and coupling coefficient are substituted into the initial values ​​of temperature rise, power, speed, and torque, respectively. The predicted values ​​for each future time are calculated by accumulating them at a fixed time step. The predicted values ​​are combined in chronological order to form a temperature rise curve, a power curve, a speed curve, and a torque curve, generating an operating prediction curve.

[0014] As a further aspect of the present invention, the specific steps for generating the factory test file are as follows: Based on the operation prediction curve, the peak value of the power curve is extracted and compared with the current limiting setting value. In case of exceeding the limit, the current limiting value is recorded and written into the current limiting register table. The casing to ground current curve is extracted and compared with the leakage threshold. In case of exceeding the limit, the abnormal state is recorded and a power failure protection command is generated, forming an operation protection command set. Based on the aforementioned operation protection instruction set, the control parameters corresponding to the customer client model are loaded into the console parameter register table, and the test records and operating condition judgment tags are organized into archive files according to batch numbers and stored in the file management unit to generate factory test files.

[0015] A motor testing system, which performs the above-described motor testing method, includes the following steps: Sampling and calibration module: Based on the unloaded data from the voltage channel, current channel, and speed detector, calculate the voltage zero-point offset, current range coefficient, and speed difference. Combine the three results with the channel number and write them into the console parameter register to establish a calibration parameter matrix. Anomaly Correction Module: Based on the calibration parameter matrix, extract the differences between temperature rise, rotation speed, vibration and reference values, obtain anomaly points in median rate of change, extract neighboring observations and weight them according to the correlation between historical error and difference, replace anomaly points, replace large differential segments with boundary mean interpolation, verify the anomaly location using the isolated forest algorithm and write it into the module, and construct a stable operating sequence. Driven analysis module: Based on the stable operating sequence, it divides the fixed-cycle voltage segment, extracts the amplitude point and phase shift, calculates the weighted value and forms the concentration vector, extracts the peak value and combines the composite features, uses a convolutional neural network to match the existing vector set, obtains the minimum distance term, and establishes the operating condition mode set; Performance calculation module: Based on the set of operating conditions, calculate the proportional coefficients of temperature rise, power, speed, and torque channels, perform vector outer product to obtain interpolation points, construct temperature rise, power, speed, and torque curves, and establish operation prediction curves; Protection and archiving module: Based on the operation prediction curve, extract the power peak value and compare it with the current limit value, record the over-limit point to the current limit register, compare the over-limit current difference to ground to trigger power-off, write the model parameters, archive the curve and status, and establish the factory test file.

[0016] Compared with the prior art, the advantages and positive effects of the present invention are as follows: In this invention, by calculating the voltage zero-point offset, current range coefficient, and speed difference from the collected no-load sampling data and outputting the correction coefficient, the parameters of different channels are matrixed and stored in the parameter register table, so that subsequent measurements are performed with a unified reference.

[0017] In this invention, the isolated forest algorithm is introduced to identify path length anomalies in the multi-channel difference set, so as to accurately locate abnormal data caused by noise interference, drift and transient impact. Furthermore, by weighted filling of gaps and continuous fluctuation correction, the signal sequence is kept stable and consistent before entering the analysis stage. In this invention, the stable operating sequence is divided into time frames at equal intervals, the amplitude and phase distribution are calculated, and a convolutional neural network is used to extract multi-scale features. This enables the correlation and identification of local energy peaks and phase weights, enhancing the expressive power and pattern distinguishability of the features. Then, the similarity is compared with existing patterns to complete the working condition determination, making the classification results under complex working conditions more reliable. In this invention, the coefficient combination is used as input to deduce the curves of various operating parameters in the future, forming a trend prediction. The peak power value and current limit value, the casing-to-ground current and leakage threshold are compared respectively to trigger over-limit protection operation and parameter writing. At the same time, test records and judgment tags are archived in batches, which significantly improves the stability of the test results, the accuracy of diagnosis, the predictive foresight and the safety protection response. Attached Figure Description

[0018] Figure 1 This is a system flowchart of the present invention; Figure 2 This is a schematic diagram of the method steps of the present invention. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0020] Example 1 Please see Figure 1 The present invention provides a technical solution: a motor testing method comprising: S1: Based on the no-load sampling data of the voltage channel, current channel and speed detector of the stator winding test unit, calculate the voltage zero point offset, current range coefficient and speed difference and output the correction coefficient. Combine the parameter matrices of each channel and write them into the parameter register table of the console to generate the calibration parameter matrix. S2: Based on the calibration parameter matrix, the power load stage temperature rise, speed, vibration output and reference value are segmented. The isolated forest algorithm is used to calculate the difference and mark the anomaly. The gap is filled by multi-channel weighted filling, the continuous fluctuation segment is smoothed and the correction value is written back to generate a stable operation sequence. S3: Based on the stable operation sequence, a convolutional neural network is used to segment the fixed frequency voltage sequence of the rotor drive unit and calculate the amplitude and phase distribution by weighting. The amplitude energy is accumulated to mark the peak value. The feature vector is generated by weighting the peak value and phase. The existing patterns are compared to select the item with the highest similarity to generate the working condition pattern set. S4: Based on the set of operating conditions, solve for the temperature rise coefficient, power coefficient, speed and torque coefficient, combine multiple coefficients to deduce the future temperature rise, power, speed and torque curves, and generate the operation prediction curve; S5: Based on the operation prediction curve, compare the peak power with the current limit value and write it into the current limit register table if it exceeds the limit. Compare the casing to ground current with the leakage current threshold and trigger the power-off protection if it exceeds the limit. Load the customer's model parameters into the console parameter register table, archive the test records and judgment tags, and generate the factory test file.

[0021] The calibration parameter matrix includes voltage offset, current range coefficient, and speed correction coefficient; the stable operation sequence includes temperature rise correction value, speed correction value, and vibration correction value; the operating mode set includes normal operation mode, light load mode, and fault mode; the operation prediction curves include temperature rise change curve, power change curve, speed change curve, and torque change curve; and the factory test file includes batch number, test data record, and operating condition judgment label.

[0022] The specific steps for generating the calibration parameter matrix are as follows: Based on the no-load sampling data of the voltage channel, current channel and speed detector of the stator winding test unit, the initial output of the voltage channel under zero load is measured and compared with the standard reference value to obtain the zero offset. The output of the current channel under different loads is measured and compared with the range calibration value to obtain the range coefficient. The output of the speed detector at multiple speed points is collected and compared with the standard speed difference to generate a multi-channel initial parameter set. Based on the multi-channel initial parameter set, the three parameters including voltage zero-point offset, current range coefficient and speed difference are arranged in channel order to form a parameter matrix, which is written into the console parameter register table as the reference for subsequent test calls. The matrix data storage and index update are completed to generate the calibration parameter matrix. Based on the no-load sampling data acquired by the voltage channel of the stator winding test unit, the Fast Fourier Transform (FFT) algorithm was used. The sampling points were set to 1024, the sampling frequency to 10000Hz, and a Hanning window was selected to window the sampled signal. The windowed signal was then input into the FFT operation process. The amplitude corresponding to the zero-Hz position in the spectrum was extracted as the DC component amplitude. The difference between the DC component amplitude and the standard reference value of 10.000V was calculated and recorded as the zero-point offset of the voltage channel. Based on the current channel of the stator winding test unit under multi-load conditions... The data is processed using a least squares regression algorithm. The regression order is set to first order, the input independent variable is the current value provided by the standard current source, and the input dependent variable is the current value collected by the sensor. Regression calculation is performed on the input data, and the slope of the regression result is extracted as the current channel range coefficient. Based on the output signal collected by the speed detector at multiple speed points, the time interval inversion method is used. The sampling time interval is set to one second, and the input is the total number of pulses output by the detector within the time. The total number of pulses is converted into a speed value and the difference is calculated with the standard speed. The difference is recorded as the speed difference, and a multi-channel initial parameter set is generated. Based on the multi-channel initial parameter set, a matrix splicing and storage method is adopted. The voltage zero-point offset, current range coefficient and speed difference are arranged in the channel order to generate a two-dimensional parameter matrix. The matrix data is then written to the console parameter register table in the order of register start address 0x2000. The index table number is set to 3 and the number of index items is updated to the total number of parameter matrix elements to generate the calibration parameter matrix.

[0023] The specific steps for generating a stable running sequence are as follows: Based on the calibration parameter matrix, the temperature rise, speed, and vibration output values ​​of the power load stage are segmented with the reference values ​​in chronological order. The difference between the temperature rise value and the reference temperature rise in each segment is calculated and recorded, the difference between the speed value and the reference speed in each segment is calculated and recorded, and the difference between the vibration value and the reference vibration in each segment is calculated and recorded, generating a multi-channel difference set. Based on a multi-channel difference set, the isolated forest algorithm is used to extract adjacent power, temperature rise, rotation speed and vibration values ​​at the marked anomaly locations. The weighted average replacement gap value is calculated according to the channel weight, and the continuity of the replaced sequence is checked. Smoothing correction is performed on the fluctuation segment to generate the corrected channel sequence. Based on the corrected channel sequence, the corrected power, temperature rise, speed, and vibration data are recombined into a multi-channel sequence according to the time index and output in time order to generate a stable operating sequence. Based on the calibration parameter matrix, a time series segmentation method is adopted, with a segment length of 60 seconds, a segment start time point of zero sampling start time, and a segment step size of 60 seconds. The power load platform's temperature rise output value, speed output value, vibration output value, and corresponding reference value are segmented in chronological order. The temperature rise output value in each segment is sequentially compared with the reference temperature rise value, and the difference is recorded. The speed output value in each segment is sequentially compared with the reference speed value, and the difference is recorded. The vibration output value in each segment is sequentially compared with the reference vibration value, and the difference is recorded, generating a multi-channel difference set. Based on a multi-channel difference set, the isolated forest algorithm is adopted, with 100 trees, a maximum sample size of 256 per tree, a feature sampling ratio of 1.0, and a random seed of 42. The difference set is trained and inferred to mark abnormal locations. At the marked abnormal locations, the sampled values ​​of power, temperature rise, speed, and vibration channels within adjacent time periods are extracted. The weight values ​​are set according to the channels: power channel 0.4, temperature rise channel 0.3, speed channel 0.2, and vibration channel 0.1. The adjacent values ​​are weighted and averaged according to the set weights, and the result is used to replace the gap value at the abnormal location. Then, the temporal continuity of the replaced complete sequence is checked, and the detected fluctuation segment is smoothed and corrected using the three-point moving average method to generate the corrected channel sequence. Based on the corrected channel sequence, a time index recombination method is used to read the corrected power, temperature rise, speed and vibration data in the order of time index. The data of each channel are recombined into a two-dimensional multi-channel data sequence at the same time point and output in the order of time to generate a stable operating sequence.

[0024] The isolated forest algorithm, after acquiring a multi-channel difference set, constructs multiple random splitting trees for each difference sample. In each tree, parameters are randomly selected and splitting thresholds are randomly set. The data space where the sample is located is continuously divided until the sample is completely isolated. The average path length of the sample in each tree is recorded, and the path length is compared with the threshold set according to the normal data distribution. Samples below the threshold are identified as outliers. After marking the outlier locations, adjacent power, temperature rise, rotation speed, and vibration values ​​are extracted. The weighted average replacement gap value is calculated according to the preset weight of the channel, and a continuity check is performed on the replaced sequence. At the same time, smoothing is performed on continuous fluctuation segments. An isolated forest, according to the formula:

[0025] in: Indicates the first The weighted average of the test channels for each motor. Indicates the first The weights of each feature Indicates the first The first motor test channel The values ​​of each feature, This represents the total number of features. The adjustment coefficient representing the temperature rise characteristic. The adjustment coefficient representing the vibration characteristics. Indicates the dynamic correction factor. Indicates the rate of change of the vibration signal; Execution process: First, abnormal data is detected and marked. Then, in each motor test channel, feature values ​​including power, temperature rise, speed, and vibration are extracted. For each channel, a weighted average method is used to calculate replacement values ​​for missing values. According to the given weight coefficients Perform weighted processing to calculate the original weighted average. Next, an adjustment coefficient for the temperature rise characteristic is introduced. Adjustment coefficient of vibration characteristics Adjustments are made based on the fluctuation amplitude and stability of temperature rise and vibration characteristics in the dataset to enhance the influence of these characteristics in the calculation. When the temperature rise data changes little or is relatively stable... It will be set to a lower value when the vibration signal fluctuates greatly. The increase reflects the importance of the test results, followed by a dynamic adjustment factor. Calculations are performed based on the time-domain changes of the motor test channel to reflect the trend changes in data during motor operation. This ensures that abnormal fluctuations can be identified and corrected by comparing data before and after the test, and the rate of change of the vibration signal is measured. To help balance the influence of vibration data on the results, the final weighted average is obtained by combining the corrected parameters with the original weighted average. This generates corrected motor test channel data, ensuring the accuracy, stability, and reliability of the motor test data.

[0026] The specific steps for generating a stable running sequence are as follows: Based on the stable operation sequence, the time series of the rotor drive unit under a fixed frequency voltage is divided into frames with equal intervals. A convolutional neural network is used to extract features of the amplitude and phase distribution of the frames. The amplitude distribution of each frame is calculated and recorded, and the phase distribution of each frame is calculated and recorded to generate an amplitude and phase distribution set. Based on the amplitude and phase distribution set, the energy of adjacent intervals is accumulated sequentially in the amplitude distribution of each frame segment to determine and mark the energy peak position. The peak position is recalculated and recorded according to the set weight in combination with the phase distribution to generate a feature vector set. Based on the feature vector set, each feature vector is compared with the existing working condition mode features for similarity item by item. The item with the highest similarity is selected and the corresponding mode label is recorded. The mode labels are then summarized to form a working condition mode set. Based on a stable operating sequence, a fixed-length time window segmentation method is adopted, with a time window length of 256 sampling points, a time window step size of 128 sampling points, and a sampling frequency of 5000 Hz. The time sequence of the rotor drive unit under a fixed frequency voltage is segmented into frames according to the set time window. A convolutional neural network is adopted, with three convolutional layers. The first layer has a kernel size of 3×3 and 32 kernels, the second layer has a kernel size of 3×3 and 64 kernels, and the third layer has a kernel size of 3×3 and 128 kernels. The activation function type of each layer is ReLU, the stride is 1, and the padding method is same. Channel-wise convolution operation is performed on the amplitude distribution of each segmented frame and the convolution output is recorded. Channel-wise convolution operation is performed on the phase distribution of each segmented frame and the convolution output is recorded. The amplitude convolution output is calculated and recorded in the order of sampling points within the frame and the phase convolution output is calculated and recorded in the order of sampling points within the frame, generating an amplitude-phase distribution set. Based on the amplitude and phase distribution set, the cumulative energy peak location method is adopted. The energy calculation interval length is set to 5 sampling points and the step size is 1 sampling point. The energy values ​​in adjacent intervals are accumulated sequentially for the amplitude distribution of each frame segment to determine and mark the energy peak position. The weighted phase correction method is adopted, with the amplitude weight set to 0.7 and the phase weight set to 0.3. The energy peak position is recalculated and recorded in combination with the phase distribution to generate a feature vector set. Based on the feature vector set, the cosine similarity comparison method is used to calculate the similarity between each feature vector and the existing working condition mode features in turn. The vector length normalization parameter is set to 1.0 in the similarity calculation, and the similarity threshold comparison is not truncated. The item with the largest similarity is selected and the corresponding mode label is recorded. All mode labels are summarized in order to form a mode label set, and the working condition mode set is generated.

[0027] After obtaining a stable operating sequence, the convolutional neural network acquires the time series of the rotor drive unit under a fixed frequency and voltage condition, and divides it into frames with equal intervals. For each frame, the amplitude distribution and phase distribution are extracted. In the convolution operation, multiple sets of convolution kernels slide along the time and parameter dimensions to calculate the local feature response. After the response result is nonlinearly mapped, sensitive information on temperature rise, power, speed and vibration feature patterns is retained. Then, through multi-layer convolution and pooling operations, the data dimension is gradually compressed to extract a high-dimensional feature vector that can reflect the combination of amplitude and phase changes. Convolutional neural networks, according to the formula:

[0028] in: Indicates the first Amplitude distribution of the frame segment Indicates the first Weighting coefficients at time points, Indicates the first The signal amplitude at time [time]. This indicates the length of a frame segment, which is the total number of sampling points within that segment. This represents the maximum amplitude correction factor. Indicates the maximum amplitude of the frame segment. This represents the average amplitude correction factor. Indicates the average amplitude of the frame segment. This represents the standard deviation correction factor. Indicates the standard deviation of the amplitude of a frame segment; Execution process: First, the signal during motor operation is divided into multiple frames with equal intervals. Each frame contains a certain number of sampling points. For each frame, the time interval within the frame is calculated. signal amplitude The absolute value, and through the weighting coefficient These amplitudes are weighted to obtain the total amplitude of the frame, and then three correction coefficients are introduced. , and It is used to correct the maximum amplitude, average amplitude, and amplitude standard deviation of a frame segment, respectively. The maximum amplitude correction factor is... Adjust the maximum amplitude within the frame segment The contribution to the calculation results reflects the most intense amplitude change in the signal; the average amplitude correction coefficient. Adjust the average amplitude of all sampling points within the frame segment The standard deviation correction factor is used to balance the overall amplitude of the signal at different times. Adjust the standard deviation of the frame amplitude To balance the impact of signal fluctuations, a weighted calculation is performed, and then combined with correction coefficients to obtain the final corrected amplitude distribution. It can not only accurately extract the amplitude characteristics of motor signals, but also effectively eliminate interference caused by noise and fluctuations, ensuring that the data during motor testing is more stable and reliable.

[0029] The specific steps for generating the running prediction curve are as follows: Based on the set of operating conditions, the original values ​​of temperature rise, power, speed and torque within a set time period are extracted. The temperature rise change is calculated sequentially according to the time index and divided by the corresponding time difference to obtain the temperature rise coefficient. The power change is calculated and divided by the corresponding time difference to obtain the power coefficient. The product of speed and torque is calculated and divided by the set base value to obtain the coupling coefficient, and a set of multi-parameter coefficients is generated. Based on a multi-parameter coefficient set, the temperature rise coefficient, power coefficient, and coupling coefficient are substituted into the initial values ​​of temperature rise, power, speed, and torque, respectively. The predicted values ​​at each future moment are calculated by accumulating them at a fixed time step. The predicted values ​​are then combined in time order to form temperature rise curve, power curve, speed curve, and torque curve, thereby generating an operation prediction curve. Based on the operating condition mode set, a fixed-length time period index extraction method is adopted. The time period length is set to 600 seconds and the time index step size is 1 second. The original sampled values ​​of the temperature rise channel, power channel, speed channel and torque channel within the set time period are extracted from the operating condition mode set. The temperature rise coefficient is obtained by calculating the difference between the temperature rise values ​​of two adjacent time points according to the time index and dividing the difference by the corresponding time difference. The power coefficient is obtained by calculating the difference between the power values ​​of two adjacent time points and dividing the difference by the corresponding time difference. The speed value of each time point is multiplied by the torque value of the corresponding time point and the product is divided by the set base value of 2000. A multi-parameter coefficient set is generated. Based on a multi-parameter coefficient set, a fixed-step cumulative prediction method is adopted. A fixed time step of 1 second is set. The temperature rise coefficient is substituted into the initial temperature rise value in sequence and the predicted temperature rise value at each future time is calculated by accumulating the values ​​according to the time step. The power coefficient is substituted into the initial power value in sequence and the predicted power value at each future time is calculated by accumulating the values ​​according to the time step. The coupling coefficient is substituted into the initial speed and torque values ​​in sequence and the predicted speed and torque values ​​at each future time are calculated by accumulating the values ​​according to the time step. All the predicted temperature rise values ​​are combined into a temperature rise curve in time order, all the predicted power values ​​are combined into a power curve in time order, all the predicted speed values ​​are combined into a speed curve in time order, and all the predicted torque values ​​are combined into a torque curve in time order, thus generating an operation prediction curve.

[0030] The specific steps for generating the factory test file are as follows: Based on the operation prediction curve, the peak value of the power curve is extracted and compared with the current limiting setting value. In case of exceeding the limit, the current limiting value is recorded and written into the current limiting register table. The casing to ground current curve is extracted and compared with the leakage threshold. In case of exceeding the limit, the abnormal state is recorded and a power failure protection command is generated, forming an operation protection command set. Based on the operation protection instruction set, the corresponding control parameters of the customer client model are loaded into the parameter register table of the console, and the test records and operating condition judgment tags are organized into archive files according to batch number and stored in the file management unit to generate the factory test file. Based on the operation prediction curve, a threshold comparison and register writing method is adopted. The sliding window length for peak power detection is set to 10 sampling points and the step size is 1 sampling point. The detection window slides in the power curve in time sequence and extracts the maximum power value within the window. The maximum power value is compared with the current limiting setting value. When the detection result is out of limit, the current current limiting setting value is recorded and a binary data frame is generated according to the register writing format. The register writing start address is set to 0x3000 and the data frame length is 2 bytes. The data frame is written to the current limiting register table. A threshold comparison and status marking method is adopted. The sampling point threshold for the casing-to-ground current curve detection is set to the leakage threshold of 50 mA. The values ​​of each sampling point of the casing-to-ground current curve are read in time sequence and compared with the leakage threshold one by one. When the detection result is out of limit, the abnormal status flag is recorded as 1 and a power-off protection command is generated. The protection command is stored in a fixed-length control frame format to generate an operation protection command set. Based on the operation protection instruction set, the parameter loading and file archiving method is adopted. The control parameters corresponding to the customer model are read from the parameter storage unit and written to the console parameter register table in register order. The register start address is set to 0x4000 and the write step size is 1 register unit. The write operation is completed in sequence according to the control parameters. The test records and operating condition judgment tags are combined into an archive file in batch number order. The archive file format is CSV, and the field order is batch number, timestamp, test record content, and operating condition judgment tag. The file encoding format is UTF-8. The archive file is written to the batch corresponding directory of the file management unit to generate the factory test file.

[0031] Please see Figure 2 A motor testing system, which performs the above-mentioned motor testing method, includes the following steps: Sampling and calibration module: Based on the unloaded data from the voltage channel, current channel, and speed detector, calculate the voltage zero-point offset, current range coefficient, and speed difference. Combine the three results with the channel number and write them into the console parameter register to establish a calibration parameter matrix. Anomaly Correction Module: Based on the calibration parameter matrix, extract the differences between temperature rise, rotation speed, vibration and reference values, obtain anomaly points of median rate of change, extract neighboring observations and weight them according to the correlation between historical error and difference, replace anomaly points, replace large differential segments with boundary mean interpolation, and verify the anomaly location using the isolated forest algorithm and write it into the system to build a stable operating sequence. Driven analysis module: Based on stable operating sequence, it divides fixed period voltage segment, extracts amplitude points and phase shift, calculates weighted values ​​and forms concentration vector, extracts peak values ​​and combines composite features, uses convolutional neural network to match existing vector set, obtains minimum distance term, and establishes operating condition mode set; Performance calculation module: Based on the operating condition mode set, calculate the proportional coefficients of temperature rise, power, speed and torque channels, perform vector outer product to obtain interpolation points, construct temperature rise, power, speed and torque curves, and establish operation prediction curves; Protection and archiving module: Based on the operation prediction curve, extract the peak power value and compare it with the current limit value, record the over-limit point to the current limit register, compare the difference between the current to ground and trigger power-off when the difference exceeds the limit, write the model parameters, archive the curve and status, and establish the factory test file.

[0032] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments that can be applied to other fields. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.

Claims

1. A method for testing an electric motor, characterized in that, The system includes: S1: Based on the no-load sampling data of the voltage channel, current channel and speed detector of the stator winding test unit, calculate the voltage zero point offset, current range coefficient and speed difference and output the correction coefficient. Combine the parameter matrices of each channel and write them into the parameter register table of the console to generate the calibration parameter matrix. S2: Based on the calibration parameter matrix, the power load stage temperature rise, speed, vibration output and reference value are segmented. The isolated forest algorithm is used to calculate the difference and mark the anomaly. The gap is filled by multi-channel weighted filling, the continuous fluctuation segment is smoothed and the correction value is written back to generate a stable operation sequence. S3: Based on the stable operation sequence, a convolutional neural network is used to segment the fixed frequency voltage sequence of the rotor drive unit and calculate the amplitude and phase distribution by weighting. The peak value of the amplitude energy is accumulated and the feature vector is generated by weighting the peak value and phase. The existing patterns are compared and the item with the highest similarity is selected to generate a set of operating conditions. S4: Based on the set of operating conditions, solve for the temperature rise coefficient, power coefficient, speed and torque coefficient, combine multiple coefficients to deduce the future temperature rise, power, speed and torque curves, and generate the operation prediction curve; S5: Based on the predicted operation curve, compare the peak power with the current limit value and write it into the current limit register table. Compare the casing to ground current with the leakage threshold and trigger the power-off protection. Load the customer model parameters into the console parameter register table. Archive the test records and judgment tags and generate the factory test file.

2. The motor testing system according to claim 1, characterized in that, The calibration parameter matrix includes voltage offset, current range coefficient, and speed correction coefficient; the stable operation sequence includes temperature rise correction value, speed correction value, and vibration correction value; the operating condition mode set includes normal operation mode, light load mode, and fault mode; the operation prediction curve includes temperature rise change curve, power change curve, speed change curve, and torque change curve; and the factory test file includes batch number, test data record, and operating condition judgment label.

3. The motor testing method according to claim 1, characterized in that, The specific steps for generating the calibration parameter matrix are as follows: Based on the no-load sampling data of the voltage channel, current channel and speed detector of the stator winding test unit, the initial output of the voltage channel under zero load is measured and compared with the standard reference value to obtain the zero offset. The output of the current channel under different loads is measured and compared with the range calibration value to obtain the range coefficient. The output of the speed detector at multiple speed points is collected and compared with the standard speed difference to generate a multi-channel initial parameter set. Based on the multi-channel initial parameter set, the three parameters, including voltage zero-point offset, current range coefficient, and speed difference, are arranged in channel order to form a parameter matrix. This matrix is ​​written into the console parameter register table as a reference for subsequent test calls, completing the matrix data storage and index update, and generating the calibration parameter matrix.

4. The motor testing method according to claim 1, characterized in that, The specific steps for generating the stable running sequence are as follows: Based on the calibration parameter matrix, the temperature rise, speed, and vibration output values ​​of the power load stage are segmented with the reference values ​​in chronological order. The difference between the temperature rise value and the reference temperature rise for each segment is calculated and recorded, the difference between the speed value and the reference speed for each segment is calculated and recorded, and the difference between the vibration value and the reference vibration for each segment is calculated and recorded, thus generating a multi-channel difference set. Based on the multi-channel difference set, the isolated forest algorithm is used to extract adjacent power, temperature rise, rotation speed and vibration values ​​at the marked anomaly locations. The weighted average replacement gap value is calculated according to the channel weight, the continuity of the replacement sequence is checked, and the fluctuation section is smoothed to generate the corrected channel sequence. Based on the corrected channel sequence, the corrected power, temperature rise, speed, and vibration data are recombined into a multi-channel sequence according to the time index, maintaining the time order for output, and generating a stable operating sequence.

5. The motor testing method according to claim 4, characterized in that, The isolated forest, after acquiring the multi-channel difference set, constructs multiple random split trees for each difference sample. In each tree, parameter items are randomly selected and a split threshold is randomly set. The data space where the sample is located is continuously divided until the sample is completely isolated. The average path length of the sample in each tree is recorded, and the path length is compared with the threshold set according to the normal data distribution. Samples below the threshold are identified as outliers. After marking the outlier location, adjacent power, temperature rise, rotation speed, and vibration values ​​are extracted. The weighted average replacement gap value is calculated according to the preset weight of the channel, and a continuity check is performed on the replaced sequence. At the same time, smoothing processing is performed on continuous fluctuation segments.

6. The motor testing method according to claim 1, characterized in that, The specific steps for generating the stable running sequence are as follows: Based on the stable operation sequence, the time series of the rotor drive unit under a fixed frequency voltage is divided into frames with equal intervals. A convolutional neural network is used to extract features from the amplitude and phase distribution of the frames. The amplitude distribution of each frame is calculated and recorded, and the phase distribution of each frame is calculated and recorded to generate an amplitude and phase distribution set. Based on the amplitude and phase distribution set, the energy of adjacent intervals is accumulated sequentially in the amplitude distribution of each frame segment to determine and mark the energy peak position. The peak position is recalculated and recorded according to the set weight in combination with the phase distribution to generate a feature vector set. Based on the feature vector set, each feature vector is compared with the existing working condition mode features for similarity item by item. The item with the highest similarity is selected and the corresponding mode label is recorded. The mode labels are then summarized to form a working condition mode set.

7. The motor testing method according to claim 6, characterized in that, After obtaining a stable operating sequence, the convolutional neural network acquires the time series of the rotor drive unit under a fixed frequency voltage condition, and divides it into frames at equal intervals. For each frame, it extracts the amplitude distribution and phase distribution. In the convolution operation, it calculates the local feature response by sliding multiple sets of convolution kernels along the time and parameter dimensions. After nonlinear mapping, the response results retain sensitive information on temperature rise, power, speed, and vibration characteristic patterns. Then, through multi-layer convolution and pooling operations, it gradually compresses the data dimension and extracts a high-dimensional feature vector that can reflect the combination change law of amplitude and phase.

8. The motor testing method according to claim 1, characterized in that, The specific steps for generating the aforementioned running prediction curve are as follows: Based on the aforementioned operating condition mode set, the original values ​​of temperature rise, power, speed, and torque within a set time period are extracted. The temperature rise change is calculated according to the time index and divided by the corresponding time difference to obtain the temperature rise coefficient. The power change is calculated and divided by the corresponding time difference to obtain the power coefficient. The product of speed and torque is calculated and divided by a set base value to obtain the coupling coefficient. A multi-parameter coefficient set is generated. The temperature rise coefficient, power coefficient, and coupling coefficient are substituted into the initial values ​​of temperature rise, power, speed, and torque, respectively. The predicted values ​​for each future time are calculated by accumulating them at a fixed time step. The predicted values ​​are combined in chronological order to form a temperature rise curve, a power curve, a speed curve, and a torque curve, generating an operating prediction curve.

9. The motor testing method according to claim 1, characterized in that, The specific steps for generating the factory test file are as follows: Based on the operation prediction curve, the peak value of the power curve is extracted and compared with the current limiting setting value. In case of exceeding the limit, the current limiting value is recorded and written into the current limiting register table. The casing to ground current curve is extracted and compared with the leakage threshold. In case of exceeding the limit, the abnormal state is recorded and a power failure protection command is generated, forming an operation protection command set. Based on the aforementioned operation protection instruction set, the control parameters corresponding to the customer client model are loaded into the console parameter register table, and the test records and operating condition judgment tags are organized into archive files according to batch numbers and stored in the file management unit to generate factory test files.

10. A motor testing system, characterized in that, The motor testing method according to any one of claims 1-9 includes the following steps: Sampling and calibration module: Based on the unloaded data from the voltage channel, current channel, and speed detector, calculate the voltage zero-point offset, current range coefficient, and speed difference. Combine the three results with the channel number and write them into the console parameter register to establish a calibration parameter matrix. Anomaly Correction Module: Based on the calibration parameter matrix, extract the differences between temperature rise, rotation speed, vibration and reference values, obtain anomaly points in median rate of change, extract neighboring observations and weight them according to the correlation between historical error and difference, replace anomaly points, replace large differential segments with boundary mean interpolation, verify the anomaly location using the isolated forest algorithm and write it into the module, and construct a stable operating sequence. Driven analysis module: Based on the stable operating sequence, it divides the fixed-cycle voltage segment, extracts the amplitude point and phase shift, calculates the weighted value and forms the concentration vector, extracts the peak value and combines the composite features, uses a convolutional neural network to match the existing vector set, obtains the minimum distance term, and establishes the operating condition mode set; Performance calculation module: Based on the set of operating conditions, calculate the proportional coefficients of temperature rise, power, speed, and torque channels, perform vector outer product to obtain interpolation points, construct temperature rise, power, speed, and torque curves, and establish operation prediction curves; Protection and archiving module: Based on the operation prediction curve, extract the power peak value and compare it with the current limit value, record the over-limit point to the current limit register, compare the over-limit current difference to ground to trigger power-off, write the model parameters, archive the curve and status, and establish the factory test file.

Images