Deep Fusion Method for Time-Frequency Characteristics of Secondary Power Supply in High-Temperature Aging Test Equipment

Abstract

The present invention discloses a method for deeply fusing time-frequency characteristics of a secondary power supply of a high-temperature aging test device. The method includes: collecting a fault data set of components of the high-temperature aging test device; preprocessing the fault data set to obtain a training data set and a test data set; performing short-time Fourier transform on the time-series data of each key parameter in the training data set to obtain a time-frequency diagram; performing random masking on each time-frequency diagram and inputting it into a preset model for training to generate a feature extraction model; testing the feature extraction model based on the test data set; after passing the test, inputting the time-series data of the key parameters of the device to be predicted into the feature extraction model to predict the target time-frequency domain characteristics. The method provided in this application can accurately extract the characteristics of the secondary switch power supply, so as to evaluate the health status of the secondary switch power supply of the high-temperature aging test device.

Description

Technical Field

[0001] The present invention relates to the technical field of integrated circuit non-destructive reliability screening, and particularly relates to a method and device for deep fusion of time-frequency characteristics of a secondary power supply in a high-temperature aging test equipment. Background Art

[0002] Existing high-temperature aging products can achieve long-term monitoring of the test environment by means of increasing in-machine long-term monitoring, overstress protection mechanisms, etc., and can respond in a timely manner when a product fails, realizing after-sales maintenance based on fault data. However, due to the lack of active guarantee technology for high-temperature aging products, it is difficult for related products to achieve the integrity of the test process and the consistency of the test environment stress. It is extremely easy to cause major property losses such as the destruction of millions of test devices (also known as test pieces to be detected) due to the forced interruption of the test process caused by product failures, or the aging test is recognized as a failed test due to adverse effects such as additional stress introduced during the test period caused by product performance degradation, resulting in ineffective waste of resources. At the same time, ensuring the quality of high-temperature aging tests can avoid over-aging and under-aging failures of integrated circuits, greatly reduce the failure rate of integrated circuits, and avoid large-scale electronic system failures and shutdowns such as new energy vehicles, civil airliners, and energy storage substations caused by integrated circuit failures.

[0003] High-temperature aging products such as integrated circuit high-temperature aging test equipment accelerate various physical and chemical reaction processes inside components by continuously applying a certain electrical stress to the components for a long time, prompting various potential failures inside the components to be exposed early, so as to eliminate early failure products and enable electronic components to enter a period with a low and relatively stable failure rate from the beginning of use. The secondary switching power supply provides stable voltage and current for the integrated circuit during the high-temperature dynamic aging test, ensuring the consistency of test conditions; it can flexibly adjust the output voltage and current according to different test requirements to meet the test requirements of different types of integrated circuits.

[0004] The secondary switching power supply plays a crucial role in the high-temperature aging test equipment. A failure of the secondary switching power supply may trigger failures in other parts of the system, such as the drive control detection board, further affecting the normal operation of the system. A failure of the secondary switching power supply may cause safety hazards such as overheating and short circuits, endangering the safety of operators and the integrity of the equipment. Therefore, it is very necessary to conduct fault prediction and health assessment on the secondary switching power supply of the aging test bench.

[0005] In the prior art, a time series extrapolation prediction method is used to predict the faults of a secondary switching power supply. This method usually adopts a time series decomposition strategy, where the time series is decomposed into a trend term, a seasonal term, a residual term, etc. for separate prediction, and finally the prediction results of each term are fused to obtain a time series extrapolation prediction sequence of the parameters. Although the output signal can reflect the degradation process of the device and even the module, this change is relatively weak. Considering the training and response time of the later model, the original signal is obviously not very suitable for the prediction model. Therefore, it is necessary to extract data that can represent the fault characteristics from a large number of original signals, that is, to perform dimensionality reduction, noise removal, etc. on the original data. For a secondary switching power supply circuit, due to the existence of tolerances, nonlinearities, etc., and a high operating frequency, there are rich high-frequency and low-frequency signals, and high requirements for feature extraction are imposed.

[0006] It can be seen that there is an urgent need for those skilled in the art to provide a solution for feature extraction of the secondary switching power supply of a high-temperature aging test equipment. Summary of the Invention

[0007] The purpose of the embodiments of the present invention is to provide a method and device for deep fusion of time-frequency features of a secondary power supply of a high-temperature aging test equipment, which can solve the above problems existing in the prior art.

[0008] To solve the above technical problems, the present invention provides the following technical solutions:

[0009] The embodiments of the present invention provide a method for deep fusion of time-frequency features of a secondary power supply of a high-temperature aging test equipment, wherein the method includes:

[0010] Collect a fault data set of the devices of the high-temperature aging test equipment; wherein, the fault data includes multiple key parameter time series data;

[0011] Preprocess the fault data set to obtain a training data set and a test data set;

[0012] Perform short-time Fourier transform on each key parameter time series data in the training data set to obtain a time-frequency diagram;

[0013] Perform random masking on each time-frequency diagram and input it into a preset model for training to generate a feature extraction model;

[0014] Test the feature extraction model according to the test data set;

[0015] After passing the test, input the key parameter time series data of the device to be predicted into the feature extraction model to predict the target time-frequency domain features.

[0016] Optionally, the step of preprocessing the fault data set to obtain a training data set and a test data set includes:

[0017] For each key parameter in the fault dataset, perform sliding window cutting on the time series data of the key parameter to construct a sample dataset;

[0018] Normalize each data sample in the sample dataset to obtain a normalized sample dataset;

[0019] Select the first preset percentage of sample data from the normalized sample dataset as the training dataset, and use the remaining sample data as the test dataset; wherein, the training dataset is used to train the preset model, and the test dataset is used to verify the prediction accuracy of the trained feature extraction model.

[0020] Optionally, the step of performing short-time Fourier transform on each time series data in the training dataset to obtain a time-frequency diagram includes:

[0021] For each time series data in the training dataset, use a window function with a finite duration to intercept the vibration signal in the time series data;

[0022] Perform Fourier transform on the intercepted vibration signal to obtain a local spectrum in a small range of the corresponding time period;

[0023] By moving the window function on the time axis, gradually analyze each vibration signal band to obtain a set of local spectra of the vibration signal, and generate a time-frequency diagram.

[0024] Optionally, the window function is a Hanning window function.

[0025] Optionally, the step of performing random masking on each time-frequency diagram and inputting it into a preset model for training to generate a feature extraction model includes:

[0026] Input the randomly masked time-frequency diagram into the preset model, send it to the encoding module after position encoding, and send the output of the encoding module into the decoder. Through the decoder, restore the feature parameters in the high-dimensional hidden layer into the time-frequency diagram before masking;

[0027] Use the restored time-frequency diagram to iteratively train the preset model multiple times to adjust the model parameters, where the model parameters include: embedding dimension, number of attention heads, number of encoder layers, and number of decoder layers;

[0028] Take out the encoder layer and decoder layer of the model after multiple iterative adjustments, and retain the weight parameters of each layer;

[0029] Generate a feature extraction model based on each encoder layer, each decoder layer, and the weight parameters of each layer.

[0030] Optionally, the device includes: a switching power supply for high-temperature aging test equipment, a high-temperature test chamber for high-temperature aging test equipment, and a high-temperature and high-humidity test chamber for high-temperature aging test equipment.

[0031] An embodiment of the present invention further provides a device for deep fusion of time-frequency characteristics of a secondary power supply of high-temperature aging test equipment. Among them, the device includes:

[0032] An acquisition module, configured to acquire a fault data set of devices of high-temperature aging test equipment; wherein, the fault data includes multiple key parameter time-series data;

[0033] A preprocessing module, configured to preprocess the fault data set to obtain a training data set and a test data set;

[0034] A transformation module, configured to perform short-time Fourier transform on each key parameter time-series data in the training data set to obtain a time-frequency diagram;

[0035] A mask training module, configured to randomly mask each of the time-frequency diagrams and input them into a preset model for training to generate a feature extraction model;

[0036] A test module, configured to test the feature extraction model based on the test data set;

[0037] A prediction module, configured to, after passing the test, input the key parameter time-series data of the device to be predicted into the feature extraction model to predict and obtain the target time-frequency domain feature.

[0038] Optionally, the preprocessing module includes:

[0039] A first sub-module, configured to perform sliding window cutting on the key parameter time-series data for each key parameter in the fault data set to construct a sample data set;

[0040] A second sub-module, configured to perform normalization processing on each data sample in the sample data set to obtain a normalized sample data set;

[0041] A third sub-module, configured to select the first preset percentage of sample data from the normalized sample data set as the training data set, and use the remaining sample data as the test data set; wherein, the training data set is used to train the preset model, and the test data set is used to verify the prediction accuracy of the trained feature extraction model.

[0042] Optionally, the transformation module includes:

[0043] A fourth sub-module, configured to, for each time-series data in the training data set, intercept the vibration signal in the time-series data by using a window function with a limited duration;

[0044] The fifth sub-module is used to perform Fourier transform on the intercepted vibration signal to obtain a local spectrum in a small range corresponding to a time period;

[0045] The sixth sub-module is used to gradually analyze each vibration signal band by moving the window function on the time axis to obtain a set of local spectra of the vibration signal and generate a time-frequency diagram.

[0046] Optionally, the window function is a Hanning window function.

[0047] Optionally, the mask training module includes:

[0048] The seventh sub-module is used to input the time-frequency diagram after random masking into the preset model, send it into the encoding module after position encoding, and send the output of the encoding module into the decoder, and restore the feature parameters in the high-dimensional hidden layer into the time-frequency diagram before masking through the decoder;

[0049] The eighth sub-module is used to iteratively train the preset model multiple times with the restored time-frequency diagram to adjust the model parameters, where the model parameters include: embedding dimension, number of attention heads, number of encoder layers, and number of decoder layers;

[0050] The ninth sub-module is used to take out the encoder layer and decoder layer of the model after multiple iterative adjustments and retain the weight parameters of each layer;

[0051] The tenth sub-module is used to generate a feature extraction model based on each encoder layer, each decoder layer, and the weight parameters of each layer.

[0052] Optionally, the device includes: a switching power supply for a high-temperature aging test equipment, a high-temperature test chamber for a high-temperature aging test equipment, and a high-temperature and high-humidity test chamber for a high-temperature aging test equipment.

[0053] An embodiment of the present invention further provides an electronic device, which is characterized in that it includes a processor, a communication interface, a memory, and a communication bus. Among them, the processor, the communication interface, and the memory complete mutual communication through the communication bus; the memory is used to store a computer program; when the processor executes the program stored on the memory, it implements the time-frequency feature depth fusion method flow of any one of the above high-temperature aging test equipment secondary power supplies.

[0054] The deep fusion scheme of time-frequency characteristics of the secondary power supply of the high-temperature aging test equipment disclosed in the embodiments of the present invention collects the fault data sets of the devices of the high-temperature aging test equipment; preprocesses the fault data sets to obtain training data sets and test data sets; performs short-time Fourier transform on the time-series data of each key parameter in the training data sets to obtain time-frequency diagrams; performs random masking on each time-frequency diagram and inputs it into a preset model for training to generate a feature extraction model; tests the feature extraction model based on the test data sets; after passing the test, inputs the time-series data of the key parameters of the device to be predicted into the feature extraction model to predict the target time-frequency domain features. The scheme provided by the embodiments of the present invention fully coordinates and refines a large amount of fault data of the devices of the high-temperature aging test equipment, performs random masking on the time-frequency diagrams obtained by performing short-time Fourier transform on the segmented time-series data, and uses the time-frequency diagrams after random masking to train the preset model, which can fully refine the local semantic information of each part of the devices of the high-temperature aging test equipment, overcome the constraints on the training model such as data scarcity and rough annotation, and the time-frequency features predicted by the trained feature extraction model are more accurate and reliable. Further, the device health state of the high-temperature aging test equipment analyzed based on the predicted time-frequency features is more reliable. Since the device health state of the high-temperature aging test equipment can be accurately analyzed, the device can be repaired in time when a device fault is found, thereby avoiding damage to the device under test caused by the device fault during the aging test process. Description of the Drawings

[0055] Figure 1 is a flowchart showing the steps of a method for deep fusion of time-frequency characteristics of the secondary power supply of a high-temperature aging test equipment according to an embodiment of the present application;

[0056] Figure 2 is a schematic diagram of the time-frequency diagram after random masking according to an embodiment of the present application;

[0057] Figure 3 is a schematic diagram of an autoencoder model based on Transformer according to an embodiment of the present application;

[0058] Figure 4 is a flowchart showing the steps of a method for deep fusion of time-frequency characteristics of the secondary power supply of a high-temperature aging test equipment according to an embodiment of the present application;

[0059] Figure 5 is a flowchart showing the steps of a method for evaluating the health state of a high-temperature aging test chamber according to an embodiment of the present application;

[0060] Figure 6 is a block diagram showing the structure of a device for deep fusion of time-frequency characteristics of the secondary power supply of a high-temperature aging test equipment according to an embodiment of the present application. Detailed Embodiments

[0061] To make the technical problems, technical solutions, and advantages to be solved by the present invention clearer, the following will be described in detail with reference to the accompanying drawings and specific embodiments.

[0062] The following will, with reference to the accompanying drawings, through specific embodiments and their application scenarios, elaborate in detail on the method for deeply fusing time-frequency characteristics of the secondary power supply of the high-temperature aging test equipment provided in the embodiments of the present application.

[0063] As shown in the attached Figure 1 figures, the method for deeply fusing time-frequency characteristics of the secondary power supply of the high-temperature aging test equipment in the embodiments of the present application includes the following steps:

[0064] Step 101: Collect the fault data set of the devices of the high-temperature aging test equipment.

[0065] The high-temperature aging test equipment provided in the present application can perform non-destructive aging detection on the test pieces. The time-frequency domain feature processing method in the embodiments of the present application can extract features from the state data, health data, etc. of the devices of the high-temperature aging test equipment. The extracted features can be used for the health assessment, fault prediction, etc. of the devices of the high-temperature aging test equipment, thereby ensuring the safety of the test pieces.

[0066] Among them, the fault data includes the time-series data of multiple key parameters, and the key parameters are different for different devices. For example: in the case of a secondary switching power supply device, the key parameters can be the power supply voltage, current signal, and the thermal resistance characteristics from the MOSFET case temperature to the junction temperature synchronously monitored by an infrared thermal imager and an embedded temperature sensor. In the case of a test chamber device, the key parameter can be the temperature monitoring signal, etc. These monitoring signals can be used to evaluate the health status of the devices of the high-temperature aging test equipment. In the actual implementation process, the directly measured data of these key parameters is large in quantity and contains a large amount of redundant information. Therefore, it is necessary to use the feature extraction model trained by the method for deeply fusing time-frequency characteristics of the secondary power supply of the high-temperature aging test equipment in the embodiments of the present application to perform preliminary feature extraction on the directly measured data to obtain a feature sequence with high information density.

[0067] The fault data set of the devices of the high-temperature aging test equipment collected in the embodiments of the present application can be the fault data set during the entire process from the start of degradation to the stop of operation (i.e., the end of the remaining life cycle) in the full life cycle of the devices.

[0068] When collecting the fault data set of the devices of the high-temperature aging test equipment, it is necessary to synchronously monitor the load change characteristics of the secondary switching power supply, including the output voltage and current transient response under dynamic load switching. For the load fluctuation when the device under test is connected, record the power supply ripple volatility caused by it. The calculation formula is:

[0069] Ripple volatility = ( / ) × 100%

[0070] Among them, is the peak-to-peak value of the output voltage, is the average value of the output voltage. The ripple waveform is captured by a high-precision oscilloscope, and its spectral characteristics are analyzed to identify abnormal harmonic components. The devices may include but are not limited to: the switching power supply of the high-temperature aging test equipment, the high-temperature test chamber of the high-temperature aging test equipment, the high-temperature and high-humidity test chamber of the high-temperature aging test equipment, etc. The switching power supply of the high-temperature aging test equipment can be further divided into: the primary switching power supply, the secondary switching power supply, and the MOSFET switching tube of the secondary switching power supply. Moreover, the devices of the high-temperature aging test equipment may also include: the drive control detection board.

[0071] The specific devices included in the high-temperature aging test equipment can be flexibly adjusted according to the type of the high-temperature aging test equipment. In the embodiments of the present application, the specific type of the high-temperature aging test equipment is not limited. For example: the high-temperature aging test equipment can be a high-temperature and high-humidity high-temperature aging test equipment, can be a constant-temperature and constant-pressure high-temperature aging test equipment, and can also be a high-temperature high-temperature aging test equipment, etc.

[0072] Step 102: Preprocess the fault data set to obtain a training data set and a test data set.

[0073] An optionally way to preprocess the fault data set to obtain a training data set and a test data set may include the following sub-steps:

[0074] Sub-step 1: For each key parameter in the fault data set, perform sliding window cutting on the key parameter time series data to construct a sample data set;

[0075] When constructing the sample data set by sliding window cutting, it is necessary to adaptively segment the data during the load mutation period. For the interval where the load change rate exceeds the threshold (such as ±10% / ms), a smaller window width and step size are adopted to capture the detailed features of the transient response. At the same time, dynamic baseline calibration is performed on the current and voltage output data to eliminate the baseline drift caused by the access of the device under test.

[0076] Sub-step 2: Normalize each data sample in the sample data set to obtain a normalized sample data set;

[0077] Sub-step 3: Select the sample data of the pre-set percentage from the normalized sample data set as the training data set, and use the remaining sample data as the test data set.

[0078] Among them, the training data set is used to train the pre-set model, and the test data set is used to verify the prediction accuracy of the trained feature extraction model. The pre-set percentage can be flexibly set by those skilled in the art, and no specific limitation is made in the embodiments of the present application. For example: set to 60%-80%, preferably 70%.

[0079] Step 103: Perform short-time Fourier transform on the time-series data of each key parameter in the training dataset to obtain a time-frequency diagram.

[0080] Most traditional time-domain and frequency-domain analysis methods are for stationary signals and can only obtain information in one aspect of the time domain or frequency domain. In the application of actual devices such as secondary switching power supplies, the collected signals are mostly non-stationary, and the analysis in one aspect can no longer meet the needs. Therefore, it is necessary to determine the relationship between the signal frequency and time. The joint time-frequency analysis method is a very effective tool in the current process of processing non-stationary signals. Therefore, short-time Fourier transform is used in this application to process variable time-series and non-stationary signals. The basic principle of STFT (i.e., short-time Fourier transform) is as follows: Use a window function h(t) with a finite duration to intercept the vibration signal, perform Fourier transform on the obtained signal to obtain the local spectrum in a small range of this time period, and gradually analyze the signal band by moving the window function h(t) on the time axis to obtain a set of local "spectrums" of the signal. The essence of STFT is the transformation of the basis function.

[0081] An optional way to perform short-time Fourier transform on the time-series data in the training dataset to obtain a time-frequency diagram may include the following sub-steps:

[0082] Sub-step 1: For each time-series data in the training dataset, use a window function with a finite duration to intercept the vibration signal in the time-series data;

[0083] Sub-step 2: Perform Fourier transform on the intercepted vibration signal to obtain the local spectrum in a small range of the corresponding time period;

[0084] The transformation formula for the signal collected by high-temperature aging test equipment devices such as secondary power supplies can be:

[0085]

[0086] In the formula: is the source signal, is the analysis window function; is the time-frequency spectrum at time t.

[0087] Sub-step 3: By moving the window function on the time axis, gradually analyze each vibration signal band to obtain a set of local spectrums of the vibration signal and generate a time-frequency diagram.

[0088] Preferably, the window function is selected as the Hanning window function. The Hanning window function has good smoothness and is suitable for reducing spectral leakage. The expression of the Hanning window function is as follows:

[0089]

[0090] Step 104: Randomly mask each time-frequency diagram and input it into a preset model for training to generate a feature extraction model.

[0091] By steps 101 to 104, time-frequency diagrams corresponding to different time segments of multiple key parameters can be obtained. For each time-frequency diagram, the time-frequency diagram can be masked in a random masking manner. The schematic diagram of the time-frequency diagram after random masking is as Figure 2 shown. Different from time series features and frequency domain features, time series features and frequency domain features are one-dimensional data, while the time-frequency diagram is a two-dimensional picture. The abscissa represents time, the ordinate represents frequency, and the different shades of color represent different frequency amplitudes.

[0092] Send the time-frequency diagrams after random masking into a preset model such as the Transformer autoencoder for training respectively. After the training is completed, a feature extraction model is generated. Figure 3 It is a schematic diagram of the autoencoding model based on Transformer according to the embodiments of the present application.

[0093] Transformer is a model established based on the Seq-to-Seq framework. Compared with classical deep learning models, the most prominent advantage of Transformer is the use of the multi-head attention mechanism. The purpose of the multi-head attention layer is to assign different importance to words / tokens in the sequence from multiple aspects. The Transformer model mainly consists of parts such as input, encoder, decoder, and output.

[0094] Position encoding: The position encoding layer is to determine the position information of the sequence. Since there are no recurrent layers and convolutional layers in RNN and CNN, and only relying on the self-attention mechanism cannot obtain the order information of the input, it is necessary to actively transmit the order information of the sequence to the model. The Transformer model uses a combination of sine and cosine functions to perform position encoding on the sequence, and the calculation method is as follows:

[0095]

[0096]

[0097] In the formula: is the position of the current sequence; is the dimension; is the dimension of the input feature.

[0098] Multi-Head Attention Mechanism: The multi-head attention mechanism performs operations in parallel using multiple attention mechanisms and then concatenates the operation results through a linear transformation. The core technology in Transformer is the multi-head attention mechanism, which is used to extract the dependency relationship features between data, capture the correlation between data, and establish a context prediction model. The calculation method is as follows:

[0099]

[0100]

[0101]

[0102]

[0103]

[0104]

[0105] In the formula: is the query matrix; is the key matrix; is the value matrix; , , is the trainable parameter matrix; is the processed input; is the dimension of the key matrix; , , , is the learnable parameter matrix.

[0106] Feed-Forward Network and Summation and Normalization: In the Transformer model, the encoding part and the decoding part also include a feed-forward network and summation and normalization. The calculation formula of the feed-forward neural network is as follows:

[0107]

[0108] In the formula: is the input; , , , are the parameters that can be obtained through training.

[0109] The calculation formula of summation and normalization is as follows:

[0110]

[0111] In the formula: is the input; is the result after being processed by the module.

[0112] In an alternative embodiment, the method for randomly masking each time-frequency diagram and inputting it into a preset model for training to generate a feature extraction model may include the following sub-steps:

[0113] Sub-step 1: Input the randomly masked time-frequency diagram into the preset model, send it to the encoding module after position encoding, and send the output of the encoding module into the decoder. The decoder restores the feature parameters in the high-dimensional hidden layer into the time-frequency diagram before masking;

[0114] Sub-step 2: Use the restored time-frequency diagram to iteratively train the preset model multiple times to adjust the model parameters;

[0115] Among them, the model parameters include: embedding dimension, number of attention heads, number of encoder layers, and number of decoder layers.

[0116] Sub-step 3: Take out the encoder layer and decoder layer of the model after multiple iterative adjustments, and retain the weight parameters of each layer.

[0117] Sub-step 4: Generate a feature extraction model based on each encoder layer, each decoder layer, and the weight parameters of each layer.

[0118] Step 105: Test the feature extraction model according to the test data set.

[0119] After training (also known as pre-training) the preset model based on the training data set to generate a feature extraction model, the feature extraction model can be tested based on the test data in the test data set to determine the accuracy of the prediction result of the feature extraction model.

[0120] Based on the pre-trained Transformer auto-encoding model for the training data set perform self-attention feature extraction to obtain a self-attention feature set .

[0121] Step 106: After passing the test, input the key parameter time series data of the device to be predicted into the feature extraction model to predict the target time-frequency domain features.

[0122] The method for deep fusion of time-frequency characteristics of the secondary power supply of the high-temperature aging test equipment provided by the embodiment of the present application collects the fault data set of the components of the high-temperature aging test equipment; preprocesses the fault data set to obtain a training data set and a test data set; performs short-time Fourier transform on the time-series data of each key parameter in the training data set to obtain a time-frequency diagram; performs random masking on each time-frequency diagram and inputs it into a preset model for training to generate a feature extraction model; tests the feature extraction model based on the test data set; after passing the test, inputs the time-series data of the key parameters of the device to be predicted into the feature extraction model to predict the target time-frequency domain feature. The method provided by the embodiment of the present invention fully coordinates and refines a large amount of fault data of the components of the high-temperature aging test equipment, performs random masking on the time-frequency diagrams obtained by performing short-time Fourier transform on the segmented time-series data, and uses the randomly masked time-frequency diagrams to train the preset model, which can fully refine the local semantic information of each part of the components of the high-temperature aging test equipment, overcome the constraints of data scarcity, rough annotation, etc. on the training model, and the time-frequency features predicted by the trained feature extraction model are more accurate and reliable. Further, the health status of the components of the high-temperature aging test equipment analyzed based on the predicted time-frequency features is more reliable.

[0123] The following combines Figure 4 a specific example to illustrate the deep fusion scheme of the time-frequency characteristics of the secondary power supply of the high-temperature aging test equipment provided by the present application.

[0124] In this specific example, the processing of the time-frequency characteristics of the secondary switching power supply of the high-temperature aging test equipment is taken as an example for illustration. During the operation of the secondary switching power supply of the high-temperature aging test equipment, monitoring signals such as current and voltage will be generated. In order to accurately extract the time-frequency domain characteristics of the secondary switching power supply, a method for deep fusion of the time-frequency characteristics of the secondary power supply of the high-temperature aging test equipment is proposed in this specific example. This method fully coordinates and refines the local semantic information of each part of the fault data of the high-temperature aging test equipment, overcomes the constraints of data scarcity, rough annotation, etc. on the training network, and further improves the performance required for the task. It provides a more practical method for the extrapolation prediction problem of the key parameter degradation time series of the secondary switching power supply of the high-temperature aging test equipment (also known as the aging test bench).

[0125] Such as Figure 4 the shown feature extraction method, which includes multiple important links: Link 1: Obtain the fault data of the secondary switching power supply of the aging test bench; Link 2: Perform comprehensive preprocessing on the fault data; Link 3: Perform short-time Fourier transform on the input time-series data to obtain a time-frequency diagram; Link 4: Perform random masking on the time-frequency diagram; Link 5: Put the randomly masked time-frequency diagram into a Transformer autoencoder for model training; Link 6: Input the monitored signals of the key parameters of the secondary switching power supply to be predicted collected into the Transformer autoencoding model for feature extraction and output the extracted features.

[0126] Step 1: Obtain the fault data of the secondary switching power supply of the aging test bench

[0127] As Figure 4 shown, this step includes two steps: obtaining the key parameters of the secondary switching power supply of the aging test bench and obtaining the historical time series data of the parameters to be predicted (i.e., the key parameters to be predicted) from them. In the actual time series process, with a complete set of integrated circuit high-temperature dynamic aging detection systems, the voltage and current signals of the secondary switching power supply are monitored through sensors to obtain the original fault data of the secondary switching power supply of the aging test bench.

[0128] Step 2: Conduct comprehensive preprocessing on the fault data

[0129] Send the collected original fault data into the fault data preprocessing unit for comprehensive processing to obtain the training data set and the test data set. As Figure 4 shown, this step specifically includes the following steps:

[0130] Step 1: Generate the corresponding sample data set by sliding window cutting of the data.

[0131] In this step, the time series data of the key parameters are cut by sliding window to construct the sample data set.

[0132] For example, any time series data of the key parameters of the secondary switching power supply collected by the sensor is X, , and X is cut by sliding window to generate the corresponding sample data set. When the window width is and the step size is S, the number of samples generated by cutting is:

[0133]

[0134] Then the corresponding data set generated is .

[0135] Step 2: Conduct min-max normalization processing on the training data set;

[0136] To improve the data expression ability and accelerate the convergence speed of the subsequent model training, it is necessary to conduct normalization processing on the training data set, mainly by scaling the amplitude of the original parameters through the min-max normalization method to complete the linear transformation of the data. For a single sample data , through the formula:

[0137]

[0138] The normalization processing is realized to obtain the normalized sample data set .

[0139] Step 3: Construct the training data set and the test data set;

[0140] Select the top of all the data as the training dataset, and the remaining data as the test dataset for verifying the model prediction performance. Generally, r is usually taken as 60 - 80, and here it is taken as 70.

[0141] Step 3: Perform short-time Fourier transform on the input time series data (i.e., the training dataset) to obtain a time-frequency diagram.

[0142] Most traditional time-domain and frequency-domain analysis methods are for stationary signals and can only obtain information in one aspect of the time domain or frequency domain. In the actual application of secondary power supplies, the collected signals are mostly non-stationary, and the analysis in one aspect can no longer meet the needs. We need to know the relationship between the signal frequency and time. The joint time-frequency analysis method is a very effective tool in the current process of dealing with non-stationary signals. Therefore, this paper selects the short-time Fourier transform to process variable time series and non-stationary signals. The basic principle of STFT, that is, the short-time Fourier transform, is as follows: Use a window function h(t) with a finite time duration to intercept the vibration signal, perform Fourier transform on the obtained signal to obtain the local spectrum in a small range of this time period, and through the movement of the window function h(t) on the time axis, gradually analyze the signal band to obtain a set of local "spectrums" of the signal. The essence of STFT is the transformation of the basis function.

[0143] The transformation formula for the signal collected by the secondary switching power supply can be:

[0144]

[0145] In the formula: is the source signal; is the analysis window function; is the time-frequency spectrum at time t

[0146] Preferably, the window function is selected as the Hanning window function. The Hanning window function has good smoothness and is suitable for reducing spectral leakage. The expression of the Hanning window function is as follows:

[0147]

[0148] Step 4: Perform random masking on the time-frequency diagram.

[0149] The schematic diagram of the masked time-frequency diagram generated after the random masking operation on the time-frequency diagram is as Figure 2 shown.

[0150] Step 5: Put the randomly masked time-frequency diagram into the Transformer autoencoder for model training (i.e., train the Transformer autoencoder).

[0151] The time-frequency diagrams after random masking are respectively fed into the Trransformer autoencoder (i.e., the preset model described above) for training. The autoencoding model based on Transformer is as shown in Figure 3 shown.

[0152] Transformer is a model built based on the Seq-to-Seq framework. Compared with classical deep learning models, the most prominent advantage of Transformer is the use of the multi-head attention mechanism. The purpose of the multi-head attention layer is to assign different importance to words / tokens in the sequence from multiple aspects. The Transformer model mainly consists of parts such as input, encoder, decoder, and output.

[0153] Position encoding: The position encoding layer is to determine the position information of the sequence. Since there are no recurrent layers and convolutional layers in RNN and CNN, and only relying on the self-attention mechanism cannot obtain the order information of the input, it is necessary to actively transmit the order information of the sequence to the model. The Transformer model uses a combination of sine and cosine functions to perform position encoding on the sequence, and the calculation method is as follows:

[0154]

[0155]

[0156] In the formula: is the position of the current sequence; is the dimension; is the dimension of the input feature.

[0157] Multi-head attention mechanism: The multi-head attention mechanism uses multiple attention mechanisms to perform operations in parallel, and then splices the operation results through a linear transformation. The core technology in Transformer is the multi-head attention mechanism. The multi-head attention mechanism is used to extract the dependency relationship features between data, capture the correlation relationship between data, and establish a context prediction model. The calculation method is as follows:

[0158]

[0159]

[0160]

[0161]

[0162]

[0163]

[0164] In the formula: is a query matrix; is a key matrix; is a value matrix; , , is a trainable parameter matrix; is the processed input; is the dimension of the key matrix; , , , is a learnable parameter matrix.

[0165] Feed - forward network and summation and normalization: In the Transformer model, the encoding part and the decoding part also contain a feed - forward network and summation and normalization. The calculation formula of the feed - forward neural network is as follows:

[0166]

[0167] In the formula: is the input; , , , are parameters that can be obtained through training.

[0168] The calculation formula of summation and normalization is as follows:

[0169]

[0170] In the formula: is the input; is the result after being processed by the module.

[0171] The model processing flow is: Send the masked data through positional encoding into the encoding part, and send the output of the encoding part into the decoder to restore the feature parameters in the high - dimensional hidden layer to the original un - masked data, so as to train the feature extraction ability of the model.

[0172] Secondly, select appropriate number of iterations and loss function, and input the constructed dataset into the feature extraction model to repeatedly execute the forward propagation and backward propagation iterative calculation process; during this process, continuously adjust the model parameters of the embedding dimension, number of attention heads, number of encoder and decoder layers to complete the pre - training of the model.

[0173] Thirdly, take out the encoding layer and decoding layer of the pre - trained model, and retain their weight parameters, and construct them into the trained Transformer auto - encoding model.

[0174] Finally, based on the pre - trained Transformer auto - encoding model for the training dataset Perform self-attention feature extraction to obtain a self-attention feature set .

[0175] Step 6: Input the collected key parameter monitoring signals of the secondary switching power supply to be predicted into the Transformer autoencoder model for feature extraction, and output the extracted features.

[0176] After the prediction performance test of the trained Transformer autoencoder model (i.e., the feature extraction model described above) based on the training data set, if the test passes, it can be used to extract the health state features of the secondary switching power supply of any aging test bench at any time, obtaining a multi-dimensional feature sequence for subsequent evaluation of the health state of the secondary switching power supply.

[0177] The time-frequency feature deep fusion method for the secondary power supply of the high-temperature aging test equipment provided by this application can be used to extract features from the key parameter monitoring signals of the high-temperature aging test equipment devices, obtaining a multi-dimensional feature sequence. The obtained multi-dimensional feature sequence can be used to train a preset network model, and the trained network model can be used to evaluate the health state of the high-temperature aging test equipment devices. Of course, when evaluating the health state of a certain high-temperature aging test equipment device at any time, the multi-dimensional characteristic sequence input into the trained network model can also be extracted.

[0178] Taking the high-temperature aging test equipment device as the high-temperature aging test chamber as an example, the multi-dimensional feature sequence extraction of the high-temperature aging test chamber is carried out by using the time-frequency feature deep fusion method for the secondary power supply of the high-temperature aging test equipment. The extracted multi-dimensional feature sequence is input into the pre-trained time convolutional network model to obtain the degradation feature state of the high-temperature aging test chamber, and the health state of the high-temperature aging test chamber is evaluated by the obtained degradation feature state as an example for illustration.

[0179] As shown in the appendix Figure 5 The health state evaluation method of the high-temperature aging test chamber in the embodiment of this application includes the following steps:

[0180] Step 501: Collect the monitoring signals of the high-temperature aging test chamber to be evaluated.

[0181] Among them, the monitoring signals include: nine-point temperature measurement signals of the high-temperature aging test chamber, compressor vibration signals, and fan speed signals.

[0182] The health state evaluation method of the high-temperature aging test chamber provided by this application can be applied to an electronic device. The electronic device is provided with a health state evaluation computer program for the high-temperature aging test chamber, and when the computer program is executed by a processor, the health state evaluation method of the high-temperature aging test chamber in the embodiment of this application is realized.

[0183] The high-temperature aging test equipment provided by this application can perform non-destructive aging detection on the test piece. The health status evaluation method of the high-temperature aging test chamber provided in the embodiments of this application can detect the health evaluation status of the high-temperature aging test chamber, thereby ensuring the safety of the test piece when the high-temperature aging test equipment detects the test piece.

[0184] In the actual implementation process, the directly measured temperature monitoring signal data of the high-temperature aging test chamber is large in quantity and contains a large amount of redundant information. Therefore, preliminary feature extraction is performed on the directly measured data to obtain a feature sequence with high information density.

[0185] Step 502: Extract features from the monitoring signal to obtain a multi-dimensional feature sequence.

[0186] The extracted features include at least two of the following: temperature fluctuation degree, temperature control accuracy, temperature uniformity, temperature overshoot, heating and cooling rate, time-frequency domain features of vibration signals, and time-domain features of fan speed, etc. The specific features to be extracted can be flexibly set by those skilled in the art according to actual needs, as long as the extracted features can characterize the state of the test chamber.

[0187] Step 503: Input the multi-dimensional feature sequence into a pre-trained time convolutional network model to obtain the degradation feature state of the high-temperature aging test chamber.

[0188] Among them, the degradation feature state includes: healthy state and degradation state. If the degradation feature state of the high-temperature aging test chamber is the healthy state, then step 504 and subsequent steps do not need to be executed. If the degradation feature state of the high-temperature aging test chamber is the degradation state, it means that the high-temperature aging test chamber has degraded. By performing adaptive analysis on the multi-dimensional feature sequence data through the time convolutional network model, the degradation progress of the high-temperature aging test chamber can be automatically identified.

[0189] In an optional embodiment, the training process of the time convolutional network model may include the following sub-steps:

[0190] Sub-step 1: Obtain the training time-series degradation feature samples of the high-temperature aging test chamber;

[0191] Among them, each training time-series degradation feature sample is labeled with a degradation feature state. The training time-series degradation feature samples of the high-temperature aging test chamber can be obtained by monitoring the temperature signal and the degradation feature state during the entire life cycle of a high-temperature aging test chamber.

[0192] In the actual implementation process, the training time-series degradation feature samples of the high-temperature aging test chamber can be expressed as , the maximum number of iterations max_iter. Mark the corresponding degradation feature state of the trained high-temperature aging test chamber , including the healthy state and the degradation state.

[0193] Sub-step 2: Input the training time-series degradation feature samples into the pre-established temporal convolutional network model to obtain the first probability space points after mapping;

[0194] In the actual implementation process, taking as the input, as the output to train the TCN model (i.e., the temporal convolutional network model); obtain the TCN model parameters obtained by iterative learning. When the number of iterations reaches the maximum number of iterations max_iter, stop the iterative learning.

[0195] In the actual implementation process, the maximum number of iterations max_iter can be flexibly set by those skilled in the art, and no specific limitation is made in the embodiments of the present application.

[0196] Sub-step 3: Discriminate the degradation starting point of the high-temperature aging test chamber according to the two-dimensional first probability space points, and generate a health baseline for the high-temperature aging test chamber.

[0197] In an optional embodiment, the training process of the temporal convolutional network model may further include the following process:

[0198] Obtain the test time-series degradation feature samples of the high-temperature aging test chamber; input the test time-series degradation feature samples into the temporal convolutional network model under training to obtain the second probability space points after mapping; calculate the distance between the second probability space and the health baseline based on a preset multi-distance metric algorithm.

[0199] The test time-series degradation feature samples of the high-temperature aging test chamber can be expressed as: . During the training and testing of the temporal convolutional network model, taking and as the input of the TCN algorithm model to obtain the probability space points and after mapping.

[0200] Automatically discriminate the degradation starting point according to the two-dimensional probability space results, and calculate the distances between the health space and the degradation space of the training and test high-temperature aging test chambers using a weighted fusion multi-distance metric algorithm . Traverse all failure modes to obtain the health assessment results corresponding to each failure mode. The finally trained temporal convolutional network model outputs and the health assessment curves corresponding to each high-temperature aging test chamber.

[0201] More specifically, based on the preset multi-distance metric algorithm, the way to calculate the distance between the second probability space and the health baseline can be as follows: Calculate the distances between the second probability space and the health baseline respectively based on the Mahalanobis distance algorithm, the cosine similarity algorithm, and the Manhattan distance algorithm to obtain the first distance, the second distance, and the third distance; perform a weighted sum of the first distance, the second distance, and the third distance to obtain the corresponding health degree.

[0202] In the actual implementation process, it is not limited to using the above three distance algorithms to calculate the distance. Any one of the above three distance algorithms can also be used, or any two of the above three distance algorithms can be used to obtain two distances and then perform a weighted sum of the two distances.

[0203] During the training process of the temporal convolutional network model, introduce the high-temperature aging test chamber to test the temporal degradation feature samples, and the accuracy of the prediction of the temporal convolutional network model can be tested through the prediction results. In the actual implementation process, when the prediction accuracy of the temporal convolutional network model reaches the accuracy threshold, it can be determined that the training of this temporal convolutional network model is completed. Subsequently, the trained temporal convolutional network model can be used to predict the state degradation curve of the high-temperature aging test chamber to be evaluated.

[0204] Step 504: When the high-temperature aging test chamber is in a degraded state, based on the probability space points after mapping of the multi-dimensional feature sequence and the health baseline in the temporal convolutional network model, determine the state degradation curve of the high-temperature aging test chamber to be evaluated.

[0205] Among them, the state degradation curve can characterize the health state of the high-temperature aging test chamber.

[0206] By analyzing the multi-dimensional features of the high-temperature aging test chamber to be evaluated through the temporal convolutional network model, the health degree of the test chamber at the current moment can be obtained, and at the same time, the state degradation curve (also known as the health degree degradation curve) up to this moment can be output, providing reliable data support for the further research on the remaining service life prediction of the high-temperature aging test chamber.

[0207] The method for evaluating the health status of a high-temperature aging test chamber disclosed in the embodiments of the present invention collects the temperature monitoring signals of the high-temperature aging test chamber to be evaluated; extracts features from the temperature monitoring signals to obtain a multi-dimensional feature sequence; inputs the multi-dimensional feature sequence into a pre-trained temporal convolutional network model to obtain the degradation feature state of the high-temperature aging test chamber; in the case where the high-temperature aging test chamber is in a degraded state, based on the probability space points after mapping of the multi-dimensional feature sequence and the health baseline in the temporal convolutional network model, determines the state degradation curve of the high-temperature aging test chamber to be evaluated, so as to determine the health status of the high-temperature aging test chamber. The method for evaluating the health status of a high-temperature aging test chamber provided by the embodiments of the present invention collects the temperature monitoring signals of the test chamber and extracts a multi-dimensional feature sequence, and predicts the multi-dimensional feature sequence based on the trained temporal convolutional network model, and can determine the state degradation curve of the high-temperature aging test chamber to be evaluated, thereby accurately determining the health status of the high-temperature aging test chamber.

[0208] The following uses a specific example to illustrate the health status evaluation solution of the high-temperature aging test chamber provided by the present application.

[0209] The method for evaluating the health status of a high-temperature aging test chamber based on the recognition of degradation points of a temporal convolutional network and the fusion of multiple distance metrics provided in this specific example has a core as follows: during the aging test process, monitor the temperature of the high-temperature aging test chamber. The directly measured monitoring signal has a large amount of data and contains a large amount of redundant information. First, perform preliminary feature extraction on the directly measured data, including features such as temperature fluctuation degree, temperature control accuracy, temperature uniformity, temperature overshoot, and heating and cooling rate, to obtain a feature sequence with high information density; use the temporal convolutional network to perform adaptive analysis on the feature sequence data to automatically identify the degradation starting point of the high-temperature aging test chamber. Use the distance metric algorithm to quantitatively calculate the health status of the feature sequence in the degradation stage to complete the health assessment of the high-temperature aging test chamber.

[0210] The process of the method for evaluating the health status of a high-temperature aging test chamber based on the recognition of degradation points of a temporal convolutional network and the fusion of multiple distance metrics specifically may include the following process:

[0211] S1: Temperature monitoring of the high-temperature aging test chamber:

[0212] During the aging test process, monitor the temperature signal of the high-temperature aging test chamber through a sensor to obtain a large amount of monitoring data for the state evaluation of the test chamber.

[0213] S2: Preliminary feature extraction.

[0214] Since the directly measured monitoring signal data is large in quantity and contains a large amount of redundant information, it is first necessary to extract features from the directly monitored data, including temperature fluctuation degree, temperature control accuracy, temperature uniformity, temperature overshoot, heating and cooling rate, etc., to obtain a feature sequence with high information density, that is, the multi-dimensional degradation features of the high-temperature aging test chamber.

[0215] S3: Input the multi-dimensional degradation features of the high-temperature aging test chamber into the temporal convolutional network model to automatically identify the degradation starting point, and obtain the health baseline and degradation features.

[0216] On the basis of the extracted feature sequence, the health assessment process is further realized: First, use the TCN (Temporal Convolutional Network) model to judge the degradation starting point of the temperature data and its characteristic data of the high-temperature aging test chamber, and eliminate the instability brought by artificially setting the health baseline to construct health indicators. The TCN model is based on causal convolution, strictly ensuring that the output of each time step depends only on the input before that time step, and then introducing dilated convolution on this basis. By inserting intervals between the convolutional kernel elements and increasing the receptive field according to specific rules, it can effectively capture the long-term dependencies existing in the time series data. The TCN model stacks multiple modules containing causal convolution, dilated convolution, and residual connections reasonably, so that after the time series data is processed layer by layer, rich and valuable feature information can be extracted. Then, connect to the fully connected layer, which will integrate the features extracted previously and map them to a two-dimensional space to realize the discrimination of the degradation starting point of the high-temperature aging test chamber.

[0217] Specifically, in the training stage of the TCN model, input the training data with health status labels (that is, the training time-series degradation feature samples and the degradation feature status labels marked for each sample) , where is the feature sample, is the corresponding label, 0 and 1 represent the healthy state and the degraded state respectively, and n is the number of feature samples obtained by resampling the entire degradation curve with a sliding window. In this process, the TCN model continuously optimizes the mapping parameters of the input data to the low-dimensional feature space, strengthens the classification ability of the healthy state data and the degraded state data, so that the TCN model obtains the ability to automatically segment the health baseline of the test data.

[0218] In the test stage of the TCN model, input the test data (that is, the test time-series degradation feature samples of the test chamber) , the TCN model can map the health baseline data and the degraded data to the two-dimensional probability space according to the model parameters optimized and learned during the training process, complete the identification of the degradation starting point of the test data, and obtain , avoiding the blindness of artificially segmenting the health baseline.

[0219] S4: Adopt a multi-distance metric fusion method to calculate the distance between the health baseline and the degradation features, and determine the state degradation curve and the current health degree of the high-temperature aging test chamber.

[0220] To further realize the health state assessment of the high-temperature aging test chamber at the current moment, use the multi-distance metric fusion method, comprehensively consider the Mahalanobis distance, cosine similarity, and Manhattan distance between sample data and perform weighted fusion to obtain the health degree at the current moment. At the same time, output the health degree degradation curve up to this moment, providing reliable data support for the further remaining service life prediction research of the high-temperature aging test chamber.

[0221] Figure 6 It is a structural block diagram of a device for deep fusion of time-frequency characteristics of the secondary power supply of a high-temperature aging test equipment in an embodiment of the present application.

[0222] The device for deep fusion of time-frequency characteristics of the secondary power supply of the high-temperature aging test equipment provided by the embodiment of the present application includes the following functional modules:

[0223] The acquisition module 601 is used to acquire the fault data set of the devices of the high-temperature aging test equipment; wherein, the fault data includes multiple key parameter time series data;

[0224] The preprocessing module 602 is used to preprocess the fault data set to obtain a training data set and a test data set;

[0225] The transformation module 603 is used to perform short-time Fourier transform on each key parameter time series data in the training data set to obtain a time-frequency diagram;

[0226] The mask training module 604 is used to randomly mask each time-frequency diagram and input it into a preset model for training to generate a feature extraction model;

[0227] The test module 605 is used to test the feature extraction model based on the test data set;

[0228] The prediction module 606 is used to, after passing the test, input the key parameter time series data of the device to be predicted into the feature extraction model to predict the target time-frequency domain features.

[0229] Optionally, the preprocessing module includes:

[0230] The first sub-module is used to, for each key parameter in the fault data set, perform sliding window cutting on the key parameter time series data to construct a sample data set;

[0231] A second sub-module, configured to perform normalization processing on each data sample in the sample data set to obtain a normalized sample data set;

[0232] A third sub-module, configured to select the first preset percentage of sample data from the normalized sample data set as a training data set, and use the remaining sample data as a test data set; wherein, the training data set is used to train the preset model, and the test data set is used to verify the prediction accuracy of the trained feature extraction model.

[0233] Optionally, the transformation module includes:

[0234] A fourth sub-module, configured to, for each time series data in the training data set, intercept the vibration signal in the time series data by using a window function with a limited duration;

[0235] A fifth sub-module, configured to perform Fourier transform on the intercepted vibration signal to obtain a local spectrum in a small range of the corresponding time period;

[0236] A sixth sub-module, configured to gradually analyze each vibration signal band by moving the window function on the time axis to obtain a set of local spectra of the vibration signal, and generate a time-frequency diagram.

[0237] Optionally, the window function is a Hanning window function.

[0238] Optionally, the mask training module includes:

[0239] A seventh sub-module, configured to input the randomly masked time-frequency diagram into the preset model, send it to the encoding module after position encoding, and send the output of the encoding module into the decoder, and restore the feature parameters in the high-dimensional hidden layer into the time-frequency diagram before masking through the decoder;

[0240] An eighth sub-module, configured to iteratively train the preset model multiple times by using the restored time-frequency diagram to adjust the model parameters, where the model parameters include: embedding dimension, number of attention heads, number of encoder layers, and number of decoder layers;

[0241] A ninth sub-module, configured to take out the encoder layer and the decoder layer of the model after multiple iterative adjustments, and retain the weight parameters of each layer;

[0242] A tenth sub-module, configured to generate a feature extraction model based on each encoder layer, each decoder layer, and the weight parameters of each layer.

[0243] Optionally, the device includes: a switching power supply of a high-temperature aging test device, a high-temperature test chamber of a high-temperature aging test device, and a high-temperature and high-humidity test chamber of a high-temperature aging test device.

[0244] Provided by the embodiments of the present applicationFigure 6 The time-frequency feature deep fusion device of the secondary power supply of the high-temperature aging test equipment shown can achieve Figure 1 each process implemented by the method embodiment of, and for the sake of avoiding repetition, it will not be elaborated here.

[0245] The time-frequency feature deep fusion method device of the secondary power supply of the high-temperature aging test equipment provided by the embodiment of the present application fully coordinates and refines a large amount of fault data of the devices of the high-temperature aging test equipment, and performs random masking on the time-frequency diagram obtained by performing short-time Fourier transform on the segmented time series data. The preset model is trained with the time-frequency diagram after random masking, which can fully refine the local semantic information of each part of the devices of the high-temperature aging test equipment, overcome the constraints on the training model such as data scarcity and rough annotation, and the time-frequency features predicted by the trained feature extraction model are more accurate and reliable. Further, the health state of the devices of the high-temperature aging test equipment analyzed based on the predicted time-frequency features is more reliable.

[0246] The embodiment of the present invention also provides an electronic device, including a processor, a communication interface, a memory, and a communication bus. Among them, the processor, the communication interface, and the memory complete mutual communication through the communication bus.

[0247] The memory is used to store a computer program;

[0248] The processor is used to implement the time-frequency feature deep fusion method of the secondary power supply of the high-temperature aging test equipment and the health state evaluation method of the high-temperature aging test chamber shown in the above method embodiment when executing the program stored in the memory.

[0249] The communication bus mentioned in the above terminal may be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. This communication bus can be divided into an address bus, a data bus, a control bus, etc.

[0250] The communication interface is used for communication between the above terminal and other devices.

[0251] The memory may include a Random Access Memory (RAM), and may also include a non-volatile memory, such as at least one disk memory. Optionally, the memory may also be at least one storage device located far from the aforementioned processor.

[0252] In another embodiment provided by the present invention, there is also provided a computer-readable storage medium storing instructions, which, when running on an electronic device, enable the electronic device to implement the method for deep fusion of time-frequency characteristics of the secondary power supply of the high-temperature aging test equipment or the method for evaluating the health status of the high-temperature aging test chamber described in any one of the above embodiments.

[0253] In another embodiment provided by the present invention, there is also provided a computer program product containing instructions, which, when running on an electronic device, enable the electronic device to implement the method for deep fusion of time-frequency characteristics of the secondary power supply of the high-temperature aging test equipment or the method for evaluating the health status of the test chamber of the high-temperature aging test equipment described in any one of the above embodiments.

[0254] It should be noted that in this article, the terms "include", "comprise" or any other variant thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements not only includes those elements, but also includes other elements not explicitly listed, or further includes elements inherent to such process, method, article or device. Without further limitation, an element defined by the statement "including one..." does not exclude the existence of additional identical elements in the process, method, article or device including the element.

[0255] The above is the preferred embodiment of the present invention. It should be pointed out that for those of ordinary skill in the art, without departing from the principle of the present invention, several improvements and refinements can be made, and these improvements and refinements should also be regarded as the protection scope of the present invention.

Claims

1. A method for deeply fusing the time-frequency characteristics of the secondary power supply of a high-temperature aging test device, characterized in that The method includes: Collecting a fault dataset of components in a high-temperature aging test device; wherein, the fault data includes multiple key parameter time-series data; Preprocessing the fault dataset to obtain a training dataset and a test dataset; Performing short-time Fourier transform on each key parameter time-series data in the training dataset to obtain a time-frequency diagram; Randomly masking each of the time-frequency diagrams and inputting them into a preset model for training to generate a feature extraction model; Testing the feature extraction model based on the test dataset; After passing the test, inputting the key parameter time-series data of the device to be predicted into the feature extraction model to predict the target time-frequency domain features; Collecting the monitoring signals of the high-temperature aging test chamber to be evaluated; Extracting features from the monitoring signals to obtain a multi-dimensional feature sequence; Inputting the multi-dimensional feature sequence into a pre-trained temporal convolutional network model to obtain the degradation feature state of the high-temperature aging test chamber; When the high-temperature aging test chamber is in a degraded state, determining the state degradation curve of the high-temperature aging test chamber to be evaluated based on the probability space points after mapping of the multi-dimensional feature sequence and the health baseline in the temporal convolutional network model; Among them, the training process of the temporal convolutional network model includes the following steps: Inputting the training time-series degradation feature samples into a pre-established temporal convolutional network model to obtain the first probability space points after mapping; Discriminating the degradation starting point of the high-temperature aging test chamber according to the first probability space points and generating the health baseline of the high-temperature aging test chamber; Obtaining the test time-series degradation feature samples of the high-temperature aging test chamber, inputting the test time-series degradation feature samples into the temporal convolutional network model under training to obtain the second probability space points after mapping, and calculating the distance between the second probability space points and the health baseline based on a preset multi-distance metric algorithm.

2. The method according to claim 1, characterized in that, The steps of preprocessing the fault dataset to obtain a training dataset and a test dataset include: For each key parameter in the fault dataset, sliding window cutting the key parameter time-series data to construct a sample dataset; Normalizing each data sample in the sample dataset to obtain a normalized sample dataset; Selecting the first preset percentage of sample data from the normalized sample dataset as the training dataset, and using the remaining sample data as the test dataset; wherein, the training dataset is used to train the preset model, and the test dataset is used to verify the prediction accuracy of the trained feature extraction model.

3. The method according to claim 1, characterized in that The steps of performing short-time Fourier transform on each time-series data in the training dataset to obtain a time-frequency diagram include: For each time-series data in the training dataset, using a window function with a finite duration to intercept the vibration signal in the time-series data; Performing Fourier transform on the intercepted vibration signal to obtain a local spectrum in a small range of the corresponding time period; By moving the window function on the time axis, gradually analyzing each vibration signal band to obtain a set of local spectra of the vibration signal and generating a time-frequency diagram.

4. The method according to claim 3, characterized in that, The window function is a Hanning window function.

5. The method according to claim 1, wherein The steps of randomly masking each of the time-frequency diagrams and inputting them into a preset model for training to generate a feature extraction model include: Input the time-frequency map after random masking into the preset model, send it to the encoding module after position encoding, and send the output of the encoding module into the decoder, and restore the feature parameters in the high-dimensional hidden layer into the time-frequency map before masking through the decoder; Use the restored time-frequency map to iteratively train the preset model multiple times to adjust the model parameters, where the model parameters include: embedding dimension, number of attention heads, number of encoder layers, and number of decoder layers; Extract the encoder layer and decoder layer of the model after multiple iterative adjustments, and retain the weight parameters of each layer; Generate a feature extraction model based on each encoder layer, each decoder layer, and the weight parameters of each layer.

6. The method according to claim 1, characterized in that, The device includes: a switching power supply for a high-temperature aging test device, a high-temperature test chamber for a high-temperature aging test device, and a high-temperature and high-humidity test chamber for a high-temperature aging test device.

7. An electronic device, characterized in that, It includes a processor, a communication interface, a memory, and a communication bus. Among them, the processor, communication interface, and memory complete mutual communication through the communication bus; The memory is used to store computer programs; When the processor is used to execute the program stored on the memory, it implements the time-frequency feature deep fusion method for the secondary power supply of the high-temperature aging test device as described in any one of claims 1-6.

Images