Double-pulse test method, device and equipment and storage medium

Abstract

The invention relates to the technical field of electronics, and discloses a double-pulse test method, device and equipment and a storage medium, and the method comprises the steps: obtaining multi-source feature information of a tested device to generate a device feature code; performing a double-pulse test based on the device feature code, and collecting time-frequency domain data of the tested device in the test process; performing oscillation feature recognition on the time-frequency domain data to obtain oscillation feature data; adjusting the operation state of the double-pulse test based on the oscillation characteristic data, and terminating the test until a preset suppression target is met or a preset repeated test threshold value is reached, so as to obtain a final test result; according to the invention, on the basis of feature recognition and data acquisition, continuous oscillation adjustment is realized, the optimal test operation state adaptive to the tested device is generated, the core problems of much manual intervention, long test period and poor oscillation suppression effect of the existing double-pulse test are solved, and the stability of the double-pulse test is improved.

Description

Technical Field

[0001] This invention relates to the field of electronic technology, and in particular to a dual-pulse testing method, apparatus, device, and storage medium. Background Technology

[0002] As power semiconductor devices rapidly evolve from Si-based IGBTs to SiC / GaN wide-bandgap devices, switching dv / dt and di / dt are significantly improved, making parasitic oscillations and EMI problems increasingly prominent. Meanwhile, application scenarios such as 800V platforms for new energy vehicles and 1500V photovoltaic systems place higher demands on the accuracy and efficiency of dual-pulse testing.

[0003] However, in existing dual-pulse testing, oscillation suppression mostly relies on manual or simple programmable adjustment of parameters such as gate resistance and drive voltage. This lacks precise adaptation to the individual characteristics of the device under test and cannot dynamically optimize the operating state based on the device's operating data collected in real time during the test. As a result, the oscillation suppression effect is poor and the test efficiency is low, making it difficult to meet the requirements of wide-bandgap power semiconductor devices for test accuracy and adaptive adjustment. Summary of the Invention

[0004] To overcome the shortcomings of existing technologies, the present invention aims to provide a dual-pulse testing method, apparatus, device, and storage medium. Based on feature recognition and data acquisition, the present invention achieves continuous oscillation adjustment and generates the optimal test operation state adapted to the device under test, solving the core problems of existing dual-pulse testing such as excessive manual intervention, long test cycle, and poor oscillation suppression effect, thereby improving the stability of dual-pulse testing.

[0005] The first aspect of the present invention provides a dual-pulse testing method, the dual-pulse testing method comprising the steps of: acquiring multi-source feature information of the device under test to generate a device feature code; performing a dual-pulse test based on the device feature code, and acquiring time-frequency domain data of the device under test during the test process; identifying oscillation features in the time-frequency domain data to obtain oscillation feature data; adjusting the running state of the dual-pulse test based on the oscillation feature data until a preset suppression target is met or a preset repetitive test threshold is reached, and terminating the test to obtain the final test result.

[0006] Optionally, in a first implementation of the first aspect of the present invention, the step of obtaining multi-source feature information of the device under test to generate a device feature code includes: obtaining surface silkscreen information, electrical fingerprint data and thermal resistance distribution characteristics of the device under test through optical recognition, microcurrent scanning and infrared thermal imaging detection, respectively; and performing weighted fusion of the surface silkscreen information, electrical fingerprint data and thermal resistance distribution characteristics to obtain a device feature code.

[0007] Optionally, in a second implementation of the first aspect of the present invention, the step of performing a double-pulse test based on the device feature code and collecting time-frequency domain data of the device under test during the test includes: retrieving a historical parameter set from a preset cloud knowledge base using the device feature code as an index; filtering the historical parameter set based on an improved k-nearest neighbor algorithm to obtain initial recommended parameters; performing interpolation and threshold adjustment processing on the initial recommended parameters to obtain initial test parameters; performing a double-pulse test on the device under test based on the initial test parameters, and simultaneously collecting the original voltage and current signals of the device under test during the test to obtain time-frequency domain data.

[0008] Optionally, in a third implementation of the first aspect of the present invention, the step of identifying oscillation features in the time-frequency domain data to obtain oscillation feature data includes: splitting the time-frequency domain data into a time-frequency image and a time-series waveform; extracting spectral local texture features from the time-frequency image using a preset CNN network model; modeling the long-range temporal dependency of the time-series waveform using a preset LSTM network model to predict the voltage overshoot amplitude; and integrating the spectral local texture features and the voltage overshoot amplitude to obtain oscillation feature data.

[0009] Optionally, in a fourth implementation of the first aspect of the present invention, adjusting the operating state of the double-pulse test based on the oscillation feature data until a preset suppression target is met or a preset repetitive test threshold is reached to terminate the test and obtain the final test result includes: optimizing the optimal combination of the gate resistance, driving voltage, and load inductance of the device under test based on a preset reinforcement learning algorithm and oscillation feature data to obtain the optimal parameter combination; converting the optimal parameter combination into a test adjustment command, and adjusting the operating state of the double-pulse test based on the test adjustment command; after adjusting the operating state, determining whether the oscillation suppression effect of the device under test meets the preset suppression target; if it does... If the preset suppression target is met, the double-pulse test is terminated, and the running data during this test is collected as the final test result. If the preset suppression target is not met but the number of repeated tests is less than the preset repetition threshold, the test returns to the step of optimizing the gate resistance, drive voltage, and load inductance of the device under test based on the preset reinforcement learning algorithm and oscillation feature data until the preset suppression target is met, at which point the double-pulse test is terminated, and the running data during this test is collected as the final test result. If the preset suppression target is not met and the current number of repeated optimizations exceeds the preset repetition threshold, the double-pulse test is terminated, and a test anomaly flag is generated as the final test result.

[0010] A second aspect of the present invention provides a dual-pulse testing device, comprising: an acquisition module for acquiring multi-source feature information of a device under test to generate a device feature code; a testing module for performing dual-pulse testing based on the device feature code and acquiring time-frequency domain data of the device under test during the testing process; an identification module for identifying oscillation features in the time-frequency domain data to obtain oscillation feature data; and an iteration module for adjusting the operating state of the dual-pulse test based on the oscillation feature data until a preset suppression target is met or a preset repetitive test threshold is reached, thereby terminating the test and obtaining the final test result.

[0011] A third aspect of the present invention provides a dual-pulse testing device, the dual-pulse testing device comprising: a memory and at least one processor, the memory storing instructions; at least one processor calling the instructions in the memory to cause the computer device to execute the various steps of any of the dual-pulse testing methods described above.

[0012] A fourth aspect of the present invention provides a computer-readable storage medium storing instructions that, when executed by a processor, implement the steps of any of the above-described double-pulse testing methods.

[0013] In the technical solution of this invention, firstly, multi-source feature information of the device under test is acquired, and a unique device feature code is generated to capture the differentiated features of individual devices, such as surface silkscreen, electrical fingerprint, and thermal resistance distribution, providing a customized adaptation basis for subsequent testing. Then, based on the device feature code, a dual-pulse test is initiated and time-frequency domain data is collected to provide a basis for oscillation feature analysis. Next, oscillation feature data is obtained by oscillation feature identification of the time-frequency domain data, which can objectively and accurately extract key information such as oscillation type, main frequency, and peak value, replacing manual subjective interpretation and solving the problem of inaccurate oscillation feature capture in traditional testing. Finally, the test running state is dynamically adjusted based on a preset reinforcement learning algorithm and oscillation feature data, realizing iterative optimization of parameters. This can effectively improve the oscillation suppression effect during the test process, complete the test efficiently and stably, and obtain the final test results. Based on feature recognition and data acquisition, this invention realizes continuous oscillation adjustment, generates the optimal test running state adapted to the device under test, solves the core problems of existing dual-pulse tests such as excessive manual intervention, long test cycle, and poor oscillation suppression effect, and improves the stability of dual-pulse tests. Attached Figure Description

[0014] The above and / or additional aspects and advantages of the present invention will become apparent and readily understood from the description of the embodiments taken in conjunction with the following drawings, in which: Figure 1 This is a first flowchart of a dual-pulse testing method provided in an embodiment of the present invention; Figure 2This is a second flowchart of the dual-pulse testing method provided in an embodiment of the present invention; Figure 3 This is a third flowchart of the dual-pulse testing method provided in the embodiments of the present invention; Figure 4 This is a fourth flowchart of the dual-pulse testing method provided in the embodiments of the present invention; Figure 5 The fifth flowchart of the dual-pulse testing method provided in the embodiments of the present invention; Figure 6 This is a schematic diagram of the structure of the dual-pulse testing device provided in an embodiment of the present invention; Figure 7 This is a schematic diagram of the structure of the dual-pulse testing device provided in an embodiment of the present invention. Detailed Implementation

[0015] This invention provides a dual-pulse testing method, apparatus, device, and storage medium. Based on feature recognition and data acquisition, it achieves continuous oscillation adjustment and generates the optimal test operation state adapted to the device under test. It solves the core problems of existing dual-pulse testing, such as excessive manual intervention, long test cycle, and poor oscillation suppression effect, and improves the stability of dual-pulse testing.

[0016] The terms "first," "second," "third," "fourth," etc. (if present) in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" or "having" and any variations thereof are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0017] For ease of understanding, the specific process of the embodiments of the present invention is described below. Please refer to [link / reference]. Figure 1 One embodiment of the dual-pulse testing method in this invention includes: The dual-pulse testing method includes the following steps: 101. Obtain multi-source feature information of the device under test to generate device feature code; In this embodiment, the surface silkscreen information, electrical fingerprint data and thermal resistance distribution characteristics of the device under test are collected by optical recognition equipment, microcurrent scanning equipment and infrared thermal imager, respectively, to form a multi-source feature information set; then, the DS evidence theory is used to weight and fuse the multi-source feature information to eliminate the uncertainty and redundancy of information from different sources, and finally generate a device feature code that uniquely identifies the characteristics of the device under test.

[0018] 102. Perform double-pulse testing based on device feature codes, and collect time-frequency domain data of the device under test during the testing process; In this embodiment, the device feature code is used as the retrieval basis to match the appropriate initial test parameters from the cloud knowledge base or local database, including key parameters such as gate resistance, drive voltage and load inductance. Based on the initial test parameters, the dual-pulse test system is started. A high-speed oscilloscope is used with a high-voltage probe and a current probe to synchronously acquire the original voltage and current signals of the device under test during the test process. After denoising and baseline correction of the original signals, they are converted into time-frequency domain data containing time-domain waveform and frequency-domain spectrum through short-time Fourier transform.

[0019] 103. Perform oscillation feature identification on the time-frequency domain data to obtain oscillation feature data; In this embodiment, the time-frequency domain data is split into two types of data: time-frequency images and time-series waveforms. These are then input into a preset CNN network model and an LSTM network model, respectively. The CNN network model extracts local spectral texture features from the time-frequency images. The LSTM network model models the long-range temporal dependencies of the time-series waveforms to predict the voltage overshoot amplitude of the next pulse. Finally, the outputs of the two models are integrated to generate oscillation feature data that includes oscillation type, peak value, attenuation coefficient, and predicted overshoot amplitude.

[0020] 104. Adjust the running state of the double pulse test based on the oscillation characteristic data until the preset suppression target is met or the preset repeated test threshold is reached, and then terminate the test to obtain the final test result. In this embodiment, the gate resistance, driving voltage, and load inductance are first optimized by combining reinforcement learning algorithm and oscillation feature data to obtain the optimal parameter combination and convert it into adjustment command. Then, the operation state adjustment of the double pulse test is completed based on the adjustment command. After adjustment, the oscillation feature is identified to determine whether the preset oscillation suppression target is met. If it is met, the test is terminated and the data of this operation is collected as the final test result. If it is not met and the repeated test threshold is not exceeded, the above optimization and adjustment process is repeated. If the repeated test threshold is exceeded, the test is terminated and a test anomaly mark is generated to form the final test result.

[0021] In this embodiment of the invention, firstly, multi-source feature information of the device under test (DUT) is acquired, and a unique device feature code is generated. This captures the differentiated features of individual devices, such as surface silkscreen printing, electrical fingerprints, and thermal resistance distribution, providing a customized adaptation basis for subsequent testing. Then, a dual-pulse test is initiated based on the device feature code, and time-frequency domain data is collected to provide a basis for oscillation feature analysis. Next, oscillation feature data is obtained by performing oscillation feature identification on the time-frequency domain data. This allows for the objective and accurate extraction of key information such as oscillation type, dominant frequency, and peak value, replacing manual subjective interpretation and solving the problem of inaccurate oscillation feature capture in traditional testing. Finally, the test running state is dynamically adjusted based on a preset reinforcement learning algorithm and oscillation feature data, achieving iterative optimization of parameters. This effectively improves the oscillation suppression effect during the test process, efficiently and stably completes the test, and obtains the final test results. Based on feature recognition and data acquisition, this invention achieves continuous oscillation adjustment, generating the optimal test running state adapted to the DUT, solving the core problems of excessive manual intervention, long test cycles, and poor oscillation suppression effect in existing dual-pulse tests, and improving the stability of dual-pulse tests.

[0022] Please see Figure 2 In the second embodiment of the dual-pulse testing method of the present invention, step 101 includes: 201. By using optical recognition, microcurrent scanning and infrared thermal imaging detection, the surface silkscreen information, electrical fingerprint data and thermal resistance distribution characteristics of the device under test are obtained respectively. In this embodiment, the Cognex In-Sight D905M vision system is used as the optical recognition module to optically recognize the surface silkscreen printing of the device under test. This system can effectively decode deformed or poorly etched characters and accurately extract key identification information such as the device's model specifications, production batch, and rated electrical parameters, providing a reliable basis for preliminary determination of the device's basic attributes. A Prosys SPA-6100 semiconductor parameter analyzer is used for micro-current scanning, applying a weak excitation signal to the device's gate. Its built-in high-precision source measurement unit (SMU) acquires the IV characteristic curve between the gate and emitter, extracting electrical fingerprint data that reflects the individual differences in the internal semiconductor structure of the device. This data is the core basis for distinguishing the characteristics of different devices. FLIR is then used to... The A655sc infrared thermal imager performs infrared thermal imaging detection on devices under test in an inactive state, capturing differences in heat distribution on the device surface to obtain the thermal resistance distribution characteristics of the device. These characteristics can help determine the device packaging quality and the bonding status of internal chips. The use of three heterogeneous detection methods can significantly improve the reliability and comprehensiveness of device feature recognition, laying a solid foundation for the accurate adaptation of subsequent test parameters and avoiding test deviations caused by inaccurate information from a single feature.

[0023] 202. Weighted fusion of surface silkscreen information, electrical fingerprint data and thermal resistance distribution characteristics is performed to obtain device feature codes; In this embodiment, the DS evidence theory is used to perform weighted fusion processing on three types of multi-source feature information. This theory can effectively handle the problem of synthesizing uncertain multi-source information and reduce the impact of misjudgment of a single feature. First, dynamic weights are assigned to various feature information based on their importance and reliability in device identification: surface silkscreen information, as a direct reflection of the device's basic attributes, is assigned a basic weight; electrical fingerprint data has the highest distinguishability of device characteristics and is assigned the highest weight; thermal resistance distribution features, as auxiliary verification information, are assigned an auxiliary weight; the weights can be dynamically adjusted according to changes in the actual test scenario. The three types of feature data are then standardized and transformed into feature vectors of a unified dimension. The confidence and likelihood between feature vectors are calculated using the synthesis rules of the DS evidence theory to eliminate contradictions and redundancies between multi-source information and finally generate a unique device feature code. For new device models not yet documented in the cloud knowledge base, a feature library of existing similar devices is used for rapid classification to improve the adaptability of feature fusion. Through feature fusion, the unique value of various feature information is fully utilized. The generated device feature code serves as the unique identifier for the initial test parameters, ensuring the accuracy of initial parameter matching and completely avoiding subjective errors caused by manual selection.

[0024] Please see Figure 3 In the third embodiment of the dual-pulse testing method of the present invention, step 102 includes: 301. Using the device feature code as an index, retrieve the historical parameter set from the preset cloud knowledge base; In this embodiment, a pre-defined cloud-based knowledge base stores historically tested device data, including the feature codes of various devices, corresponding gate resistance, drive voltage, load inductance, and other test parameters, as well as information such as oscillation suppression effects and environmental conditions during the testing process. Using device feature codes as search keywords, all historical test parameter records corresponding to similar feature codes are quickly matched, forming a historical parameter set containing multiple sets of different parameter combinations and corresponding test effects. The uniqueness of device feature codes ensures the accuracy of the search, enabling rapid location of historical data with characteristics similar to the currently tested device, providing a reliable data foundation for subsequent initial parameter recommendations.

[0025] 302. Based on the improved k-nearest neighbor algorithm, the historical parameter set is filtered to obtain the initial recommended parameters; In this embodiment, the improved k-nearest neighbor algorithm introduces a dynamic weighting mechanism based on the traditional algorithm. Differential weights are assigned to each historical parameter based on factors such as the number of tests corresponding to each record in the historical parameter set, the similarity between the test environment and the current test environment, and the degree of oscillation suppression performance. Parameter records with more test counts, higher environmental similarity, and better suppression performance are given higher weights, while those with lower weights are given lower weights. By calculating the similarity between the current device feature code and each feature code in the historical parameter set, and combining this with weighting factors for comprehensive sorting, the top k groups of parameters with the highest similarity and optimal weights are selected as the candidate set. Statistical analysis is then performed on the parameters in the candidate set, and the intersection of the optimal intervals of each group of parameters is taken to finally obtain the initial recommended parameters for the gate resistance, drive voltage, and load inductance.

[0026] 303. Interpolate and adjust the threshold of the initial recommended parameters to obtain the initial test parameters; In this embodiment, it is first determined whether there are missing data or large ranges in the initial recommended parameters. If so, a quadratic interpolation algorithm is used to complete and smooth the parameters, ensuring their continuity and rationality, and avoiding excessive stress on the device during the test due to parameter mutations. Subsequently, based on the rated electrical parameters of the device under test and the hardware safety threshold of the test system, the interpolated initial recommended parameters are threshold-verified and adjusted to limit the parameter values ​​within a safe operating range and set a conservative initial test starting point. For example, if the initial recommended parameters are close to the upper limit of the safety threshold, they are appropriately lowered to a reasonable range within the threshold range to ensure that the first double-pulse test will not damage the device under test. Through interpolation and threshold adjustment, it is ensured that the initial parameters can reflect the device characteristics, while the test risks are avoided through threshold constraints.

[0027] 304. Perform a double-pulse test on the device under test based on the initial test parameters, and simultaneously acquire the original voltage and current signals of the device under test during the test to obtain time-frequency domain data; In this embodiment, the dual-pulse test system consists of a core test unit, a drive unit, an acquisition unit, and a synchronization control unit. The core test unit uses a Keysight PD1500A dual-pulse test host as the core platform, paired with a programmable high-voltage DC power supply to provide a stable bus voltage for testing. The intelligent load unit uses a Chroma 63200 series programmable electronic load to achieve precise switching of load inductance. The drive unit uses a TI UCC21520 isolated gate driver board, which can accurately output drive signals and switch the corresponding gate resistance level according to the drive voltage and gate resistance setting values ​​in the initial test parameters. The acquisition unit consists of a Tektronix MDO3024 mixed-domain oscilloscope, a Tektronix P5200A high-voltage probe, and a TCPA300 current probe. The synchronization control unit achieves time base synchronization of the entire system through a GPS disciplined clock module. After the test is started, the dual-pulse test host triggers the first set of pulses according to the preset timing sequence. After receiving the command, the gate driver board drives the device under test to turn on and off. During the device's turn-off process, the first voltage overshoot and oscillation are generated. After a preset time interval, the second set of pulses is triggered, repeating the turn-on and turn-off process to form a complete dual-pulse test cycle. Throughout the two sets of pulses, the high-voltage probe is in close contact with the voltage output terminal of the device under test to capture the original voltage signal in real time. The current probe is placed in the current loop of the device to synchronously acquire the original current signal. The oscilloscope samples the two signals synchronously at a sampling rate matched with the initial test parameters (adaptively adjusted according to the device's switching characteristics) to ensure that the timestamps of the voltage and current signals are completely aligned. After sampling, the oscilloscope's built-in data processing module first performs wavelet threshold denoising and baseline correction on the original signal to eliminate the influence of environmental noise and DC drift. Then, it converts the time-domain waveform into frequency-domain data through short-time Fourier transform, finally forming complete time-frequency domain data containing the time-domain waveform, frequency spectrum, and key frequency component distribution.

[0028] Please see Figure 4 In the fourth embodiment of the dual-pulse testing method of the present invention, step 103 includes: 401. Decompose the time-frequency domain data into time-frequency images and time-series waveforms; In this embodiment, the time-frequency domain data includes two core types of information: time-domain waveform and frequency-domain spectrum. In order to adapt to the processing characteristics of different network models, it is necessary to perform targeted splitting. On the one hand, the time-frequency domain data is mapped into a two-dimensional visualized time-frequency image according to the three-dimensional relationship of time-frequency-amplitude. This image can intuitively present the energy distribution and time evolution of the oscillation signal in different frequency bands. On the other hand, continuous data sequences of original voltage and current changes over time without frequency domain conversion are extracted from time-frequency domain data as time-series waveforms, preserving the real-time trend of signal changes and time-domain features such as peaks and overshoots; by splitting the data, the two types of data are matched respectively with the ability of subsequent CNN networks to extract image features and the modeling advantages of LSTM networks for time-series data, providing a data adaptation basis for accurate identification of oscillation features.

[0029] 402. Extract local spectral texture features from time-frequency images using a pre-defined CNN network model; In this embodiment, the CNN network model adopts an architecture of "input layer - convolutional layer - activation layer - pooling layer - fully connected layer - output layer". The model has five convolutional layers: the first three layers use 3×3 kernels to capture fine-grained spectral texture, and the last two layers use 5×5 kernels to aggregate global features. Each convolutional layer is followed by a batch normalization layer to reduce the risk of gradient vanishing. The activation layer uses the ReLU function to introduce nonlinear mapping, enhancing the model's ability to express complex oscillation patterns. The pooling layer uses max pooling with a stride of 2×2. Two fully connected layers are set, containing 1024 and 256 neurons respectively. Finally, the output layer outputs feature vectors corresponding to the oscillation feature categories. During the model training phase, different types of power devices (IGBT, SiC) are collected. Time-frequency image samples of MOSFETs under different oscillation conditions were labeled with oscillation frequency, harmonic energy ratio, and spectral peak distribution, and divided into training, validation, and test sets in a 7:2:1 ratio. A stochastic gradient descent (SGD) optimizer was used with a dynamic learning rate (initial learning rate of 0.001, decaying by 10% every 10 rounds) to iteratively train the dataset with the goal of minimizing the cross-entropy loss function. The split time-frequency images are input into the trained CNN network. After the above-mentioned layer processing, the output is a spectral local texture feature vector containing key information such as oscillation frequency, harmonic energy ratio, and spectral peak distribution. The core advantage of using the CNN network is that the strong ability of the CNN network to capture local features of the image can accurately identify the differences in oscillation modes hidden in the time-frequency image, replacing the traditional subjective interpretation of frequency domain data by humans and improving the objectivity and accuracy of feature extraction.

[0030] 403. Use a pre-defined LSTM network model to model the long-range time-series dependencies of the time-series waveform in order to predict the voltage overshoot amplitude. In this embodiment, the LSTM network model adopts an architecture of "input layer - embedding layer - LSTM layer - dropout layer - fully connected layer - output layer". During model construction, the input layer receives time-series data sequences segmented into fixed time windows, and the embedding layer maps the input data into low-dimensional dense vectors. Three stacked LSTM layers are set, each containing 256 neurons. The outputs of the first two LSTM layers are connected to the dropout layer (dropout rate set to 0.2) to prevent overfitting. The output of the third LSTM layer serves as the core expression of the time-series features. Two fully connected layers are set, which map the time-series features to the predicted output dimension through linear transformation. Finally, the output layer outputs the predicted voltage overshoot amplitude. During model training, voltage and current time-series waveform data of different devices under different test parameters are collected, segmented into fixed-length sequence samples according to time order, and labeled with the actual voltage overshoot amplitude of the next pulse corresponding to each sequence to construct a training dataset. The Adam optimizer is used with a learning rate of 0.0005, and the mean squared error loss function is used to measure the deviation between the predicted and actual values. The model parameters are optimized through multiple rounds of iterative training. The decomposed voltage and current time-series waveforms are segmented into continuous data sequences according to fixed time windows and input into a trained LSTM network. A forget gate filters and discards irrelevant historical time-series information, while the input gate selectively receives signal features at the current moment and updates the cell state. The output gate generates the current moment's feature output based on the updated cell state. Through multiple iterations, the long-range dependencies of the time-series waveforms are modeled, capturing dynamic features such as the decay trend and peak occurrence patterns of the oscillation signal. Ultimately, the voltage overshoot amplitude that may occur in the next pulse test is predicted. The LSTM network can uncover hidden trend information in the time-series waveforms, overcoming the limitations of traditional methods that can only perform feature analysis based on the current signal, and improving the comprehensiveness and predictive ability of oscillation feature recognition.

[0031] 404. Integrate the local texture features of the spectrum and the voltage overshoot amplitude to obtain oscillation feature data; In this embodiment, the spectral local texture feature vector output by the CNN network and the voltage overshoot amplitude predicted by the LSTM network are first standardized to convert the two types of data into feature parameters with the same dimension and unit, so as to avoid integration deviation caused by differences in data format. Subsequently, a feature fusion algorithm was used to dynamically assign weights to the two types of features according to their importance. Among them, the local spectral texture features, as the core identifier of the oscillation mode, were assigned a higher weight, while the voltage overshoot amplitude, as a key quantitative indicator of the severity of oscillation, was assigned a lower weight. The weight ratio could be dynamically adjusted according to the priority of oscillation suppression in the actual test scenario. Through weighted summation, the two types of features were integrated into oscillation feature data containing multi-dimensional information such as oscillation frequency, harmonic energy ratio, spectral distribution, and predicted voltage overshoot amplitude, forming a comprehensive and systematic description of oscillation features.

[0032] Please see Figure 5 In the fifth embodiment of the dual-pulse testing method of the present invention, step 104 includes: 501. Based on a preset reinforcement learning algorithm and oscillation feature data, the optimal combination of gate resistance, driving voltage and load inductance of the device under test is searched to obtain the optimal parameter combination. In this embodiment, the preset reinforcement learning algorithm adopts an improved DDQN (Dual Deep Q Network) architecture. During the model construction phase, the state space integrates 18-dimensional oscillation feature data, 3-dimensional current test parameters (gate resistance, drive voltage, and real-time load inductance values), and 4-dimensional environmental variables (test environment temperature, bus voltage stability, humidity, and equipment operating status). The action space adopts a continuous-discrete hybrid encoding, where the gate resistance and drive voltage are adjusted continuously to ensure accuracy, and the load inductance is switched discretely according to preset levels to balance efficiency. The reward function is designed as a multi-objective weighted form, comprehensively considering voltage overshoot amplitude, switching loss, oscillation decay time, and safety penalty. The weight coefficients are dynamically adjusted according to the oscillation type in the oscillation feature data. For example, the weight ratio of load inductance adjustment can be increased for oscillations dominated by parasitic inductance, while the weight of gate resistance and drive voltage is increased for oscillations caused by improper gate drive parameters. During the model training phase, test data under different devices and working conditions are collected first. An experience replay pool containing state-action-reward-next state is constructed. The Adam optimizer is used to iteratively update the network parameters until the model's parameter convergence speed and suppression effect on the validation set meet the standards. During the optimization process, the oscillation characteristic data, the current test state, and environmental variables are input into the trained DDQN model. The model quickly calculates the Q value of each parameter combination through forward inference, selects the parameter combination with the optimal Q value as the output, and achieves the coordinated optimization of gate resistance, drive voltage, and load inductance.

[0033] 502. Convert the optimal parameter combination into test adjustment instructions, and adjust the running state of the double pulse test based on the test adjustment instructions; In this embodiment, the optimal parameter combination output by the DDQN model (target gate resistance value, target drive voltage value, and target load inductance level) is converted into standardized test adjustment commands according to a preset communication protocol. The commands include key information such as parameter type, target value, adjustment accuracy requirements, and execution time limit. The adjustment commands are then sent to the corresponding execution devices: the gate resistance and drive voltage adjustment commands are sent to the TI UCC21520 isolated gate driver board, which uses a built-in 16-bit DAC and relay matrix to switch the gate resistance and adjust the drive voltage output according to the preset step accuracy as required by the commands; the load inductance adjustment command is sent to the Chroma 63200 series programmable electronic load, which uses tapped relays to achieve rapid switching of the load inductance level, ensuring timely adjustment. During execution, each execution device collects the actual parameter values ​​after adjustment in real time, forming status feedback information. The feedback information is then compared and verified with the target values. If there is a deviation, a secondary fine-tuning command is triggered until the actual parameters match the optimal parameter combination, completing the precise adjustment of the dual-pulse test operation state.

[0034] 503. After adjusting the operating status, determine whether the oscillation suppression effect of the device under test meets the preset suppression target; In this embodiment, after the operating state adjustment is completed, the voltage and current raw signals of the device under test in the new operating state are re-synchronized and acquired using a high-speed oscilloscope with a high-voltage probe and a current probe. After denoising and baseline correction of the raw signals, they are converted into time-frequency domain data containing time-domain waveforms and frequency-domain spectra by short-time Fourier transform. Then, key feature data such as oscillation type, dominant frequency, peak value, and attenuation coefficient are extracted by a CNN-LSTM hybrid network model. The preset suppression target is set based on the power semiconductor device test standard and actual application requirements. It is clear that the peak value of the oscillation peak must be lower than the preset ratio of the rated bus voltage, and the oscillation attenuation time must not exceed the limit value. At the same time, the increase in switching loss is required to be controlled within a reasonable range to avoid excessive suppression of oscillation leading to a decrease in device efficiency. The suppression effect is determined by comparing the extracted oscillation feature data with the suppression target item by item.

[0035] 504. If the preset suppression target is met, the double pulse test will be terminated, and the running data during this test will be collected as the final test result. In this embodiment, when the oscillation suppression effect is determined to meet the preset target, the test system issues a test termination command, stops the output of the dual-pulse signal and cuts off the bus voltage of the test circuit; then the data acquisition process is started to comprehensively collect key operating data during this test, including the final determined combination of gate resistance, drive voltage and load inductance parameters, voltage and current waveform data throughout the process, oscillation characteristic evolution records, switching loss calculation results, etc.; after standardizing the format of these data, they are integrated according to the preset structure to form a complete final test result.

[0036] 505. If the preset suppression target is not met and the value is less than the preset repeated test threshold, the process returns to the step of performing the optimal combination optimization of the gate resistance, drive voltage and load inductance of the device under test based on the preset reinforcement learning algorithm and oscillation feature data until the preset suppression target is met, at which point the double pulse test is terminated, and the running data during this test is obtained as the final test result. In this embodiment, if the oscillation suppression effect does not meet the standard and does not exceed the repeated test threshold, the system does not terminate the test process. Instead, it uses the adjusted oscillation characteristic data, current operating parameters, and suppression effect evaluation results as updated input information and calls the preset reinforcement learning algorithm again. The algorithm combines the new state data to perform a new round of collaborative optimization of the parameter combination of gate resistance, driving voltage, and load inductance, generating the optimal parameter combination that fits the current oscillation state. Then, the operating state adjustment is completed according to the process in step 502, and the suppression effect is judged again through step 503. This closed-loop iterative process continues until the oscillation suppression effect meets the preset target. At this time, the test is terminated, and the complete operating data in this round of testing is collected and integrated as the final test result to ensure that the test result can truly reflect the characteristics of the device in the optimal suppression state.

[0037] 506. If the preset suppression target is not met and the current number of repeated optimizations exceeds the preset repeated test threshold, the double pulse test will be terminated and a test anomaly flag will be generated as the final test result. In this embodiment, when the number of repeated optimization attempts reaches a preset threshold but the oscillation suppression effect still fails to meet the target, it is determined that the device under test may have quality problems such as packaging defects or internal structural abnormalities, or that the current test conditions are extremely incompatible with the device characteristics. At this time, the test system issues a termination command to stop the double-pulse test and cut off the bus voltage to avoid damage to the device and test equipment caused by continuous abnormal oscillation. At the same time, a test anomaly identifier is generated, which includes information such as the anomaly type, the unmet oscillation index, the cumulative number of optimization attempts, and the parameter adjustment records of each round. This identifier is integrated with all the running data and waveform files collected during the test to form the final test result. This result can assist staff in conducting subsequent failure analysis, troubleshooting device quality problems or test system compatibility issues, and providing a reference for subsequent process improvement or test scheme optimization.

[0038] The above describes the dual-pulse testing method in the embodiments of the present invention. The following describes the dual-pulse testing device in the embodiments of the present invention. Please refer to [link / reference]. Figure 6 One embodiment of the dual-pulse testing device in this invention includes: The acquisition module 601 is used to acquire multi-source feature information of the device under test to generate device feature codes; Test module 602 is used to perform dual-pulse testing based on device feature codes and to acquire time-frequency domain data of the device under test during the testing process; The identification module 603 is used to identify oscillation characteristics in time-frequency domain data to obtain oscillation characteristic data; The iteration module 604 is used to adjust the running state of the double pulse test based on the oscillation characteristic data until the preset suppression target is met or the preset repeated test threshold is reached, and then the test is terminated to obtain the final test result. In this embodiment, firstly, the acquisition module 601 acquires multi-source feature information of the device under test (DUT) and generates a unique device feature code, capturing the differentiated features of individual devices such as surface silkscreen, electrical fingerprints, and thermal resistance distribution, providing a customized adaptation basis for subsequent testing. Then, the testing module 602 initiates a dual-pulse test based on the device feature code and collects time-frequency domain data, providing a basis for oscillation feature analysis. Next, the identification module 603 performs oscillation feature identification on the time-frequency domain data to obtain oscillation feature data, which can objectively and accurately extract key information such as oscillation type, main frequency, and peak value, replacing manual subjective interpretation and solving the problem of inaccurate oscillation feature capture in traditional testing. Finally, the iteration module 604 dynamically adjusts the test operation state based on a preset reinforcement learning algorithm and oscillation feature data, achieving iterative optimization of parameters. This effectively improves the oscillation suppression effect during the test, efficiently and stably completing the test and obtaining the final test results. Based on feature recognition and data acquisition, this invention achieves continuous oscillation adjustment, generating the optimal test operation state adapted to the DUT, solving the core problems of excessive manual intervention, long test cycles, and poor oscillation suppression in existing dual-pulse tests, and improving the stability of dual-pulse testing.

[0039] Figure 7 This is a schematic diagram of a dual-pulse testing device 900 provided in an embodiment of the present invention. The dual-pulse testing device 900 can vary significantly due to different configurations or performance. It may include one or more central processing units (CPUs) 910 (e.g., one or more processors) and a memory 920, and one or more storage media 930 (e.g., one or more mass storage devices) storing application programs 933 or data 932. The memory 920 and storage media 930 can be temporary or persistent storage. The program stored in the storage media 930 may include one or more modules (not shown in the diagram), each module may include a series of instruction operations on the dual-pulse testing device 900. Furthermore, the processor 910 may be configured to communicate with the storage media 930 and execute the series of instruction operations in the storage media 930 on the dual-pulse testing device 900 to implement the steps of the trimming tool block mounting base plate design method provided in the above-described method embodiments.

[0040] The dual-pulse test device 900 may also include one or more power supplies 940, one or more wired or wireless network interfaces 950, one or more input / output interfaces 960, and / or one or more operating systems 931, such as Windows Server, Mac OS X, Unix, Linux, FreeBSD, etc. Those skilled in the art will understand that... Figure 7 The illustrated dual-pulse test device structure does not constitute a limitation on the dual-pulse test device, and may include more or fewer components than illustrated, or combine certain components, or have different component arrangements.

[0041] The present invention also provides a computer-readable storage medium, which can be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium, wherein the computer-readable storage medium stores instructions that, when executed on a computer, cause the computer to perform the steps of the double-pulse test method.

[0042] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working process of the system, device, or unit described above can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0043] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0044] Finally, it should be noted that the above descriptions are merely preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A double-pulse testing method, characterized in that, The dual-pulse testing method includes the following steps: Acquire multi-source feature information of the device under test to generate device feature code; Dual-pulse testing is performed based on device signature codes, and time-frequency domain data of the device under test are acquired during the testing process. Oscillation feature identification is performed on time-frequency domain data to obtain oscillation feature data; The operation state of the double pulse test is adjusted based on the oscillation characteristic data until the preset suppression target is met or the preset repeated test threshold is reached, at which point the test is terminated to obtain the final test result.

2. The double-pulse testing method according to claim 1, characterized in that, The step of acquiring multi-source feature information of the device under test to generate device feature codes includes: The surface silkscreen information, electrical fingerprint data and thermal resistance distribution characteristics of the device under test are obtained by optical recognition, microcurrent scanning and infrared thermal imaging detection, respectively. The surface silkscreen information, electrical fingerprint data and thermal resistance distribution characteristics are weighted and fused to obtain the device feature code.

3. The double-pulse testing method according to claim 1, characterized in that, The dual-pulse test based on device signature, and the acquisition of time-frequency domain data of the device under test during the test, include: Using the device feature code as an index, historical parameter sets are retrieved from a pre-set cloud knowledge base; The historical parameter set is filtered based on the improved k-nearest neighbor algorithm to obtain the initial recommended parameters; The initial recommended parameters are interpolated and thresholded to obtain the initial test parameters; A dual-pulse test is performed on the device under test based on the initial test parameters, and the original voltage and current signals of the device under test are acquired synchronously during the test to obtain time-frequency domain data.

4. The double-pulse testing method according to claim 1, characterized in that, The process of identifying oscillation features from time-frequency domain data to obtain oscillation feature data includes: The time-frequency domain data is split into time-frequency images and time-series waveforms; The spectral local texture features are extracted from time-frequency images using a pre-defined CNN network model; A pre-defined LSTM network model is used to model the long-range time-series dependencies of the time-series waveform in order to predict the voltage overshoot amplitude. The local texture features of the spectrum and the voltage overshoot amplitude are integrated to obtain oscillation feature data.

5. The double-pulse testing method according to claim 1, characterized in that, The process of adjusting the operating state of the double-pulse test based on the oscillation characteristic data until a preset suppression target is met or a preset repetitive test threshold is reached, and then terminating the test to obtain the final test result, includes: Based on a preset reinforcement learning algorithm and oscillation feature data, the optimal combination of gate resistance, driving voltage and load inductance of the device under test is searched to obtain the optimal parameter combination. The optimal parameter combination is converted into test adjustment instructions, and the running state of the double pulse test is adjusted based on the test adjustment instructions; After adjusting the operating status, determine whether the oscillation suppression effect of the device under test meets the preset suppression target; If the preset suppression target is met, the double pulse test is terminated, and the running data during this test is collected as the final test result. If the preset suppression target is not met and the value is less than the preset repeated test threshold, the process returns to the step of performing the optimal combination optimization of the gate resistance, drive voltage and load inductance of the device under test based on the preset reinforcement learning algorithm and oscillation feature data until the preset suppression target is met, at which point the double pulse test is terminated, and the running data during this test is obtained as the final test result. If the preset suppression target is not met and the current number of repeated optimization attempts exceeds the preset repeated test threshold, the double pulse test will be terminated, and a test anomaly flag will be generated as the final test result.

6. A dual-pulse testing device, characterized in that, include: The acquisition module is used to acquire multi-source feature information of the device under test to generate device feature codes; The test module is used to perform dual-pulse testing based on device signature codes and to acquire time-frequency domain data of the device under test during the testing process. The identification module is used to identify oscillation characteristics in time-frequency domain data to obtain oscillation characteristic data; The iterative module is used to adjust the running state of the double pulse test based on the oscillation characteristic data until the preset suppression target is met or the preset repeated test threshold is reached, and then the test is terminated to obtain the final test result.

7. A dual-pulse testing device, characterized in that, The dual-pulse testing device includes: a memory and at least one processor, wherein the memory stores instructions; At least one of the processors invokes the instructions in the memory to cause the dual-pulse test apparatus to perform the steps of the dual-pulse test method as described in any one of claims 1-5.

8. A computer-readable storage medium storing instructions thereon, characterized in that, When the instructions are executed by the processor, they implement the steps of the double-pulse testing method as described in any one of claims 1-5.

Images