Multi-channel parallel power-on test and micro-leakage current detection method for circuit board

Abstract

The present application relates to the technical field of circuit board testing, more particularly to a circuit board multi-channel parallel power-on testing and micro leakage current detection method, and specifically comprises the following steps: S1, a to-be-tested circuit board enters a preset working state; S2, current values, temperature values and vibration values of each test channel are synchronously collected according to a set sampling period to obtain multi-modal data; S3, multi-modal features are extracted, and a dynamic baseline of each to-be-tested circuit board is automatically learned in an initial power-on stable stage of the to-be-tested circuit board; S4, a weighted probability model is established to calculate a leakage probability and perform twice determination; S5, a leakage pre-judgment model is constructed according to a leakage fault confirmation result, test parameters are optimized based on the leakage pre-judgment model, and maintenance is guided; S6, all test results are stored into a database, and parameters in the model are periodically iteratively optimized; the present application effectively reduces a misjudgment rate, improves maintenance positioning efficiency, does not need frequent manual parameter adjustment, and ensures that a scheme is long-term adapted to production requirements.

Description

Technical Field

[0001] This invention relates to the field of circuit board testing technology, and more specifically to a method for multi-channel parallel power-on testing and micro-leakage current detection of circuit boards. Background Technology

[0002] As electronic devices develop towards higher integration and higher reliability, such as in consumer electronics, industrial control, automotive electronics, and aerospace, circuit boards, as core components, are crucial. Leakage faults in circuit boards can directly affect the stability of equipment performance and even cause safety accidents such as overheating, short circuits, and fires. Therefore, leakage testing of circuit boards before they leave the factory has become a key quality control step in the production process.

[0003] Traditional circuit board leakage testing methods often focus on a single or few parameters, failing to cover the multi-physical field-related leakage characteristics of current, temperature, and vibration. They use a unified static standard as a fixed baseline, and the judgment logic is based on a single parameter exceeding the threshold. Without multi-parameter fusion and multiple verifications, they can only confirm leakage faults, but cannot trace the source of the fault or early signs, resulting in low repair efficiency.

[0004] Traditional solutions often focus on a single or few parameters, failing to cover the leakage characteristics associated with multiple physical fields such as current, temperature, and vibration. The lack of vibration data collection in traditional solutions makes creepage faults undetectable; some micro-leakages only manifest as fluctuations in the standard deviation of current, and monitoring only the absolute current will miss these faults, ultimately leading to a high false negative rate.

[0005] Traditional solutions rely on a uniform static standard, with all circuit boards of the same model sharing a fixed baseline. This does not take into account the individual differences between circuit boards under test. For example, different circuit boards under test may have natural differences in current fluctuation range and temperature rise rate during normal operation due to differences in component batches and soldering processes.

[0006] Traditional solutions often rely on a single parameter exceeding a threshold to determine leakage, without considering multi-parameter fusion or multiple verifications. For example, fluctuations in the power grid during testing can cause the current on the circuit board under test to temporarily exceed the threshold, which traditional solutions would directly classify as a fault. In the early stages of leakage, when the parameter only slightly exceeds the threshold, a single judgment may easily overlook it due to a loosely set threshold.

[0007] Traditional solutions can only confirm leakage faults, but cannot trace the source of the fault or its early signs. After confirming the leakage, it is necessary to manually check the circuits and components one by one, which cannot locate the cause of the fault, resulting in low maintenance efficiency.

[0008] Therefore, a method for multi-channel parallel power-on testing and micro-leakage current detection of circuit boards is needed to solve the above problems. Summary of the Invention

[0009] In order to overcome the above-mentioned defects of the prior art, the present invention provides a method for multi-channel parallel power-on testing and micro-leakage current detection of circuit boards, so as to solve the problems existing in the background art.

[0010] To achieve the above objectives, the present invention provides the following technical solution: a method for multi-channel parallel power-on testing and micro-leakage current detection of circuit boards, characterized by comprising the following steps:

[0011] S1. Install all the circuit boards under test into the test fixture, complete the system initialization and load the test configuration files of each circuit board under test, and then send a command to power on all test channels in parallel, so that the circuit boards under test can enter the preset working state.

[0012] S2. Under the premise that the circuit board under test is in the preset working state, the current value, temperature value and vibration value of each test channel are collected synchronously according to the set sampling period to obtain multimodal data;

[0013] S3. Based on the collected multimodal data, extract multimodal features and automatically learn the dynamic baseline of each circuit board under test during the initial power-on stabilization phase.

[0014] S4. Based on the multimodal characteristics and dynamic baseline, establish a weighted probability model to calculate the leakage probability and make two judgments based on the leakage probability, and output the leakage fault confirmation result.

[0015] The leakage probability is calculated using the following expression:

[0016]

[0017] Where P(t) is the leakage probability, ΔI is the current deviation, and σ I ΔT / Δt is the standard deviation of the current, ΔT / Δt is the rate of temperature rise, and V is the amplitude of a specific frequency band; w1, w2, and w3 are the weighting coefficients corresponding to the current characteristics, temperature rise characteristics, and vibration characteristics, respectively; F, G, and H are the normalization functions corresponding to the current characteristics, temperature rise characteristics, and vibration characteristics, respectively. With the dynamic baseline as the normal reference, the degree of abnormality of each characteristic is mapped to a probability between 0 and 1.

[0018] S5. Construct a leakage current prediction model based on the leakage current fault confirmation results, optimize test parameters based on the leakage current prediction model and guide maintenance.

[0019] S6. Store all test results in the database and periodically iterate and optimize the weight coefficients and normalization function in the weighted probability model.

[0020] The method for extracting multimodal features is as follows: extract the instantaneous current value I(t) from the current data of each sampling period, and obtain the instantaneous current deviation ΔI(t) by subtracting it from the rated current loaded by S1; calculate the standard deviation σ of ΔI(t) within the window using a dynamic short-time sliding window with a step size of 1 sampling. I Based on the temperature values ​​T(t) and T(t-Δt) of two consecutive adjacent sampling periods, the temperature difference ΔT is calculated and then divided by the time interval Δt to obtain the temperature rise rate ΔT / Δt. The original vibration data collected by S2 is segmented and a Hanning window is applied. After fast Fourier transform, it is converted into a frequency domain frequency amplitude spectrum. The amplitude is selected according to the specific frequency band preset in the S1 configuration file, and its effective value is calculated as the vibration characteristic value V of the sampling period.

[0021] The method of making two judgments based on the leakage probability is as follows: if the leakage probability exceeds the preset first-level threshold n times consecutively, a preliminary leakage alarm is triggered; after the preliminary alarm is triggered, the system will not immediately determine failure, but will start a high-frequency acquisition mode, continuously monitor for m seconds, and use a preset second-level threshold to verify whether it is caused by environmental factors. If the verification is successful, it is attributed to environmental noise; if the verification fails, it is confirmed as a leakage fault.

[0022] The method for constructing the leakage current prediction model is as follows: once a channel is finally confirmed to have leakage current, the system automatically starts data tracing analysis: retrieve all time series data of the channel from the start of power-on to the triggering of the initial alarm; use the time series anomaly detection algorithm to backtrack and find the earliest time point when abnormal signs appear; analyze which feature first appears abnormal at the abnormal time.

[0023] The method for optimizing test parameters and guiding maintenance based on the leakage current prediction model is as follows: For optimized test parameters: the circuit board to be tested is classified according to its model, production batch, and core component model. New circuit boards to be tested are automatically matched to the category. For the earliest abnormal feature with the highest proportion in the historical samples of this category, the initial judgment threshold is optimized to: dynamic baseline value × (1 + historical average abnormal amplitude × 0.8) and written into the parameter table. If the real-time value of this feature exceeds the threshold during the test, an early warning is triggered. If there is no leakage current, the default threshold is restored. After each batch of tests, the effect is verified based on the early interception rate and false alarm rate. If the target is not met, the threshold is reverted or new abnormal features are analyzed for iterative optimization.

[0024] For guided repair: The system first constructs a mapping table of "earliest abnormal feature - cause of failure" based on historical fault data and repair experience. After the circuit board under test is confirmed to be leaking, it matches the suspected cause of failure according to its earliest abnormal feature and generates a structured guide containing the suspected location and recommended tools. After repair, the actual cause is entered, and the prediction is compared with the actual result. If the accuracy is low, the mapping table is updated.

[0025] The method for iteratively optimizing the weight coefficients and normalization functions within the model is as follows: historical test data is extracted from the database to construct a training sample set containing input features and labels, where the input features are the degree of feature anomalies extracted during the testing process for each circuit board under test, and the labels are the true states labeled based on the actual test results; the optimization model is trained based on the training sample set, and the weight coefficients w1, w2, w3 and the normalization functions F, G, and H within the model are optimized iteratively, where w1, w2, w3 are the weight coefficients corresponding to the current feature, temperature rise feature, and vibration feature, respectively; and F, G, and H are the normalization functions corresponding to the current feature, temperature rise feature, and vibration feature, respectively.

[0026] The optimization method for the weight coefficients is as follows: For leakage cases in linear relationships, the anomaly level of historical samples is used as input and the true label is used as output to construct a logistic regression model; during model training, the optimal coefficients are solved by maximum likelihood estimation or gradient descent, and these coefficients are the optimized w1, w2, w3; for leakage cases in nonlinear relationships, a shallow neural network is constructed, the anomaly level of features is input into the network, and the leakage probability is output; the weights of the first layer of the network are w1, w2, w3; false positives and false negatives are assigned higher training weights than ordinary cases, forcing the model to focus on correcting these errors during optimization.

[0027] The optimization method of the normalization function is as follows: For current deviation and current standard deviation, the deviation distribution of normal samples and the deviation distribution of leakage samples in historical data are statistically analyzed. When the current deviation is less than the deviation distribution of normal samples, the normalization function is 0; when the current deviation is greater than the deviation distribution of leakage samples, the normalization function is 1. The intermediate interval is smoothly transitioned by linear interpolation or the Sigmoid function. The function increases monotonically with the degree of abnormality. For amplitude and temperature rise rate in a specific frequency band, kernel density estimation or decision tree regression is used to learn the mapping relationship between the degree of feature abnormality and the probability of leakage. The output curve of the tree is used as the optimized normalization function.

[0028] The technical effects and advantages of this invention are as follows:

[0029] 1. This invention synchronously collects multi-modal data of current, temperature, and vibration, covering multi-physical field characteristics related to leakage current. At the same time, it automatically learns the dynamic baseline of each circuit board under test during the initial stabilization stage, eliminating individual difference interference from the source. It can capture the current standard deviation fluctuation caused by micro leakage current and identify specific frequency band vibrations caused by leakage current, making leakage current feature identification more comprehensive, effectively reducing the false judgment rate, and greatly improving the test accuracy.

[0030] 2. This invention constructs a weighted probability model to calculate the leakage probability by integrating multiple features, and sets up a multiple verification mechanism, which can not only avoid misjudgment based on a single feature, but also filter out temporary noise interference, thereby further reducing the false alarm rate and still outputting accurate results stably in complex testing environments.

[0031] 3. After confirming leakage, this invention uses time-series anomaly detection to trace back the earliest abnormal time point and characteristics, identifying the earliest abnormal indicator. Based on this, subsequent similar circuit boards under test can use sensitive thresholds to achieve early interception based on key characteristics, shortening the average testing time; at the same time, it provides directional guidance for maintenance, such as current anomalies pointing to circuit insulation problems, and temperature anomalies pointing to component defects, improving maintenance location efficiency.

[0032] 4. This invention stores all test data in a database and periodically optimizes the model weights and normalization function through logistic regression or neural networks. As test cases accumulate, the model can autonomously adapt to the characteristics of new types of circuit boards under test and cope with environmental changes such as temperature and humidity fluctuations, so that the judgment accuracy gradually improves over time without the need for frequent manual parameter adjustments, ensuring that the solution is adaptable to production needs in the long term. Attached Figure Description

[0033] Figure 1 This is a flowchart of the method steps of the present invention. Detailed Implementation

[0034] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0035] Reference Figure 1 This invention provides a method for multi-channel parallel power-on testing and minute leakage current detection of circuit boards, specifically including the following steps:

[0036] S1. Install all the circuit boards under test into the test fixture, complete the system initialization and load the test configuration files of each circuit board under test, and then send a command to power on all test channels in parallel, so that the circuit boards under test can enter the preset working state.

[0037] S2. Under the premise that the circuit board under test is in the preset working state, the current value, temperature value and vibration value of each test channel are collected synchronously according to the set sampling period to obtain multimodal data;

[0038] S3. Based on the collected multimodal data, extract multimodal features and automatically learn the dynamic baseline of each circuit board under test during the initial power-on stabilization phase.

[0039] S4. Based on the multimodal characteristics and dynamic baseline, establish a weighted probability model to calculate the leakage probability and make a judgment, and output the leakage fault confirmation result.

[0040] S5. Construct a leakage current prediction model based on the leakage current fault confirmation results, optimize test parameters based on the leakage current prediction model and guide maintenance.

[0041] S6. Store all test results in the database and periodically iterate and optimize the parameters in the model.

[0042] In a more specific application of the present invention, S1, after installing all the circuit boards under test onto the test fixture, completing system initialization and loading the test configuration files of each circuit board under test, the main controller sends a command to power on all test channels in parallel, so that the circuit boards under test enter the preset working state.

[0043] In practice, the following steps are required: First, the positioning components of the test fixture must be replaced according to the model and size of the circuit board under test. The integrity of the probes and the contact pressure must be checked. Then, the circuit board under test is pushed in along the guide groove and fixed by the pressing mechanism. Subsequently, the installation quality is verified by comparing with the standard image to eliminate misalignment and omissions. Next, after the main controller is powered on, it sends self-test commands to the multimodal data acquisition card and power module hardware. The rated parameters are compared to confirm that the hardware is normal. Then, the test system operating system and control program are loaded, the data buffer and communication protocol are initialized, and a database connection is established. After that, the test configuration file in XML format is retrieved from the database through the identification of the circuit board under test. The integrity and logical rationality of the file parameters are verified. After parsing, the parameters are assigned to the corresponding modules. Finally, the main controller sends parallel power-on commands to each channel power module synchronously through the industrial Ethernet. The power modules provide power in a stepped voltage boost manner, monitor the current in real time and trigger overcurrent protection. After power-on preheating, the real-time electrical parameters and hardware status of the circuit board under test are read. Those that meet the preset requirements are marked as "test status". Abnormalities are investigated by power-off. After all circuit boards under test are ready, the system enters step S2.

[0044] S2. Under the premise that the circuit board under test is in the preset working state, the current value, temperature value and vibration value of each test channel are collected synchronously according to the set sampling period to obtain multimodal data.

[0045] The current value is acquired through a high-precision current sensor, such as a Hall current sensor or a shunt, connected in series in the power supply circuit of each circuit board under test. The sensor converts the real-time operating current into a corresponding analog voltage signal. This analog signal is then transmitted to the dedicated current acquisition channel of the multi-mode data acquisition card, and converted into a digital signal by an analog-to-digital converter with a resolution of 16 bits or higher. At the same time, the low-pass filter circuit built into the acquisition card filters out power grid fluctuations and electromagnetic interference noise, ensuring that the accuracy of the current data is controlled within ±0.1%FS. Furthermore, the current acquisition time of each channel is strictly aligned with the synchronous clock signal to avoid time offset.

[0046] Temperature data acquisition targets key heat-generating points on the circuit board under test, such as the surface of the CPU chip and the vicinity of the power management chip pins. These locations are preset in the S1 configuration file. Surface-mount thermocouples or NTC thermistors are used as sensing elements. Thermally conductive adhesive is used to tightly attach the sensor to the surface of the test point to ensure efficient heat conduction. The sensor converts temperature changes into weak resistance or voltage signals, which are then processed by the signal conditioning circuit and converted into digital temperature data by the ADC. The accuracy is controlled within ±0.5℃, and the temperature and current data share the same synchronous clock to ensure time consistency within the same sampling period.

[0047] Vibration data acquisition focuses on the structural vibration signals of the circuit board under test during operation. Near the casing of the circuit board under test or the pins of key components in each test channel, a piezoelectric accelerometer is fixed by magnetic attraction or adhesive. The sensor converts the vibration acceleration signal into a charge signal, which is then converted into a measurable voltage signal by a charge amplifier and transmitted to the vibration acquisition channel of the acquisition card. The built-in wideband filter circuit of the channel retains the effective vibration frequency band, which is then converted into a digital signal by an ADC. At the same time, the real-time calculation function of the acquisition module processes the raw vibration data into effective values ​​or spectral characteristic data of a specific frequency band. The start and stop times of vibration acquisition are matched with a synchronous clock to ensure synchronization with current and temperature data in the time dimension.

[0048] Ultimately, all the collected digital data on current, temperature, and vibration will be temporarily stored in the system's data cache, forming a multimodal dataset for each test channel, providing a foundation for feature extraction in the subsequent S3 step.

[0049] S3. Based on the collected multimodal data, extract multimodal features and automatically learn the dynamic baseline of each circuit board under test during the initial power-on stabilization phase.

[0050] The method for extracting multimodal features is as follows: For current features, firstly, the instantaneous current value I(t) is extracted from the current data of each sampling period. This instantaneous current value is then directly subtracted from the rated current applied by S1 to obtain the instantaneous current deviation ΔI(t). The deviation result retains the same accuracy as the original current data; a positive number represents a current higher than the rated value, and a negative number represents a current lower than the rated value, intuitively reflecting the degree to which the current deviates from the normal reference. Based on this, to capture current fluctuation characteristics, the system uses a dynamic short-time sliding window. The average value of all ΔI(t) within the window is first calculated, then the squared difference between each ΔI(t) and the average value is summed and averaged. Finally, the square root is taken to obtain the current standard deviation σ within the window. I The window is updated in real time with a step size of 1 sample to ensure that it can dynamically track subtle changes in current fluctuations and promptly identify abnormal fluctuations caused by micro-leakage.

[0051] For temperature characteristics, the temperature rise rate ΔT / Δt is calculated based on the temperature data of two consecutive adjacent sampling periods: first, the temperature value T(t) of the current sampling period and the temperature value T(t-Δt) of the previous sampling period are obtained, and the difference between the two is calculated as ΔT=T(t)-T(t-Δt). Then, it is divided by the time interval Δt to obtain the temperature rise rate per unit time. To avoid interference from single-point temperature fluctuations on the rate calculation, the system performs a moving average filter on the temperature rise rate results of five consecutive sampling periods to filter out instantaneous noise, and finally outputs a smoothed ΔT / Δt, with the unit uniformly set to ℃ / s. If a sudden drop in temperature occurs, it is marked as abnormal data and removed during subsequent baseline learning.

[0052] For vibration characteristics, the system first segments the raw vibration data collected by S2, taking 512 sampling points for each segment, and applying a Hanning window to each segment to reduce spectral leakage. Then, the time-domain vibration signal is converted into a frequency-amplitude spectrum in the frequency domain through a fast Fourier transform. According to the specific frequency band preset in the S1 configuration file, the amplitude corresponding to all frequency points in the spectrum is selected, and the effective values ​​of these amplitudes are calculated as the vibration characteristic value V of the sampling period. If there is no effective amplitude in the frequency band, it is assigned a value of 0 to ensure the continuity and effectiveness of the characteristic value.

[0053] In the dynamic baseline learning phase, the system first sets the first x seconds after the circuit board under test (PCB) is initially powered on as the stabilization phase. This duration is verified through historical test data to ensure that most PCBs under test complete power-on preheating and their parameters tend to stabilize during this phase. During this phase, the system continuously extracts the four types of feature data for each PCB under test. For each type of feature, the 3σ principle is first used to remove outliers in the stabilization phase, and then the arithmetic mean of the remaining valid data is calculated as the dynamic baseline value of the PCB under test, namely the current deviation baseline, current standard deviation baseline, temperature rise rate baseline, and vibration characteristic baseline. The baseline data of each PCB under test is stored separately and bound to its physical identity to avoid interference from individual differences between different PCBs under test. In subsequent steps, the anomaly determination of all feature data is based on the dynamic baseline of the PCB under test itself, realizing personalized anomaly identification.

[0054] S4. Based on the multimodal characteristics and dynamic baseline, establish a weighted probability model to calculate the leakage probability and make a judgment, and output the leakage fault confirmation result.

[0055] The probability of leakage current is calculated by establishing a weighted probability model to comprehensively determine the probability of leakage current, using the expression:

[0056]

[0057] Where P(t) is the leakage probability, ΔI is the current deviation, and σ I ΔT / Δt is the current standard deviation, ΔT / Δt is the temperature rise rate, V(t) is the amplitude of a specific frequency band; F, G, H are normalization functions that map the degree of abnormality of each feature to a probability between 0 and 1, with the dynamic baseline as the normal reference; w1, w2, w3 are the weighting coefficients set for them.

[0058] The determination method is as follows: if the leakage probability exceeds the preset first-level threshold n times consecutively, a preliminary leakage alarm is triggered; after the preliminary alarm is triggered, the system will not immediately determine failure, but will start a high-frequency acquisition mode, continuously monitor for m seconds, and verify using a preset second-level threshold. If the verification passes, it is attributed to environmental noise; if the verification fails, it is confirmed as a leakage fault.

[0059] By constructing a weighted probability model and calculating the leakage probability by integrating multiple features, and setting up a multiple verification mechanism, we can avoid misjudgment based on a single feature and filter out temporary noise interference, thereby further reducing the false alarm rate and ensuring stable output of accurate results even in complex testing environments.

[0060] S5. Construct a leakage current prediction model based on the leakage current fault confirmation results, optimize test parameters based on the leakage current prediction model, and guide maintenance.

[0061] The method for constructing the leakage current prediction model is as follows: once a channel is finally confirmed to have leakage current, the system automatically starts data tracing analysis: retrieve all time series data of the channel from the start of power-on to the triggering of the initial alarm; use the time series anomaly detection algorithm to backtrack and find the earliest time point when abnormal signs appear; analyze which feature first appears abnormal at the abnormal moment.

[0062] When optimizing test parameters based on the leakage current prediction model, the system first categorizes the circuit boards under test according to their model, production batch, and core component model. When testing a new circuit board, the system automatically matches the corresponding category. If the historical sample size of the category in the prediction model training library is ≥20, optimization is enabled; otherwise, the default parameters are temporarily used. Subsequently, for the earliest abnormal feature with the highest proportion in the historical samples of the category, the judgment threshold is adjusted in the early stage of testing. The optimization logic is to reduce the default threshold to "dynamic baseline value × (1 + historical average abnormal amplitude × 0.8)" and write the optimized threshold into the feature judgment parameter table. When the real-time value of the feature exceeds the threshold during testing, an early alarm is triggered and the sampling rate is increased to 500Hz to achieve early interception. If there is no leakage current, the default threshold is restored. After each batch of tests, the early interception rate and false alarm rate are statistically analyzed. If the early interception rate is ≥40% and the false alarm rate is ≤3%, the parameters are retained. If the false alarm rate exceeds 3%, the threshold is reverted. If the interception rate is below 20%, it is analyzed whether new abnormal features have appeared and the adjustment basis is updated, forming a closed loop of classification matching, parameter adjustment, effect verification, and iterative optimization.

[0063] When using a leakage current prediction model to guide repairs, a mapping table of "earliest abnormal feature - fault cause" is first constructed based on historical fault data and repair experience to identify typical faults corresponding to the earliest abnormal features. When a circuit board under test is confirmed to have a leakage current, the system matches the suspected fault cause from the mapping table based on its earliest abnormal feature, automatically generating a structured guide containing the suspected location, recommended testing tools, operation steps, and repair suggestions, which is output through pop-up windows on the operation interface and mobile terminals of repair personnel. After the repair is completed, the repair personnel enter the actual fault cause, and the system compares the prediction with the actual result. If the prediction accuracy is lower than the threshold, the mapping table is updated to continuously optimize the guidance accuracy, achieving a precise connection from feature anomaly to fault location and improving repair efficiency.

[0064] S6. Store all test results in the database and periodically iterate and optimize the parameters in the model.

[0065] The method for iteratively optimizing the parameters within the model is as follows: historical test data is extracted from the database to construct a training sample set containing input features and labels, where the input features are the degree of feature anomalies extracted during the testing process for each circuit board under test, and the labels are the true states labeled based on the actual test results; the optimization model is trained based on the training sample set, and the weight coefficients and normalization functions within the model are optimized iteratively.

[0066] The optimization method for the weight coefficients is as follows: For leakage cases in linear relationships, the anomaly level of historical samples is used as input and the true label is used as output to construct a logistic regression model; during model training, the optimal coefficients are solved by maximum likelihood estimation or gradient descent, and these coefficients are the optimized w1, w2, w3; for leakage cases in nonlinear relationships, a shallow neural network is constructed, the anomaly level of features is input into the network, and the leakage probability is output; the weights of the first layer of the network are w1, w2, w3; false positives and false negatives are assigned higher training weights than ordinary cases, forcing the model to focus on correcting these errors during optimization.

[0067] The optimization method for the normalization function is as follows: For current deviation and current standard deviation, the deviation distributions of normal samples and leakage samples in historical data are statistically analyzed. When the current deviation is less than the deviation distribution of normal samples, the normalization function is 0; when the current deviation is greater than the deviation distribution of leakage samples, the normalization function is 1. The intermediate interval is smoothly transitioned through linear interpolation or the Sigmoid function. The function monotonically increases with the degree of anomaly. For amplitude and temperature rise rate in specific frequency bands, kernel density estimation or decision tree regression is used to learn the mapping relationship between the degree of feature anomaly and the probability of leakage. The output curve of the tree is used as the optimized normalization function.

[0068] Through the above description of the embodiments, those skilled in the art can clearly understand that the various embodiments of this application can be implemented by means of software or software combined with necessary general-purpose hardware platforms, and of course, they can also be implemented by hardware functions; based on this understanding, the technical solution of this application, in essence or the part that contributes to the prior art, can be embodied in the form of a software product, which is stored in a storage medium and includes several instructions to cause a computer device, such as including but not limited to a personal computer, server, or network device, to execute all or part of the steps of the method described in any embodiment of this application.

[0069] The foregoing describes exemplary embodiments of this application. It should be understood that the above exemplary embodiments are not restrictive but illustrative, and the scope of protection of this application is not limited thereto. It should be understood that those skilled in the art can make modifications and variations to the embodiments of this application without departing from the spirit and scope of this application, and such modifications and variations should be within the scope of protection of this application.

Claims

1. A method for multi-channel parallel power-on test and micro-current detection of a circuit board, characterized in that, Specifically comprising the following steps: S1, install all the test circuit boards to the test fixture, complete the system initialization and load the test configuration file of each test circuit board, and then the main controller sends a command to make all test channels parallel power on, so that the test circuit boards enter the preset working state; S2, under the premise that the test circuit boards are in the preset working state, the current value, temperature value and vibration value of each test channel are synchronously collected at a set sampling period to obtain multi-modal data; S3, based on the collected multi-modal data, multi-modal features are extracted, and the dynamic baseline of each test circuit board is automatically learned in the initial power-on stable stage of the test circuit board; S4, according to the multi-modal features and the dynamic baseline, a weighted probability model is established to calculate the electric leakage probability and make two determinations according to the electric leakage probability, and an electric leakage fault confirmation result is output; The calculation method of the electric leakage probability is: through the expression: ； where P(t) is the leakage probability, ΔI is the current deviation, σ I is the current standard deviation, ΔT / Δt is the temperature rise rate, V is the amplitude of a specific frequency band; w1, w2, and w3 are weight coefficients corresponding to the current feature, the temperature rise feature, and the vibration feature, respectively; F, G, and H are normalization functions corresponding to the current feature, the temperature rise feature, and the vibration feature, respectively, taking the dynamic baseline as the normal baseline, and mapping the abnormal degree of each feature to a probability between 0 and 1; S5, constructing an electric leakage pre-judgment model according to the electric leakage fault confirmation result, optimizing test parameters based on the electric leakage pre-judgment model and guiding maintenance; S6, storing all test results in a database, and periodically iterating and optimizing the weight coefficients and normalization functions in the weighted probability model.

2. The method of claim 1, wherein the method further comprises: The mode of extracting the multi-modal features is: extracting the instantaneous current value I(t) from each sampling period current data, and obtaining the instantaneous current deviation ΔI(t) by subtracting the rated current of S1; calculating the standard deviation σ of ΔI(t) in the window by a dynamic short-time sliding window with 1 sampling as a step I ; calculating the temperature difference ΔT based on the temperature values T(t) and T(t-Δt) of two adjacent sampling periods in succession, and then dividing the temperature difference ΔT by the time interval Δt to obtain the temperature rise rate ΔT / Δt; segmenting the original vibration data collected by S2 and applying a Hanning window, converting into a frequency domain frequency amplitude spectrum by fast Fourier transform, screening the amplitude according to the specific frequency band preset by the S1 configuration file, and calculating the effective value as the vibration characteristic value V of the sampling period.

3. The method of claim 1, wherein the method further comprises: The way of making two determinations according to the electric leakage probability is: if the electric leakage probability continuously exceeds the preset first-level threshold for n times, a preliminary electric leakage alarm is triggered; After triggering the preliminary alarm, the system does not immediately determine failure, but starts a high-frequency collection mode, continuously monitors for m seconds, and uses a preset second-level threshold to verify whether it is caused by environmental factors, if the verification is passed, it is attributed to environmental noise; if the verification fails, it is confirmed as an electric leakage fault.

4. The method of claim 1, wherein the method further comprises: applying a voltage to the plurality of channels of the circuit board; and measuring the current flowing through the plurality of channels of the circuit board. The way of constructing the electric leakage pre-judgment model is: once a channel is finally confirmed to have electric leakage, the system automatically starts data trace analysis: all time series data of the channel from power-on to triggering the preliminary alarm are retrieved; Using time series anomaly detection algorithm, the earliest abnormal sign time point is found back; analyze which feature is the first to appear abnormal at the abnormal time.

5. The method for multi-channel parallel power-on testing and micro-leakage current detection of circuit boards according to claim 1, characterized in that: The way of optimizing test parameters and guiding maintenance based on the electric leakage pre-judgment model is: for optimizing test parameters: according to the type, production batch and core component type of the test circuit board, classify, and automatically match the category of the new test circuit board; for the most abnormal feature with the highest proportion in the historical samples of the category, the test initial determination threshold is optimized to: dynamic baseline value × (1+ historical abnormal amplitude average × 0.8), and written into the parameter table; if the real-time value of the feature exceeds the threshold during the test, a warning is triggered, and if there is no electric leakage, the default threshold is restored, and the effect is verified according to the early interception rate and false alarm rate after each batch of test, and if it does not meet the standard, the threshold is adjusted back or new abnormal features are analyzed for iterative optimization; For guiding maintenance: the system first constructs a "first abnormal feature-fault reason" mapping table according to historical fault data and maintenance experience, after the test circuit board is confirmed to have electric leakage, the suspected fault reason is matched according to the first abnormal feature and a structured guide containing the suspected position and recommended tools is generated, the actual cause is recorded after maintenance, and the predicted and actual results are compared, and if the accuracy is low, the mapping table is updated.

6. The method of claim 1, wherein: The manner of iteratively optimizing the weight coefficients in the model and the normalization function is: extracting historical test data from the database, constructing a training sample set containing input features and labels, wherein the input features are the feature abnormality degrees of each circuit board extracted during the test process, and the labels are the real states labeled based on the actual test results; According to the training sample set, the optimization model is trained, and the weight coefficients w1, w2, and w3 in the model and the normalization functions F, G, and H are iteratively optimized, wherein w1, w2, and w3 are the weight coefficients corresponding to the current features, temperature rise features, and vibration features, respectively; F, G, and H are the normalization functions corresponding to the current features, temperature rise features, and vibration features, respectively.

7. The method of claim 6, wherein the method further comprises: applying a voltage to the plurality of channels of the circuit board; and measuring the current flowing through the plurality of channels of the circuit board. The optimization manner of the weight coefficients is: for the linear relationship of the leakage, the feature abnormality degree of the historical sample is taken as the input, and the real label is taken as the output, and a logistic regression model is constructed; during the model training process, the optimal coefficients are solved by maximum likelihood estimation or gradient descent method, and these coefficients are the optimized w1, w2, and w3; for the nonlinear relationship of the leakage, a shallow neural network is constructed, the feature abnormality degree is input into the network, and the leakage probability is output; the first layer weight of the network is w1, w2, and w3; the false positive and false negative cases are given higher training weights than ordinary cases, and the model is forced to focus on correcting these errors during optimization.

8. The method of claim 6, wherein the method further comprises: applying a voltage to the plurality of channels of the circuit board; and measuring the current flowing through the plurality of channels of the circuit board. The optimization manner of the normalization function is: for the current deviation and current standard deviation, the deviation distribution of normal samples and the deviation distribution of leakage samples in historical data are counted, when the current deviation is less than the deviation distribution of normal samples, the normalization function is 0, when the current deviation is greater than the deviation distribution of leakage samples, the normalization function is 1, the intermediate interval is smoothly transitioned by linear interpolation or Sigmoid function, and the function monotonically increases with the abnormality degree; for the amplitude of a specific frequency band and the temperature rise rate, kernel density estimation or decision tree regression is used to learn the mapping relationship between the feature abnormality degree and the leakage probability, and the output curve of the tree is taken as the optimized normalization function.

Images