Device and method for testing comprehensive performance of logic board

Abstract

The invention discloses a logic board comprehensive performance testing device and testing method, and relates to the technical field of electronic equipment testing. A time sequence signal, a power consumption curve and infrared temperature data are synchronously acquired through a logic board interface, a time-space aligned fusion data set is generated by using a dynamic time warping algorithm, a power consumption coupling analysis model is input, a dynamic incidence matrix of a signal integrity parameter and power consumption fluctuation is acquired, and key coupling characteristics are extracted through singular value decomposition. And constructing an adaptive neighborhood density clustering model according to the correlation mode parameters and the temperature gradient distribution, generating an abnormal region probability graph, and marking electrothermal coupling abnormal coordinates and confidence. And inputting the fusion data, the correlation parameters and the abnormal probability graph into a graph neural network, taking a time sequence signal spectrum entropy, a local temperature mean value and a power consumption fluctuation variance as node features, taking an electrothermal coupling coefficient as an edge weight, iteratively updating a node state through a graph attention mechanism, and outputting a defect type and a three-dimensional coordinate. And the defect detection accuracy is effectively improved.

Description

Technical Field

[0001] The present invention relates to the technical field of electronic device testing, and specifically provides a comprehensive performance testing device and method for a logic board. Background Art

[0002] With the accelerating advancement of the intelligent and digital transformation of electronic devices, the logic board, as the core control unit, has been widely used in fields such as new energy vehicles, industrial automation, communication devices, and consumer electronics. Driven by technologies such as 5G communication and edge computing, the signal frequency and complexity that the logic board needs to process have increased significantly, and at the same time, it faces challenges in harsh working conditions such as high and low temperatures, vibration, and shock. Its performance stability is directly related to the reliability of the device and the safety of the system. Especially in scenarios such as automotive-grade electronics and high-speed communication modules, the comprehensive performance testing of the logic board has become the core link of product quality control.

[0003] Currently, the testing of logic boards mainly relies on single means such as signal integrity analysis, power consumption testing, and thermal imaging detection. These methods often have limitations when detecting local anomalies. For example, analyzing timing signals alone cannot reveal the root cause of power consumption fluctuations, and it is difficult to distinguish electrothermal coupling from environmental noise only relying on thermal imaging. In addition, some methods use fixed thresholds to judge anomalies and are difficult to adapt to dynamic changes under different load conditions. At the same time, traditional fault location methods rely on rule setting or empirical judgment and fail to effectively combine multi-modal data for intelligent analysis, resulting in insufficient defect detection accuracy and difficulty in meeting the testing requirements of high-complexity logic boards. Summary of the Invention

[0004] Aiming at the deficiencies of the prior art, the present invention provides a comprehensive performance testing device and method for a logic board, which solves the problems in the above background art.

[0005] To achieve the above objectives, the present invention is implemented through the following technical solutions: A comprehensive performance testing method for a logic board, comprising the following steps: S1. Synchronously collect the running timing signal, dynamic power consumption curve, and infrared temperature distribution data through the logic board interface, eliminate the minute timing offsets between multiple sensors through the dynamic time warping algorithm, and generate a spatio-temporally aligned fusion data set; S2. Input the fusion data set into a pre-set power consumption coupling analysis model, obtain the dynamic correlation matrix of signal integrity parameters and power consumption fluctuations, extract key coupling features through singular value decomposition, and output the correlation mode parameters of signal ringing and instantaneous overcurrent; S3. According to the correlation mode parameters and temperature gradient distribution, construct a density clustering model with an adaptive neighborhood radius, generate an abnormal area probability map, and mark the electrothermal coupling abnormal coordinates and confidence levels; S4. Input the fusion data set, correlation mode parameters, and abnormal area probability map into a graph neural network model, use the spectral entropy of the timing signal, local temperature mean, and power consumption fluctuation variance as node features, and the electrothermal coupling coefficient as the edge weight, iteratively update the node states through the graph attention mechanism, and output the defect type and three-dimensional coordinates.

[0006] Further, the specific process of eliminating the minute timing offsets between multiple sensors through the dynamic time warping algorithm and generating a spatio-temporally aligned fusion data set is as follows: Perform multi-channel timestamp alignment on the timing signal, dynamic power consumption curve, and infrared temperature distribution data, based on the dynamic window length adjustment strategy, obtain the minimum bending path of the multi-source data sequence, and map the data of each sensor to a unified time reference through cubic spline interpolation to generate a spatio-temporally aligned fusion data set.

[0007] Further, the specific process of inputting the fusion data set into a pre-set power consumption coupling analysis model and obtaining the dynamic correlation matrix of signal integrity parameters and power consumption fluctuations is as follows: Perform sliding window segmentation on the signal integrity parameters and power consumption fluctuation data, obtain the exponential decay integral of the power consumption change rate and the signal time-domain waveform within each window, and construct a dynamic correlation matrix to characterize the coupling relationship between signal distortion and transient power consumption.

[0008] Further, the specific process of extracting key coupling features through singular value decomposition and outputting the correlation mode parameters of signal ringing and instantaneous overcurrent is as follows: Perform singular value decomposition on the dynamic correlation matrix, select the time-domain eigenvector corresponding to the principal component in the left singular vector and the frequency-domain eigenvector in the right singular vector, and map them to a pre-set ringing-overcurrent mode space to generate correlation mode parameters.

[0009] Further, the specific process of constructing a density clustering model with an adaptive neighborhood radius according to the correlation mode parameters and temperature gradient distribution is as follows: Dynamically obtain the local neighborhood radius according to the signal distortion rate and temperature gradient distribution in the correlation mode parameters, and divide the high-density abnormal areas through an improved density clustering algorithm to generate candidate defect clusters.

[0010] Further, the specific process of generating the abnormal area probability map and marking the electro-thermal coupling abnormal coordinates and confidence levels is as follows: Based on the electro-thermal coupling strength and the consistency of the temperature gradient direction of the data points within the candidate defect clusters, the abnormal confidence level is obtained, and a probability map is generated and the center coordinates of the abnormal area are marked.

[0011] Further, the process with the spectral entropy of the time-series signal, the local temperature mean, and the variance of the power consumption fluctuation as node features and the electro-thermal coupling coefficient as the edge weight is as follows: The logical board area covered by the abnormal area probability map is divided into grids, and the spectral entropy value, the local temperature mean, and the variance of the power consumption fluctuation of the time-series signal within each grid cell are extracted as node features. According to the electro-thermal coupling coefficient in the correlation pattern parameters, the edge weights between adjacent nodes are dynamically calculated, and a topological graph structure representing the electro-thermal coupling relationship is constructed.

[0012] Further, the specific process of iteratively updating the node states through the graph attention mechanism and outputting the defect type and three-dimensional coordinates is as follows: A multi-layer graph attention network is deployed on the topological graph structure. By aggregating the electro-thermal features of the neighborhood nodes through the attention weights, after iteratively updating the node state vectors, through the combined output of the classifier and the regressor, the node states are mapped to the preset defect type space and the physical coordinate space of the logical board, generating the defect type labels and three-dimensional coordinate information.

[0013] A comprehensive performance testing device for a logic board includes the following modules: a multi-source synchronous fusion module, a parameter coupling analysis module, an electro-thermal anomaly clustering module, and a graph network defect determination module; the multi-source synchronous fusion module is used to synchronously collect the running time-series signal, the dynamic power consumption curve, and the infrared temperature distribution data through the logic board interface, eliminate the minute time-series offset between multiple sensors through the dynamic time warping algorithm, and generate a spatio-temporally aligned fusion data set; the fusion data set is input into a preset power consumption coupling analysis model to obtain the dynamic correlation matrix of the signal integrity parameters and the power consumption fluctuation, extract the key coupling features through singular value decomposition, and output the correlation pattern parameters of the signal ringing and the instantaneous overcurrent; the parameter coupling analysis module is used to construct a density clustering model with an adaptive neighborhood radius according to the correlation pattern parameters and the temperature gradient distribution, generate an abnormal area probability map, and mark the electro-thermal coupling abnormal coordinates and confidence levels; the electro-thermal anomaly clustering module is used to input the fusion data set, the correlation pattern parameters, and the abnormal area probability map into a graph neural network model, with the spectral entropy of the time-series signal, the local temperature mean, and the variance of the power consumption fluctuation as node features and the electro-thermal coupling coefficient as the edge weight, iteratively update the node states through the graph attention mechanism, and output the defect type and three-dimensional coordinates.

[0014] The present invention has the following beneficial effects:

[0015] (1) A comprehensive performance testing method for a logic board eliminates the time - sequence offset of multi - sensors through the dynamic time warping algorithm, realizes the spatio - temporal alignment of time - sequence signals, power consumption, and temperature data, and provides a high - consistency data basis for subsequent analysis. Based on the dynamic correlation matrix and singular value decomposition technology, it reveals the hidden correlations between signal ringing, instantaneous over - current, and power - consumption fluctuations, and improves the transient anomaly detection ability. By combining the temperature gradient and the signal distortion rate, it dynamically adjusts the clustering neighborhood radius, enhancing the robustness of anomaly - area recognition under complex working conditions. Through the graph neural network to fuse electrical, thermal, and time - sequence features, it realizes the accurate classification and three - dimensional positioning of hidden defects such as micro - solder joint failures, reducing the dependence on manual re - inspection.

[0016] (2) A comprehensive performance testing device for a logic board. The multi - source synchronous fusion module, parameter coupling analysis module, electro - thermal anomaly clustering module, and graph - network defect determination module form a closed - loop testing link, which is compatible with the interfaces of mainstream testing equipment. An integrated programmable load generator is supported to simulate micro - second - level impedance mutations, meeting the transient response test requirements of vehicle - grade products. Using a distributed task scheduling mechanism, multiple logic boards can be tested in parallel at one time, significantly improving the throughput efficiency in mass - production scenarios. Based on the electro - thermal coupling anomaly probability map and the output results of the graph neural network, it automatically generates a defect - location report, shortening the fault - analysis cycle.

[0017] Of course, it is not necessary for any product implementing the present invention to achieve all the above - mentioned advantages simultaneously. Brief Description of the Drawings

[0018] Figure 1 It is a flowchart of a comprehensive performance testing method for a logic board of the present invention.

[0019] Figure 2 It is a flowchart of a comprehensive performance testing device for a logic board of the present invention. Detailed Embodiments

[0020] In the embodiments of the present application, through a comprehensive performance testing device and method for a logic board, problems such as asynchronous analysis of multiple parameters, missed detection of transient anomalies, insufficient adaptability to complex working conditions, and low - efficiency positioning of hidden defects in traditional testing are solved. Through the multi - source synchronous fusion module, accurate alignment of time - sequence, power - consumption, and temperature data is realized. Combining the dynamic correlation matrix and singular value decomposition technology to capture hidden coupling relationships, dynamically adjusting clustering parameters, and through the graph neural network to fuse multi - modal features, automatic classification and spatial positioning of micron - level solder joint failures are realized, significantly improving the test coverage rate, parallel efficiency, and dynamic load accuracy.

[0021] The general idea of the solution in the embodiments of the present application is as follows:

[0022] Synchronously collect the operation timing signals, dynamic power consumption curves, and infrared temperature distribution data through the logic board interface, and eliminate the minute timing offsets between multiple sensors through the dynamic time warping algorithm to generate a spatio-temporally aligned fusion data set.

[0023] Input the fusion data set into a pre-set power consumption coupling analysis model, obtain the dynamic correlation matrix of signal integrity parameters and power consumption fluctuations, extract key coupling features through singular value decomposition, and output the correlation mode parameters of signal ringing and instantaneous overcurrent.

[0024] Construct a density clustering model with an adaptive neighborhood radius based on the correlation mode parameters and temperature gradient distribution, generate an abnormal area probability map, and mark the electro-thermal coupling abnormal coordinates and confidence levels.

[0025] Input the fusion data set, correlation mode parameters, and abnormal area probability map into a graph neural network model. Use the spectral entropy of the timing signal, local temperature mean, and variance of power consumption fluctuations as node features, and the electro-thermal coupling coefficient as the edge weight. Iteratively update the node states through the graph attention mechanism and output the defect type and three-dimensional coordinates.

[0026] Please refer to Figure 1 , an embodiment of the present invention provides a technical solution: The present invention is realized through the following technical solutions: A comprehensive performance test method for a logic board, including the following steps: S1. Synchronously collect the operation timing signals, dynamic power consumption curves, and infrared temperature distribution data through the logic board interface, and eliminate the minute timing offsets between multiple sensors through the dynamic time warping algorithm to generate a spatio-temporally aligned fusion data set; S2. Input the fusion data set into a pre-set power consumption coupling analysis model, obtain the dynamic correlation matrix of signal integrity parameters and power consumption fluctuations, extract key coupling features through singular value decomposition, and output the correlation mode parameters of signal ringing and instantaneous overcurrent; S3. Construct a density clustering model with an adaptive neighborhood radius based on the correlation mode parameters and temperature gradient distribution, generate an abnormal area probability map, and mark the electro-thermal coupling abnormal coordinates and confidence levels; S4. Input the fusion data set, correlation mode parameters, and abnormal area probability map into a graph neural network model. Use the spectral entropy of the timing signal, local temperature mean, and variance of power consumption fluctuations as node features, and the electro-thermal coupling coefficient as the edge weight. Iteratively update the node states through the graph attention mechanism and output the defect type and three-dimensional coordinates.

[0027] In this implementation scheme, step S1: Synchronously collect the operation timing signal, dynamic power consumption curve, and infrared temperature distribution data through the logic board interface to ensure the time consistency of multi-source data. The dynamic time warping (DTW) algorithm is used to eliminate the minute timing offset between multiple sensors and generate a spatio-temporally aligned fusion data set. Dynamic time warping (DTW): A time series alignment algorithm that eliminates the time deviation of data collection from different sensors by calculating the minimum bending path; Spatio-temporal alignment: Refers to the consistency of multi-source data in the time and space dimensions, facilitating subsequent comprehensive analysis. Step S2: Analyze the dynamic correlation between signal integrity parameters and power consumption fluctuations, and extract key coupling features. Input the fusion data set into a pre-set power consumption coupling analysis model, calculate the dynamic correlation matrix, and extract key coupling features through singular value decomposition (SVD), and output the correlation mode parameters of signal ringing and instantaneous overcurrent. Dynamic correlation matrix: A mathematical matrix representing the non-linear relationship between signal integrity parameters and power consumption fluctuations; Singular value decomposition (SVD): A matrix decomposition method used to extract the principal component features in data; Signal ringing: Refers to the oscillation phenomenon generated when the signal changes rapidly, usually caused by impedance mismatch; Instantaneous overcurrent: Refers to the phenomenon that the current exceeds the rated value within a short time. Step S3: Locate the abnormal area on the logic board according to the correlation characteristics between signal distortion and temperature distribution. Based on the correlation mode parameters and temperature gradient distribution, construct a density clustering model with an adaptive neighborhood radius, generate an abnormal area probability map, and mark the electro-thermal coupling abnormal coordinates and confidence levels. Adaptive neighborhood radius: Dynamically adjusts the neighborhood range of the clustering algorithm according to the data distribution, improving the adaptability under complex working conditions; Density clustering model: A clustering algorithm based on data density used to divide high-density abnormal areas; Abnormal area probability map: Represents the distribution of abnormal areas on the logic board in the form of probability. Step S4: Achieve precise classification and spatial positioning of defects by fusing multi-modal data; Implementation method: Input the fusion data set, correlation mode parameters, and abnormal area probability map into a graph neural network model. Using the spectral entropy of the timing signal, local temperature mean, and variance of power consumption fluctuations as node features, and the electro-thermal coupling coefficient as the edge weight, iteratively update the node state through the graph attention mechanism (GAT), and output the defect type and three-dimensional coordinates. Graph neural network (GNN): A deep learning model for processing graph-structured data, suitable for multi-modal data fusion; Graph attention mechanism (GAT): Aggregates neighborhood node information through attention weights, improving the model's focusing ability on key features; Spectral entropy: An index representing the complexity of the signal frequency domain, used to describe the spectral characteristics of the timing signal; Electro-thermal coupling coefficient: A parameter representing the correlation strength between the electrical signal and the temperature distribution.

[0028] Specifically, the specific process of eliminating the minute temporal offset between multiple sensors through the dynamic time warping algorithm and generating a spatio-temporally aligned fusion dataset is as follows: Perform multi-channel timestamp alignment on the temporal signal, dynamic power consumption curve, and infrared temperature distribution data. Based on the dynamic window length adjustment strategy, obtain the minimum bending path of the multi-source data sequence, and map the data of each sensor to a unified time reference through cubic spline interpolation to generate a spatio-temporally aligned fusion dataset.

[0029] In this implementation, for multi-channel timestamp alignment, input data: Let S1(t), S2(t), and S3(t) represent the logical board operation temporal signal, dynamic power consumption curve, and infrared temperature distribution data respectively, and their timestamp sets be T1, T2, and T3. Since the sampling rates of different sensors are different, it may lead to timestamp misalignment, so time alignment is required. Objective: Establish a unified time reference T ref , so that all data is aligned on this time axis. For the dynamic window length adjustment strategy, let d(i,j) represent the Euclidean distance metric between S1(t) and S2(t): where i and j are the indices of the data points in S1 and S2 respectively. Calculate the cumulative distance matrix D(i,j): D(i,j) = d(i,j) + min{D(i - 1,j), D(i,j - 1), D(i - 1,j - 1)}; where: D(i - 1,j) represents stretching S1 along the time axis; D(i,j - 1) represents stretching S2 along the time axis; D(i - 1,j - 1) represents aligning the two signals. By finding the minimum bending path W = {(i1,j1),(i2,j2),…,(i L ,j L )} of the cumulative distance matrix D(i,j), find the best matching relationship of the data points. Mapping to a unified time reference through spline interpolation, due to different sensor sampling frequencies, even after DTW alignment, the data points may still be unevenly distributed. Therefore, cubic spline interpolation is used to map all data points to a unified time reference Tref. Let the interpolated signal be Then: where: a g , b g , c g , d g are the coefficients to be solved; t g ,, t g+1 are adjacent time nodes. By constraining the continuity and boundary conditions of the interpolation curve, solve the interpolation coefficients to keep the data smoothly distributed on the unified time axis. Generating a spatio-temporally aligned fusion dataset Finally, all sensor data and are aligned on Tref to form a fusion dataset: This fused dataset is used for subsequent power consumption coupling analysis and anomaly detection.

[0030] Specifically, the process of inputting the fused dataset into a pre-set power consumption coupling analysis model to obtain the dynamic correlation matrix of signal integrity parameters and power consumption fluctuations is as follows: The signal integrity parameters and power consumption fluctuation data are segmented by a sliding window to obtain the power consumption change rate and the exponential decay integral of the signal time-domain waveform within each window, and a dynamic correlation matrix is constructed to characterize the coupling relationship between signal distortion and transient power consumption.

[0031] In this implementation scheme, due to the strong time-varying nature of the signal transmission on the logic board, directly calculating global statistical features may ignore local anomalies. Therefore, a sliding window method is used to segment the timing signal, and each window is used to analyze the signal integrity and power consumption fluctuation characteristics within a short period of time. Window division rule: S(t) represents the timing signal on the logic board, and P(t) represents the power consumption curve; the window length ΔT is selected, and the number of data points within the window is Nw; the window step size is Δt (usually Δt < ΔT) to achieve sliding update of the data; let the window index be k, then the time range of the k-th window is: T k =[t k ,t k +ΔT]; Calculation of power consumption change rate Transient power consumption fluctuations will affect the integrity of the logic board signal, so the power consumption change rate within the window is calculated: where: Rk is the power consumption change rate of the k-th window; P(t q ) is the power consumption at time t q ; Nw is the number of data points within the window; the absolute value is used to measure the intensity of instantaneous power consumption fluctuations. Exponential decay integral of the signal time-domain waveform The signal integrity is greatly affected by power consumption fluctuations, especially transient power consumption overshoot may cause signal ringing and distortion. Therefore, the exponential decay integral of the signal waveform is calculated within each window to quantify its decay characteristics: where: I k is the signal decay integral within window k; S(t) is the signal time-domain waveform; λ is the decay factor, which controls the exponential decay speed; the exponential term gives greater weight to the closer time points, emphasizing the transient change characteristics. Construction of the dynamic correlation matrix To quantify the coupling relationship between signal integrity and power consumption fluctuations, a dynamic correlation matrix A is defined: where: A m,n represents the dynamic correlation strength between signal m and power consumption fluctuation n; K is the total number of windows; R k is the power consumption change rate; I k is the signal decay integral. The correlation matrix A reflects the coupling relationship between signal distortion (such as ringing, overshoot) and transient power consumption fluctuations, and is used for subsequent feature extraction and anomaly detection.

[0032] Specifically, the specific process of extracting key coupling features through singular value decomposition and outputting the correlation mode parameters of signal ringing and instantaneous overcurrent is as follows: Perform singular value decomposition on the dynamic correlation matrix, select the time-domain eigenvectors corresponding to the principal components in the left singular vectors and the frequency-domain eigenvectors in the right singular vectors, and map them to a preset ringing-overcurrent mode space to generate correlation mode parameters.

[0033] In this implementation, perform singular value decomposition on the dynamic correlation matrix, given the dynamic correlation matrix Perform singular value decomposition (SVD): A = UΣV T ; where: is the left singular vector matrix, representing the time-domain eigenvectors of the correlation matrix; is a diagonal matrix containing the singularities σ i (measuring the importance of different features); is the right singular vector matrix, representing the frequency-domain eigenvectors of the correlation matrix; V T is the transpose of V. Left singular vector (U): Characterizes the main mode of signal ringing in the time domain. Right singular vector (V): Characterizes the main mode of instantaneous overcurrent in the frequency domain. Singular value (Σ): Represents the relative importance of different modes, and larger singular values correspond to the main ringing-overcurrent coupling features. Select the principal component eigenvectors. Since some noise modes in the vectors obtained by singular value decomposition are irrelevant to the analysis, we only retain the principal components. Define the singular values in descending order: σ1 ≥ σ2 ≥... ≥ σ r ; Select the first a principal components (a is determined by the cumulative energy percentage threshold): where: a represents the number of eigenvectors retained; η is the energy threshold (usually set to 90% - 95%). Finally, obtain: Time-domain principal eigenvector: U a = [u1, u2,..., u a ; Frequency-domain principal eigenvector: V a = v1, v2,..., v a . Map to the preset ringing-overcurrent mode space, which is used to characterize the correlation features between signal ringing and instantaneous overcurrent. Let the preset mode space basis function be: Ψ = {ψ1, ψ2,..., ψ P}; Project onto the mode space through the principal eigenvectors: Θ = U a T Ψ; where: Θ is the correlation mode parameter matrix, representing the characteristic components of the ringing-overcurrent mode; Ψ is the space basis function, usually constructed from experimental data or empirical models. Finally, Θ is used as the key feature parameter to characterize the correlation mode between signal ringing and instantaneous overcurrent.

[0034] Specifically, the specific process of constructing a density clustering model with an adaptive neighborhood radius according to the association pattern parameters and the temperature gradient distribution is as follows: Based on the signal distortion rate and the temperature gradient distribution in the association pattern parameters, the local neighborhood radius is dynamically obtained, and the high-density abnormal area is divided by an improved density clustering algorithm to generate candidate defect clusters.

[0035] In this implementation scheme, the goal of this step is to use the signal distortion rate and temperature gradient information to dynamically adjust the neighborhood range, identify the abnormal area, and classify the high-density abnormal points through an improved density clustering method, so as to generate candidate defect clusters. The neighborhood radius is usually a fixed value, resulting in uneven recognition capabilities in different density regions. This method dynamically adjusts the neighborhood radius based on the signal distortion rate and temperature gradient distribution, thereby improving the recognition ability of abnormal areas. Signal distortion rate: From the association pattern parameters, it represents the degree of interference to signal integrity. For example, a high ringing amplitude or an instantaneous overcurrent peak may indicate an abnormal local current impact. Temperature gradient distribution: The local gradient in the infrared temperature data reflects the electrothermal effect, and temperature anomalies are often accompanied by current mutations or signal distortions. In areas with large signal distortions and steep temperature gradients, the neighborhood radius is reduced to accurately cluster abnormal points; in normal signal areas, the neighborhood radius is increased to avoid false detections. This method realizes the adjustment of the adaptive neighborhood radius, enabling the algorithm to maintain a high resolution in abnormal areas with different densities. After determining the adaptive neighborhood radius, the density clustering algorithm is used to divide the high-density abnormal points into multiple clustering regions. Compared with the traditional density clustering method, this method has been improved, mainly including: comprehensively considering the signal distortion rate and the temperature change rate, calculating the abnormality score of each point, and improving the sensitivity to abnormal points. For the initially unclassified isolated abnormal points, this method adopts the neighborhood abnormal mean compensation strategy to judge whether they should be classified into an abnormal cluster, avoiding the generation of excessive isolated noise points. Adopting a boundary expansion mechanism based on density gradient makes the clustering boundary more conform to the real distribution of the physical abnormal area, avoiding the problem of unreasonable clustering shapes. After clustering, each high-density abnormal area is marked as a candidate defect cluster, and the characteristics of these clusters include: central point coordinates: representing the core position of the abnormal area; cluster density value: used to quantify the degree of abnormality, and the higher the density, the more serious the abnormality; abnormal category probability: According to the clustering result and the historical defect data matching, it provides a possible defect type prediction. These candidate defect clusters will be input into the graph neural network model in the next step for further analysis of the defect type and three-dimensional coordinates.

[0036] Specifically, the specific process of generating an abnormal area probability map and marking the electrothermal coupling abnormal coordinates and confidence is as follows: Based on the electrothermal coupling intensity and the temperature gradient direction consistency of the data points in the candidate defect cluster, the abnormal confidence is obtained, and a probability map is generated and the central coordinates of the abnormal area are marked.

[0037] In this implementation plan, the core objective of this step is to calculate the anomaly confidence, generate a probability distribution map, and mark the central coordinates of the electro-thermal coupling anomaly based on the data of the candidate defect clusters. The candidate defect clusters consist of multiple high-density anomaly data points, and each point contains: Electro-thermal coupling strength: indicating the degree of association between electrical signal anomalies and temperature changes (such as the synchronization of signal ringing and local overheating). Temperature gradient direction consistency: used to determine whether the anomaly area has a stable heat diffusion pattern (to avoid misjudgment due to local random temperature fluctuations). The evaluation process of the anomaly strength is as follows: Calculate the mean value of the electro-thermal coupling strength of the data points within the cluster. Statistically analyze the electro-thermal coupling characteristics of all points in the defect cluster and calculate the overall mean value to measure the anomaly degree of this area. Statistically analyze the differences in the temperature gradient directions of all points within the cluster. If the differences are small, it indicates that the anomalies in this area are consistent. Calculate the anomaly confidence. The anomaly confidence (C f ) reflects the possibility of a certain data point being an anomaly point. The calculation method combines: the electro-thermal coupling strength (H f ) of this point, and the temperature gradient direction consistency (G f ) of the area where it is located. When calculating the anomaly confidence, we adopt an exponential decay model to emphasize the core points in the high-anomaly area and weaken the contribution of the boundary points: where: C f : The anomaly confidence of the f-th point; H f : The electro-thermal coupling strength of the f-th point; H max : The maximum electro-thermal coupling strength among all points (used for normalization); G f : The temperature gradient direction consistency of the f-th point (the value range is [0, 1], and the closer to 1, the stronger the direction consistency); α: The adjustment factor (controlling the influence degree of the temperature gradient direction on the confidence). Generate the probability map of the anomaly area. Based on the anomaly confidence C f of all data points, generate a probability density distribution through two-dimensional interpolation: Adopt Gaussian kernel density estimation to smooth the data points, making the probability distribution of the anomaly area clearer. The numerical range in the probability map is [0, 1], indicating the possibility of electro-thermal coupling anomaly in this area. Visual presentation: High-confidence area (red): indicating an area with obvious anomalies, where defects may exist. Medium-confidence area (yellow): There may be slight anomalies, but further analysis is still required. Low-confidence area (blue): Normal area, or weak electro-thermal coupling. Mark the central coordinates of the anomaly area. To accurately mark the anomaly area, use the centroid calculation method to determine the center of each defect cluster: where: (x c , y c ): The central coordinates of the anomaly area; (x j , y j ): The coordinates of the f-th data point; C f: The anomaly confidence of the f-th data point. This method can ensure that data points with high confidence contribute more to the central position, making the center point more conform to the actual anomaly area, excluding the influence of low-confidence points, and avoiding the deviation of the anomaly area.

[0038] Specifically, taking the spectral entropy of the time-series signal, the local temperature mean, and the variance of power consumption fluctuation as node features, and the electro-thermal coupling coefficient as the edge weight, the process is as follows: The logic board area covered by the anomaly area probability map is divided into grids, and the spectral entropy value, the local temperature mean, and the variance of power consumption fluctuation of the time-series signal in each grid cell are extracted as node features. According to the electro-thermal coupling coefficient in the association mode parameter, the edge weights between adjacent nodes are dynamically calculated to construct a topological graph structure representing the electro-thermal coupling relationship.

[0039] In this implementation, the construction process taking the spectral entropy of the time-series signal, the local temperature mean, and the variance of power consumption fluctuation as node features, and the electro-thermal coupling coefficient as the edge weight: In this process, first, the logic board area covered by the anomaly area probability map is divided into grids, and the time-series signal, temperature distribution, and power consumption fluctuation data in each grid cell are extracted as node features. The specific operations are as follows: The spectral entropy of the time-series signal: used to characterize the uncertainty or complexity of the signal. The higher the spectral entropy value, the more complex the signal. The local temperature mean: represents the temperature level of this area and can reflect the electro-thermal coupling intensity. The variance of power consumption fluctuation: The variance value of power consumption fluctuation reflects the stability of power consumption in this area. A larger variance means that the power consumption fluctuates violently in this area. According to the electro-thermal coupling coefficient in the association mode parameter, the edge weights between adjacent nodes can be dynamically calculated. The electro-thermal coupling coefficient reflects the electro-thermal coupling intensity between two nodes. In this way, a topological graph can be constructed, where the features of each node include the spectral entropy of the time-series signal, the local temperature mean, and the variance of power consumption fluctuation, and the weight of the edge is determined by the electro-thermal coupling coefficient.

[0040] Specifically, the specific process of iteratively updating the node state through the graph attention mechanism and outputting the defect type and three-dimensional coordinates is as follows: Deploy multiple layers of graph attention networks on the topological graph structure, aggregate the electro-thermal features of neighboring nodes through attention weights, and after iteratively updating the node state vector, through the combined output of the classifier and the regressor, map the node state to the preset defect type space and the physical coordinate space of the logic board to generate the defect type label and three-dimensional coordinate information.

[0041] In this implementation, the process of iteratively updating the node state through the graph attention mechanism and outputting the defect type and three-dimensional coordinates is as follows: After constructing the topological structure of the graph, the graph attention mechanism (GAT) is applied to update the state of each node. The main role of the graph attention mechanism is to make the influence of important neighbor nodes greater when updating the state vector of each node by learning the weights (i.e., "attention") between nodes. The specific process is as follows: In each layer of the graph neural network, the neighbor weights of each node are calculated through the attention mechanism. The influence of neighbor nodes is no longer fixed, but dynamically calculated according to the features between nodes and the edge weights. The updated node state vector will reflect the electro-thermal coupling characteristics of the node. Finally, through the joint output of the classifier and the regressor, the defect type and three-dimensional coordinate information of the node are obtained. Formula representation: Where: The updated state vector of the z-th node at the l+1 layer. N(z): The set of neighbor nodes of node z. α zv : The attention weight between node z and node v, indicating the influence degree of node v on node z. W (l) : The weight matrix of the l-th layer, used to transform the node features. σ: The activation function, usually a non-linear activation function such as ReLU. Through the graph attention mechanism, the state vector of the node is gradually updated and the electro-thermal characteristics of its neighbors are integrated, thereby effectively capturing the electro-thermal coupling relationship of the entire topological graph.

[0042] A comprehensive performance test device for a logic board includes the following modules: a multi-source synchronous fusion module, a parameter coupling analysis module, an electro-thermal anomaly clustering module, and a graph network defect determination module; the multi-source synchronous fusion module is used to synchronously collect the running timing signal, dynamic power consumption curve, and infrared temperature distribution data through the logic board interface, eliminate the minute timing offset between multi-sensors through the dynamic time warping algorithm, and generate a spatio-temporally aligned fusion data set; input the fusion data set into a pre-set power consumption coupling analysis model, obtain the dynamic correlation matrix of signal integrity parameters and power consumption fluctuations, extract key coupling features through singular value decomposition, and output the correlation mode parameters of signal ringing and instantaneous overcurrent; the parameter coupling analysis module is used to construct a density clustering model with an adaptive neighborhood radius according to the correlation mode parameters and temperature gradient distribution, generate an abnormal area probability map, and mark the electro-thermal coupling abnormal coordinates and confidence levels; the electro-thermal anomaly clustering module is used to input the fusion data set, correlation mode parameters, and abnormal area probability map into the graph neural network model, use the time series signal spectrum entropy, local temperature mean, and power consumption fluctuation variance as node features, and the electro-thermal coupling coefficient as the edge weight, iteratively update the node state through the graph attention mechanism, and output the defect type and three-dimensional coordinates.

[0043] In this implementation scheme, the multi-source synchronous fusion module: By synchronously collecting and spatio-temporally aligning data from different sources (such as time-series signals, power consumption, and temperature data), it ensures data consistency during the analysis process and avoids incorrect analysis caused by sensor time differences. This is the basis for high-precision performance testing and defect detection. The dynamic time warping algorithm is used to solve the problem of minute temporal offsets between multiple sensors, ensuring accurate data fusion and real-time alignment, with high precision and reliability, and is applicable to complex multi-source data fusion scenarios. The parameter coupling analysis module: This module can extract the relationship between signal integrity and power consumption fluctuations from multi-source data, providing effective correlation information for subsequent analysis and helping to identify potential problems in the circuit, such as the coupling problem between power consumption fluctuations and signal quality. Using singular value decomposition technology to extract key coupling features, it can automatically discover the deep-seated correlation between signals and power consumption, and then infer complex electrothermal phenomena such as signal ringing and overcurrent. The electrothermal anomaly clustering module: This module can effectively identify and mark electrothermal anomaly regions through a density clustering algorithm with an adaptive neighborhood radius, providing accurate data support for subsequent defect localization. Combining the signal distortion rate and temperature gradient information to dynamically adjust the neighborhood radius, this adaptive clustering method can flexibly handle density changes in different regions and provide a more accurate division of anomaly regions than traditional methods. The graph network defect determination module: Through a graph neural network model, this module can combine various node features (such as spectral entropy, temperature mean, and variance of power consumption fluctuations) with the electrothermal coupling coefficient to automatically identify and locate the types and spatial positions of defects on the logic board, greatly improving the accuracy and efficiency of defect detection. Using the graph attention mechanism, it can focus on the most important nodes and edges in the graph neural network, enhancing the accuracy of defect detection and prediction. At the same time, jointly outputting the defect type and three-dimensional coordinates breaks the limitations of traditional detection methods and has strong innovation and practical application value.

[0044] In summary, this application has at least the following effects:

[0045] A logic board comprehensive performance testing device and testing method. Through the multi-source synchronous fusion module, it can accurately synchronize and align data from different sensors, eliminate minute timing offsets, ensure the high precision and consistency of data fusion, and provide a reliable basis for subsequent analysis and testing. In the parameter coupling analysis module, the singular value decomposition technology is used to extract key coupling features, which can effectively reveal the relationship between signal integrity and power consumption fluctuations, and improve the ability to identify the system operation status and potential problems. The electrothermal anomaly clustering module combines the density clustering algorithm with an adaptive neighborhood radius, which can flexibly identify and divide electrothermal anomaly regions, and generate an accurate probability map of the anomaly regions, improving the accuracy and reliability of anomaly detection. The graph network defect determination module combines graph neural network technology, uses features such as time series signals, local temperature means, and power consumption fluctuation variances, accurately judges the defect type and determines its three-dimensional coordinates, effectively improving the automation level and accuracy of defect detection and positioning.

[0046] Those skilled in the art should understand that the embodiments of the present invention can be provided as a method, a system, or a computer program product. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present invention can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0047] The present invention is described with reference to the flowcharts and / or block diagrams of systems, devices (systems), and computer program products according to the embodiments of the present invention. It should be understood that each process and / or block in the flowcharts and / or block diagrams, and the combination of processes and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing devices generate a device for implementing the functions specified in Figure 1 one process or multiple processes and / or blocks Figure 1 one block or multiple blocks.

[0048] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer-readable memory generate a manufactured article including an instruction device, and the instruction device implements the functions specified in Figure 1 one process or multiple processes and / or blocks Figure 1 one block or multiple blocks.

[0049] These acquisition machine program instructions can also be loaded onto an acquisition machine or other programmable data processing devices, so that a series of operation steps are executed on the acquisition machine or other programmable devices to generate the processing implemented by the acquisition machine. Thus, the instructions executed on the acquisition machine or other programmable devices provide for implementing the steps in the process Figure 1 One process or multiple processes and / or blocks Figure 1 Steps for the functions specified in one block or multiple blocks.

[0050] Although the preferred embodiments of the present invention have been described, those skilled in the art can make additional changes and modifications to these embodiments once they learn the basic creative concept. Therefore, the appended claims are intended to be construed as including the preferred embodiments and all changes and modifications falling within the scope of the present invention.

[0051] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalent technologies, the present invention is also intended to include these modifications and variations.

Claims

1. A comprehensive performance testing method for a logic board, characterized in that It includes the following steps: S1. Synchronously collect the operation timing signal, dynamic power consumption curve and infrared temperature distribution data through the logic board interface, eliminate the minute timing offset between multiple sensors through the dynamic time warping algorithm, and generate a spatio-temporally aligned fusion data set; S2. Input the fusion data set into a preset power consumption coupling analysis model, obtain the dynamic correlation matrix of signal integrity parameters and power consumption fluctuations, extract key coupling features through singular value decomposition, and output the correlation mode parameters of signal ringing and instantaneous overcurrent; S3. According to the correlation mode parameters and temperature gradient distribution, construct a density clustering model with an adaptive neighborhood radius, generate an abnormal area probability map, and mark the electrothermal coupling abnormal coordinates and confidence level; S4. Input the fusion data set, correlation mode parameters and abnormal area probability map into a graph neural network model. Using the spectral entropy of the timing signal, local temperature mean and power consumption fluctuation variance as node features and the electrothermal coupling coefficient as the edge weight, iteratively update the node state through the graph attention mechanism, and output the defect type and three-dimensional coordinates.

2. The comprehensive performance testing method of a logic board according to claim 1, wherein: The specific process of eliminating the minute timing offset between multiple sensors through the dynamic time warping algorithm and generating a spatio-temporally aligned fusion data set is as follows: Perform multi-channel timestamp alignment on the timing signal, dynamic power consumption curve and infrared temperature distribution data. Based on the dynamic window length adjustment strategy, obtain the minimum bending path of the multi-source data sequence, and map the data of each sensor to a unified time reference through cubic spline interpolation to generate a spatio-temporally aligned fusion data set.

3. A method for comprehensively testing the performance of a logic board according to claim 2, characterized in that: The specific process of inputting the fusion data set into a preset power consumption coupling analysis model and obtaining the dynamic correlation matrix of signal integrity parameters and power consumption fluctuations is as follows: Perform sliding window segmentation on the signal integrity parameters and power consumption fluctuation data, obtain the exponential decay integral of the power consumption change rate and the signal time-domain waveform within each window, and construct a dynamic correlation matrix to characterize the coupling relationship between signal distortion and transient power consumption.

4. A method for comprehensively testing the performance of a logic board according to claim 3, characterized in that: The specific process of extracting key coupling features through singular value decomposition and outputting the correlation mode parameters of signal ringing and instantaneous overcurrent is as follows: Perform singular value decomposition on the dynamic correlation matrix, select the time-domain eigenvector corresponding to the principal component in the left singular vector and the frequency-domain eigenvector in the right singular vector, map them to a preset ringing-overcurrent mode space, and generate correlation mode parameters.

5. A method for comprehensively testing the performance of a logic board according to claim 4, characterized in that: The specific process of constructing a density clustering model with an adaptive neighborhood radius according to the correlation mode parameters and temperature gradient distribution is as follows: Dynamically obtain the local neighborhood radius according to the signal distortion rate and temperature gradient distribution in the correlation mode parameters, and divide the high-density abnormal area through an improved density clustering algorithm to generate candidate defect clusters.

6. A method for comprehensively testing the performance of a logic board according to claim 5, characterized in that: The specific process of generating an abnormal area probability map and marking the electrothermal coupling abnormal coordinates and confidence level is as follows: Obtain the abnormal confidence level according to the electrothermal coupling intensity and temperature gradient direction consistency of the data points in the candidate defect cluster, generate a probability map and mark the center coordinates of the abnormal area.

7. A method for comprehensively testing the performance of a logic board according to claim 6, characterized in that: The process of using the spectral entropy of the timing signal, local temperature mean and power consumption fluctuation variance as node features and the electrothermal coupling coefficient as the edge weight is as follows: The logical board area covered by the abnormal area probability map is divided into grids, and the spectral entropy value, local temperature mean, and power consumption fluctuation variance of the timing signal in each grid cell are extracted as node features. According to the electrothermal coupling coefficient in the correlation pattern parameters, the edge weights between adjacent nodes are dynamically calculated to construct a topological graph structure representing the electrothermal coupling relationship.

8. A method for comprehensively testing the performance of a logic board according to claim 7, characterized in that: The specific process of iteratively updating the node state through the graph attention mechanism and outputting the defect type and three-dimensional coordinates is as follows: Deploy a multi-layer graph attention network on the topological graph structure. Aggregate the electrothermal characteristics of neighboring nodes through attention weights. After iteratively updating the node state vector, map the node state to the preset defect type space and the physical coordinate space of the logical board through the combined output of a classifier and a regressor to generate defect type labels and three-dimensional coordinate information.

9. A comprehensive performance testing device for a logic board, which is applied to the comprehensive performance testing method for a logic board described in any one of claims 1-8, and is characterized in that, It includes the following modules: multi-source synchronous fusion module, parameter coupling analysis module, electrothermal anomaly clustering module, and graph network defect determination module; The multi-source synchronous fusion module is used to synchronously collect the running timing signal, dynamic power consumption curve, and infrared temperature distribution data through the logical board interface, eliminate the minute timing offset between multiple sensors through the dynamic time warping algorithm, and generate a spatio-temporally aligned fusion data set; Input the fusion data set into a preset power consumption coupling analysis model to obtain the dynamic correlation matrix of signal integrity parameters and power consumption fluctuations, extract key coupling features through singular value decomposition, and output the correlation pattern parameters of signal ringing and instantaneous overcurrent; The parameter coupling analysis module is used to construct a density clustering model with an adaptive neighborhood radius according to the correlation pattern parameters and temperature gradient distribution, generate an abnormal area probability map, and mark the electrothermal coupling abnormal coordinates and confidence levels; The electrothermal anomaly clustering module is used to input the fusion data set, correlation pattern parameters, and abnormal area probability map into a graph neural network model. Using the spectral entropy of the timing signal, local temperature mean, and power consumption fluctuation variance as node features and the electrothermal coupling coefficient as the edge weight, iteratively update the node state through the graph attention mechanism, and output the defect type and three-dimensional coordinates.

Images