Electronic component failure early warning method, device, equipment and medium

Abstract

The application relates to the technical field of artificial intelligence. By providing an electronic component failure early warning method, device, equipment and medium, wherein the method comprises: based on a preset electronic component historical failure waveform library, performing nonlinear relationship modeling on acquired model compression parameters, sampling strategy parameters and failure detection indexes to generate a missed detection rate mutation boundary surface; calculating task criticality according to real-time monitoring signal characteristics, pre-stored historical failure probability database data and residual power data collected by a battery management unit to generate a task criticality index; based on the task criticality index and the missed detection rate mutation boundary surface, adaptively adjusting the calculation resource configuration to generate a calculation mode switching instruction; and according to the calculation mode switching instruction, maintaining the early warning model calculation continuity through a feature cache reuse mechanism, so that the technical effects of improving the real-time risk response, reducing the missed detection rate mutation risk and enhancing the calculation continuity are achieved.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, and in particular to methods, devices, equipment and media for early warning of electronic component failures. Background Technology

[0002] With the deep application of artificial intelligence technology in industrial equipment monitoring, embedded intelligent diagnostic systems are becoming increasingly important in critical infrastructure operation and maintenance, mobile testing terminals, and remote monitoring platforms. Efficient real-time fault early warning is a core element in ensuring the safe operation of equipment and is also a key focus and challenge in intelligent diagnostic technology research.

[0003] Traditional technologies use fixed-threshold decision models for fault diagnosis. However, these models cannot adapt to dynamic risk changes under complex operating conditions, resulting in delayed response and high false alarm rates. In resource-constrained scenarios, traditional technologies employ static model compression strategies to reduce computational load. However, this strategy lacks performance boundary constraints on model compression parameters and sampling strategies. Increasing the model compression rate can easily trigger sudden changes in the false negative rate, leading to missed detections of critical faults. Furthermore, traditional technologies do not establish cross-model feature transfer mechanisms when switching computation modes, often neglecting the issue of maintaining feature continuity, resulting in interruptions in the early warning model's state and creating monitoring blind spots. Summary of the Invention

[0004] Therefore, it is necessary to provide electronic component fault early warning methods, devices, equipment and media to address the above-mentioned technical problems, so as to achieve the technical effects of improving the real-time performance of risk response, reducing the risk of sudden changes in the missed detection rate, and enhancing the continuity of calculation.

[0005] In a first aspect, this application provides a method for early warning of electronic component failures, the method comprising:

[0006] Based on a pre-set library of historical fault waveforms of electronic components, nonlinear relationship modeling is performed on the acquired model compression parameters, sampling strategy parameters and fault detection indicators to generate a boundary surface for sudden changes in the missed detection rate.

[0007] The task criticality is calculated based on the characteristics of real-time monitoring signals, the pre-stored historical fault probability database data, and the remaining power data collected by the battery management unit, and a task criticality index is generated.

[0008] Based on the task criticality index and the boundary surface of the sudden change in the missed detection rate, the computing resource configuration is adaptively adjusted to generate computing mode switching instructions.

[0009] Based on the computation mode switching instruction, the continuity of early warning model computation is maintained through a feature cache reuse mechanism.

[0010] In one embodiment, based on a preset historical fault waveform library of electronic components, nonlinear relationship modeling is performed on the acquired model compression parameters, sampling strategy parameters, and fault detection indicators to generate a boundary surface for the sudden change in the missed detection rate, including:

[0011] Multi-scale transient feature extraction is performed on a pre-defined database of historical fault waveforms of electronic components to obtain a fault feature tensor set;

[0012] Based on the fault feature tensor set, boundary detection is performed on the correlation between model compression parameters, sampling strategy parameters and fault detection indicators to generate parameter sensitivity distribution data;

[0013] Nonlinear response reconstruction is performed on the parameter sensitivity distribution data to establish a boundary surface for the sudden change in the false negative rate.

[0014] In one embodiment, based on the task criticality index and the boundary surface of the false negative rate mutation, the computing resource allocation is adaptively adjusted to generate a computing mode switching instruction, including:

[0015] Dynamic threshold decision analysis is performed on task criticality indicators to obtain mode switching requirement identifiers;

[0016] Gradient obstacle avoidance path planning is performed based on the boundary surface of the sudden change in the missed detection rate to generate safe switching trajectory data;

[0017] The mode switching requirement identifier and the safe switching trajectory data are combined to generate a calculated mode switching command.

[0018] In one embodiment, gradient obstacle avoidance path planning is performed based on the boundary surface of the sudden change in the missed detection rate to generate safe switching trajectory data, including:

[0019] A gradient safety channel is obtained by extracting the safety corridor from the boundary surface of the sudden change in the false negative rate.

[0020] Within the gradient safety channel, an energy-constrained path search is performed to generate a sequence of switching paths that satisfy the energy consumption threshold. The expression for the sequence of switching paths that satisfy the energy consumption threshold is:

[0021]

[0022] in, This represents a sequence of switching paths that meet the energy consumption threshold. This represents the set of feasible paths within the gradient safe channel. Representing path points The curvature energy dissipation function at that point, Representing path points The real-time false negative rate gradient value, This represents the energy consumption-performance balance coefficient. Represents the differential arc length of the path. Indicates candidate paths, Indicates the path parameter index. Indicates the first path The coordinates of the points Represents the coordinate vector of the path points;

[0023] Dynamic trajectory encoding is performed on the switching path sequence that meets the energy consumption threshold to obtain safe switching trajectory data.

[0024] In one embodiment, according to the computation mode switching instruction, the continuity of the early warning model computation is maintained through a feature cache reuse mechanism, including:

[0025] The source and target computation models are determined according to the computation mode switching instructions. Tensor space mapping is performed on the intermediate feature layers of the source and target computation models to generate a feature alignment matrix.

[0026] Based on the feature alignment relation matrix, the output feature tensors generated by the source computing model in the current computing cycle are incrementally cached to generate incremental feature cache data.

[0027] Incremental feature cache data is loaded through the target computation model, and state recovery is performed based on the incremental feature cache data.

[0028] In one embodiment, the task criticality is calculated based on real-time monitoring signal characteristics, pre-stored historical fault probability database data, and remaining power data collected by the battery management unit, and a task criticality index is generated, including:

[0029] Multimodal transient analysis was performed on the characteristics of the real-time monitoring signal to obtain the signal risk entropy value. The expression for the signal risk entropy value is as follows:

[0030]

[0031] in, This represents the signal risk entropy value. Indicates the first The time-frequency characteristic matrix of each mode, Indicates the first Transient weight matrix for each modality It represents the Hadamah accumulation. Denotes the Frobenius norm of a matrix. Represents the total number of modes. This represents the total number of modes in the signal analysis. Indicates the current modality index. Modal index indicating summation of denominators;

[0032] Based on pre-stored historical fault probability database data, time-varying weights are calculated to generate dynamic fault probability coefficients. The expression for the dynamic fault probability coefficients is as follows:

[0033]

[0034] in, This represents the dynamic failure probability coefficient. This represents the historical failure probability baseline value. Indicates the cumulative operating time of the device. Indicates the aging degradation factor. This represents the gradient change in the failure probability over time. Indicates the mutation sensitivity coefficient. Represents a symbolic function;

[0035] Based on the remaining power data collected by the battery management unit, a sustainable computing power assessment is performed to generate a system range index;

[0036] By integrating signal risk entropy, dynamic failure probability coefficient, and system endurance index, a criticality metric is obtained. The expression for the mission criticality metric is as follows:

[0037]

[0038] in, Indicates the criticality of the task. This represents the signal risk entropy value. This represents the dynamic failure probability coefficient. This indicates the system's battery life index. Indicates the prevention of zero constant, The exponential weighting factor represents the signal risk entropy value. The exponential weighting factor represents the failure probability coefficient. This represents the index weighting factor of the range index.

[0039] In one embodiment, based on the fault feature tensor set, boundary probing is performed on the correlation between model compression parameters, sampling strategy parameters, and fault detection indicators to generate parameter sensitivity distribution data, including:

[0040] Multi-dimensional feature decoupling analysis is performed on the fault feature tensor set to obtain an independent feature component set;

[0041] Based on the independent feature component set, sensitivity propagation calculation is performed on the coupling relationship between model compression parameters, sampling strategy parameters and fault detection indicators to generate parameter sensitivity gradient field;

[0042] The system detects critical points of performance abrupt changes in the parameter sensitivity gradient field and generates parameter sensitivity distribution data.

[0043] Secondly, this application also provides an electronic component fault early warning device, which includes:

[0044] The missed detection rate surface modeling module is used to perform nonlinear relationship modeling on the acquired model compression parameters, sampling strategy parameters and fault detection indicators based on a preset historical fault waveform library of electronic components, and generate a boundary surface for the sudden change of missed detection rate.

[0045] The task criticality calculation module is used to calculate the task criticality based on real-time monitoring signal characteristics, pre-stored historical fault probability database data, and remaining power data collected by the battery management unit, and generate task criticality indicators.

[0046] The mode switching instruction generation module is used to adaptively adjust the computing resource configuration based on the task criticality index and the boundary surface of the false detection rate change, and generate computing mode switching instructions.

[0047] The computation continuity maintenance module is used to maintain the computation continuity of the early warning model through a feature cache reuse mechanism according to the computation mode switching instruction.

[0048] Thirdly, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of any of the methods in the first aspect of this application.

[0049] Fourthly, this application also provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of any of the methods in the first aspect of this application.

[0050] This application provides a method, apparatus, device, and medium for early warning of electronic component faults. The method includes: modeling the nonlinear relationship between model compression parameters, sampling strategy parameters, and fault detection indicators based on a preset historical fault waveform library of electronic components, generating a boundary surface for sudden changes in the missed detection rate, which can capture the nonlinear correlation between parameters and construct performance constraint boundaries, providing a reliable basis for subsequent adjustment of computing resource configuration, thereby reducing the risk of sudden changes in the missed detection rate; combining real-time monitoring signal characteristics, pre-stored historical fault probability database data, and remaining power data collected by the battery management unit to calculate the criticality of the task, which can dynamically integrate multi-source data to reflect the risk level and system endurance status of the current monitoring scenario, making the adaptive adjustment of computing resources more in line with actual operating needs, thereby improving the real-time performance of risk response.

[0051] Based on the task criticality index and the boundary surface of the false negative rate mutation, the computing resource allocation is adaptively adjusted and a computing mode switching instruction is generated. Under the premise of meeting performance constraints, the allocation of computing resources can be optimized and the risk of false negative rate mutation can be further reduced. Furthermore, through the feature cache reuse mechanism, the intermediate features of the source computing model are retained and reused when the computing mode is switched, avoiding the interruption of the early warning model state and effectively enhancing the continuity of computing. Attached Figure Description

[0052] To more clearly illustrate the technical solutions in the embodiments or related technologies of this application, the accompanying drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0053] Figure 1 This is a flowchart of an electronic component fault early warning method in one embodiment of the present invention;

[0054] Figure 2 In one embodiment of the present invention, a flowchart is generated by performing nonlinear relationship modeling on the acquired model compression parameters, sampling strategy parameters and fault detection indicators based on a preset historical fault waveform library of electronic components to generate a boundary surface of the missed detection rate mutation.

[0055] Figure 3 This is a structural diagram of an electronic component fault early warning device according to one embodiment of the present invention. Detailed Implementation

[0056] To make the above-mentioned objects, features, and advantages of this application more apparent and understandable, the specific embodiments of this application will be described in detail below with reference to the accompanying drawings. Many specific details are set forth in the following description to provide a thorough understanding of this application. However, this application can be implemented in many other ways different from those described herein, and those skilled in the art can make similar modifications without departing from the spirit of the application. Therefore, this application is not limited to the specific embodiments disclosed below.

[0057] First, the application scenarios of the embodiments of this application are described. In the embodiments of this application, methods, devices, equipment, and media for electronic component fault early warning are provided, applicable to scenarios such as monitoring core electronic components in industrial manufacturing production lines, early warning of electronic components in rail transit traction control systems, and status perception of key components in converters of new energy power plants.

[0058] In illustrative purposes, the electronic component fault early warning method, device, equipment and medium provided in the embodiments of this application can also be applied to other application scenarios such as the operation and maintenance of core electronic modules of medical equipment, the status early warning of power grid terminal components, and the health management of airborne electronic systems in aerospace. This is only an example and does not limit the specific application scenarios.

[0059] like Figure 1 As shown, this application provides a method for early warning of electronic component faults, the method comprising:

[0060] S101: Based on a pre-set library of historical fault waveforms of electronic components, nonlinear relationship modeling is performed on the acquired model compression parameters, sampling strategy parameters and fault detection indicators to generate a boundary surface for sudden changes in the missed detection rate.

[0061] For example, the fault early warning terminal retrieves a preset historical fault waveform library of electronic components, and filters out fault waveform reference information related to model compression parameters, sampling strategy parameters and fault detection indicators from the library; at the same time, the fault early warning terminal obtains the model compression parameters, sampling strategy parameters and fault detection indicators to be analyzed, analyzes the nonlinear correlation between the three types of parameters and between the three types of parameters and the electronic component false detection rate, and sorts out the influence law of various parameter combinations on the change of false detection rate.

[0062] Based on the correlation patterns obtained from the analysis and the reference information of the filtered fault waveforms, the fault early warning terminal constructs a nonlinear relationship modeling model. By optimizing the model, it describes the corresponding logic between the three types of parameters and the sudden change in the missed detection rate. Then, based on the model, it completes the nonlinear relationship modeling between the three types of parameters and the missed detection rate of electronic components, and generates the boundary surface of the sudden change in the missed detection rate.

[0063] S102: Calculate the task criticality based on the characteristics of real-time monitoring signals, the pre-stored historical fault probability database data, and the remaining power data collected by the battery management unit, and generate the task criticality index.

[0064] For example, the fault early warning terminal acquires real-time monitoring signal characteristics, pre-stored historical fault probability database data, and remaining power data collected by the battery management unit. It analyzes the core information related to the operational risks of electronic components from the real-time monitoring signal characteristics, filters fault probability-related information from the historical fault probability database that matches the current monitoring scenario, and verifies valid information reflecting the system's endurance from the remaining power data. Based on the collaborative calculation rules of multi-source data, the fault early warning terminal performs a comprehensive task criticality calculation on the processed information to generate a task criticality index.

[0065] S103: Based on the task criticality index and the boundary surface of the false negative rate change, adaptively adjust the computing resource configuration and generate computing mode switching instructions.

[0066] For example, the fault early warning terminal acquires the task criticality index and the boundary surface of the missed detection rate change, analyzes the resource configuration requirements corresponding to the task criticality index in combination with the performance constraint analysis of the boundary surface of the missed detection rate change, adaptively plans and adjusts the computing resource configuration accordingly, and then determines the matching computing mode switching logic based on the adjusted computing resource configuration and generates a computing mode switching instruction.

[0067] S104: Based on the calculation mode switching instruction, maintain the continuity of early warning model calculation through the feature cache reuse mechanism.

[0068] For example, the fault early warning terminal receives a computing mode switching instruction, parses the instruction to determine the switching relationship of the early warning model, extracts and caches the intermediate feature data of the current early warning model through the feature caching reuse mechanism, establishes a cross-model feature association mapping, allows the target early warning model to load cached features and realize the connection of computing states, and maintains the computing continuity of the early warning model.

[0069] One embodiment of this application provides an electronic component fault early warning method, including: based on a preset historical fault waveform library of electronic components, performing nonlinear relationship modeling on model compression parameters, sampling strategy parameters, and fault detection indicators to generate a boundary surface for sudden changes in the missed detection rate. This step can capture the nonlinear correlation between various parameters and construct performance constraint boundaries, providing a reliable basis for subsequent adjustment of computing resource configuration, thereby reducing the risk of sudden changes in the missed detection rate; combining real-time monitoring signal characteristics, pre-stored historical fault probability database data, and remaining power data collected by the battery management unit to calculate the criticality of the task. This step can dynamically integrate multi-source data to reflect the fault risk level and system endurance status of the current monitoring scenario, making the adaptive adjustment of computing resources more in line with the actual operating needs of the equipment, thereby improving the real-time performance of fault risk response.

[0070] Based on the task criticality index and the boundary surface of the false negative rate mutation, the computing resource configuration is adaptively adjusted and a computing mode switching instruction is generated. This step can optimize the computing resource allocation strategy under the premise of meeting performance constraints and further reduce the risk of false negative rate mutation. Then, through the feature cache reuse mechanism, the intermediate feature data of the source computing model is retained and reused during the computing mode switching process, avoiding the interruption of the computing state of the early warning model and effectively improving the computing continuity of the early warning model.

[0071] like Figure 2 As shown, based on a pre-set historical fault waveform library of electronic components, nonlinear relationship modeling is performed on the acquired model compression parameters, sampling strategy parameters, and fault detection indicators to generate a boundary surface for the sudden change in the missed detection rate, including:

[0072] S201: Perform multi-scale transient feature extraction on the preset historical fault waveform library of electronic components to obtain the fault feature tensor set.

[0073] For example, the fault early warning terminal retrieves a preset historical fault waveform library of electronic components, performs preprocessing operations such as denoising and normalization on the electronic component fault waveform data stored in the library, and then performs hierarchical decomposition of the preprocessed fault waveform data in terms of time and frequency dimensions according to a preset multi-scale division standard. For each scale waveform segment, it captures transient features such as voltage change, frequency drift, and phase shift. The transient features extracted from each scale are dimensionally aligned and information fused. Through tensor structured encapsulation processing, a fault feature tensor set is obtained.

[0074] The multi-scale transient feature extraction includes a complete process of fault waveform data preprocessing, multi-dimensional scale division, multi-scale transient feature capture, feature dimension alignment and fusion, and tensor structured encapsulation.

[0075] S202: Based on the fault feature tensor set, boundary detection is performed on the correlation between model compression parameters, sampling strategy parameters and fault detection indicators to generate parameter sensitivity distribution data.

[0076] For example, the fault early warning terminal performs dimensional decomposition on the fault feature tensor set, identifies the feature dimensions related to model compression parameters, sampling strategy parameters, and fault detection indicators, and then analyzes the correlation strength between model compression parameters and each feature dimension, the influence of sampling strategy parameters on feature extraction performance, and the matching relationship between fault detection indicators and feature data. A boundary detection scheme with multiple parameter combinations is designed, and the fault detection performance changes under different parameter combinations are tested sequentially according to the detection scheme. The critical performance state is located, and the sensitivity fluctuation of each parameter before and after the critical state is recorded. All fluctuation data are structured and organized to generate parameter sensitivity distribution data.

[0077] The parameter sensitivity distribution data includes feature dimension correlation information, parameter combination detection records, performance critical state data, and structured results of sensitivity fluctuations of each parameter.

[0078] S203: Reconstruct the nonlinear response of the parameter sensitivity distribution data to establish the boundary surface of the sudden change in the false negative rate.

[0079] For example, the fault early warning terminal performs outlier removal and effective data screening on parameter sensitivity distribution data. Based on the screened effective data, it establishes a mapping relationship between parameter change and false negative rate fluctuation. It quantifies the nonlinear response law between the two through a nonlinear fitting algorithm. Based on the quantified response law, it determines the critical parameter threshold range of false negative rate mutation. It then uses a surface construction algorithm to fit the critical threshold range and the corresponding false negative rate change data to establish a boundary surface for false negative rate mutation.

[0080] The nonlinear response reconstruction includes the entire process of parameter sensitivity data screening, establishing the mapping relationship between parameters and false negative rate, quantifying the nonlinear response law, and fitting the critical threshold range.

[0081] In one embodiment, based on the task criticality index and the boundary surface of the false negative rate mutation, the computing resource allocation is adaptively adjusted to generate a computing mode switching instruction, including:

[0082] (1) Perform dynamic threshold decision analysis on the task criticality index to obtain the mode switching requirement identifier.

[0083] For example, the fault warning terminal acquires the task criticality index, collects historical criticality data of electronic components and related information such as environmental parameters and load status of the current operating scenario, thereby determining the initial benchmark and adjustment range of the dynamic threshold, establishing a dynamic threshold update mechanism that includes update cycle, adjustment range rules and scenario adaptation coefficient, and calibrating the dynamic threshold in real time according to the mechanism.

[0084] The fault early warning terminal compares the task criticality index with the calibrated dynamic threshold in multiple dimensions, distinguishes the monitoring needs corresponding to high, medium and low task criticality levels, determines whether the computing power supply and response speed of the current computing mode meet the corresponding level requirements, and if not, clarifies the necessity of mode switching. At the same time, it determines the core type of the target computing mode by combining the operating priority of electronic components, marks the urgency of the switch trigger, and generates a mode switching requirement identifier.

[0085] The mode switching requirement identifier includes the result of the mode switching necessity determination, the target computing mode type information, the switching trigger priority information, the task criticality level matching description, and the description of the current computing mode adaptation defects.

[0086] (2) Gradient obstacle avoidance path planning is performed based on the boundary surface of the sudden change in the missed detection rate to generate safe switching trajectory data.

[0087] For example, the fault early warning terminal retrieves the boundary surface of the missed detection rate change, analyzes the performance constraint parameters such as the parameter threshold range, gradient change rate, and change critical point contained in the surface, sorts out the gradient distribution law of the missed detection rate as the parameters change, and divides the gradient safe region, gradient transition region and gradient risk region according to the gradient change magnitude and change risk level.

[0088] The fault early warning terminal prioritizes avoiding sudden changes in the missed detection rate and ensuring fault detection performance. Combining the adjustable dimensions and range of computing resource configuration with energy consumption constraints, it designs a gradient obstacle avoidance path planning criterion that covers risk avoidance priority, resource consumption upper limit, and switching efficiency requirements. According to the planning criterion, it searches for candidate switching paths in the gradient safety region and gradient transition region. The missed detection rate control effect, resource adaptability, and execution feasibility of the candidate paths are verified by simulating the parameter adjustment process. The optimal candidate path is selected, the parameter configuration of the path nodes is refined, the execution order of the path is clarified, and safe switching trajectory data is generated.

[0089] The safe switching trajectory data includes the model compression parameter configuration of path nodes, the sampling strategy parameter configuration, the fault detection index threshold, the path execution sequence instructions, the risk control threshold for the missed detection rate of each node, the estimated data of computing resource consumption, and the path fault tolerance adjustment scheme.

[0090] (3) The mode switching requirement identifier and the safe switching trajectory data are combined to obtain the calculation mode switching instruction.

[0091] For example, the fault warning terminal verifies the validity of the mode switching requirement identifier and the safe switching trajectory data, checks the logical integrity and scenario matching degree of the mode switching requirement identifier, verifies the parameter rationality and path feasibility of the safe switching trajectory data, confirms the logical consistency of the two types of data in terms of target calculation mode pointing, switching priority and performance constraint requirements, and extracts the core decision information in the mode switching requirement identifier and the key content such as path node parameters, execution order and risk control threshold in the safe switching trajectory data.

[0092] The fault warning terminal, referring to the preset computing mode switching instruction format specification, divides the instruction into modules such as instruction header, core parameter area, execution rule area, and safety constraint area. The extracted key content is structured and integrated by module. Combined with the real-time operating status of electronic components, the specific execution time of computing mode switching is determined. The core basis for path selection and the effective rules of parameter configuration are clarified. The integrated instruction content is formatted and its integrity is checked to ensure that the instruction is unambiguous and executable, and the computing mode switching instruction is synthesized.

[0093] The computing mode switching instruction includes a mode switching execution signal, a target computing mode parameter configuration set, a path execution instruction sequence, a description of the switching execution timing, a description of the path selection basis, a description of the security constraints, fault tolerance adjustment trigger rules, and the instruction effective time range.

[0094] In one embodiment, gradient obstacle avoidance path planning is performed based on the boundary surface of the sudden change in the missed detection rate to generate safe switching trajectory data, including:

[0095] (1) Extract the safety corridor from the boundary surface of the sudden change in the false negative rate to obtain the gradient safety channel.

[0096] For example, the fault early warning terminal retrieves the boundary surface of the sudden change in the missed detection rate, analyzes the global gradient distribution characteristics of the surface, locates the high-risk area where the gradient change rate of the missed detection rate exceeds a preset threshold, and collects the boundary coordinates, gradient change trend and impact range data of the high-risk area.

[0097] The fault warning terminal delineates a continuous feasible region within the safe range of the missed detection rate gradient based on a preset safe gradient threshold. It then performs polygon boundary fitting and smoothing on this region to clarify the horizontal and vertical boundary range of the channel and marks the high-risk avoidance points within the channel, thus obtaining a gradient safe channel.

[0098] The gradient safety channel includes the range of the gradient safety threshold for the missed detection rate, the channel boundary fitting parameters, the range of feasible area coordinates, the high-risk avoidance point markers, and the channel width parameters.

[0099] (2) Perform energy-constrained path search within the gradient safety channel to generate a switching path sequence that satisfies the energy consumption threshold. The expression for the switching path sequence that satisfies the energy consumption threshold is:

[0100]

[0101] in, This represents a sequence of switching paths that meet the energy consumption threshold. This represents the set of feasible paths within the gradient safe channel. Representing path points The curvature energy dissipation function at that point, Representing path points The real-time false negative rate gradient value, This represents the energy consumption-performance balance coefficient. Represents the differential arc length of the path. Indicates candidate paths, Indicates the path parameter index. Indicates the first path The coordinates of the points This represents the coordinate vector of the path point.

[0102] For example, the fault warning terminal determines the set of feasible paths consisting of all feasible paths within the gradient safety channel, defines the calculation dimension of the curvature energy consumption function, clarifies the adaptation rules of the energy consumption-performance balance coefficient, and sets the upper limit constraint of the energy consumption threshold.

[0103] The fault warning terminal traverses each candidate path in the set of feasible paths, calculates the total curvature energy consumption and the maximum missed detection rate gradient value of each candidate path, and performs a comprehensive evaluation in combination with the energy consumption-performance balance rule. It then selects the candidate path with the lowest total energy consumption and the lowest missed detection rate gradient risk, organizes the coordinates and execution order of the path nodes, and generates a switching path sequence that meets the energy consumption threshold.

[0104] The switching path sequence that meets the energy consumption threshold includes a set of path node coordinates, curvature energy consumption data of each node, missed detection rate gradient value of each node, path execution order, comprehensive evaluation score, energy consumption threshold compliance indicator, and risk level label.

[0105] (3) Dynamic trajectory encoding is performed on the switching path sequence that meets the energy consumption threshold to obtain safe switching trajectory data.

[0106] For example, the fault early warning terminal extracts core information such as node coordinates, curvature energy consumption data, missed detection rate gradient value, and execution order from the switching path sequence that meets the energy consumption threshold, and determines the field definition, data type and encoding rules of the dynamic trajectory encoding.

[0107] The fault early warning terminal converts the path information into a structured encoding format according to the encoding rules, adds trajectory timestamps, status verification identifiers, energy consumption and performance constraint parameters, and generates binary or character encoded data that can be directly parsed by the early warning model. The integrity of the encoding result is verified to obtain safe switching trajectory data.

[0108] The safe switching trajectory data includes encoded path node information, trajectory execution time series, status verification identifier, energy consumption and performance constraint parameters, format verification code, and decoding mapping rules.

[0109] In one embodiment, according to the computation mode switching instruction, the continuity of the early warning model computation is maintained through a feature cache reuse mechanism, including:

[0110] (1) Determine the source computing model and the target computing model according to the computing mode switching instruction, perform tensor space mapping on the intermediate feature layers of the source computing model and the target computing model, and generate a feature alignment relation matrix.

[0111] For example, the fault warning terminal receives a computing mode switching command, parses the command header, parameter area and execution rules of the computing mode switching command, extracts the model switching identifier field, clarifies the unique device identifier, current running stage and computing progress of the source computing model and the target computing model, and retrieves the structural metadata of the source computing model and the target computing model stored locally.

[0112] The fault early warning terminal parses the structural metadata of the source and target computation models, sorts out the tensor dimensions, feature encoding rules, and output time series of the intermediate feature layers, constructs the dimension matching rules of the tensor space mapping, calculates the association weights of the corresponding feature layers of the source and target computation models, generates a structured matrix containing dimension mapping relationships and weight parameters, and generates a feature alignment relationship matrix.

[0113] The computation mode switching instruction includes an instruction header, parameter area, execution rules, and model switching identifier field. The feature alignment relation matrix includes the intermediate feature layer dimension information of the source computation model, the intermediate feature layer dimension information of the target computation model, feature dimension association weights, mapping rule identifier, and matrix checksum.

[0114] (2) Based on the feature alignment relation matrix, the output feature tensor generated by the source computing model in the current computing cycle is incrementally cached to generate incremental feature cache data.

[0115] For example, the fault warning terminal retrieves the feature alignment relationship matrix, parses the feature mapping rules and dimension association weights in the matrix, determines the incremental feature dimensions that need to be cached in the output feature tensor of the source computing model, sets the trigger conditions, data update frequency and cache validity period for incremental caching, and creates a storage index for the preset feature cache area.

[0116] The fault early warning terminal obtains the output feature tensor generated by the source computing model in the current computing cycle, extracts effective feature data according to the determined incremental feature dimension, encapsulates the extracted feature data in a structured manner, adds cache timestamp, data verification identifier and mapping association information, and writes the encapsulated data into the corresponding storage index of the preset feature cache area to generate incremental feature cache data.

[0117] The incremental feature dimensions include dimensions in the source computation model's output feature tensor that match the intermediate feature layer of the target computation model, dimensions whose feature change rate exceeds a threshold, and dimensions whose association weight is higher than a preset value. The incremental feature cache data includes incremental dimension data of the source computation model's output feature tensor, cache timestamp, data verification identifier, feature mapping association information, storage index address, and cache validity period.

[0118] (3) Load incremental feature cache data through the target calculation model and perform state recovery based on the incremental feature cache data.

[0119] For example, the fault warning terminal triggers the feature loading process of the target computing model, retrieves incremental feature cache data, parses the feature dimensions, mapping association information and storage index address in the data, matches the dimension requirements of the intermediate feature layer input interface of the target computing model, and constructs the feature data loading path and verification rules.

[0120] The fault early warning terminal controls the target computing model to read incremental feature cache data according to the loading path, maps the feature data to the corresponding input port of the intermediate feature layer of the target computing model, restores the computing state of the target computing model based on the cached feature data, performs continuity verification on the restored computing state, ensures that the output of the target computing model is continuously connected with the computing results of the source computing model, and maintains the computing continuity of the early warning model.

[0121] The feature loading path includes the read address of incremental feature cache data, data format conversion rules, and the input port mapping relationship of the intermediate feature layer of the target computation model. State recovery includes feature filling of the intermediate feature layer of the target computation model, computation state verification, output result continuity verification, and anomaly tolerance mechanisms.

[0122] In one embodiment, the task criticality is calculated based on real-time monitoring signal characteristics, pre-stored historical fault probability database data, and remaining power data collected by the battery management unit, and a task criticality index is generated, including:

[0123] (1) Multimodal transient analysis is performed on the characteristics of the real-time monitoring signal to obtain the signal risk entropy value. The expression for the signal risk entropy value is:

[0124]

[0125] in, This represents the signal risk entropy value. Indicates the first The time-frequency characteristic matrix of each mode, Indicates the first Transient weight matrix for each modality It represents the Hadamah accumulation. Denotes the Frobenius norm of a matrix. Represents the total number of modes. This represents the total number of modes in the signal analysis. Indicates the current modality index. The modal index represents the summation of the denominator.

[0126] For example, the fault early warning terminal acquires real-time monitoring signal characteristics and performs multi-dimensional mode division on the real-time monitoring signal characteristics. The division is based on factors such as the time-frequency domain characteristics, energy distribution, and transient characteristics of the signal. For each divided mode, the time-frequency domain feature information is extracted to generate a time-frequency feature matrix corresponding to each mode. The preset transient weight matrix corresponding to each mode is retrieved. This matrix is ​​used to quantify the risk contribution of signal transients under different modes.

[0127] The fault early warning terminal performs element-wise operations on the time-frequency feature matrix and transient weight matrix of each mode, calculates the norm of the matrix after the operation, normalizes the norm results of all modes, and calculates the entropy value based on the normalization result. The entropy value reflects the uncertainty of the risk distribution of the signal under multiple modes, and the signal risk entropy value is obtained.

[0128] The multi-dimensional modal classification includes modal categories based on signal time-frequency domain characteristics, energy distribution, and transient characteristics. The signal risk entropy value includes the transient contribution ratio of each modal's time-frequency characteristics, normalized risk distribution information, the signal risk level quantified by entropy value, and the uncertainty characterization of multi-modal risk.

[0129] (2) Based on the pre-stored historical fault probability database data, time-varying weights are calculated to generate dynamic fault probability coefficients. The expression for the dynamic fault probability coefficients is:

[0130]

[0131] in, This represents the dynamic failure probability coefficient. This represents the historical failure probability baseline value. Indicates the cumulative operating time of the device. Indicates the aging degradation factor. This represents the gradient change in the failure probability over time. Indicates the mutation sensitivity coefficient. Represents a symbolic function.

[0132] For example, the fault early warning terminal retrieves pre-stored historical fault probability database data, extracts historical fault probability benchmark values, which are reference values ​​for the fault probability of the equipment under standard operating conditions, obtains the cumulative running time of the equipment, and calculates the attenuation effect of equipment aging on the fault probability by combining the preset aging attenuation factor. The aging attenuation factor is used to quantify the rate of change of fault probability caused by the increase in equipment running time.

[0133] The fault early warning terminal calculates the gradient change of the fault probability over time. This change reflects the dynamic rate of change of the fault probability over time. The mutation sensitivity coefficient and the sign function are combined to determine the mutation adjustment range. The mutation sensitivity coefficient is used to amplify or reduce the impact of the fault probability mutation. The aging attenuation effect and the mutation adjustment range are fused together to generate a dynamic fault probability coefficient.

[0134] The historical failure probability benchmark includes the failure probability reference value of the equipment under standard operating conditions and the failure probability statistics of different operating stages. The dynamic failure probability coefficient includes the historical failure probability benchmark value, the impact of aging and attenuation, the magnitude of abrupt adjustment, the quantified value of the time-varying failure probability after fusion, and the dynamic trend of failure probability.

[0135] (3) Based on the remaining power data collected by the battery management unit, a sustainable computing power assessment is performed to generate a system endurance index.

[0136] For example, the fault warning terminal obtains the remaining power data collected by the battery management unit. This data is the current remaining capacity information of the battery. Combined with the computing power consumption rate of the current computing mode, which is the amount of computing power resources consumed per unit time under the current computing mode, the terminal assesses the continuous computing time that the remaining power can support. The continuous computing time is the sustainable running time of the remaining power under the current computing power consumption rate.

[0137] The fault warning terminal compares the continuous calculation time with the preset battery life threshold, which is the minimum battery life requirement to ensure stable operation of the equipment. The battery life threshold is converted into a quantitative battery life index. The higher the index value, the stronger the system's battery life.

[0138] The remaining power data includes the battery's current remaining capacity information and battery health status related data. The system's battery life index includes remaining power data, computing power consumption rate, continuous computing time evaluation results, quantified battery life level, and battery life threshold comparison results.

[0139] (4) The criticality index is obtained by merging the signal risk entropy value, dynamic failure probability coefficient, and system endurance index. The expression for the mission criticality index is as follows:

[0140]

[0141] in, Indicates the criticality of the task. This represents the signal risk entropy value. This represents the dynamic failure probability coefficient. This indicates the system's battery life index. Indicates the prevention of zero constant, The exponential weighting factor represents the signal risk entropy value. The exponential weighting factor represents the failure probability coefficient. This represents the index weighting factor of the range index.

[0142] For example, the fault warning terminal acquires the signal risk entropy value, dynamic fault probability coefficient and system endurance index, retrieves the preset index weight factor and zero-prevention constant corresponding to each parameter, the zero-prevention constant is used to avoid calculation abnormalities when the system endurance index is zero, and sums the system endurance index and the zero-prevention constant to obtain the corrected value of the endurance.

[0143] The fault warning terminal performs exponential operations on the signal risk entropy value and dynamic fault probability coefficient according to the corresponding index weight factors. The results of the operation are then combined with the sum of the system endurance index according to the weights. The combined calculation comprehensively considers the synergistic effects of signal risk, fault probability and system endurance to obtain the mission criticality index.

[0144] The index weighting factors include the index weighting factors for signal risk entropy, dynamic failure probability coefficient, and system endurance index, used to quantify the contribution of different parameters to mission criticality. Mission criticality indicators include the weighted contribution of signal risk entropy, the weighted contribution of dynamic failure probability coefficient, the weighted contribution of system endurance index, the fused criticality quantification, and the criteria for classifying mission criticality levels.

[0145] In one embodiment, based on the fault feature tensor set, boundary probing is performed on the correlation between model compression parameters, sampling strategy parameters, and fault detection indicators to generate parameter sensitivity distribution data, including:

[0146] (1) Perform multi-dimensional feature decoupling analysis on the fault feature tensor set to obtain the independent feature component set.

[0147] For example, the fault early warning terminal retrieves the fault feature tensor set, analyzes the tensor dimension composition, feature association rules and physical meaning of the fault feature tensor set, and sorts out the coupling association types between different feature dimensions, including linear coupling and nonlinear coupling.

[0148] The fault early warning terminal designs multi-dimensional decoupling rules based on the physical meaning of features and fault correlation. It performs orthogonal decomposition and dimension separation on the coupled feature dimensions. After separation, the independence of each feature dimension is verified. The verification content includes the cross-correlation coefficient and information entropy redundancy between features. The verified features are structured and encapsulated to obtain an independent feature component set.

[0149] The fault feature tensor set includes multi-scale transient features extracted from historical fault waveforms of electronic components, tensor dimension composition information, feature association rules, and feature physical meaning annotations. The independent feature component set includes decoupled single-dimensional features, feature independence verification results, feature physical meaning annotations, fault correlation degree classification information, and orthogonal decomposition processing records.

[0150] (2) Based on the independent feature component set, the sensitivity propagation calculation is performed on the coupling relationship between the model compression parameters, sampling strategy parameters and fault detection indicators to generate the parameter sensitivity gradient field.

[0151] For example, the fault early warning terminal retrieves an independent feature component set, extracts feature components related to model compression parameters, sampling strategy parameters, and fault detection indicators, establishes a one-to-one mapping relationship between parameters and feature components, defines the calculation rules for sensitivity propagation, and clarifies the transmission path and attenuation coefficient of the influence of parameter changes on feature components.

[0152] The fault early warning terminal simulates minute changes in model compression parameters and sampling strategy parameters according to the calculation rules of sensitivity propagation, tracks the response change amplitude of corresponding feature components, calculates the correlation strength between parameter changes and feature responses, and integrates the correlation strength in a grid according to the parameter dimension to generate a parameter sensitivity gradient field.

[0153] The independent feature component set includes decoupled single-dimensional features, the mapping relationship between features and parameters, and the annotation of the physical meaning of features. The parameter sensitivity gradient field includes the mapping relationship between parameters and feature components, sensitivity propagation path records, correlation strength quantization values, gradient field dimensional distribution information, and gridded gradient data.

[0154] (3) Detect the critical point of performance change in the parameter sensitivity gradient field and generate parameter sensitivity distribution data.

[0155] For example, the fault early warning terminal retrieves the parameter sensitivity gradient field, analyzes the correlation strength change trend in the gradient field, identifies the region of gradient value abrupt change, determines the candidate critical point based on the preset performance mutation threshold, and performs multi-dimensional verification on the candidate critical point. The verification content includes the correlation consistency between the parameter change amplitude, the feature response amplitude and the fault detection performance.

[0156] The fault early warning terminal clusters and structures the verified performance mutation critical points, statistically analyzes the sensitivity change patterns under different parameter combinations, integrates the critical point information and sensitivity change patterns, and generates structured data containing critical point distribution and sensitivity trends, thus generating parameter sensitivity distribution data.

[0157] The parameter sensitivity gradient field includes the correlation strength change trend, gradient value mutation region identifier, performance mutation threshold, and candidate critical point coordinates. The parameter sensitivity distribution data includes the performance mutation critical point coordinates, sensitivity change trend statistics, parameter combination clustering results, performance mutation threshold matching information, and structured sensitivity data.

[0158] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0159] In one embodiment, such as Figure 3 As shown, this application also provides an electronic component fault early warning device 300, which includes:

[0160] The missed detection rate surface modeling module 301 is used to perform nonlinear relationship modeling on the acquired model compression parameters, sampling strategy parameters and fault detection indicators based on a preset historical fault waveform library of electronic components, and generate a boundary surface for the sudden change of missed detection rate.

[0161] The task criticality calculation module 302 is used to calculate the task criticality based on the characteristics of real-time monitoring signals, the pre-stored historical fault probability database data and the remaining power data collected by the battery management unit, and generate task criticality indicators.

[0162] The mode switching instruction generation module 303 is used to adaptively adjust the computing resource configuration based on the task criticality index and the boundary surface of the false negative rate change, and generate computing mode switching instructions.

[0163] The computation continuity maintenance module 304 is used to maintain the computation continuity of the early warning model through a feature cache reuse mechanism according to the computation mode switching instruction.

[0164] Specifically, the fault early warning terminal includes: a missed detection rate surface modeling module 301. This module retrieves a preset historical fault waveform library of electronic components, obtains model compression parameters, sampling strategy parameters and fault detection indicators, performs nonlinear correlation feature extraction on the three types of parameters, sorts out the mapping rules between parameters and missed detection rate, and uses a surface fitting algorithm to construct an initial surface structure that can characterize the critical range of sudden change in missed detection rate.

[0165] The false negative rate surface modeling module 301 performs multiple rounds of performance verification on the initial surface structure, compares the consistency between the surface boundary and the actual false negative rate mutation critical state, corrects the surface fitting parameters, and performs structured encapsulation on the verified surface to generate the false negative rate mutation boundary surface.

[0166] The boundary surface of the false negative rate mutation includes the nonlinear mapping relationship of parameters, the critical condition of the false negative rate mutation, the fitting parameters of the boundary surface, and the performance verification correction record.

[0167] The task criticality calculation module 302 acquires real-time monitoring signal characteristics, pre-stored historical fault probability database data, and remaining power data collected by the battery management unit. It performs preprocessing operations such as noise reduction and normalization on the three types of data to extract core feature information related to electronic component failure risk, historical fault probability trend, and system endurance support capability.

[0168] The task criticality calculation module 302 retrieves the preset weight allocation rules, performs weighted fusion calculation on the three types of core feature information, quantifies the fault risk level and resource support matching degree during task execution, and generates a quantitative indicator that can comprehensively reflect the task priority, namely the task criticality indicator.

[0169] Among them, the mission criticality indicators include the fault risk weighted score, the contribution value of historical fault probability, the quantitative value of endurance support capability, and the comprehensive criticality level indicator.

[0170] The mode switching instruction generation module 303 obtains the task criticality index and the boundary surface of the missed detection rate mutation, analyzes the performance constraint range defined by the boundary surface of the missed detection rate mutation, and formulates an adaptive adjustment scheme for computing resource allocation based on the computing power requirement level corresponding to the task criticality index, clarifying the adjustment direction of resource allocation ratio and scheduling priority.

[0171] The mode switching instruction generation module 303 determines the corresponding computing mode switching logic according to the adjustment scheme, clarifies the switching trigger conditions, execution timing and safety constraints, and integrates the adjustment scheme, switching logic and constraints in a structured manner, encapsulates them according to the preset instruction format, and generates computing mode switching instructions.

[0172] The computing mode switching instruction includes computing resource configuration adjustment details, computing mode switching logic, execution timing parameters, security constraints, and instruction format verification code.

[0173] The computation continuity maintenance module 304 receives the computation mode switching instruction, parses the model switching identifier in the instruction, determines the currently running source computation model and the target computation model to be switched, retrieves the intermediate feature layer structure information of the two types of models, and establishes feature mapping association rules.

[0174] The computational continuity maintenance module 304 extracts intermediate feature data from the source computational model based on feature mapping association rules and performs incremental caching. It controls the target computational model to load the cached feature data, restores the computational state of the target computational model, verifies the continuity of the output results of the two types of models, and maintains the computational continuity of the early warning model.

[0175] The continuity of the early warning model calculation includes feature data cache records, cross-model feature mapping relationships, calculation state recovery results, and output continuity verification reports.

[0176] The missed detection rate surface modeling module 301 is also used for:

[0177] Multi-scale transient feature extraction is performed on a pre-defined database of historical fault waveforms of electronic components to obtain a fault feature tensor set;

[0178] Based on the fault feature tensor set, boundary detection is performed on the correlation between model compression parameters, sampling strategy parameters and fault detection indicators to generate parameter sensitivity distribution data;

[0179] Nonlinear response reconstruction is performed on the parameter sensitivity distribution data to establish a boundary surface for the sudden change in the false negative rate.

[0180] The mode switching instruction generation module 303 is also used for:

[0181] Dynamic threshold decision analysis is performed on task criticality indicators to obtain mode switching requirement identifiers;

[0182] Gradient obstacle avoidance path planning is performed based on the boundary surface of the sudden change in the missed detection rate to generate safe switching trajectory data;

[0183] The mode switching requirement identifier and the safe switching trajectory data are combined to generate a calculated mode switching command.

[0184] The mode switching instruction generation module 303 is also used for:

[0185] A gradient safety channel is obtained by extracting the safety corridor from the boundary surface of the sudden change in the false negative rate.

[0186] Within the gradient safety channel, an energy-constrained path search is performed to generate a sequence of switching paths that satisfy the energy consumption threshold. The expression for the sequence of switching paths that satisfy the energy consumption threshold is:

[0187]

[0188] in, This represents a sequence of switching paths that meet the energy consumption threshold. This represents the set of feasible paths within the gradient safe channel. Representing path points The curvature energy dissipation function at that point, Representing path points The real-time false negative rate gradient value, This represents the energy consumption-performance balance coefficient. Represents the differential arc length of the path. Indicates candidate paths, Indicates the path parameter index. Indicates the first path The coordinates of the points Represents the coordinate vector of the path points;

[0189] Dynamic trajectory encoding is performed on the switching path sequence that meets the energy consumption threshold to obtain safe switching trajectory data.

[0190] The continuity maintenance module 304 is also used for:

[0191] The source and target computation models are determined according to the computation mode switching instructions. Tensor space mapping is performed on the intermediate feature layers of the source and target computation models to generate a feature alignment matrix.

[0192] Based on the feature alignment relation matrix, the output feature tensors generated by the source computing model in the current computing cycle are incrementally cached to generate incremental feature cache data.

[0193] Incremental feature cache data is loaded through the target computation model, and state recovery is performed based on the incremental feature cache data.

[0194] The task criticality calculation module 302 is also used for:

[0195] Multimodal transient analysis was performed on the characteristics of the real-time monitoring signal to obtain the signal risk entropy value. The expression for the signal risk entropy value is as follows:

[0196]

[0197] in, This represents the signal risk entropy value. Indicates the first The time-frequency characteristic matrix of each mode, Indicates the first Transient weight matrix for each modality It represents the Hadamah accumulation. Denotes the Frobenius norm of a matrix. Represents the total number of modes. This represents the total number of modes in the signal analysis. Indicates the current modality index. Modal index indicating summation of denominators;

[0198] Based on pre-stored historical fault probability database data, time-varying weights are calculated to generate dynamic fault probability coefficients. The expression for the dynamic fault probability coefficients is as follows:

[0199]

[0200] in, This represents the dynamic failure probability coefficient. This represents the historical failure probability baseline value. Indicates the cumulative operating time of the device. Indicates the aging degradation factor. This represents the gradient change in the failure probability over time. Indicates the mutation sensitivity coefficient. Represents a symbolic function;

[0201] Based on the remaining power data collected by the battery management unit, a sustainable computing power assessment is performed to generate a system range index;

[0202] By integrating signal risk entropy, dynamic failure probability coefficient, and system endurance index, a criticality metric is obtained. The expression for the mission criticality metric is as follows:

[0203]

[0204] in, Indicates the criticality of the task. This represents the signal risk entropy value. This represents the dynamic failure probability coefficient. This indicates the system's battery life index. Indicates the prevention of zero constant, The exponential weighting factor represents the signal risk entropy value. The exponential weighting factor represents the failure probability coefficient. This represents the index weighting factor of the range index.

[0205] The missed detection rate surface modeling module 301 is also used for:

[0206] Multi-dimensional feature decoupling analysis is performed on the fault feature tensor set to obtain an independent feature component set;

[0207] Based on the independent feature component set, sensitivity propagation calculation is performed on the coupling relationship between model compression parameters, sampling strategy parameters and fault detection indicators to generate parameter sensitivity gradient field;

[0208] The system detects critical points of performance abrupt changes in the parameter sensitivity gradient field and generates parameter sensitivity distribution data.

[0209] In one embodiment, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in the above-described method embodiments.

[0210] In one embodiment, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps in the above-described method embodiments.

[0211] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to in the description of the method embodiments. The device embodiments described above are merely illustrative. The components described as separate parts may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this disclosure according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0212] The above-described embodiments are merely illustrative of several implementation methods of the embodiments of this application, and their descriptions are relatively specific and detailed. However, they should not be construed as limiting the scope of the patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the embodiments of this application, and these modifications and improvements all fall within the protection scope of the embodiments of this application.

Claims

1. A method for early warning of electronic component faults, characterized in that, The method includes: Based on a pre-set library of historical fault waveforms of electronic components, nonlinear relationship modeling is performed on the acquired model compression parameters, sampling strategy parameters and fault detection indicators to generate a boundary surface for sudden changes in the missed detection rate. The task criticality is calculated based on the characteristics of real-time monitoring signals, the pre-stored historical fault probability database data, and the remaining power data collected by the battery management unit, and a task criticality index is generated. Based on the task criticality index and the boundary surface of the false negative rate mutation, the computing resource configuration is adaptively adjusted to generate a computing mode switching instruction. According to the calculation mode switching instruction, the continuity of the early warning model calculation is maintained through the feature cache reuse mechanism.

2. The electronic component fault early warning method according to claim 1, characterized in that, The system, based on a pre-set historical fault waveform library of electronic components, performs nonlinear relationship modeling on the acquired model compression parameters, sampling strategy parameters, and fault detection indicators to generate a boundary surface for sudden changes in the missed detection rate, including: Multi-scale transient feature extraction is performed on the preset historical fault waveform library of electronic components to obtain a fault feature tensor set; Based on the fault feature tensor set, boundary detection is performed on the correlation between the model compression parameters, the sampling strategy parameters, and the fault detection index to generate parameter sensitivity distribution data; The parameter sensitivity distribution data is reconstructed using a nonlinear response to establish the boundary surface of the sudden change in the missed detection rate.

3. The electronic component fault early warning method according to claim 1, characterized in that, The adaptive adjustment of computing resource allocation based on the task criticality index and the boundary surface of the false negative rate mutation, and the generation of computing mode switching instructions, includes: Dynamic threshold decision analysis is performed on the task criticality index to obtain the mode switching requirement identifier; Gradient obstacle avoidance path planning is performed based on the boundary surface of the sudden change in the missed detection rate to generate safe switching trajectory data. The mode switching requirement identifier and the safe switching trajectory data are combined to obtain the calculation mode switching instruction.

4. The electronic component fault early warning method according to claim 3, characterized in that, The step of performing gradient obstacle avoidance path planning based on the boundary surface of the missed detection rate mutation to generate safe switching trajectory data includes: The safety corridor is extracted from the boundary surface of the sudden change in the false negative rate to obtain a gradient safety channel; Within the gradient safety channel, an energy-constrained path search is performed to generate a switching path sequence that satisfies an energy consumption threshold. The expression for the switching path sequence that satisfies the energy consumption threshold is: in, This represents a sequence of switching paths that meet the energy consumption threshold. This represents the set of feasible paths within the gradient safe channel. Representing path points The curvature energy dissipation function at that point, Representing path points The real-time false negative rate gradient value, This represents the energy consumption-performance balance coefficient. Represents the differential arc length of the path. Indicates candidate paths, Indicates the path parameter index. Indicates the first path The coordinates of the points Represents the coordinate vector of the path points; The switching path sequence that meets the energy consumption threshold is dynamically coded to obtain the safe switching trajectory data.

5. The electronic component fault early warning method according to claim 1, characterized in that, The step of maintaining the continuity of early warning model calculation through a feature cache reuse mechanism according to the calculation mode switching instruction includes: The source computing model and the target computing model are determined according to the computing mode switching instruction. Tensor space mapping is performed on the intermediate feature layers of the source computing model and the target computing model to generate a feature alignment relationship matrix. Based on the feature alignment relation matrix, the output feature tensor generated by the source computing model in the current computing cycle is incrementally cached to generate incremental feature cache data; The incremental feature cache data is loaded through the target computing model, and state recovery is performed based on the incremental feature cache data.

6. The electronic component fault early warning method according to claim 1, characterized in that, The task criticality is calculated based on real-time monitoring signal characteristics, pre-stored historical fault probability database data, and remaining power data collected by the battery management unit, generating task criticality indicators, including: Multimodal transient analysis is performed on the characteristics of the real-time monitoring signal to obtain the signal risk entropy value. The expression for the signal risk entropy value is as follows: in, This represents the signal risk entropy value. Indicates the first The time-frequency characteristic matrix of each mode, Indicates the first Transient weight matrix for each modality It represents the Hadamah accumulation. Denotes the Frobenius norm of a matrix. Represents the total number of modes. This represents the total number of modes in the signal analysis. Indicates the current modality index. Modal index indicating summation of denominators; Based on the pre-stored historical fault probability database data, time-varying weights are calculated to generate dynamic fault probability coefficients. The expression for the dynamic fault probability coefficients is as follows: in, This represents the dynamic failure probability coefficient. This represents the historical failure probability baseline value. Indicates the cumulative operating time of the device. Indicates the aging degradation factor. This represents the gradient change in the failure probability over time. Indicates the mutation sensitivity coefficient. Represents a symbolic function; Based on the remaining power data collected by the battery management unit, a sustainable computing power assessment is performed to generate a system endurance index; By integrating the signal risk entropy value, the dynamic failure probability coefficient, and the system endurance index, a criticality co-calculation is performed to obtain the task criticality index. The expression for the task criticality index is as follows: in, Indicates the criticality of the task. This represents the signal risk entropy value. This represents the dynamic failure probability coefficient. This indicates the system's battery life index. Indicates the prevention of zero constant, The exponential weighting factor represents the signal risk entropy value. The exponential weighting factor represents the failure probability coefficient. This represents the index weighting factor of the range index.

7. The electronic component fault early warning method according to claim 2, characterized in that, The step of performing boundary probing on the correlation between the model compression parameters, the sampling strategy parameters, and the fault detection index based on the fault feature tensor set, and generating parameter sensitivity distribution data, includes: Multi-dimensional feature decoupling analysis is performed on the fault feature tensor set to obtain an independent feature component set; Based on the independent feature component set, sensitivity propagation calculation is performed on the coupling relationship between the model compression parameters, the sampling strategy parameters, and the fault detection index to generate a parameter sensitivity gradient field; The critical point of performance mutation is detected in the parameter sensitivity gradient field, and the parameter sensitivity distribution data is generated.

8. An electronic component fault early warning device, characterized in that, The device includes: The missed detection rate surface modeling module is used to perform nonlinear relationship modeling on the acquired model compression parameters, sampling strategy parameters and fault detection indicators based on a preset historical fault waveform library of electronic components, and generate a boundary surface for the sudden change of missed detection rate. The task criticality calculation module is used to calculate the task criticality based on real-time monitoring signal characteristics, pre-stored historical fault probability database data, and remaining power data collected by the battery management unit, and generate task criticality indicators. The mode switching instruction generation module is used to adaptively adjust the computing resource configuration based on the task criticality index and the boundary surface of the missed detection rate mutation, and generate computing mode switching instructions. The computation continuity maintenance module is used to maintain the computation continuity of the early warning model through a feature cache reuse mechanism according to the computation mode switching instruction.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the electronic component fault early warning method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the electronic component fault early warning method according to any one of claims 1 to 7.

Images