Fault prediction methods, devices, electronic equipment and storage media

Abstract

This invention provides a fault prediction method, apparatus, electronic device, and storage medium. The method includes: performing fault prediction based on component attribute data of the product under test (BUT) during the design phase; applying a fault hierarchy model to infer the root cause of faults based on the predicted fault probability distribution; and obtaining the fault propagation path corresponding to the fault modes involved in the BUT. The fault hierarchy model is configured with a multi-layer topology from the product component level to the product whole-machine level, and the transmission relationship of faults in the multi-layer topology from the product component level to the product whole-machine level. This not only realizes the leap from qualitative experience analysis to quantitative data prediction, but also transforms isolated component risks into visualized fault propagation paths. This allows designers to clearly see the specific impact chain of underlying design defects on the reliability of the whole machine, thereby enabling targeted optimization before product manufacturing and greatly improving R&D efficiency and the final quality of the product.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, and in particular to a fault prediction method, apparatus, electronic device, and storage medium. Background Technology

[0002] With the rapid development of integration and intelligence in consumer electronics products, their functional complexity is increasing, software and hardware are highly coupled, and iteration cycles are shortening, posing challenges to ensuring product reliability and requiring improvements in fault prediction methods.

[0003] Currently, reliability analysis and prediction methods mainly include traditional FMEA analysis and statistical fault prediction methods. The former relies on engineers' experience to identify faults and assess their probability and severity, while the latter uses statistical models such as the Weibull distribution to analyze historical data to predict product lifespan. However, when faced with complex consumer electronics R&D scenarios, both of these approaches have significant limitations. Traditional FMEA is time-consuming and cannot adapt to the rapid iteration pace of products, nor can it quantify the risk of multiple concurrent faults. Statistical methods are mostly used for post-event evaluation and are unable to systematically identify the impact of faults across hardware, software, and levels, thus affecting the effectiveness of fault prediction. Summary of the Invention

[0004] This invention provides a fault prediction method, apparatus, electronic device, and storage medium to address the problems in the prior art, such as the time-consuming and inefficient reliance on manual experience for assessment, the difficulty in quantifying the risks of multiple failure modes, and the lack of effective system-level fault propagation analysis methods.

[0005] This invention provides a fault prediction method, comprising: Determine the component attribute data of the product under test during the design phase; Fault prediction is performed based on the component attribute data to obtain a fault probability distribution; the fault probability distribution contains the fault probabilities of multiple fault modes. Based on the fault probability distribution, a fault stratification model is applied to infer the root cause of the fault and obtain the fault propagation path corresponding to the fault mode involved in the product under test. The fault stratification model is configured with a multi-layer topology from the product component level to the product whole machine level, and the transmission relationship of the fault being passed from the product component level to the product whole machine level in the multi-layer topology.

[0006] According to a fault prediction method provided by the present invention, the fault hierarchical model is used to deduce the fault origin based on the fault probability distribution, obtain the fault propagation origin, and based on the propagation relationship, traverse upwards in the multi-layer topology from the level where the fault propagation origin is located until the product whole machine level to obtain the fault propagation path.

[0007] According to a fault prediction method provided by the present invention, the step of performing fault prediction based on the component attribute data to obtain a fault probability distribution further includes: When the failure probability distribution indicates the presence of a risky failure mode among the failure modes involved in the product under test, physical field simulation is performed based on the component attribute data. The simulation cloud map obtained from the physical field simulation is correlated and displayed with the fault probability distribution.

[0008] According to a fault prediction method provided by the present invention, after obtaining the fault propagation path corresponding to the fault mode involved in the product under test, the method further includes: Based on the failure probability distribution and the failure propagation path, a visual risk map is generated; the visual risk map includes at least one of a risk matrix heat map, a component risk radar map, and a failure propagation path map. The risk matrix heatmap uses the product model or project to which the product under test belongs, as well as the relevant failure modes, as dimensions to display the failure probability in the failure probability distribution. The component risk radar chart shows the failure probability of a single component in the product under test under the relevant failure modes. The fault propagation path diagram illustrates the complete fault propagation chain reflected by the fault propagation path, with the fault mode involved as the root cause, and the faults being passed step by step from the intermediate level of the multi-layer topology to the product whole-machine level.

[0009] According to a fault prediction method provided by the present invention, the step of performing fault prediction based on the component attribute data to obtain a fault probability distribution includes: Based on the component attribute data, a fault prediction model is applied to predict faults and obtain the fault probability distribution. The fault prediction model includes multiple fault classifiers, each of which corresponds to a fault mode. The fault prediction model is used to determine the fault probability distribution based on the fault probability obtained from the fault prediction of each fault classifier. The fault prediction model is trained based on a component attribute knowledge graph. The component attribute knowledge graph is constructed based on the following steps: Knowledge is extracted from the sample product manual to obtain sample triples; Based on the sample triples, construct the component attribute knowledge graph; The component attribute knowledge graph includes component nodes, attribute nodes, and fault mode nodes, as well as attribute associations between component nodes and attribute nodes, causal associations between component nodes and fault mode nodes, and attribution associations between component nodes.

[0010] According to a fault prediction method provided by the present invention, the step of applying a fault prediction model to predict faults based on the component attribute data to obtain the fault probability distribution further includes: Determine the actual operating data of the product under test throughout its entire lifecycle, which includes the manufacturing stage, testing stage, after-sales stage, and usage stage; Real fault data is extracted from the real operating data, and the prediction deviation is determined based on the real fault data and the fault probability distribution. If the prediction deviation exceeds the deviation threshold, and / or the amount of actual operating data exceeds the data volume threshold, the fault prediction model and the component attribute knowledge graph are updated based on the actual operating data.

[0011] According to a fault prediction method provided by the present invention, updating the fault prediction model and the component attribute knowledge graph based on the actual operating data includes: Based on the data volume characteristics of the actual operational data, the update mode is determined; When the update mode is incremental learning, the parameters of the fault prediction model are fine-tuned based on the real operating data. When the update mode is retraining, the fault prediction model is retrained based on the real running data and the component attribute knowledge graph. Identify new fault modes and their relationships in the real operating data, and update the new fault modes and their relationships in the component attribute knowledge graph.

[0012] The present invention also provides a fault prediction device, comprising: The data determination unit is used to determine the component attribute data of the product under test during the design phase. The fault prediction unit is used to predict faults based on the component attribute data to obtain a fault probability distribution; the fault probability distribution contains the fault probabilities of multiple fault modes. The root cause deduction unit is used to perform root cause deduction of the fault based on the fault probability distribution and the fault hierarchical model to obtain the fault propagation path corresponding to the fault mode involved in the product under test. The fault hierarchical model is configured with a multi-layer topology from the product component level to the product whole machine level, and the propagation logic of the fault being passed step by step from the product component level to the product whole machine level in the multi-layer topology.

[0013] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and running on the processor, wherein the processor executes the computer program to implement the fault prediction method as described above.

[0014] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the fault prediction method as described above.

[0015] The fault prediction method, device, electronic device, and storage medium provided by this invention directly utilize component attribute data to predict the probability of various fault modes during the design phase, and combine the predicted probability distribution with a fault hierarchical model with a multi-layer topology for system-level deduction. This not only achieves a leap from qualitative experience analysis to quantitative data prediction, but more importantly, it can transform isolated component risks into visualized fault propagation paths. This allows designers to clearly see the specific impact chain of underlying design defects on the overall reliability of the machine, and thus perform targeted optimizations before product manufacturing, greatly improving R&D efficiency and the final quality of the product. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0017] Figure 1 This is a flowchart illustrating the fault prediction method provided by the present invention; Figure 2 This is an example diagram of the multi-layer topology provided by the present invention; Figure 3 This is a strategy example diagram of the problem transformation strategy provided by the present invention; Figure 4 This is a partial simplified diagram of the component attribute knowledge graph provided by the present invention; Figure 5 This is a general framework diagram of the fault prediction method provided by the present invention; Figure 6 This is an overall architecture diagram of the fault prediction system provided by the present invention; Figure 7 This is a schematic diagram of the fault prediction device provided by the present invention; Figure 8 This is a schematic diagram of the structure of the electronic device provided by the present invention. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0019] As consumer electronics products, such as smartphones and educational devices, rapidly develop towards integration and intelligence, their functional complexity is constantly increasing, hardware and software are highly coupled, and product iteration cycles are becoming increasingly shorter. To ensure product reliability, the traditional Failure Mode and Effect Analysis (FMEA) method is widely used in the industry.

[0020] Currently, reliability analysis and prediction mainly employ the following methods: First, the traditional FMEA analysis method, which uses spreadsheets or dedicated software to record failure modes, relying primarily on engineers' personal experience to identify failures, assess their probability of occurrence and severity, and rank risks using the Risk Priority Number (RPN); second, statistical failure prediction methods, which utilize statistical models such as the Weibull distribution to analyze historical data to predict product lifespan; and third, machine learning-based failure detection methods, which often use sensor data to train models, focusing on monitoring and diagnosing failures during equipment operation.

[0021] However, the aforementioned solutions have significant limitations when facing the complex R&D scenarios of today's consumer electronics. Specifically, firstly, traditional FMEA methods rely heavily on engineers' personal experience, are time-consuming, and are difficult to adapt to the rapid pace of product iteration. They also cannot quantify the risks of multiple concurrent failures, such as the probability of a component simultaneously exhibiting bulging and short circuits. Secondly, existing statistical methods and operational detection methods are mostly used for post-event evaluation or real-time monitoring, lacking the ability to proactively predict based on component attributes during the design phase. Furthermore, for products like learning machines that integrate high-performance processors, large-capacity batteries, and complex software systems, traditional methods struggle to systematically identify fault propagation paths across hardware, software, and levels. This makes it difficult for engineers to clarify, from the initial design stage, how the failure of underlying components propagates tier by tier and ultimately affects the overall functionality of the device.

[0022] To address this issue, the present invention provides a fault prediction method that aims to solve the problems of time-consuming and inefficient manual assessment, difficulty in quantifying the risks of multiple failure modes, and lack of effective system-level fault propagation analysis methods in existing technologies. This method utilizes component attribute data of the product under test during the design phase to predict the probability of multiple failure modes, and performs root cause inference through a fault hierarchy model containing multi-layered topology. This enables the visualization and systematic analysis of fault propagation paths from underlying components to the entire machine level, thereby providing accurate and objective decision-making basis for improving reliability during the product design phase.

[0023] Figure 1 This is a flowchart illustrating the fault prediction method provided by the present invention, which is applied to a fault prediction system. This method / system is primarily used in the research and development phase of consumer electronics products, such as smartphones, tablets, and educational devices, to address the problems of data fragmentation and prediction lag in traditional design processes. Figure 1 As shown, the method includes: Step 110: Determine the component attribute data of the product under test during the design phase; Step 120: Perform fault prediction based on component attribute data to obtain the fault probability distribution; the fault probability distribution contains the fault probabilities of multiple fault modes. Step 130: Based on the fault probability distribution, apply the fault stratification model to perform fault root cause inference and obtain the fault propagation path corresponding to the fault modes involved in the product under test. The fault stratification model is configured with a multi-layer topology from the product component level to the product whole machine level, and the transmission relationship of the fault in the multi-layer topology from the product component level to the product whole machine level.

[0024] Specifically, during product development, when the product under test (BUT) is in the design stage after completing its conceptual definition but before entering mass production, a Bill of Materials (BOM) and Computer-Aided Design (CAD) drawings have typically been generated. At this point, it is necessary to determine the component attribute data of the BUT. That is, the fault prediction system can parse the BOM or interface with the Product Lifecycle Management (PLM) system to extract the detailed characteristics of each component constituting the BUT. This is the component attribute data of the BUT during the design stage. This data includes not only the physical attributes of the components, such as whether the battery uses lithium polymer or lithium iron phosphate materials, whether the packaging process is soft-pack or hard-pack, and the specifications and dimensions of capacitors and resistors, but also supply chain attributes, such as supplier name, production batch, and place of origin, to distinguish the quality differences between different suppliers, as well as design parameters, such as the component's rated voltage, operating temperature range, and design life.

[0025] After obtaining the component attribute data, the system does not simply query historical fault records, but performs a quantitative evaluation of the components in the product under test. This is because in practical applications, a component, such as a lithium battery, may exhibit multiple fault modes simultaneously; for example, it may experience battery bulging, as well as cycle life degradation and short circuits at the same time.

[0026] Therefore, in this embodiment of the invention, when using component attribute data for fault prediction, it does not simply predict whether there is a fault risk or whether a fixed fault mode, such as battery bulging, exists. Instead, it performs multi-dimensional predictions, predicting whether any one or more of various fault modes exist and what the specific fault probability is, thus obtaining a fault probability distribution. This fault probability distribution can be represented as a probability set or vector containing multiple dimensions, each indicating the likelihood of the corresponding component experiencing various fault modes. For example, for a specific component, such as a lithium battery, the fault probability distribution output by the system may explicitly indicate that the probability of battery bulging is 12%, the probability of cycle life degradation is 29%, and the probability of a short circuit is 3%. This data-driven output method achieves accurate quantification of product design reliability risks.

[0027] Subsequently, in order to assess the specific impact of these underlying faults on the final product, this embodiment of the invention can utilize a pre-built fault hierarchy model for in-depth simulation. This model is essentially an architecture describing the internal logical and physical connections of the product, configured with a multi-layered topology. Figure 2 This is an example diagram of the multi-layer topology provided by the present invention, such as... Figure 2 As shown, this topology, based on the complexity of the product, clearly divides the hierarchy from bottom to top, from the product component level (such as chips, battery cells, etc.) to the intermediate component module level (such as batteries, camera modules, etc.), subsystem level (such as impact systems, power systems, etc.), and system level (such as hardware motherboards, software systems, etc.), ultimately converging at the overall product level (such as smartphones). Simultaneously, the fault hierarchy model also establishes the transmission relationships between these levels, clarifying how a failure at a lower level leads to functional abnormalities at a higher level. For example, a short circuit in a battery cell causes the battery to bulge, which in turn causes the entire back cover of the device to lift.

[0028] Based on this, the system uses the fault probability distribution predicted in the previous step as a basis to perform bottom-up fault root cause deduction using a fault hierarchy model. By traversing the multi-layered topology, the system can finally obtain the complete fault propagation path. This path clearly shows how a fault starts from a component at the bottom layer, is amplified or propagated through modules, subsystems, and the system, and finally causes specific failure consequences at the whole machine level, such as system crashes, fires, or functional loss.

[0029] The fault prediction method provided by this invention directly utilizes component attribute data to predict the probability of various fault modes during the design phase, and combines the predicted probability distribution with a fault hierarchical model with a multi-layer topology for system-level deduction. This not only achieves a leap from qualitative experience analysis to quantitative data prediction, but more importantly, it can transform isolated component risks into visualized fault propagation paths. This allows designers to clearly see the specific impact chain of underlying design defects on the overall reliability of the machine, and thus perform targeted optimizations before product manufacturing, greatly improving R&D efficiency and the final quality of the product.

[0030] Based on the above embodiments, the fault hierarchical model is used to deduce the fault origin based on the fault probability distribution, obtain the fault propagation origin, and based on the transmission relationship, traverse upwards from the level where the fault propagation origin is located in the multi-layer topology structure until the product whole machine level, to obtain the fault propagation path.

[0031] Specifically, in practical applications, fault root cause deduction is not a random search, but rather a process that first determines the starting point of the deduction based on the fault probability distribution, and then performs a hierarchical traversal in a multi-layered topology to ultimately generate a complete fault propagation path.

[0032] In detail, the system analyzes the failure probability distribution to determine the starting point for failure inference. That is, by using set failure thresholds, such as 5% or 10%, or pre-set filtering mechanisms, it can accurately locate those failure modes and corresponding specific failure events with a high probability of occurrence from numerous components and potential failure modes, such as power management chip overheating failure or connector contact failure. These filtered failure modes or high-risk failure events serve as the starting point for failure inference in the fault hierarchy model. Determining this starting point not only locks in the specific physical location (which component) but also the specific physical state (which failure mode).

[0033] Next, the system will initiate an upward traversal mechanism based on the multi-layered topology and transitive relationships in the fault hierarchy model. This multi-layered topology can be understood as an inverted fault tree, where the leaf nodes correspond to the bottom-level product component level, the root node to the top-level product system level, and the branches correspond to the component module level, subsystem level, and system level, respectively. The transitive relationships are the logical connections between these nodes, reflecting the causal logic of lower-level failures leading to upper-level anomalies; for example, chip overheating causing the motherboard to shut down protectively.

[0034] Under this mechanism, the system starts from the level where the fault propagation originates, usually the lowest-level product component level, and logically deduces upwards along the transmission relationship. This process is like following a vine to find a melon; the system sequentially determines which module's function will be abnormal due to the component's failure, which will then lead to the collapse of which subsystem and the system, until finally traversing upwards to the overall product level. Through this rigorous step-by-step deduction, the system connects scattered failure points, ultimately generating a complete fault propagation path. For example, capacitor short circuit [product component level] - power module output interruption [component module level] - power system power supply abnormality [subsystem level] - hardware motherboard cannot start [system level] - the entire machine cannot be powered on [product level].

[0035] In this embodiment of the invention, the failure risk of components at the micro level can be accurately mapped to the overall performance of the machine at the macro level. By deducing the fault starting point, the high-risk source can be quickly focused. Combined with the upward traversal mechanism, the chain reaction process of the fault can be restored. This not only helps designers understand what is wrong, but also intuitively shows what the consequences will be. Thus, the key issues that may lead to serious failure of the whole machine can be accurately identified in the early stage of design, thereby providing highly targeted logical support for the optimization of reliability design.

[0036] Based on the above embodiments, fault prediction is performed based on component attribute data to obtain a fault probability distribution, and then the process further includes: When the failure probability distribution indicates that there are risky failure modes among the failure modes involved in the product under test, physical field simulation is performed based on component attribute data. The simulation cloud map obtained from the physical field simulation is correlated and displayed with the fault probability distribution.

[0037] Specifically, in actual product development processes, data prediction alone is sometimes insufficient to completely eliminate engineers' concerns about complex physical phenomena. Therefore, in this embodiment of the invention, after obtaining the failure probability distribution through failure prediction, a physical verification mechanism is further introduced to achieve complementary advantages between data prediction and physical simulation.

[0038] Specifically, when the failure probability distribution indicates the presence of a risky failure mode among the failure modes involved in the product under test, the system will trigger the verification process. Here, a risky failure mode refers to a failure mode whose failure probability exceeds a preset safety threshold, such as 20%, and whose consequences are extremely serious, such as the risk of fire or explosion. For example, if failure prediction determines that the probability of overheating failure of the cooling module of a new processor model is high, this mode is a risky failure mode.

[0039] Once such a risky failure mode is identified, the system goes beyond simply providing a failure probability value. It further utilizes component attribute data, such as component dimensions, material thermal conductivity, specific heat capacity, and Young's modulus, to drive computer-aided engineering (CAE) software to perform targeted physical field simulations. This process is triggered automatically or semi-automatically. The system converts the component attribute data into the boundary conditions and material parameters required for the simulation, performing simulations such as thermal fluid simulations, structural mechanics simulations, and electromagnetic field simulations. For example, for the aforementioned overheating failure risk, the system will call thermal simulation tools to calculate the temperature field distribution of the processor's heat dissipation module under specific operating conditions.

[0040] After the simulation is completed, the system will generate intuitive simulation cloud maps, such as temperature distribution cloud maps and stress deformation cloud maps. It is worth noting that the system does not display these cloud maps separately, but rather displays them in conjunction with each other. This means that on the system interface, the failure probability distribution based on data-driven prediction, such as the failure probability of 85% corresponding to overheating failure, will be presented side by side with the simulation cloud map obtained based on the physical field simulation, such as the core area temperature displayed in red, reaching as high as 95℃, or displayed in the same view through layer overlay.

[0041] In this embodiment of the invention, fault prediction enables rapid initial screening, identifying high-risk points from a massive amount of design data. This is then verified using physical field simulation. The accuracy of the fault prediction is validated by a simulation cloud map that conforms to physical laws. The results of both are displayed together, allowing designers to cross-reference fault risks from both statistical (probability) and physical (cloud map) dimensions, avoiding potential misjudgments from a single algorithm. This also intuitively assists in quickly locating the physical root causes of design defects, thereby guiding design optimization more efficiently.

[0042] Based on the above embodiments, the fault propagation path corresponding to the fault modes involved in the product under test is obtained, and then the following steps are also included: Based on the failure probability distribution and failure propagation path, a visual risk map is generated; the visual risk map includes at least one of the following: risk matrix heat map, component risk radar map, and failure propagation path map. The risk matrix heatmap displays the failure probability in the failure probability distribution by the product model or project to which the product belongs and the failure modes involved. The component risk radar chart displays the failure probability of a single component in the product under test under the relevant failure modes; The fault propagation path diagram shows the complete fault propagation chain reflected by the fault propagation path, with the fault mode involved as the root cause, and the faults are transmitted step by step from the intermediate level of the multi-layer topology to the whole product level.

[0043] Specifically, in order to transform the failure probability distribution and failure propagation path into easily understandable decision support information, in this embodiment of the invention, after obtaining the failure propagation path, the system automatically generates an intuitive visual risk map based on the failure probability distribution and failure propagation path. This visual risk map is not uniform but provides multiple representations based on different analytical perspectives, including at least one of the following: risk matrix heatmap, component risk radar chart, and failure propagation path map, to meet different needs from macro-management to micro-diagnosis.

[0044] For macro-level risk management across product lines or projects, the system can generate a risk matrix heatmap, a two-dimensional data display chart. Specifically, the risk matrix heatmap uses the product model or project to which the product belongs (e.g., the X series of learning machines, the Y series of mobile phones) as one dimension (horizontal axis) and the relevant failure modes (e.g., screen cracking, battery swelling) as another dimension (vertical axis). At the intersection cells of the matrix, the probability of failure is visually displayed through color intensity (e.g., dark red for high risk, light green for low risk). This heatmap allows managers to instantly identify which project or which failure is the most problematic area for the current product line.

[0045] For in-depth analysis of specific components, the system provides a component risk radar chart. That is, when engineers need to focus on a single component in the product under test, such as a power management chip on a motherboard for detailed evaluation, this chart displays the failure probability of that component under various failure modes on multiple radial axes. For example, by observing the shape of the radar chart, engineers can immediately see that while the chip has a very low short-circuit risk (recessed), it has an extremely high overheat risk (protruding), thus quickly identifying the component's performance weaknesses.

[0046] To visually represent the causal chain of a fault, the system generates a fault propagation path diagram. This diagram is not merely a static structural diagram, but a dynamically generated logical flow diagram based on a fault hierarchy model. That is, it clearly depicts the complete fault propagation chain, using the fault mode involved as the root cause (i.e., the fault starting point, such as capacitor breakdown). It shows how the fault propagates domino-like from the intermediate levels of the multi-layered topology (i.e., component module level, subsystem level, and system level) to the overall product level (e.g., the entire device fails to power on). This visualized path diagram helps designers clearly see how a small, underlying design flaw amplifies step by step and ultimately destroys the user experience.

[0047] In this embodiment of the invention, by constructing a multi-dimensional visualization system including heatmaps, radar charts, and path maps, the threshold for data understanding is greatly reduced. Specifically, the risk matrix heatmap helps managers conduct horizontal comparisons and resource allocation across multiple projects; the component risk radar chart helps engineers accurately pinpoint the weak points of individual components; and the fault propagation path map provides a panoramic perspective for causal analysis.

[0048] Based on the above embodiments, fault prediction is performed based on component attribute data to obtain a fault probability distribution, including: Based on component attribute data, a fault prediction model is applied to predict faults and obtain the fault probability distribution. The fault prediction model includes multiple fault classifiers, each of which corresponds to a fault mode. The fault prediction model is used to determine the fault probability distribution based on the fault probability obtained from the fault prediction of each fault classifier. The fault prediction model is trained based on a component attribute knowledge graph; The component attribute knowledge graph is constructed based on the following steps: Knowledge is extracted from the sample product manual to obtain sample triples; Construct a component attribute knowledge graph based on sample triples; The component attribute knowledge graph includes component nodes, attribute nodes, and failure mode nodes, as well as attribute associations between component nodes and attribute nodes, cause associations between component nodes and failure mode nodes, and affiliation associations between component nodes.

[0049] Specifically, when performing fault prediction, the system does not use general simple regression analysis, but instead uses a fault prediction model based on machine learning to achieve multi-dimensional prediction, thereby obtaining the fault probability distribution.

[0050] The fault prediction model here has a specially designed internal architecture to handle complex scenarios where components may experience multiple faults simultaneously. That is, it is not a single black-box model, but rather a structure comprising multiple fault classifiers. Logically, each fault classifier corresponds to a fault mode. For example, for a battery component, the model might encapsulate a fault classifier specifically for predicting bulging, and another specifically for predicting short circuits. These fault classifiers work in parallel, calculating the probability of their respective fault modes occurring. Finally, the results of these parallel calculations are aggregated to obtain the fault probability distribution.

[0051] In detail, when performing fault prediction, the component attribute data can be processed before being input into the model. For numerical data, preprocessing such as missing value imputation and standardization can be performed, while for categorical data, one-hot encoding can be used. Then, the results of the two processing steps can be concatenated to form a feature vector. This feature vector can then be input into the fault prediction model, which employs a problem transformation strategy. Figure 3 This is a strategy example diagram of the problem transformation strategy provided by the present invention, such as... Figure 3 As shown, this model uses MultiOutputClassifier as the meta-estimator. Internally, the model encapsulates multiple (e.g., n) parallel fault classifiers. Each classifier predicts the probability of a fault mode and outputs a probability value. Combining the probability values ​​output by these parallel classifiers yields a probability vector, i.e., the fault probability distribution. This strategy ensures the model can accurately handle multi-label classification tasks and output vectorized results.

[0052] However, it is worth noting that before applying the fault prediction model for fault prediction, in order to improve the model's performance and thus optimize its performance in practical applications, this embodiment of the invention also requires pre-training the model. Specifically, the fault prediction model is pre-trained based on a component attribute knowledge graph, which structurally stores a massive amount of historical engineering experience.

[0053] The construction process of the component attribute knowledge graph here may specifically include the following steps: First, the system needs to address the issue of data sources. That is, it needs to extract knowledge based on sample product manuals. Here, sample product manuals refer to various unstructured historical technical documents, such as FMEA reports, component specifications, failure analysis reports, etc., from historical projects. The system can use natural language processing technologies, such as entity recognition and entity relation extraction, to extract core information from these documents in order to construct sample triples, i.e., entity-relationship-entity structures.

[0054] The system can then use this sample triple to construct a component attribute knowledge graph. Figure 4 This is a partial simplified diagram of the component attribute knowledge graph provided by the present invention, such as... Figure 4 As shown, this constructed graph network contains three types of core nodes: component nodes (such as lithium batteries) representing specific component entities, attribute nodes (such as capacity 4000mAh) representing parameter characteristics, and fault mode nodes (such as battery bulging) representing failure types.

[0055] At the same time, these nodes are also connected by explicit semantic relationships, namely the attribute association between component nodes and attribute nodes, such as lithium battery - attribute - capacity 4000mAh; the cause association between component nodes and fault mode nodes, such as lithium battery - may cause - battery swelling; and the affiliation association between component nodes, such as lithium battery - belongs to - power system, which reflects the hierarchical structure between components.

[0056] In this embodiment of the invention, by constructing a structured component attribute knowledge graph, unstructured historical documents are transformed into a machine-readable knowledge network, solving the problem of the difficulty in accumulating and reusing engineering experience. Furthermore, the fault prediction model trained based on this graph, which includes a multi-classifier, can make full use of this historical knowledge to perform multi-dimensional concurrent fault prediction for new design schemes. This ensures that the fault prediction model has a strong historical data support and endows the system with the computational ability to handle complex multi-fault modes, significantly improving the accuracy and comprehensiveness of fault prediction.

[0057] Based on the above embodiments, based on component attribute data, a fault prediction model is applied to predict faults and obtain a fault probability distribution. The process then further includes: Determine the actual operating data of the product under test throughout its entire lifecycle, which includes the manufacturing stage, testing stage, after-sales stage, and usage stage; Extract real fault data from real operational data, and determine the prediction bias based on the real fault data and fault probability distribution; If the prediction deviation exceeds the deviation threshold, and / or the amount of actual operating data exceeds the data volume threshold, the fault prediction model and component attribute knowledge graph are updated based on the actual operating data.

[0058] Specifically, during the product development and design phase, the predicted probability distribution provided by the fault prediction model is essentially a prediction based on historical experience. To ensure that this prediction becomes more accurate over time, a closed-loop optimization mechanism is constructed in this embodiment of the invention.

[0059] In detail, Figure 5 This is a general framework diagram of the fault prediction method provided by the present invention, as shown below. Figure 5 As shown, after completing fault prediction, the system enters the continuous monitoring and optimization phase. Specifically, the system continuously tracks and acquires the actual operating data of the product under test throughout its entire lifecycle. This "entire lifecycle" breaks down the traditional data barriers between R&D and after-sales, covering the manufacturing phase (such as yield data in the Manufacturing Execution System), the testing phase (such as reliability test data in the laboratory), the after-sales phase (such as maintenance work orders in the Customer Relationship Management System), and the usage phase (such as operating logs returned by user-end devices).

[0060] As this data is continuously fed back, the system utilizes data mining, keyword matching, and other technologies to extract real-world fault data from this actual operational data. For example, the system might identify from repair work orders that a certain batch of mobile phones did indeed experience 1,000 battery swelling incidents six months after their release. Next, the system compares this real-world fault data with the fault probability distribution obtained from fault prediction during the design phase to calculate the difference between the two, i.e., the prediction bias. For instance, if the initial fault prediction model predicted a battery swelling probability of only 1%, but the actual fault data shows a probability as high as 10%, this results in a significant bias.

[0061] Next, the system determines whether to trigger the update mechanism based on the magnitude of the prediction deviation and the scale of the actual operational data. Specifically, the system will trigger the update mechanism when the prediction deviation exceeds a deviation threshold (indicating the model is inaccurate) and / or when the amount of actual operational data exceeds a data volume threshold (indicating sufficient new samples have been accumulated). At this point, the system will use a closed-loop iterative engine to update the fault prediction model and component attribute knowledge graph using the actual operational data. This not only updates the model's parameters, allowing it to learn new fault patterns, but also updates the structure of the knowledge graph, writing the newly added fault modes and their relationships into the graph, thus expanding the knowledge base.

[0062] Conversely, if the prediction deviation does not exceed the deviation threshold and the actual data volume does not exceed the data volume threshold, the system will not trigger the update mechanism but will continue to monitor.

[0063] In this embodiment of the invention, real data generated during the manufacturing, testing, after-sales, and usage stages are used to verify the prediction results in the early design stage, and deviations are calculated and updates are triggered accordingly. This makes the fault prediction model and knowledge graph no longer static and unchanging, but can continuously self-correct and self-evolve like human experts as the product is on the market and real cases are accumulated. This not only ensures that the model maintains a high accuracy rate when facing new products and new working conditions, but also realizes the dynamic value-added and long-term reuse of enterprise engineering knowledge.

[0064] Based on the above embodiments, and based on real operational data, the fault prediction model and component attribute knowledge graph are updated, including: Determine the update mode based on the data volume characteristics of real operational data; When the update mode is incremental learning, the parameters of the fault prediction model are fine-tuned based on real operating data. When the update mode is retraining, the fault prediction model is retrained based on real running data and component attribute knowledge graph. Identify new fault modes and their relationships in real-world operational data, and update the new fault modes and their relationships in the component attribute knowledge graph.

[0065] Specifically, the process of updating the fault prediction model and component attribute knowledge graph based on real operational data may include: When the update mechanism is triggered, the system first determines the update mode based on the characteristics of the actual operational data. These characteristics primarily refer to the size, distribution density, and degree of difference between the actual operational data and the data from the initial design phase. Based on these characteristics, the system intelligently decides whether to adopt a fast or slow update strategy.

[0066] Specifically, if the data scale is small or the data distribution does not change drastically, such as only a small number of maintenance records have been added, the system will adopt a fast update strategy, i.e., incremental learning. In this case, the system uses a lightweight online learning algorithm and a closed-loop iterative engine to fine-tune the parameters of the fault prediction model using real-world data. This method requires very little computing power to quickly adjust the model weights, adapting them to the latest data trends.

[0067] Conversely, if the accumulated data volume is enormous, or if a disruptive data distribution emerges, such as the adoption of a completely new material system that causes the model to completely fail, the system will adopt a slow update strategy, i.e., retraining. In this case, the system will initiate a full update process, merging the accumulated real-world operational data and component attribute knowledge graph (which contains all historically accumulated knowledge) with the old data. Through a closed-loop iterative engine, the merged dataset will be used to retrain the fault prediction model. This is equivalent to a complete overhaul of the model; although time-consuming, it ensures that the model thoroughly grasps the new patterns.

[0068] Meanwhile, regardless of the mode adopted, the component attribute knowledge graph is updated synchronously. That is, the system will deeply scan the accumulated real-world operational data to identify new fault modes and their relationships. For example, the system may discover a fault mode that has never been recorded before, such as low-temperature foldable screen breakage. Once it is confirmed that this is a new knowledge point, the system will immediately write it into the knowledge graph, that is, update the new fault mode and its relationships into the component attribute knowledge graph to achieve knowledge expansion.

[0069] In this embodiment of the invention, a dual-mode update mechanism combining incremental learning and retraining resolves the contradiction between the timeliness and accuracy of model updates. Incremental learning enables rapid response and maintains model accuracy at low cost during daily operation; retraining enables thorough iteration at critical nodes, preventing model aging. Simultaneously, the dynamic expansion mechanism of the knowledge graph ensures that every failure is transformed into knowledge assets, truly making the fault prediction system smarter with use.

[0070] Figure 6 This is an overall architecture diagram of the fault prediction system provided by the present invention, as shown below. Figure 6 As shown, the system constructs an integrated human-machine collaboration platform, aiming to achieve intelligent management throughout the entire process from design input to risk control. At the core application level, the system deploys an intelligent FMEA auxiliary interface. Engineers only need to fill in component attribute data in the interface, and the background can automatically call the fault prediction model, providing real-time feedback on the predicted fault probability distribution and matching similar historical cases on the interface, thereby assisting designers in quickly identifying potential hazards.

[0071] Furthermore, the platform is equipped with a virtual simulation verification interface and a visualization dashboard. Through the simulation interface, the system can link with CAE software to trigger physical field simulations with a single click for predicted risks, and display the generated simulation cloud map in association with the predicted failure probability distribution, achieving dual verification of data and physical mechanisms. At the same time, through risk matrix heatmaps, component risk radar charts, and failure propagation path diagrams, the system transforms complex risk data into intuitive charts, helping designers clearly control the reliability status of products from different dimensions.

[0072] The fault prediction device provided by the present invention is described below. The fault prediction device described below and the fault prediction method described above can be referred to in correspondence.

[0073] Figure 7 This is a schematic diagram of the fault prediction device provided by the present invention, as shown below. Figure 7 As shown, the device includes: The data determination unit 710 is used to determine the component attribute data of the product under test during the design phase. The fault prediction unit 720 is used to perform fault prediction based on the component attribute data to obtain a fault probability distribution; the fault probability distribution includes the fault probabilities of multiple fault modes. The root cause inference unit 730 is used to perform root cause inference based on the fault probability distribution and the fault hierarchy model to obtain the fault propagation path corresponding to the fault mode involved in the product under test. The fault hierarchy model is configured with a multi-layer topology from the product component level to the product whole machine level, and the propagation logic of the fault being passed step by step from the product component level to the product whole machine level in the multi-layer topology.

[0074] The fault prediction device provided by this invention directly uses component attribute data to predict the probability of various fault modes during the design stage, and combines a fault hierarchical model with a multi-layer topology to perform system-level inference based on the predicted probability distribution. This not only achieves a leap from qualitative experience analysis to quantitative data prediction, but more importantly, it can transform isolated component risks into visualized fault propagation paths. This allows designers to clearly see the specific impact chain of underlying design defects on the overall reliability of the machine, and thus carry out targeted optimizations before product manufacturing, greatly improving R&D efficiency and the final quality of the product.

[0075] Based on the above embodiments, the fault hierarchical model is used to deduce the fault origin based on the fault probability distribution, obtain the fault deduction origin, and based on the transmission relationship, traverse upwards from the level where the fault deduction origin is located in the multi-layer topology until the product whole machine level to obtain the fault propagation path.

[0076] Based on the above embodiments, the device further includes a simulation verification unit, used for: When the failure probability distribution indicates the presence of a risky failure mode among the failure modes involved in the product under test, physical field simulation is performed based on the component attribute data. The simulation cloud map obtained from the physical field simulation is correlated and displayed with the fault probability distribution.

[0077] Based on the above embodiments, the device further includes a risk visualization unit, used for: Based on the failure probability distribution and the failure propagation path, a visual risk map is generated; the visual risk map includes at least one of a risk matrix heat map, a component risk radar map, and a failure propagation path map. The risk matrix heatmap uses the product model or project to which the product under test belongs, as well as the relevant failure modes, as dimensions to display the failure probability in the failure probability distribution. The component risk radar chart shows the failure probability of a single component in the product under test under the relevant failure modes. The fault propagation path diagram illustrates the complete fault propagation chain reflected by the fault propagation path, with the fault mode involved as the root cause, and the faults being passed step by step from the intermediate level of the multi-layer topology to the product whole-machine level.

[0078] Based on the above embodiments, the fault prediction unit 720 is used for: Based on the component attribute data, a fault prediction model is applied to predict faults and obtain the fault probability distribution. The fault prediction model includes multiple fault classifiers, each of which corresponds to a fault mode. The fault prediction model is used to determine the fault probability distribution based on the fault probability obtained from the fault prediction of each fault classifier. The fault prediction model is trained based on a component attribute knowledge graph. The component attribute knowledge graph is constructed based on the following steps: Knowledge is extracted from the sample product manual to obtain sample triples; Based on the sample triples, construct the component attribute knowledge graph; The component attribute knowledge graph includes component nodes, attribute nodes, and fault mode nodes, as well as attribute associations between component nodes and attribute nodes, causal associations between component nodes and fault mode nodes, and attribution associations between component nodes.

[0079] Based on the above embodiments, the device further includes an update and optimization unit, used for: Determine the actual operating data of the product under test throughout its entire lifecycle, which includes the manufacturing stage, testing stage, after-sales stage, and usage stage; Real fault data is extracted from the real operating data, and the prediction deviation is determined based on the real fault data and the fault probability distribution. If the prediction deviation exceeds the deviation threshold, and / or the amount of actual operating data exceeds the data volume threshold, the fault prediction model and the component attribute knowledge graph are updated based on the actual operating data.

[0080] Based on the above embodiments, the update and optimization unit is used for: Based on the data volume characteristics of the actual operational data, the update mode is determined; When the update mode is incremental learning, the parameters of the fault prediction model are fine-tuned based on the real operating data. When the update mode is retraining, the fault prediction model is retrained based on the real running data and the component attribute knowledge graph. Identify new fault modes and their relationships in the real operating data, and update the new fault modes and their relationships in the component attribute knowledge graph.

[0081] Figure 8 An example is a schematic diagram of the physical structure of an electronic device, such as... Figure 8 As shown, the electronic device may include: a processor 810, a communication interface 820, a memory 830, and a communication bus 840, wherein the processor 810, the communication interface 820, and the memory 830 communicate with each other through the communication bus 840. The processor 810 can call logical instructions in the memory 830 to execute a fault prediction method, which includes: determining component attribute data of the product under test in the design stage; performing fault prediction based on the component attribute data to obtain a fault probability distribution; the fault probability distribution includes the fault probabilities of multiple fault modes; based on the fault probability distribution, applying a fault hierarchy model to perform fault root cause inference to obtain the fault propagation path corresponding to the fault modes involved in the product under test; the fault hierarchy model is configured with a multi-layer topology from the product component level to the product whole level, and the transmission relationship of the fault being passed step by step from the product component level to the product whole level in the multi-layer topology.

[0082] Furthermore, the logical instructions in the aforementioned memory 830 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0083] On the other hand, the present invention also provides a computer program product, the computer program product including a computer program stored on a non-transitory computer-readable storage medium, the computer program including program instructions, when the program instructions are executed by a computer, the computer is able to execute the fault prediction method provided by the above methods, the method including: determining component attribute data of the product under test in the design stage; performing fault prediction based on the component attribute data to obtain a fault probability distribution; the fault probability distribution includes fault probabilities of multiple fault modes; based on the fault probability distribution, applying a fault hierarchical model to perform fault root cause inference to obtain the fault propagation path corresponding to the fault modes involved in the product under test; the fault hierarchical model is configured with a multi-layer topology from the product component level to the product whole machine level, and a transmission relationship in the multi-layer topology from the product component level to the product whole machine level step by step.

[0084] In another aspect, the present invention also provides a non-transitory computer-readable storage medium storing a computer program thereon, which, when executed by a processor, is implemented to perform the fault prediction method provided by the methods described above. The method includes: determining component attribute data of the product under test during the design phase; performing fault prediction based on the component attribute data to obtain a fault probability distribution; the fault probability distribution containing fault probabilities of multiple fault modes; and applying a fault hierarchy model to perform root cause inference based on the fault probability distribution to obtain the fault propagation path corresponding to the fault modes involved in the product under test; the fault hierarchy model is configured with a multi-layer topology from the product component level to the product whole-machine level, and a transmission relationship in the multi-layer topology from the product component level to the product whole-machine level.

[0085] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. 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 embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0086] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0087] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A fault prediction method, characterized in that, include: Determine the component attribute data of the product under test during the design phase; Based on the component attribute data, fault prediction is performed to obtain the fault probability distribution; The failure probability distribution contains failure probabilities for multiple failure modes. Based on the fault probability distribution, a fault stratification model is applied to infer the root cause of the fault and obtain the fault propagation path corresponding to the fault modes involved in the product under test. The fault hierarchy model is configured with a multi-layer topology from the product component level to the product whole machine level, and a transmission relationship in which faults are passed step by step from the product component level to the product whole machine level in the multi-layer topology.

2. The fault prediction method according to claim 1, characterized in that, The fault hierarchical model is used to deduce the fault origin based on the fault probability distribution, obtain the fault propagation origin, and based on the propagation relationship, traverse upwards from the level where the fault propagation origin is located in the multi-layer topology until the product whole machine level to obtain the fault propagation path.

3. The fault prediction method according to claim 1, characterized in that, The step of performing fault prediction based on the component attribute data to obtain a fault probability distribution further includes: When the failure probability distribution indicates the presence of a risky failure mode among the failure modes involved in the product under test, physical field simulation is performed based on the component attribute data. The simulation cloud map obtained from the physical field simulation is correlated and displayed with the fault probability distribution.

4. The fault prediction method according to any one of claims 1 to 3, characterized in that, After obtaining the fault propagation path corresponding to the fault modes involved in the product under test, the method further includes: Based on the failure probability distribution and the failure propagation path, a visual risk map is generated; the visual risk map includes at least one of a risk matrix heat map, a component risk radar map, and a failure propagation path map. The risk matrix heatmap uses the product model or project to which the product under test belongs, as well as the relevant failure modes, as dimensions to display the failure probability in the failure probability distribution. The component risk radar chart shows the failure probability of a single component in the product under test under the relevant failure modes. The fault propagation path diagram illustrates the complete fault propagation chain reflected by the fault propagation path, with the fault mode involved as the root cause, and the faults being passed step by step from the intermediate level of the multi-layer topology to the product whole-machine level.

5. The fault prediction method according to any one of claims 1 to 3, characterized in that, The step of performing fault prediction based on the component attribute data to obtain a fault probability distribution includes: Based on the component attribute data, a fault prediction model is applied to predict faults and obtain the fault probability distribution. The fault prediction model includes multiple fault classifiers, each of which corresponds to a fault mode. The fault prediction model is used to determine the fault probability distribution based on the fault probability obtained from the fault prediction of each fault classifier. The fault prediction model is trained based on a component attribute knowledge graph. The component attribute knowledge graph is constructed based on the following steps: Knowledge is extracted from the sample product manual to obtain sample triples; Based on the sample triples, construct the component attribute knowledge graph; The component attribute knowledge graph includes component nodes, attribute nodes, and fault mode nodes, as well as attribute associations between component nodes and attribute nodes, causal associations between component nodes and fault mode nodes, and attribution associations between component nodes.

6. The fault prediction method according to claim 5, characterized in that, The step of applying a fault prediction model to predict faults based on the component attribute data to obtain the fault probability distribution further includes: Determine the actual operating data of the product under test throughout its entire lifecycle, which includes the manufacturing stage, testing stage, after-sales stage, and usage stage; Real fault data is extracted from the real operating data, and the prediction deviation is determined based on the real fault data and the fault probability distribution. If the prediction deviation exceeds the deviation threshold, and / or the amount of actual operating data exceeds the data volume threshold, the fault prediction model and the component attribute knowledge graph are updated based on the actual operating data.

7. The fault prediction method according to claim 6, characterized in that, The step of updating the fault prediction model and the component attribute knowledge graph based on the actual operating data includes: Based on the data volume characteristics of the actual operational data, the update mode is determined; When the update mode is incremental learning, the parameters of the fault prediction model are fine-tuned based on the real operating data. When the update mode is retraining, the fault prediction model is retrained based on the real running data and the component attribute knowledge graph. Identify new fault modes and their relationships in the real operating data, and update the new fault modes and their relationships in the component attribute knowledge graph.

8. A fault prediction device, characterized in that, include: The data determination unit is used to determine the component attribute data of the product under test during the design phase. The fault prediction unit is used to predict faults based on the component attribute data to obtain a fault probability distribution; the fault probability distribution contains the fault probabilities of multiple fault modes. The root cause inference unit is used to perform root cause inference based on the fault probability distribution and the fault stratification model to obtain the fault propagation path corresponding to the fault mode involved in the product under test. The fault hierarchy model is configured with a multi-layer topology from the product component level to the product whole machine level, and a transmission logic for the fault to be passed step by step from the product component level to the product whole machine level in the multi-layer topology.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the fault prediction method as described in any one of claims 1 to 7.

10. A non-transitory 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 fault prediction method as described in any one of claims 1 to 7.

Images