Engine fault prediction method and system
By constructing a hierarchical fault association chain knowledge base and performing targeted, time-series progressive identification, the problems of large data processing volume and difficulty in identifying early and subtle faults in existing technologies have been solved, enabling accurate prediction and early warning of engine faults.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- CENT SOUTH UNIV
- Filing Date
- 2026-01-29
- Publication Date
- 2026-05-15
Smart Images
Figure CN122046128A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of engine fault prediction technology, and in particular to an engine fault prediction method and an engine fault prediction system. Background Technology
[0002] Engines are the core power units of construction machinery and heavy vehicles, and their operating status directly determines the reliability and safety of the equipment. With the development of industrial intelligence, fault prediction technology based on monitoring operating parameters has become a core means of engine operation and maintenance.
[0003] Existing engine fault prediction methods mostly employ a multi-parameter parallel monitoring combined with threshold judgment: collecting a large number of operating parameters such as engine speed, oil pressure, vibration, and exhaust gas composition, and setting abnormal thresholds for each parameter. When several parameters exceed the threshold, a fault is predicted. However, this type of method has significant drawbacks: First, it requires high-frequency, indiscriminate collection of all parameters, and all collected data needs to be compared and analyzed with the thresholds, resulting in a large amount of data processing, consuming significant computing resources, and making it difficult to deploy on low-cost edge controllers; Second, the judgment logic is only a comparison between parameters and thresholds, which is easily affected by random parameter fluctuations under varying load and speed conditions, leading to a high false positive rate; Third, when a single parameter is abnormal, it is generally marked as exceeding the limit, making it impossible to trace the root cause of the abnormality and making it difficult to accurately identify early, subtle faults.
[0004] Therefore, existing engine fault prediction methods, due to their large data processing volume and the fact that they only compare each data point with the corresponding threshold, are unable to accurately identify early and minor faults, thus failing to meet the actual needs of engine operation and maintenance. Summary of the Invention
[0005] The main objective of this invention is to provide an engine fault prediction method and system, which aims to solve the technical problem that existing engine fault prediction methods, due to their large data processing volume and the fact that they only compare each data point with the corresponding threshold, are unable to accurately identify early and minor faults, thus failing to meet the actual needs of engine operation and maintenance.
[0006] To achieve the above objectives, the present invention provides an engine fault prediction method, comprising the following steps: Based on engine failure mechanisms and full lifecycle operation data, a hierarchical fault association chain corresponding to different fault types is sorted out to build an engine fault association chain knowledge base. The fault association chain includes the main driving parameter and several associated node parameters, as well as the temporal transmission constraints and anomaly judgment conditions between each parameter. The main driving parameter is the core initiating parameter of the corresponding fault type, the associated node parameters are the characteristic parameters on the fault transmission path, the temporal transmission constraint is the time interval threshold between adjacent parameter anomalies, and the anomaly judgment condition is the abnormal amplitude and fluctuation frequency threshold of each parameter. Real-time acquisition of engine operating data, extraction of monitoring values of main drive parameters, and comparison of monitoring values with preset main drive parameter thresholds; When the monitored value reaches the driving threshold, the fault prediction process is triggered, and the process enters the directional time-series progressive identification mode of the associated node parameters. Based on whether all the parameters of the associated nodes in the fault association chain meet the corresponding anomaly judgment conditions and whether the anomaly timing of each associated node parameter conforms to the propagation constraint, a complete closed-loop confirmation of the fault association chain is performed. When the confirmation is successful, the fault type and fault association chain corresponding to the fault association chain are extracted, and fault prediction and reporting are performed. If any level of associated node parameters does not meet the anomaly judgment conditions, a fault association chain interruption signal is generated. Based on the interruption signal, it is identified whether there is a fault association chain skipping transmission, and engine latent fault prediction is performed.
[0007] Optionally, after the step of constructing an engine fault association chain knowledge base by sorting out hierarchical fault association chains corresponding to different fault types based on engine fault mechanisms and full life cycle operation data, the method further includes: Establish an adaptation strategy between engine operating conditions and main drive parameter thresholds so that the main drive parameter thresholds can be adaptively adjusted according to engine operating conditions; Based on the threshold driven by the main driving parameter, a coupling adjustment rule for the parameters of the associated nodes is established.
[0008] Optionally, after the step of constructing an engine fault association chain knowledge base by sorting out hierarchical fault association chains corresponding to different fault types based on engine fault mechanisms and full life cycle operation data, the method further includes: Regularly collect engine maintenance records and new operational anomaly data; The parameter composition, temporal propagation constraints, and anomaly determination conditions of the fault correlation chain are optimized through incremental learning algorithms. When a new fault type is added, a corresponding new fault association chain is constructed.
[0009] Optionally, the step of triggering the fault prediction process and entering the directional temporal progressive identification mode of associated node parameters when the monitored value reaches the driving threshold includes: Based on the fault association chain knowledge base, the parameters of each association node are collected in a targeted high-frequency manner according to the transmission order from the main driving parameters, the first-level association node parameters to the end association node parameters. For each associated node parameter, determine whether the corresponding monitoring value meets the corresponding anomaly judgment condition, and whether the anomaly occurrence time conforms to the temporal propagation constraint with the preceding parameters; If the parameters of the current associated node complete the anomaly detection, the identification of the parameters of the next level associated node will be triggered.
[0010] Optionally, the step of confirming the complete closed loop of the fault correlation chain includes: Obtain the weight coefficients assigned to each associated node parameter based on the importance of each node parameter in the fault propagation path; The overall matching degree of the fault association chain is calculated based on the weight coefficients of the parameters of each associated node in the fault association chain. The complete closed loop of the fault association chain is confirmed based on whether the overall matching degree of the fault association chain reaches the matching degree threshold.
[0011] Optionally, the method further includes: From the start time of directional high-frequency acquisition of parameters of each associated node to the end time of determining whether to report fault prediction, reduce the acquisition frequency of other engine parameters besides the main drive parameters and associated node parameters.
[0012] Optionally, the step of generating a fault association chain interruption signal if any first-level associated node parameter does not meet the anomaly determination condition, identifying whether there is a fault association chain skipping propagation based on the interruption signal, and performing engine latent fault prediction includes: Obtain the interrupted node and downstream node of the fault association chain; Interrupt node correction data is generated based on the periodic sampling data of the interrupt node parameters; The fault association chain is corrected by interrupted nodes, and the directional time-series progressive identification mode is continued to be executed to the downstream nodes of the interrupted nodes to predict the hidden faults of the engine.
[0013] Optionally, the step of constructing an engine fault association chain knowledge base by sorting out hierarchical fault association chains corresponding to different fault types based on engine fault mechanisms and full life cycle operation data includes: For the target engine model, based on the physical structure and control logic, a set of independent potential fault types is decoupled. ,in, m is the number of fault types; For each fault type Starting from the failure mechanism, an initial set of monitoring parameters consisting of several key performance indicators is determined. ,in, Let j be the j-th initial monitoring parameter in the initial monitoring parameter set for the i-th fault type. , where n is the number of initial monitoring parameters included in the initial monitoring parameter set for the i-th fault type; Obtain the historical full life cycle operation dataset of the target engine, which includes normal operation data and fault operation data labeled with fault type and occurrence time; Using the dataset, for each fault type The corresponding initial monitoring parameter set was subjected to time-series correlation analysis and causal inference analysis to screen out parameters related to the fault type. A subset of parameters exhibits strong causal temporal correlations. Based on the correlation strength and propagation direction, the parameters within this subset are sorted to form fault types. The candidate parameter propagation sequence; Based on fault operation data, the time interval distribution between the first abnormality of the former and the first abnormality of the latter in the candidate parameter propagation sequence is statistically analyzed, and the time interval determined by a specific percentile in the time interval distribution is used as the temporal propagation constraint threshold between the two adjacent parameters. Each fault type The corresponding candidate parameter propagation sequence, time-series propagation constraint threshold, and anomaly judgment conditions for each parameter are stored in a searchable hierarchical graph structure to form a fault association chain knowledge base; among them, the first parameter of the candidate parameter propagation sequence is defined as the main driving parameter, and the remaining parameters are defined as associated node parameters.
[0014] Optionally, the method for each fault type After the step of storing the corresponding candidate parameter propagation sequences, temporal propagation constraint thresholds, and anomaly determination conditions for each parameter in a searchable hierarchical graph structure to form a fault association chain knowledge base, the following steps are also included: Obtain the fault reporting rate, false alarm rate, and fault location accuracy of the fault prediction method. Based on the fault reporting rate, false alarm rate, and fault location accuracy, the parameter subset, temporal propagation constraint threshold, and anomaly judgment conditions that have a strong causal and temporal correlation with the fault type are optimized in reverse until the prediction performance of the fault association chain knowledge base meets the preset reliability index.
[0015] To achieve the above objectives, the present invention also proposes an engine fault prediction system, wherein the engine fault prediction system applies the engine fault prediction method; the system includes: The fault association chain construction module is used to sort out the hierarchical fault association chains corresponding to different fault types based on engine fault mechanisms and full life cycle operation data, so as to build an engine fault association chain knowledge base. The fault association chain includes the main driving parameters and several associated node parameters, as well as the temporal transmission constraints and anomaly judgment conditions between each parameter. The main driving parameters are the core initiating parameters of the corresponding fault type, the associated node parameters are the characteristic parameters on the fault transmission path, the temporal transmission constraints are the time interval thresholds between adjacent parameter anomalies, and the anomaly judgment conditions are the abnormal amplitude and fluctuation frequency thresholds of each parameter. The fault acquisition module is used to collect engine operating data in real time, extract the monitoring values of the main drive parameters, and compare the monitoring values with the preset drive thresholds of the main drive parameters. When the monitoring value reaches the drive threshold, the fault prediction process is triggered, and the directional time-series progressive identification mode of the associated node parameters is entered. The fault identification module performs a complete closed-loop confirmation of the fault association chain based on whether all associated node parameters in the fault association chain meet the corresponding anomaly judgment conditions and whether the anomaly timing of each associated node parameter conforms to the propagation constraints. When the confirmation is successful, the module extracts the fault type and fault association chain corresponding to the fault association chain and performs fault prediction reporting. If any level of associated node parameter does not meet the anomaly judgment conditions, a fault association chain interruption signal is generated. Based on the interruption signal, the module identifies whether there is a fault association chain skipping propagation and performs engine latent fault prediction.
[0016] The technical solution of this invention helps to solve the technical problem that existing engine fault prediction methods, due to their large data processing volume and the fact that they only compare each data point with a corresponding threshold, are unable to accurately identify early and minor faults, thus failing to meet the actual needs of engine operation and maintenance. A detailed analysis follows: This invention, based on engine failure mechanisms and full lifecycle operational data, establishes hierarchical fault association chains corresponding to different fault types. The main driving parameters within these chains serve as the driving force for fault prediction. Specifically, when a main driving parameter reaches its threshold, the fault prediction process is triggered, entering a directional, time-series progressive identification mode for associated node parameters. In this mode, depending on whether the parameters of each associated node in the fault association chain meet the corresponding anomaly judgment conditions and the propagation constraints of the anomaly sequence, if a complete closed-loop confirmation of the fault association chain can be achieved, the corresponding fault type and fault association chain are extracted, and a fault prediction report is submitted. Therefore, this invention does not require judging whether each parameter exceeds a threshold during fault prediction. Instead, it individually detects the main driving parameters with obvious fault-series driving characteristics. When an anomaly occurs in the main driving parameter, fault confirmation is performed one by one through the time-related associated node parameters, ultimately achieving a complete closed-loop confirmation of the fault chain based on the time-series progressive identification method.
[0017] Therefore, this invention not only significantly reduces the amount of data processing, but also enables accurate fault identification by verifying and confirming data with temporal causal relationships one by one. Especially when making early fault predictions, when the various engine parameters do not deviate significantly from the normal range, weak faults can be detected through this temporal progressive identification method, thereby meeting the technical requirements for rapid early fault identification in engine operation and maintenance.
[0018] Furthermore, if the parameters of any first-level associated node in the fault chain do not meet the conditions, a fault association chain interruption signal is generated. The interruption signal can also be used to identify whether there is a fault association chain skipping transmission and to predict engine latent faults. This allows latent fault identification to still be performed even when the temporal progression identification of the fault association chain is interrupted due to data delays, sensor failures, or other issues. Attached Figure Description
[0019] Figure 1 This is a flowchart illustrating the engine fault prediction method in the first embodiment of the present invention. Figure 2 This is a schematic diagram illustrating the construction process of the fault association chain knowledge base in this invention; Figure 3 This is a schematic diagram of the functional modules of the engine fault prediction system in this invention.
[0020] The objectives, features, and advantages of this invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation
[0021] It should be understood that the specific embodiments described herein are for illustrative purposes only and are not intended to limit the scope of the invention.
[0022] In the following description, the use of suffixes such as "unit," "component," or "element" to denote elements is solely for the purpose of illustrative purposes and has no specific meaning in itself. Therefore, "unit," "component," or "element" may be used interchangeably.
[0023] Please see Figures 1 to 3 The first embodiment of the present invention provides an engine fault prediction method, comprising the following steps: Step S10: Based on engine failure mechanisms and full lifecycle operation data, a hierarchical fault association chain corresponding to different fault types is sorted out to construct an engine fault association chain knowledge base. The fault association chain includes the main driving parameter and several associated node parameters, as well as the temporal transmission constraints and anomaly judgment conditions between each parameter. The main driving parameter is the core initiating parameter of the corresponding fault type, the associated node parameters are the characteristic parameters on the fault transmission path, the temporal transmission constraint is the time interval threshold between abnormal adjacent parameters, and the anomaly judgment condition is the abnormal amplitude and fluctuation frequency threshold of each parameter. Step S20: Collect engine operating data in real time, extract the monitoring values of the main drive parameters, and compare the monitoring values with the preset main drive parameter drive thresholds. Step S30: When the monitored value reaches the driving threshold, the fault prediction process is triggered, and the directional time-series progressive identification mode of the associated node parameters is entered. Step S40: Based on whether all the parameters of the associated nodes in the fault association chain meet the corresponding anomaly judgment conditions and whether the anomaly timing of each associated node parameter meets the propagation constraints, a complete closed-loop confirmation of the fault association chain is performed. When the confirmation is passed, the fault type and fault association chain corresponding to the fault association chain are extracted, and a fault prediction report is submitted. Step S50: If any first-level associated node parameter does not meet the anomaly judgment condition, a fault association chain interruption signal is generated. Based on the interruption signal, it is identified whether there is a fault association chain skipping transmission, and engine latent fault prediction is performed.
[0024] The technical solution of this invention helps to solve the technical problem that existing engine fault prediction methods, due to their large data processing volume and the fact that they only compare each data point with a corresponding threshold, are unable to accurately identify early and minor faults, thus failing to meet the actual needs of engine operation and maintenance. A detailed analysis follows: This invention, based on engine failure mechanisms and full lifecycle operational data, establishes hierarchical fault association chains corresponding to different fault types. The main driving parameters within these chains serve as the driving force for fault prediction. Specifically, when a main driving parameter reaches its threshold, the fault prediction process is triggered, entering a directional, time-series progressive identification mode for associated node parameters. In this mode, depending on whether the parameters of each associated node in the fault association chain meet the corresponding anomaly judgment conditions and the propagation constraints of the anomaly sequence, if a complete closed-loop confirmation of the fault association chain can be achieved, the corresponding fault type and fault association chain are extracted, and a fault prediction report is submitted. Therefore, this invention does not require judging whether each parameter exceeds a threshold during fault prediction. Instead, it individually detects the main driving parameters with obvious fault-series driving characteristics. When an anomaly occurs in the main driving parameter, fault confirmation is performed one by one through the time-related associated node parameters, ultimately achieving a complete closed-loop confirmation of the fault chain based on the time-series progressive identification method.
[0025] Therefore, this invention not only significantly reduces the amount of data processing, but also enables accurate fault identification by verifying and confirming data with temporal causal relationships one by one. Especially when making early fault predictions, when the various engine parameters do not deviate significantly from the normal range, weak faults can be detected through this temporal progressive identification method, thereby meeting the technical requirements for rapid early fault identification in engine operation and maintenance.
[0026] Furthermore, if the parameters of any first-level associated node in the fault chain do not meet the conditions, a fault association chain interruption signal is generated. The interruption signal can also be used to identify whether there is a fault association chain skipping transmission and to predict engine latent faults. This allows latent fault identification to still be performed even when the temporal progression identification of the fault association chain is interrupted due to data delays, sensor failures, or other issues.
[0027] An engine is a highly coupled and complex system, and the occurrence and evolution of its faults exhibit clear causality, temporality, and transmissibility. Therefore, this invention uses a fault association chain to characterize the causality and temporality of engine fault propagation: Take a fuel system malfunction as an example: The fault originated from slight wear of the fuel injector, which was characterized by increased fluctuations in fuel injection pressure (i.e., abnormal main drive parameters).
[0028] The primary propagation of the fault (associated node 1) is as follows: pressure fluctuations cause a decrease in fuel injection quantity accuracy (e.g., abnormal injection pulse width feedback). This typically occurs within seconds to tens of seconds after the pressure anomaly (i.e., timing constraints).
[0029] The secondary transmission of the fault (related node 2) is: inaccurate injection leads to incomplete combustion in the cylinder (e.g., increased concentration of unburned hydrocarbons in the exhaust and distortion of the in-cylinder combustion pressure curve).
[0030] The three-level transmission of the fault (related node 3) is as follows: deterioration of combustion leads to fluctuations in engine output torque and a decrease in power (abnormal performance parameters).
[0031] The ultimate manifestation of the malfunction: After long-term operation, it may cause blockage of the aftertreatment system or catalytic converter poisoning.
[0032] This process forms a fault correlation chain: an initial, minor, localized fault (such as wear) will affect other related parameters along a specific path and in a specific time sequence, according to physical laws (fluid dynamics, combustion science, mechanical dynamics) and control system logic, ultimately resulting in performance degradation or functional loss.
[0033] An engine has numerous sensors with complex relationships between their parameters. Fault correlation chains are used to extract the fault propagation links that are most relevant to a specific fault and have the clearest transmission relationship from the complex parameter network by combining mechanisms and data.
[0034] In the technical solution of this invention, the system does not process all data. Under the condition of the most limited computing power, it can only continuously and frequently monitor the starting point of each fault association chain, that is, the main driving parameters.
[0035] Once a certain main driving parameter reaches the driving threshold, the fault prediction process is immediately initiated, which only initiates high-frequency, targeted collection and judgment of parameters of subsequent related nodes on this specific link.
[0036] The technical solution of this invention not only checks whether subsequent parameters are abnormal, but more importantly, it checks whether the time sequence of the abnormality occurs conforms to the timing constraints defined in the knowledge base. For example, if a significant drop in torque is observed within 0.1 seconds after an abnormal fuel pressure, this does not conform to the physical time constant of combustion conduction, and the system will determine it as a false anomaly, thus not reporting a fault and greatly reducing the false alarm rate.
[0037] If all nodes on the link fail sequentially and on time, it is equivalent to gradually confirming the occurrence of the fault mode. Therefore, after the end node of the fault association chain is also confirmed to be abnormal, not only can an alarm be triggered, but the fault type location can also be triggered when the fault association chain is started, which greatly improves the efficiency of early weak fault identification.
[0038] Therefore, the technical solution of the present invention achieves the following advantages: First, the computing power requirement is greatly reduced: it realizes the transformation from all-weather, all-parameter monitoring to a normal low-power monitoring state, and only triggers targeted monitoring by checking whether a limited number of main drive parameters are abnormal. It is particularly suitable for deployment on edge computing devices with limited resources (such as vehicle computing devices). Second, it has strong anti-interference capabilities and a low false alarm rate: a single parameter fluctuation (such as random vibration) is unlikely to simultaneously and timely satisfy the abnormal conditions of multiple parameters in the entire fault correlation chain. Moreover, timing constraints are simple and effective filters that can effectively eliminate transient interference.
[0039] Third, it enables fault root cause localization and early warning: it can not only report the fault type, but also indicate the possible fault types and propagation paths through the triggered fault correlation chains. It can issue early warnings (such as at the component level or subsystem level) before the fault has fully propagated to the performance level (such as power degradation).
[0040] Specifically, the abnormal judgment conditions of the main driving parameter threshold and the associated node parameters are not necessarily values that exceed the normal parameter range. In fault prediction, these data may even be higher or lower values within the normal parameter range. This allows fault prediction to be achieved in the early stage of a fault, the early stage of a fault, or when the fault is weak, through the abnormal judgment conditions of the fault association chain and the temporal propagation constraints. Therefore, the fault prediction reporting in this invention includes fault prediction when the fault has not occurred significantly, which is different from the existing technology that only reports faults when the fault has occurred significantly.
[0041] According to the first embodiment of the engine fault prediction method of the present invention, in the second embodiment of the engine fault prediction method of the present invention, after step S10, the method further includes: Step S60: Establish an adaptation strategy between engine operating conditions and main drive parameter thresholds so that the main drive parameter thresholds can be adaptively adjusted according to engine operating conditions. Step S70: Based on the driving threshold of the main driving parameter, establish the coupling adjustment rules for the parameters of the associated nodes.
[0042] Specifically, the engine's operating thresholds are not fixed but dynamically change with operating conditions (such as speed, load, and temperature). An adaptation strategy is established to dynamically adjust the drive thresholds of the main drive parameters based on real-time operating conditions. For example, the normal vibration amplitude at high speeds is much higher than at idle; a fixed threshold would lead to false alarms. Furthermore, a coupling adjustment rule is established to adjust the judgment conditions of subsequent related node parameters based on the adjustment of the main drive parameter thresholds. This ensures that the sensitivity and specificity of the entire fault correlation chain remain consistent under different operating conditions, avoiding prediction logic failures caused by operating condition switching, and enabling the system to have reliable fault prediction capabilities in all weather and all operating conditions. These two steps together constitute a key optimization of the core fault prediction process, enhancing the practicality and technical contribution of the solution.
[0043] According to the first embodiment of the engine fault prediction method of the present invention, in the third embodiment of the engine fault prediction method of the present invention, after step S10, the method further includes: Step S80: Regularly collect engine maintenance records and new operational anomaly data; Step S90: Optimize the parameter composition, temporal propagation constraints, and anomaly determination conditions of the fault association chain through incremental learning algorithm; Step S100: When a new fault type is added, construct the corresponding new fault association chain.
[0044] Steps S80 to S100 constitute a closed loop of adaptive evolution and knowledge expansion for the fault prediction system, enabling the knowledge base of the fault prediction model to dynamically adjust to individual engine differences, long-term performance degradation, and new unknown faults. This solution obtains feedback information from real-world scenarios by periodically collecting maintenance records and new anomaly data in step S80. Step S90 utilizes an incremental learning algorithm to dynamically adjust and optimize the structure (e.g., node composition), rules (e.g., time-series thresholds), and judgment conditions of existing fault association chains without disrupting the original knowledge base, allowing its fault prediction logic to track the natural degradation of engine performance. Step S100 directly supplements the system with newly discovered fault modes, expanding the coverage of the knowledge base. These three steps together realize the system's evolution from initial experience solidification to continuous dynamic optimization, providing the core technical guarantee for maintaining long-term fault prediction accuracy.
[0045] In a fourth embodiment of the engine fault prediction method of the present invention, based on the first embodiment of the present invention, step S30 includes: Step S31: Based on the fault association chain knowledge base, the parameters of each association node are collected in a directional high-frequency manner according to the transmission order from the main driving parameters, the first-level association node parameters to the end association node parameters. Step S32: For each associated node parameter, determine whether the corresponding monitoring value meets the corresponding anomaly judgment condition, and whether the anomaly occurrence time conforms to the temporal propagation constraint with the preceding parameters. Step S33: If the current associated node parameter completes the anomaly determination, then the identification of the next level associated node parameter is triggered.
[0046] Once the parameters of the current associated node have been determined to be abnormal, regardless of whether the result is normal or abnormal, the identification of the parameters of the next level associated node will continue to be triggered.
[0047] Furthermore, whenever a parameter of a current associated node completes an anomaly determination, the evaluation score of the current associated node parameter is obtained based on the weight coefficient assigned to the current associated node parameter and the evaluation score of the current associated node parameter. The evaluation score of the current associated node parameter is then compared with the evaluation scores of the parameters of the preceding nodes in the fault association chain to obtain the evaluation score of the current fault association chain. When the evaluation score of the current fault association chain reaches the warning score, an alarm is triggered in advance.
[0048] According to the first embodiment of the engine fault prediction method of the present invention, in the fifth embodiment of the engine fault prediction method of the present invention, step S40 includes: Step S41: Obtain the weight coefficient assigned to each associated node parameter based on the importance of each node parameter in the fault propagation path; Step S42: Calculate the comprehensive matching degree of the fault association chain based on the weight coefficients of the parameters of each associated node in the fault association chain. Step S43: Confirm the complete closed loop of the fault association chain based on whether the overall matching degree of the fault association chain reaches the matching degree threshold.
[0049] Based on existing time-series propagation constraints and anomaly detection conditions, a comprehensive matching degree threshold based on weighted coefficients is introduced for closed-loop verification. This is mainly to address the uncertainty and non-idealization of fault propagation signals under complex operating conditions, thereby further improving the accuracy of fault prediction and its industrial applicability.
[0050] Specifically, in actual operation, fault propagation may occur due to operating condition interference, system redundancy, or individual component differences, causing some minor nodes on a certain correlation chain to fail to fully meet the preset anomaly judgment conditions, or to exhibit slight timing deviations. If strict timing propagation constraints and AND logic for anomaly judgment conditions (i.e., all nodes must fully meet the requirements) are used, it may lead to missed alarms. The weighted system in this embodiment allows for quantitative evaluation of such situations. When high-weight core nodes meet the requirements and the overall matching degree is up to standard, an alarm can still be triggered, improving the system's fault tolerance to non-ideal fault modes.
[0051] Furthermore, this embodiment helps distinguish between fault modes and noise: random disturbances or coupling effects may cause individual low-weight, non-critical parameters to accidentally trigger their abnormal conditions, and timing constraints alone may not be sufficient to completely eliminate such occasional disturbances. The comprehensive matching degree threshold introduces a global quantitative assessment, meaning that in a genuine fault propagation, high-weight core parameters on the critical path should have a higher probability of simultaneously exhibiting abnormalities. By requiring the overall matching degree to reach a high threshold, it is possible to more effectively distinguish between genuine fault propagation modes and spurious associations caused by random noise, further reducing the false alarm rate.
[0052] Furthermore, the weighting coefficients facilitate the quantification of confidence levels and support tiered early warnings: the overall matching degree is a continuous value ranging from 0% to 100%, providing quantified confidence for fault prediction. The system can implement tiered early warnings based on the matching degree (for example, fault tiered early warnings can be "attention," "suspected," or "confirmed"), providing more refined basis for operation and maintenance decisions, rather than just binary "yes / no" judgments, thus enhancing the system's decision support capabilities.
[0053] According to the fifth embodiment of the engine fault prediction method of the present invention, and the sixth embodiment of the engine fault prediction method of the present invention, the method further includes: Step S110: From the start time of directional high-frequency acquisition of parameters of each associated node to the end time of determining whether to report fault prediction, reduce the acquisition frequency of other engine parameters besides the main drive parameters and associated node parameters.
[0054] Specifically, this embodiment represents one of the core optimization strategies of this invention for achieving efficient and energy-saving operation at the edge. When the fault correlation chain is not triggered, the system monitors the main driving parameters at a low frequency, resulting in extremely low computational load. Once the main driving parameters exceed the limit, the fault prediction process is initiated, and system resources (such as ADC sampling channels, CPU processing cycles, and bus bandwidth) are prioritized and concentrated for targeted high-frequency acquisition and real-time analysis of parameters of all associated nodes involved in the currently suspected fault chain. At this time, to ensure the real-time performance and accuracy of the critical fault prediction task, the system actively reduces (e.g., switches to low-frequency sampling or suspends) the acquisition frequency of other engine parameters unrelated to the current fault prediction. This strategy, while ensuring fault prediction accuracy, minimizes the average computing power consumption and power consumption of the system, enabling this solution to be deployed economically and reliably on resource-constrained edge controllers.
[0055] In the first embodiment of the engine fault prediction method of the present invention, and in the seventh embodiment of the engine fault prediction method of the present invention, step S50 includes: Step S51: Obtain the interrupted node and downstream node of the fault association chain; Step S52: Generate interrupt node correction data based on the periodic sampling data of the interrupt node parameters; Step S53: The interrupted node is used to correct the input fault association chain, and the directional time-series progressive identification mode is continued to be executed to the downstream nodes of the interrupted node in order to predict the engine latent fault.
[0056] This embodiment elevates fault prediction from point-based judgment based on fixed thresholds to process analysis based on feature evolution, enabling the system to discover latent faults that have not yet fully manifested but have already shown signs, thus achieving more effective early warning.
[0057] The specific steps for generating correction data include: The periodic sampling sequence of interrupt node parameters at a preset time before the interruption occurs is obtained in order to extract healthy sample data of interrupt node parameters under similar operating conditions from historical normal operation data. Time-domain and frequency-domain analysis is performed on the periodic sampling sequence of the interrupted node to identify features that are statistically significantly different from healthy sample data but have a slight amplitude. For example, the energy of a certain frequency band has a continuous but slight increase, or the waveform has a periodic distortion that is difficult to detect. This suspected feature frequency band or pattern is the target that needs to be enhanced. By using a bandpass filter centered on the suspected fault characteristic frequency band, the current periodic sampling sequence is filtered to obtain an enhanced signal. The filtered signal amplifies the energy in this frequency band and suppresses interference from other irrelevant frequency bands. Using healthy sample data, we learn the typical signal shape of the interrupted node parameters after passing through the same bandpass filter in a healthy state, and use it as a healthy baseline reconstructed signal. The enhanced signal is compared with the reconstructed health baseline signal, and the resulting difference signal is the reconstructed residual signal, which is used to reflect the degree to which the enhanced signal deviates from the health baseline in the suspected fault frequency band. The reconstructed residual signal generated in the above steps is used as correction data and output to subsequent processes. This correction data replaces or supplements the original interruption node data and is used to continue performing time-series progressive identification to downstream nodes in the fault correlation chain, thereby determining whether there are hidden faults that were missed by the initial threshold due to their extremely weak abnormal features.
[0058] Therefore, this embodiment separates and amplifies the weak abnormal patterns submerged in background noise through two stages: enhancement and comparison. These patterns are then transformed into correction data that is more easily identified by subsequent processes, greatly improving the system's sensitivity to detecting early, atypical faults.
[0059] In the eighth embodiment of the engine fault prediction method of the present invention, based on the first to seventh embodiments, step S10 includes: Step S11: For the target engine model, based on the physical structure and control logic, decouple the sets of mutually independent potential fault types. ,in, m is the number of fault types; Step S12, for each fault type Starting from the failure mechanism, an initial set of monitoring parameters consisting of several key performance indicators is determined. ,in, Let j be the j-th initial monitoring parameter in the initial monitoring parameter set for the i-th fault type. , where n is the number of initial monitoring parameters included in the initial monitoring parameter set for the i-th fault type; Step S13: Obtain the historical full life cycle operation dataset of the target engine, wherein the dataset contains normal operation data and fault operation data labeled with fault type and occurrence time; Step S14: Using the dataset, for each fault type The corresponding initial monitoring parameter set was subjected to time-series correlation analysis and causal inference analysis to screen out parameters related to the fault type. A subset of parameters exhibits strong causal temporal correlations. Based on the correlation strength and propagation direction, the parameters within this subset are sorted to form fault types. The candidate parameter propagation sequence; Step S15: Based on the fault operation data, statistically analyze the time interval distribution between the first abnormality of the former and the first abnormality of the latter in the candidate parameter propagation sequence, and use the time interval determined by a specific percentile in the time interval distribution as the temporal propagation constraint threshold between the two adjacent parameters. Step S16, classify each fault type The corresponding candidate parameter propagation sequence, time-series propagation constraint threshold, and anomaly judgment conditions for each parameter are stored in a searchable hierarchical graph structure to form a fault association chain knowledge base; among them, the first parameter of the candidate parameter propagation sequence is defined as the main driving parameter, and the remaining parameters are defined as associated node parameters.
[0060] The abnormal judgment condition for each parameter is the abnormal judgment threshold, which can be lower than the abnormal value of the parameter in the prior art. For example, it can be a value that is higher or lower than the normal range.
[0061] Specifically, the time interval distribution between the first anomaly of each of two adjacent parameters, such as the time interval distribution between parameter k and parameter k+1, will have an interval time data set based on a large amount of historical full life cycle running dataset. The interval time of a specific percentile (e.g., the 95th percentile) in the interval time data set is used as the temporal propagation constraint threshold between two adjacent parameters (parameter k and parameter k+1).
[0062] By using the interval time of a specific percentile (e.g., the 95th percentile) in the interval time dataset as the temporal propagation constraint threshold between two adjacent parameters (parameter k and parameter k+1), abnormal signals with unusually long propagation times in historical data (which may not be caused by fault modes) can be filtered out.
[0063] According to the eighth embodiment of the engine fault prediction method of the present invention, and the ninth embodiment of the engine fault prediction method of the present invention, step S16 is further included as follows: Step S17: Obtain the fault reporting rate, false alarm rate, and fault location accuracy of the fault prediction method. Step S18: Based on the fault reporting rate, false alarm rate, and fault location accuracy, perform reverse optimization of the fault type. There are parameter subsets with strong causal temporal correlations, temporal propagation constraint thresholds, and anomaly judgment conditions, until the predictive performance of the fault association chain knowledge base meets the preset reliability index.
[0064] To achieve the above objectives, the present invention also proposes an engine fault prediction system, wherein the engine fault prediction system applies the engine fault prediction method; the system includes: The fault association chain construction module is used to sort out the hierarchical fault association chains corresponding to different fault types based on engine fault mechanisms and full life cycle operation data, so as to build an engine fault association chain knowledge base. The fault association chain includes the main driving parameters and several associated node parameters, as well as the temporal transmission constraints and anomaly judgment conditions between each parameter. The main driving parameters are the core initiating parameters of the corresponding fault type, the associated node parameters are the characteristic parameters on the fault transmission path, the temporal transmission constraints are the time interval thresholds between adjacent parameter anomalies, and the anomaly judgment conditions are the abnormal amplitude and fluctuation frequency thresholds of each parameter. The fault acquisition module is used to collect engine operating data in real time, extract the monitoring values of the main drive parameters, and compare the monitoring values with the preset drive thresholds of the main drive parameters. When the monitoring value reaches the drive threshold, the fault prediction process is triggered, and the directional time-series progressive identification mode of the associated node parameters is entered. The fault identification module performs a complete closed-loop confirmation of the fault association chain based on whether all associated node parameters in the fault association chain meet the corresponding anomaly judgment conditions and whether the anomaly timing of each associated node parameter conforms to the propagation constraints. When the confirmation is successful, the module extracts the fault type and fault association chain corresponding to the fault association chain and performs fault prediction reporting. If any level of associated node parameter does not meet the anomaly judgment conditions, a fault association chain interruption signal is generated. Based on the interruption signal, the module identifies whether there is a fault association chain skipping propagation and performs engine latent fault prediction.
[0065] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms, or by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present invention, 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 is stored in a readable storage medium (such as ROM / RAM, magnetic disk, optical disk) as described above, and includes several instructions to cause a terminal device to enter the methods described in the various embodiments of the present invention.
[0066] In the description of this specification, references to terms such as "one embodiment," "another embodiment," "other embodiments," or "first embodiment to Xth embodiment," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, method steps, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.
[0067] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or system that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or system. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or system that includes that element.
[0068] The sequence numbers of the above embodiments of the present invention are for description only and do not represent the superiority or inferiority of the embodiments.
[0069] The above are merely preferred embodiments of the present invention and do not limit the scope of the patent. Any equivalent structural or procedural transformations made based on the description and drawings of the present invention, or direct or indirect applications in other related technical fields, are similarly included within the scope of patent protection of the present invention.
Claims
1. An engine fault prediction method, characterized in that, Includes the following steps: Based on engine failure mechanisms and full lifecycle operation data, a hierarchical fault association chain corresponding to different fault types is sorted out to build an engine fault association chain knowledge base. The fault association chain includes the main driving parameter and several associated node parameters, as well as the temporal transmission constraints and anomaly judgment conditions between each parameter. The main driving parameter is the core initiating parameter of the corresponding fault type, the associated node parameters are the characteristic parameters on the fault transmission path, the temporal transmission constraint is the time interval threshold between adjacent parameter anomalies, and the anomaly judgment condition is the abnormal amplitude and fluctuation frequency threshold of each parameter. Real-time acquisition of engine operating data, extraction of monitoring values of main drive parameters, and comparison of monitoring values with preset main drive parameter thresholds; When the monitored value reaches the driving threshold, the fault prediction process is triggered, and the process enters the directional time-series progressive identification mode of the associated node parameters. Based on whether all the parameters of the associated nodes in the fault association chain meet the corresponding anomaly judgment conditions and whether the anomaly timing of each associated node parameter conforms to the propagation constraint, a complete closed-loop confirmation of the fault association chain is performed. When the confirmation is successful, the fault type and fault association chain corresponding to the fault association chain are extracted, and fault prediction and reporting are performed. If any level of associated node parameters does not meet the anomaly judgment conditions, a fault association chain interruption signal is generated. Based on the interruption signal, it is identified whether there is a fault association chain skipping transmission, and engine latent fault prediction is performed.
2. The engine fault prediction method according to claim 1, characterized in that, Following the step of constructing an engine fault association chain knowledge base by organizing hierarchical fault association chains corresponding to different fault types based on engine fault mechanisms and full life cycle operational data, the following steps are also included: Establish an adaptation strategy between engine operating conditions and main drive parameter thresholds so that the main drive parameter thresholds can be adaptively adjusted according to engine operating conditions; Based on the threshold driven by the main driving parameter, a coupling adjustment rule for the parameters of the associated nodes is established.
3. The engine fault prediction method according to claim 1, characterized in that, Following the step of constructing an engine fault association chain knowledge base by organizing hierarchical fault association chains corresponding to different fault types based on engine fault mechanisms and full life cycle operational data, the following steps are also included: Regularly collect engine maintenance records and new operational anomaly data; The parameter composition, temporal propagation constraints, and anomaly determination conditions of the fault correlation chain are optimized through incremental learning algorithms. When a new fault type is added, a corresponding new fault association chain is constructed.
4. The engine fault prediction method according to claim 1, characterized in that, The step of triggering the fault prediction process and entering the directional temporal progressive identification mode of associated node parameters when the monitored value reaches the driving threshold includes: Based on the fault association chain knowledge base, the parameters of each association node are collected in a targeted high-frequency manner according to the transmission order from the main driving parameters, the first-level association node parameters to the end association node parameters. For each associated node parameter, determine whether the corresponding monitoring value meets the corresponding anomaly judgment condition, and whether the anomaly occurrence time conforms to the temporal propagation constraint with the preceding parameters; If the parameters of the current associated node complete the anomaly detection, the identification of the parameters of the next level associated node will be triggered.
5. The engine fault prediction method according to claim 1, characterized in that, The steps for confirming the complete closed loop of the fault correlation chain include: Obtain the weight coefficients assigned to each associated node parameter based on the importance of each node parameter in the fault propagation path; The overall matching degree of the fault association chain is calculated based on the weight coefficients of the parameters of each associated node in the fault association chain. The complete closed loop of the fault association chain is confirmed based on whether the overall matching degree of the fault association chain reaches the matching degree threshold.
6. The engine fault prediction method according to claim 5, characterized in that, The method further includes: From the start time of directional high-frequency acquisition of parameters of each associated node to the end time of determining whether to report fault prediction, reduce the acquisition frequency of other engine parameters besides the main drive parameters and associated node parameters.
7. The engine fault prediction method according to claim 1, characterized in that, The step of generating a fault association chain interruption signal if any first-level associated node parameter does not meet the anomaly determination condition, identifying whether there is a fault association chain skipping propagation based on the interruption signal, and performing engine latent fault prediction includes: Obtain the interrupted node and downstream node of the fault association chain; Interrupt node correction data is generated based on the periodic sampling data of the interrupt node parameters; The fault association chain is corrected by interrupted nodes, and the directional time-series progressive identification mode is continued to be executed to the downstream nodes of the interrupted nodes to predict the hidden faults of the engine.
8. The engine fault prediction method according to any one of claims 1 to 7, characterized in that, The steps involved in constructing an engine fault association knowledge base by organizing hierarchical fault association chains corresponding to different fault types based on engine fault mechanisms and full lifecycle operational data include: For the target engine model, based on the physical structure and control logic, a set of independent potential fault types is decoupled. ,in, m is the number of fault types; For each fault type Starting from the failure mechanism, an initial set of monitoring parameters consisting of several key performance indicators is determined. ,in, Let j be the j-th initial monitoring parameter in the initial monitoring parameter set for the i-th fault type. , where n is the number of initial monitoring parameters included in the initial monitoring parameter set for the i-th fault type; Obtain the historical full life cycle operation dataset of the target engine, which includes normal operation data and fault operation data labeled with fault type and occurrence time; Using the dataset, for each fault type The corresponding initial monitoring parameter set was subjected to time-series correlation analysis and causal inference analysis to screen out parameters related to the fault type. A subset of parameters exhibits strong causal temporal correlations. Based on the correlation strength and propagation direction, the parameters within this subset are sorted to form fault types. The candidate parameter propagation sequence; Based on fault operation data, the time interval distribution between the first abnormality of the former and the first abnormality of the latter in the candidate parameter propagation sequence is statistically analyzed, and the time interval determined by a specific percentile in the time interval distribution is used as the temporal propagation constraint threshold between the two adjacent parameters. Each fault type The corresponding candidate parameter propagation sequence, time-series propagation constraint threshold, and anomaly judgment conditions for each parameter are stored in a searchable hierarchical graph structure to form a fault association chain knowledge base; among them, the first parameter of the candidate parameter propagation sequence is defined as the main driving parameter, and the remaining parameters are defined as associated node parameters.
9. The engine fault prediction method according to claim 8, characterized in that, The term will be used for each fault type After the step of storing the corresponding candidate parameter propagation sequences, temporal propagation constraint thresholds, and anomaly determination conditions for each parameter in a searchable hierarchical graph structure to form a fault association chain knowledge base, the following steps are also included: Obtain the fault reporting rate, false alarm rate, and fault location accuracy of the fault prediction method. Based on the fault reporting rate, false alarm rate, and fault location accuracy, the parameter subset, temporal propagation constraint threshold, and anomaly judgment conditions that have a strong causal and temporal correlation with the fault type are optimized in reverse until the prediction performance of the fault association chain knowledge base meets the preset reliability index.
10. An engine fault prediction system, characterized in that, The engine fault prediction system applies the engine fault prediction method according to any one of claims 1 to 9; the system comprises: The fault association chain construction module is used to sort out the hierarchical fault association chains corresponding to different fault types based on engine fault mechanisms and full life cycle operation data, so as to build an engine fault association chain knowledge base. The fault association chain includes the main driving parameters and several associated node parameters, as well as the temporal transmission constraints and anomaly judgment conditions between each parameter. The main driving parameters are the core initiating parameters of the corresponding fault type, the associated node parameters are the characteristic parameters on the fault transmission path, the temporal transmission constraints are the time interval thresholds between adjacent parameter anomalies, and the anomaly judgment conditions are the abnormal amplitude and fluctuation frequency thresholds of each parameter. The fault acquisition module is used to collect engine operating data in real time, extract the monitoring values of the main drive parameters, and compare the monitoring values with the preset drive thresholds of the main drive parameters. When the monitoring value reaches the drive threshold, the fault prediction process is triggered, and the directional time-series progressive identification mode of the associated node parameters is entered. The fault identification module performs a complete closed-loop confirmation of the fault association chain based on whether all associated node parameters in the fault association chain meet the corresponding anomaly judgment conditions and whether the anomaly timing of each associated node parameter conforms to the propagation constraints. When the confirmation is successful, the module extracts the fault type and fault association chain corresponding to the fault association chain and performs fault prediction reporting. If any level of associated node parameter does not meet the anomaly judgment conditions, a fault association chain interruption signal is generated. Based on the interruption signal, the module identifies whether there is a fault association chain skipping propagation and performs engine latent fault prediction.