Method for determining device performance degradation start time, electronic device, and vehicle

Abstract

The application provides a device performance degradation starting time determination method, an electronic device and a vehicle. The method comprises the following steps: obtaining an after-sales maintenance record of a target device; determining a fault time anchor point of the target device based on the after-sales maintenance record; determining a backtracking analysis sample within a preset backtracking window based on the fault time anchor point; stratifying the backtracking analysis sample according to a working condition variable of a running state of the target device to obtain a corresponding working condition stratification space; determining a multi-dimensional degradation deviation sequence based on an observation sample of each to-be-analyzed time and a dynamic health reference baseline matched with the corresponding working condition within the backtracking window; determining a plurality of candidate degradation starting points of the target device based on the multi-dimensional degradation deviation sequence; comprehensively scoring each candidate degradation starting point, performing pseudo-exception elimination and time reversal, screening out a candidate degradation starting point with the highest comprehensive score and capable of forming a complete evolution closed loop as a real degradation starting time. The application improves the accuracy of device performance degradation identification.

Description

Technical Field

[0001] This application relates to the field of equipment fault diagnosis and analysis technology, and in particular to a method for determining the start time of equipment performance degradation, electronic equipment, and vehicles. Background Technology

[0002] In related technologies, equipment fault identification commonly employs methods such as sensor threshold judgment, Diagnostic Trouble Code (DTC) diagnosis, and physical model prediction to achieve basic fault detection. Specifically, the sensor threshold judgment method sets fixed thresholds for equipment operating parameters such as temperature, voltage, and current, determining a fault when these parameters exceed the threshold range. The DTC diagnosis method relies on the fault diagnosis mechanism built into the equipment control system to identify and diagnose DTCs, thus determining the fault status. The physical model prediction method constructs mechanistic models such as thermal and electrochemical models of the equipment, combining them with equipment operating data to predict trends in equipment status.

[0003] However, the aforementioned fault identification methods generally suffer from monitoring lag and difficulty in effectively capturing the gradual degradation process of equipment. They can only identify the final trigger time of the fault, but the trigger time is not the true starting time of equipment performance degradation. They cannot effectively mark the degradation initiation node, resulting in low identification accuracy and poor universality of the relevant methods. They are difficult to achieve accurate early fault prediction, causing the initiation process of equipment performance degradation to remain unobservable and unquantifiable in engineering for a long time. Summary of the Invention

[0004] This application provides a method for determining the start time of equipment performance degradation, an electronic device, and a vehicle, to at least solve the problems in related technologies where monitoring lag makes it difficult to effectively capture the gradual degradation process of equipment, resulting in low identification accuracy, poor versatility, and difficulty in achieving accurate early fault prediction. The technical solution of this application is as follows: According to a first aspect of the embodiments of this application, a method for determining the start time of device performance degradation is provided, comprising: Obtain the after-sales maintenance records of the target equipment; The failure time anchor point of the target equipment is determined based on the after-sales maintenance records; Based on the fault time anchor point, determine the backtracking analysis sample within the preset backtracking window; The backtracking analysis samples are stratified according to the operating condition variables of the target equipment's operating status to obtain the corresponding operating condition stratification space; A multidimensional degradation deviation sequence is determined based on the observation samples at each time point to be analyzed within the backtracking window and the dynamic health reference baseline under the corresponding operating conditions. The dynamic health reference baseline is constructed based on the operating condition hierarchical space and the reference samples of the corresponding historical health stages. Based on the multidimensional degradation deviation sequence, multiple candidate degradation initiation points of the target device are determined; Each candidate degradation starting point is comprehensively scored according to multi-dimensional evaluation indicators. False anomalies are eliminated and time inversion is performed to select the candidate degradation starting point with the highest comprehensive score and the ability to form a complete evolutionary closed loop, which is then used as the true degradation start time.

[0005] According to a second aspect of the embodiments of this application, a device for determining the start time of device performance degradation is provided, comprising: The first acquisition module is used to acquire the after-sales maintenance records of the target device; The first determining module is used to determine the fault time anchor point of the target equipment based on the after-sales maintenance record; The second determining module is used to determine the backtracking analysis sample within the preset backtracking window based on the fault time anchor point; The stratification module is used to stratify the backtracking analysis samples according to the operating condition variables of the target equipment's operating status, and establish a corresponding operating condition stratification space. The third determination module is used to determine the multidimensional degradation deviation sequence based on the observation samples at each time to be analyzed within the backtracking window and the dynamic health reference baseline under the corresponding working condition matching. The dynamic health reference baseline is constructed based on the working condition hierarchical space and the reference samples of the corresponding historical health stage. The fourth determining module is used to determine multiple candidate degradation starting points of the target device based on the multidimensional degradation deviation sequence; The first screening module is used to comprehensively score each candidate degradation starting point according to multi-dimensional evaluation indicators, and select the candidate degradation starting point with the highest comprehensive score and the ability to form a complete evolutionary loop as the actual degradation start time.

[0006] According to a third aspect of the embodiments of this application, an electronic device is provided, comprising: It includes a processor, a memory; and a program or instructions stored in the memory and executable on the processor, wherein when the program or instructions are executed by the processor, they implement the steps of the device performance degradation start time determination method as described above.

[0007] According to a fourth aspect of the embodiments of this application, a vehicle is provided, including the electronic equipment described above.

[0008] According to a fifth aspect of the present application, a readable storage medium is provided, on which a program or instructions are stored, which, when executed by a processor of an electronic device, implement the steps of the device performance degradation start time determination method as described above.

[0009] According to a sixth aspect of the present application, a computer program product is provided, including a computer program or instructions, which, when executed by a processor of an electronic device, implement the steps of the device performance degradation start time determination method as described above.

[0010] The technical solutions provided by the embodiments of this application have at least the following beneficial effects: In this embodiment, after-sales maintenance records of the target equipment are obtained; the failure time anchor point of the target equipment is determined based on the after-sales maintenance records; a backtracking analysis sample within a preset backtracking window is determined based on the failure time anchor point; the backtracking analysis sample is stratified according to the operating condition variables of the target equipment's operating state to obtain a corresponding operating condition stratification space; a multidimensional degradation deviation sequence is determined based on the observation sample at each time to be analyzed within the backtracking window and the dynamic health reference baseline under the corresponding operating condition matching, wherein the dynamic health reference baseline is constructed based on the operating condition stratification space and reference samples of the corresponding historical health stage; multiple candidate degradation starting points of the target equipment are determined based on the multidimensional degradation deviation sequence; each candidate degradation starting point is comprehensively scored according to multidimensional evaluation indicators, and the candidate degradation starting point with the highest comprehensive score and capable of forming a complete evolutionary closed loop is selected as the true degradation start time. This application's embodiments determine the fault time anchor point using after-sales maintenance records, establish a dynamic health reference baseline using operating condition matching, and infer the true degradation starting point using a continuous evolution chain. This enables the inversion of the degradation starting point using weak after-sales labels, even when the true fault start time cannot be directly observed, thus upgrading the capability from fault occurrence identification to fault evolution starting point identification. Furthermore, the constructed dynamic health reference baseline ensures that deviation calculations are comparable under operating conditions, effectively suppressing misjudgments caused by load changes, mode switching, and environmental changes, thereby improving the accuracy of equipment performance degradation identification.

[0011] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and do not limit this application. Attached Figure Description

[0012] The accompanying drawings, incorporated in and forming part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application. They do not constitute an undue limitation of this application. To more clearly illustrate the technical solutions in the embodiments of this application 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 application. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0013] Figure 1This is a flowchart of a method for determining the start time of device performance degradation provided in an embodiment of this application.

[0014] Figure 2 This is an application flowchart of a method for determining the start time of device performance degradation provided in an embodiment of this application.

[0015] Figure 3 This is a schematic diagram of a device degradation evolution timeline provided in an embodiment of this application.

[0016] Figure 4 This is a block diagram of a device for determining the start time of device performance degradation, as provided in an embodiment of this application.

[0017] Figure 5 This is a block diagram of an electronic device provided in an embodiment of this application.

[0018] Figure 6 This is a block diagram of a device for determining the start time of device performance degradation, provided in an embodiment of this application. Detailed Implementation

[0019] To enable those skilled in the art to better understand the technical solutions of this application, the technical solutions in the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings.

[0020] It should be noted that the terms "first," "second," etc., used in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this application as detailed in the appended claims.

[0021] Figure 1 This is a flowchart of a method for determining the start time of device performance degradation provided in an embodiment of this application, such as... Figure 1 As shown, the method includes the following steps: Step 101: Obtain the after-sales maintenance records of the target device.

[0022] Step 102: Determine the fault time anchor point of the target equipment based on the after-sales maintenance record.

[0023] Step 103: Determine the backtracking analysis samples within the preset backtracking window based on the fault time anchor point.

[0024] Step 104: Perform condition stratification on the backtracking analysis sample according to the operating condition variables of the target equipment's operating status to obtain the corresponding operating condition stratification space.

[0025] Step 105: Determine the multidimensional degradation deviation sequence based on the observation samples at each time to be analyzed within the backtracking window and the dynamic health reference baseline under the corresponding operating conditions. The dynamic health reference baseline is constructed based on the operating condition hierarchical space and the reference samples of the corresponding historical health stages.

[0026] Step 106: Determine multiple candidate degradation starting points for the target device based on the multidimensional degradation deviation sequence.

[0027] Step 107: For each candidate degradation starting point, a comprehensive score is given according to multi-dimensional evaluation indicators. The candidate degradation starting point with the highest comprehensive score and the ability to form a complete evolutionary loop is selected as the actual degradation start time.

[0028] The method for determining the start time of device performance degradation described in this application can be applied to vehicles, servers, etc., without limitation. The vehicle-side implementation equipment can be an in-vehicle terminal, vehicle control platform, industrial control computer, engine, battery, motor, and other electronic equipment. The server can be an independent server, a server cluster, or a server that provides cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content delivery networks, or big data and artificial intelligence platforms, etc., without limitation.

[0029] The following is combined Figure 1 The specific implementation steps of a method for determining the start time of equipment performance degradation provided in the embodiments of this application will be described in detail.

[0030] In step 101, the after-sales maintenance records of the target device are obtained.

[0031] The acquisition method can be local or remote, and this embodiment does not impose any restrictions. The acquired after-sales records may include at least one of the following: after-sales maintenance records, maintenance work order records, fault confirmation records, fault code records, and key component replacement records. Of course, in addition to these, other records related to equipment faults may also be included, and this example does not impose any restrictions.

[0032] In step 102, the failure time anchor point of the target device is determined based on the after-sales maintenance record.

[0033] In this step, multiple candidate times related to the fault are extracted from the after-sales records; each candidate time is comprehensively scored based on the fault category matching degree, time sequence, record consistency and repair action intensity; the candidate time with the highest comprehensive score is selected as the fault time anchor point.

[0034] In other words, after acquiring at least one type of after-sales record corresponding to the target equipment, including after-sales maintenance records, maintenance work order records, fault confirmation records, fault code records (or fault code trigger records), and key component replacement records, multiple fault candidate times are extracted from the after-sales records, such as the maintenance entry time, work order creation time, first fault code reporting time, first confirmed fault time, and key component replacement time. Based on fault category matching degree, time sequence, record consistency, and maintenance action intensity, each fault candidate time is comprehensively scored, and the fault candidate time with the highest comprehensive score is selected as the fault time anchor point (or fault time anchor point interval). The fault time anchor point may include: target equipment ID, fault category, candidate time set, candidate time score, final anchor point time or anchor point interval, and anchor point confidence level. In this step, the fault time anchor point serves as the endpoint of the backtracking analysis, used to limit the extraction range and backtracking window of subsequent multi-source time series data.

[0035] In step 103, backtracking analysis samples within a preset backtracking window are determined based on the fault time anchor point.

[0036] In this step, the backtracking endpoint is set as the fault time anchor point, and a backtracking window is preset forward; multi-source time-series operation data of the target device are extracted in each preset backtracking window; the multi-source time-series operation data of each time point in the backtracking window are analyzed and filtered to obtain the backtracking analysis sample after analysis and filtering.

[0037] In other words, the system acquires multi-source time-series operational data of the target device prior to the fault time anchor point. This multi-source time-series operational data can include status signals, behavioral signals, environmental signals, control signals, and fault signals. The system then performs unified time reference alignment on data from different sampling frequencies, filters out missing segments, abnormal transition segments, and effective coverage, and obtains the filtered backtracking analysis samples.

[0038] The backtracking analysis sample can include: timestamps, valid signal sets, missing data markers, abnormal segment markers, coverage metrics, etc. It can serve as the basic input for subsequent operating condition stratification, baseline construction, and deviation calculation. Furthermore, the backtracking analysis sample can also be directly used by downstream health assessment, anomaly pattern recognition, and predictive model training methods.

[0039] In step 104, the backtracking analysis sample is stratified according to the operating condition variables of the target equipment's operating status to obtain the corresponding operating condition stratification space.

[0040] In this step, based on the backtracking analysis samples within the backtracking window, for each time point, operating condition variables reflecting the operating status of the target equipment are extracted; the operating condition vector is discretized; and the discretized operating condition variables are layered to establish an operating condition layering space.

[0041] Operating conditions typically refer to the overall working status, operating conditions, load, environment, and operating mode of equipment during operation, such as load size, speed, voltage, temperature, operating gear, business traffic, and work scenarios. Operating condition stratification refers to grouping, classifying, and hierarchizing the backtesting analysis samples according to different operating conditions, which is essentially splitting the samples according to operating status.

[0042] In this embodiment, the operational condition stratification refers to grouping the discretized operational condition variables of the target equipment according to characteristics such as operating mode, load level, and environmental conditions. Data with the same operating characteristics are grouped together, thereby constructing mutually distinguishable operational condition data sets. Simply put, it involves grouping and classifying the parameters / data corresponding to different operating states, loads, and environments of the target equipment according to the predefined categories, forming different state partitions.

[0043] The process of establishing a working condition hierarchical space involves selecting sample data within a preset backtracking window, collecting equipment operating parameters at each time step to form a working condition vector, performing discrete processing on continuous working condition parameters, and then completing data hierarchical classification based on the discrete working condition categories to construct a working condition hierarchical space.

[0044] In this embodiment, the backtracking analysis samples are stratified by operating condition to distinguish the data differences caused by different operating states, reduce interference from operating condition differences, and improve the accuracy of data analysis and fault identification.

[0045] In this step, a backtracking window is preset forward, with the fault time anchor point (or anchor point interval) as the endpoint. Within this preset backtracking window, for each time point, operating condition variables reflecting the operating status of the target equipment (such as load, speed, temperature, mode, environmental conditions, etc.) are extracted. Simultaneously, based on equipment type and fault mechanism, the backtracking analysis samples are stratified according to the operating condition variables to form a subsequent health baseline matching space. In other words, this embodiment uses operating condition variables as the core dimension, combined with equipment type and fault mechanism as auxiliary constraints, to classify, segment, and cluster the backtracking analysis samples, ensuring consistent operating condition characteristics among samples in the same layer and guaranteeing the accuracy of subsequent health baseline matching. Specifically, this may include the following steps: 1) Organize and filter operating condition variables.

[0046] In this step, the effective operating condition dimensions are first determined, distinguishing between core variables and auxiliary variables. Operating condition variables can include: load, speed, temperature, pressure, runtime, environmental parameters, start / stop status, and operating gear. The selection rules for the effective operating condition dimensions can eliminate variables with minimal fluctuations and no impact on equipment status, retaining indicators strongly correlated with faults and equipment performance. Then, constraints are bound, that is, the equipment type and fault mechanism corresponding to the back-analysis samples are synchronously associated as stratified pre-screening conditions.

[0047] 2) Preprocessing of sample data for retrospective analysis.

[0048] A unified preprocessing is performed on the backtracking analysis samples. The preprocessing may include: 21) Data cleaning: removing outliers, missing values, and invalid noise data; 22) Dimension unification: normalizing / standardizing operating condition data of different magnitudes; 23) Sample classification: first, roughly classifying into major categories according to equipment type and failure mechanism, and then carrying out operating condition stratification within the major categories.

[0049] 3) Working condition stratification. This embodiment will use the following three methods as examples to illustrate.

[0050] Method 1: Threshold range division, applicable to equipment with clearly defined rated values, gears, and range standards for operating conditions.

[0051] First, this method defines a grading threshold for each operating condition variable (low / medium / high load, rated / overspeed, normal temperature / high temperature zone, etc.) based on equipment technical manuals, operating specifications, and historical experience. Second, it uses single-variable stratification: the samples are divided according to a single operating condition threshold. Finally, it uses multi-variable combination stratification: multiple operating condition variables are cross-combined to form multi-dimensional operating condition intervals, with each interval being an independent operating condition layer.

[0052] For example, the operating conditions of wind turbine equipment include: speed and load; the layered ranges are: (low speed + low load), (low speed + high load), (high speed + low load), and (high speed + high load), for a total of 4 operating condition layers.

[0053] Method 2: Clustering Stratification, suitable for scenarios with no clear threshold for operating conditions and large continuous fluctuations. It uses an unsupervised clustering algorithm for automatic grouping and is suitable for complex, variable-condition equipment. Includes: First, extract the working condition feature vectors of all samples; Secondly, algorithms such as K-Means and DBSCAN can be used to cluster based on the similarity of working conditions; Finally, each cluster corresponds to a working condition layer, and working condition labels are defined for each layer.

[0054] Method 3: Layered matching of working conditions, suitable for periodic and fixed working conditions.

[0055] For equipment with a fixed operating mode (such as assembly line equipment and construction machinery), including: 31) Extract typical standard operating condition modes.

[0056] 32) Perform similarity matching between the backtracked samples and the standard pattern.

[0057] 33) Samples that match the same pattern are classified into the same working condition layer.

[0058] 34) Intra-layer verification and merging / splitting.

[0059] 341) Verification, that is, checking the deviation of the working condition characteristics of samples in the same layer. If the deviation is too large, the samples are re-splittered.

[0060] 342) Merge thin layers that have highly overlapping operating characteristics and no difference in baseline matching to reduce redundancy.

[0061] 343) Tagging service, which involves assigning a unique number and working condition description label to each working condition layer. Then, the baseline matching spatial clustering hierarchical structure is constructed.

[0062] 35) Construct a baseline matching space.

[0063] After stratification, each working condition layer is treated as an independent unit, and corresponding health baselines are constructed to form an overall health baseline matching space. Subsequent new samples can first determine their working condition layer and then call the corresponding baseline to complete the comparison analysis.

[0064] The backtracking window can be a fixed-duration window or an adaptively determined window based on the equipment type and fault mechanism; this embodiment does not impose any limitations. The fault mechanism refers to the inherent causes and evolution patterns of the failure of the target equipment components, including failure types such as mechanical wear, fatigue damage, component loosening, media corrosion, and electrical aging. For example, bearing wear and winding overheating aging in a motor.

[0065] In this step, a working condition stratification mapping table can be used to stratify working conditions. This table can include: time segment ID, associated backtracking window, working condition layer number, working condition feature vector, and working condition coverage. Based on this table, a one-to-one mapping is achieved between backtracking analysis samples, backtracking windows, and working condition layers. Establishing a working condition stratification space provides working condition matching constraints for the subsequent construction of a dynamic health baseline, ensuring that subsequent deviations are calculated under comparable working conditions. In other words, it eliminates data interference from different operating conditions, making analysis, modeling, fault identification, and trend judgment more accurate, and avoiding misjudgments across working conditions.

[0066] In step 105, a multidimensional degradation deviation sequence is determined based on the observation samples at each time to be analyzed within the backtracking window and the dynamic health reference baseline under the corresponding operating conditions. The dynamic health reference baseline is constructed based on the operating condition hierarchical space and the reference samples of the corresponding historical health stages.

[0067] In this step, a dynamic health reference baseline under working condition matching is first constructed based on the working condition stratification space and the reference samples of the corresponding historical health stages. Then, a multidimensional degradation deviation sequence is determined based on the observation samples of each time to be analyzed within the backtracking window and the dynamic health reference baseline under the corresponding working condition matching.

[0068] Among them, the dynamic health reference baseline constructed based on the hierarchical space of operating conditions and reference samples of corresponding historical health stages includes: In this step, based on the operating condition hierarchical space, the operating condition layer to which each backtracking moment belongs within the backtracking window is determined; samples of the historical health stages of the target equipment within each operating condition layer are extracted to obtain an individual health sample set; when the individual health sample set is insufficient, group reference samples of similar health equipment under the same or similar operating conditions are supplemented to obtain a group health sample set; the individual health sample baseline and the group health sample baseline of the individual health sample set and the group health sample set are calculated; the individual health sample baseline and the group health sample baseline are weighted and fused according to sample completeness, operating condition similarity and time proximity to obtain a dynamic health reference baseline under operating condition matching.

[0069] In this step, for each backtracking moment within the backtracking window, reference samples from the historical health stages of the equipment are extracted first within the corresponding operating condition layer. If the historical health samples for the equipment are insufficient, group reference samples from similar healthy equipment under the same or similar operating conditions are introduced. Then, individual health sample baselines and group health sample baselines are calculated, and based on these baselines, a weighted fusion is performed according to sample completeness, operating condition similarity, and temporal proximity to generate a dynamic health reference baseline. The dynamic health reference baseline may include: operating condition layer number, individual baseline value, group baseline value, fused baseline value, baseline sample size, and baseline confidence level. This dynamic health reference baseline can serve as a comparison benchmark for subsequent multidimensional degradation deviation sequence calculations.

[0070] The multidimensional degradation deviation sequence is determined based on the observation samples at each time point to be analyzed within the backtracking window and the dynamic health reference baseline under the corresponding operating conditions, including: For each time point to be analyzed within the backtracking window, the set of observation values ​​corresponding to that time point is extracted from the observation samples of the target device; the set of observation samples for each time point to be analyzed is compared with the dynamic health reference baseline under the corresponding operating condition time step by time; based on the comparison results, the deviation value of each key signal at the corresponding time point is calculated to obtain the one-dimensional degradation deviation value at the corresponding time point; at the same time point to be analyzed, the one-dimensional degradation deviation values ​​corresponding to all key signals are aggregated to obtain the multi-dimensional degradation deviation vector at the corresponding time point; the multi-dimensional degradation deviation vectors of each time point to be analyzed are stacked sequentially according to the time order of each time point to be analyzed to obtain a time-seriesd multi-dimensional degradation deviation sequence.

[0071] In other words, the observed samples (including state variables, behavioral variables, or fusion indicators, etc., at each time point to be analyzed within the backtracking window, which can be understood as a set of actual observed values ​​at a certain time within the backtracking window) are compared time-by-time with the dynamic health reference baseline under the corresponding operating conditions. Based on the time-by-time comparison results, the deviation value (including univariate deviation or multivariate comprehensive deviation) of each key signal at the corresponding time is calculated; the unidimensional degradation deviation value at the corresponding time is obtained; at the same time point to be analyzed, the unidimensional degradation deviation values ​​are aggregated to obtain the multidimensional degradation deviation vector at the corresponding time; the multidimensional degradation deviation vectors of each time point to be analyzed are stacked sequentially according to the time order of each time point to be analyzed to obtain a time-seriesd multidimensional degradation deviation sequence. That is, deviation subsequences are formed on multiple key signals respectively, and a multidimensional degradation deviation sequence is formed according to preset rules.

[0072] In this step, the multidimensional degradation deviation sequence is represented by a deviation sequence table, which includes timestamps, deviation values ​​for each signal, a comprehensive deviation value, deviation direction, and deviation amplitude levels. This multidimensional degradation deviation sequence can serve as direct input for persistence determination, trend consistency determination, and candidate degradation initiation point identification.

[0073] This multidimensional degradation bias sequence can also be mapped by downstream methods to risk indicators, state indicators, or judgment indices in specific scenarios. For example, it can be mapped to thermal behavior bias and compensating behavior bias.

[0074] It should be noted that, in this embodiment, within the backtracking window, the set of observations of the target device at each time to be analyzed is a set of real-time observations.

[0075] In step 106, multiple candidate degradation initiation points of the target device are determined based on the multidimensional degradation deviation sequence.

[0076] In this step, based on the multidimensional degradation deviation sequence, sliding windows are constructed on short-term, medium-term, and long-term time scales respectively; within each sliding window, it is determined whether the degradation deviation in the multidimensional degradation deviation sequence meets the set conditions (such as continuously existing on multiple time scales and showing a continuous increase, cumulative deterioration, or development trend in the same direction); the time point that meets the set conditions is identified as a candidate degradation starting point.

[0077] In other words, in this step, it is determined whether the degradation deviation in the multidimensional degradation deviation sequence continuously exceeds the normal fluctuation range at short time scale, medium time scale, and long time scale; at the same time, it is determined whether the degradation deviation has the trend characteristics of continuous increase, cumulative deterioration, or synchronous deviation of multiple signals; and the time point or time period that meets the conditions is marked as the candidate degradation start point (or candidate degradation abnormal interval).

[0078] The list of candidate degradation initiation points in this step can include candidate time, abnormal interval, persistence score, trend consistency score, and multi-signal synergy score. This list of candidate degradation initiation points provides a candidate set for final degradation start time selection and pseudo-anomaly removal. This candidate set can serve as candidate input for downstream methods in early anomaly modeling, weakly supervised label generation, or stage start point identification.

[0079] In step 107, each candidate degradation starting point is comprehensively scored according to multi-dimensional evaluation indicators, and the candidate degradation starting point with the highest comprehensive score and capable of forming a complete evolutionary loop is selected as the actual degradation start time.

[0080] In this step, each candidate degradation starting point is comprehensively scored according to at least two of the set multi-dimensional evaluation indicators (such as abnormal duration, abnormal amplitude, growth slope, temporal correlation with the fault anchor point, subsequent evolution continuity, multi-signal consistency, etc. Of course, in practical applications, the multi-dimensional evaluation indicators are not limited to these and may include other indicators, which are not limited in this embodiment). Based on the comprehensive scoring results, candidate degradation starting points with false anomalies (candidate degradation starting points caused by environmental changes, operating condition switching, maintenance events, short-term noise, or data loss) are eliminated. The candidate degradation starting point with the highest comprehensive score after elimination and which can form a complete evolutionary closed loop with the subsequent significant degradation stage and fault stage is selected as the true degradation start time. The true degradation start time of the equipment can be used to divide the healthy operation stage and the performance degradation stage of the equipment, provide a time series benchmark for degradation law modeling, remaining service life prediction, and fault early warning threshold calibration, but is not limited to this.

[0081] In this step, in an optimal embodiment, multiple candidate degradation initiation points are comprehensively scored according to multi-dimensional evaluation indicators such as abnormal duration, abnormal amplitude, growth slope, temporal correlation with the fault anchor point, subsequent evolution continuity, and multi-signal consistency; at the same time, false anomalies caused by environmental changes, operating condition switching, maintenance events, short-term noise, or data loss are identified and eliminated; the optimal candidate initiation point after screening is taken as the true degradation start time of equipment performance degradation.

[0082] In this step, the inversion results of the true degradation start time can include the final start time, candidate ranking results, reasons for elimination, comprehensive score, and confidence level of the start time. This true degradation start time provides a starting benchmark for subsequent stage division and evolution path output. Furthermore, this true degradation start time can be directly used as a key label input in downstream fault prediction, early warning model training, and stage modeling.

[0083] In this embodiment, after-sales maintenance records of the target device are obtained; the failure time anchor point of the target device is determined based on the after-sales maintenance records; a backtracking analysis sample within a preset backtracking window is determined based on the failure time anchor point; the backtracking analysis sample is stratified according to the operating condition variables of the target device's operating state to obtain a corresponding operating condition stratification space; a multidimensional degradation deviation sequence is determined based on the observation sample at each time to be analyzed within the backtracking window and the dynamic health reference baseline under the corresponding operating condition matching, wherein the dynamic health reference baseline is constructed based on the operating condition stratification space and the reference sample of the corresponding historical health stage; multiple candidate degradation starting points of the target device are determined based on the multidimensional degradation deviation sequence; each candidate degradation starting point is comprehensively scored according to multidimensional evaluation indicators, and the candidate degradation starting point with the highest comprehensive score and capable of forming a complete evolutionary closed loop is selected as the true degradation start time. In this embodiment, the failure time anchor point is determined by after-sales maintenance records, the health reference baseline is determined by operating condition matching, and the true degradation starting point is deduced by using a continuous evolution chain. Under the condition that the true failure start time cannot be directly observed, the degradation starting point can be deduced by using weak after-sales labels, realizing the capability upgrade from failure occurrence identification to failure evolution starting point identification; and through the constructed dynamic health reference baseline, the deviation calculation is made comparable to the operating conditions, thereby effectively suppressing misjudgments caused by load changes, mode switching and environmental changes, and improving the accuracy of equipment performance degradation identification.

[0084] Furthermore, this application, through multi-scale persistence, trend consistency, and multi-signal collaborative judgment mechanisms, can effectively distinguish between short-term disturbances and true degradation processes, improving the ability to identify chronic and progressive faults. Additionally, through candidate starting point ranking, pseudo-anomaly removal, and start time confidence output mechanisms, this application provides not only time points in the inversion results but also the strength and credibility of the evidence, facilitating engineering review, post-sales analysis, and subsequent model training.

[0085] Optionally, in another embodiment, based on the above embodiment, the method may further include: taking the actual degradation start time as the starting point, dividing the deviation evolution trend in the subsequent time period of the multidimensional degradation deviation sequence into an initial degradation stage, a significant degradation stage, and a failure stage; and outputting the initial degradation start time, stage boundary time, failure time, and complete evolution path result.

[0086] In other words, this step automatically divides the initial degradation stage, significant degradation stage, and fault stage based on the actual degradation start time, as well as the amplitude classification, growth rate change, and multi-signal synchronous deterioration relationship of degradation deviation in the degradation deviation sequence, and outputs degradation start time, stage boundary time, fault time, degradation rate, start confidence, multi-signal consistency score, and fault evolution path results.

[0087] The failure evolution path results may include the initial degradation time, significant degradation time, failure time, stage label, stage duration, degradation rate, explanatory notes, and confidence index.

[0088] The degradation start time, stage boundary time, failure time, and failure evolution path results output in this step can also serve as structured inputs for downstream failure prediction, health assessment, stage identification, pattern matching, and early warning models. For example, in engine coolant leakage prediction, the above stages can be further mapped to compensation period, critical period, and failure period, but this mapping is only one application example of this application.

[0089] This application can automatically output the complete fault evolution path, including the initial degradation stage, the significant degradation stage, and the failure stage, thus elevating equipment fault analysis from single-point judgment to a structured description of the entire process, enhancing interpretability and application value. Furthermore, this application requires no additional hardware sensors and does not rely on complex physical mechanism reconstruction; it can be directly deployed based on existing multi-source time-series data from vehicle networks or industrial IoT, demonstrating good engineering feasibility and cross-device promotion capabilities.

[0090] Please also see Figure 2 The above is an application flowchart of a method for determining the start time of device performance degradation provided in an embodiment of this application; the method includes: Step 201: Construct a candidate set of weak after-sales tags for the target device and generate fault time anchors.

[0091] This step includes: 1) Obtaining the after-sales maintenance records corresponding to the target equipment, including but not limited to: after-sales repair records, repair work order records, fault confirmation records, fault code triggering records, and key component replacement records. 2) Extracting multiple fault candidate times from the after-sales maintenance records, including at least: repair entry time, work order creation time, first fault code reporting time, first confirmed fault time, and key component replacement time.

[0092] In this embodiment, to avoid anchor point shift caused by directly using a single after-sales time, a comprehensive score is proposed for each candidate time. The comprehensive score includes at least: Fault Category Matching Score: Determines whether the record corresponding to the candidate time contains a fault description, code category, repair object, or replacement part that matches the target fault category; Time sequence logic scoring: Determine whether the candidate time satisfies the causal order of "report code earlier than confirmation, confirmation earlier than replacement, and repair later than anomaly occurrence"; Multi-record consistency scoring: Determine whether fault codes, confirmation records, maintenance records, or replacement records appear simultaneously around the same time to corroborate each other; Maintenance action intensity rating: Determines whether high-intensity maintenance actions such as replacement of critical components, system overhaul, coolant replenishment, and disassembly and assembly of critical assemblies have occurred. For each candidate time of failure, the comprehensive score is calculated according to the following formula: The comprehensive score for fault candidate time = a × category matching degree + b × time logic consistency + c × record consistency + d × maintenance action intensity.

[0093] Where a, b, c, and d are preset weights, and their sum is 1. These preset weights can be adaptively adjusted according to the fault type.

[0094] If the fault candidate time with the highest comprehensive score is significantly higher than other fault candidate times, then the fault candidate time with the highest comprehensive score will be used as the target fault time anchor point; if the comprehensive scores of multiple fault candidate times are close within a continuous time range, then that continuous time range will be defined as the fault time anchor point interval.

[0095] It should be noted that the fault time anchor point (fault time anchor point interval) output in this step is not used as the final degradation start time, but as the endpoint of the backtracking analysis. Both steps 202 and 203 use this backtracking analysis endpoint as a benchmark to extract multi-source time series data backward and establish a backtracking window.

[0096] For example, in the engine coolant leak prediction scenario, this fault time anchor point corresponds to the fault endpoint of "the leak has reached the repair or confirmation stage", which is used to backward estimate when the cooling system enters the compensation anomaly, when it enters the critical deterioration stage, and when it is close to failure.

[0097] Step 202: Obtain multi-source time series data and perform time unification and sample validity screening.

[0098] This step, based on the fault time anchor point obtained in step 201, extracts multi-source time-series data within a predetermined backtracking window of the target device. The multi-source time-series data includes at least: status signals; behavioral signals; environmental signals; control signals; and fault signals, but is not limited to these in specific applications. In this embodiment, since different signals have different sampling frequencies, this application first unifies all signals to the same time reference. The time reference can be a time axis with a fixed sampling frequency, such as 1Hz, 10Hz, or other suitable frequencies; this embodiment does not impose any limitations.

[0099] In this embodiment, for short-term missing segments, interpolation, forward hold, or the most recent valid value can be used to fill them; for long-term continuous missing segments, obviously abnormal jump segments, and segments with effective coverage below the threshold, they can be removed or marked as low-confidence segments.

[0100] Furthermore, in this embodiment, in order to avoid invalid operating conditions interfering with subsequent degradation judgment, sample segments that are unrelated to the fault mechanism of the target equipment can be removed in advance, such as severe data interruption segments; discontinuous sample segments after maintenance operations; abnormal operating mode segments that are obviously not comparable, etc. Finally, a retrospective analysis sample under a unified time axis is obtained.

[0101] The backtracking analysis samples output in this step are further mapped to different operating condition layers in step 203; in steps 204 and 205, the backtracking analysis samples directly participate in the construction of dynamic health reference baseline and deviation calculation.

[0102] For example, in the scenario of predicting engine coolant leakage, the backtracking analysis samples output in step 202 may include: corresponding thermal state signals, engine operating condition variables, ambient temperature signals, and cooling system actuator states, which are used to subsequently build thermal behavior baselines and actuator compensation deviations.

[0103] In this embodiment, 1) thermal state signal corresponds to state class signal.

[0104] Thermal state signals are essentially representations of the current internal physical state of a device. For example, T_eng(t) and T_coolant(t) reflect the thermal state of the engine and cooling system at a certain moment. These signals do not describe "what action was taken," but rather "what state is currently in," so they should be classified as state signals.

[0105] 2) Engine operating condition variables, mainly corresponding behavioral signals, some of which may also have state attributes.

[0106] In the context of a coolant leak, the engine operating condition variables N_eng(t), Load(t), and V_vehicle(t) primarily describe how the equipment is operating, its operating intensity, and its load conditions. Therefore, they are categorized as behavioral signals because they reflect the equipment's operating behavior and working conditions, rather than a purely static state.

[0107] 3) Ambient temperature signal, corresponding to environmental signals. For example, T_amb(t) is an environmental variable.

[0108] 4) Cooling system actuator status, corresponding control signals.

[0109] For example, A_f(t), A_h(t), and A_p(t) are essentially the working states or control outputs of the cooling system actuators, and are subsequently vectorized into A(t) = [A_f(t), A_h(t), A_p(t)], and used for system type identification, cooling capacity characterization, and compensation deviation calculation. Therefore, in the general classification, it appropriately corresponds to control class signals. Because it represents the action output after the control system drives the actuator, such as fan opening, water pump operating intensity, valve action, etc., it should be classified as a control class in terms of methodology.

[0110] 5) Fault information, corresponding to fault type signals.

[0111] For example, DTC(t) and historical maintenance records are fault signals.

[0112] Step 203: Construct the backtracking window and the working condition hierarchical space.

[0113] In this step, a backtracking window is set forward from the fault time anchor point (or fault time anchor point interval) obtained in step 201 as the endpoint. The backtracking window can be a fixed duration window or a window adaptively determined based on the device type and fault mechanism; this embodiment does not impose any restrictions.

[0114] Within this backtracking window, for each time point, operating condition variables reflecting the operating status of the target equipment are extracted. These operating condition variables may include load, speed, temperature, operating mode, environmental conditions, and equipment-specific state variables, but are not limited to these in specific applications. In other words, within the backtracking window, variables that can characterize the current operating condition are selected from multi-source time-series data and extracted uniformly as "operating condition hierarchical input variables." Then, at the same time point, these variables are jointly combined to form an operating condition vector, used to determine, for example, the current operating condition, which historical healthy samples should be compared with, and under what conditions the subsequent baseline should be constructed.

[0115] Then, the operating condition variables are converted into operating condition vectors, and the vector conversion formula is as follows: C(t) = [c1(t), c2(t), ..., cn(t)].

[0116] Where c1(t), c2(t), ..., cn(t) represent the operating condition variables at time t, and C(t) represents the operating condition vector at time t.

[0117] Next, the working condition vector is discretized, which can be done in any way: Regular compartmentation method: For example, compartments can be divided into equal or non-equal intervals according to speed, load, and temperature range. Clustering stratification: For example, using KMeans or other clustering methods, similar operating condition samples are divided into several operating condition layers. The input of the clustering method is the operating condition vector at each time point, such as [speed, load, ambient temperature, vehicle speed]. The several operating condition layers output by clustering are several groups of similar operating condition samples formed after clustering, each group representing a typical operating condition.

[0118] For example, given 8 time points (t1 to t8), we can use k-means clustering to create 3 operating condition layers: Operating condition layer A, t1, t2, characteristics: low speed, low load, low speed / idle speed; Operating condition layer B, t3, t4, characteristics: medium speed, medium load, city driving; Operating condition layer C, t5, t6, t7, t8, characteristics: high speed, high load, high temperature, medium to high speed operation. In this case, "several operating condition layers" are layers A, B, and C.

[0119] Hybrid operating condition coding method: First, the core variables are binned, and then the pattern variables are enumerated and coded. To ensure that each working condition layer has a sample size available for statistical comparison, a minimum sample size threshold N_min can be set. When the sample size of a certain working condition layer is insufficient, it can be merged with the adjacent working condition layers, or a population health sample can be introduced in step 204 to make up the difference.

[0120] It should be noted that the operating condition layer generated in step 203 is not the final result, but is used in step 204 to determine "from which operating condition layer should the health reference sample be extracted"; that is, step 203 determines the matching range of the subsequent dynamic health reference baseline.

[0121] For example, in the engine coolant leakage prediction scenario, step 203 is equivalent to the process of constructing and discretizing the operating condition space, so that the thermal behavior and actuator behavior under different cooling demand conditions can be compared under comparable conditions.

[0122] Step 204: Construct a dynamic health reference baseline under matching operating conditions.

[0123] First, for each time point t to be analyzed within the backtracking window in step 203, extract the set of real-time observations (i.e., a set of observations) of the target device at each time point t. This is done by extracting the corresponding row of data from the backtracking analysis sample after unified alignment in step 203 (for example, in an engine coolant leakage scenario, the row of data corresponding to a certain time point to be analyzed may include: thermal state signals, operating condition variables, environmental signals, actuator states, etc.), and determining the operating condition layer C_k to which the corresponding time point belongs. Based on the above example, "several operating condition layers" are layers A, B, and C. Each operating condition layer will contain variables for time point t. For example, if the time point to be analyzed is t=t5, then that time point t corresponds to operating condition C.

[0124] Then, within the operating condition layer C_k, samples from the historical health stages of the target equipment are extracted first to form an individual health sample set; if the number of individual health samples is insufficient, the operating condition coverage is incomplete, or the time span is insufficient, then group health samples of similar health equipment in the same or similar operating condition layers are further extracted to form a group health sample set.

[0125] Next, individual health sample sets and group health sample sets are constructed to obtain individual health baselines and group health baselines. The individual health sample baseline is denoted by B_ind(C_k), and the group health sample baseline is denoted by B_grp(C_k). The baselines can be constructed using any of the following methods: mean model; median model; quantile model, such as P50, P75, P95; regression model; or local fitting model. The specific construction process is well-known to those skilled in the art and will not be elaborated upon here.

[0126] In this embodiment, when both individual health sample baselines and group health sample baselines exist, a weighted fusion can be performed based on sample completeness, work condition similarity, and temporal proximity to obtain a dynamic health reference baseline: B_dyn(t) =λ1×B_ind(C_k) +λ2×B_grp(C_k).

[0127] Where λ1 and λ2 are dynamic weights, and λ1 + λ2 = 1.

[0128] The dynamic health baseline output in step 204 is directly invoked in step 205 to convert the "current observation" into a "deviation value relative to the health reference." Without this step, it is impossible to distinguish between normal operating condition fluctuations and true degradation.

[0129] For example, in the engine coolant leakage prediction scenario, the dynamic health reference baseline obtained in step 204 can be further specified as the temperature response baseline, actuator behavior baseline, and cooling capacity reference baseline, providing a comparative basis for subsequent thermal behavior deviation and compensation deviation calculations.

[0130] It should be noted that in the engine coolant leakage prediction scenario, the dynamic health reference baseline obtained in step 204 can be further specified as the temperature response baseline, actuator behavior baseline, and cooling capacity reference baseline. This is because in this scenario, the "health reference baseline" inherently has three different levels: observing the phenomenon: temperature; observing the means: actuators; and observing the essence: cooling capacity.

[0131] Among them, 1) the temperature response baseline refers to the normal level or normal variation pattern that the temperature signals of a healthy device should exhibit under a certain operating condition. In other words, under this operating condition, what should the temperature of the healthy system be, or how should it change? For example, at the operating condition level... 2200–2500 rpm 60-80% load Ambient temperature 30–35℃ Speed ​​80–100 km / h Ck After statistical analysis of the healthy samples, the following results were obtained: median coolant temperature: 93℃; coolant temperature P95: 97℃; median engine temperature rise rate: 0.08℃ / s. These can all constitute the "temperature response baseline". In other words, the temperature response baseline can be: a reference value, such as: median healthy coolant temperature = 93℃; a reference range, such as: normal range of healthy coolant temperature = 91–97℃; or a response function, such as: a function predicting the healthy temperature under different speeds and loads.

[0132] In subsequent deviation calculations, the current temperature is compared with this baseline: current coolant temperature - healthy temperature baseline; current temperature rise rate - healthy temperature rise rate baseline; this forms the "thermal behavior deviation" mentioned later.

[0133] Therefore: the temperature response baseline is equal to how the temperature should respond to the current operating conditions under healthy conditions.

[0134] 2) The actuator behavior baseline refers to the normal operating level or operating mode that actuators such as fans, pumps, and valves should typically exhibit in order to maintain normal thermal balance under a certain operating condition.

[0135] Under these operating conditions, what actuator output does a health system typically require, for example, within the same operating condition layer? 2200–2500 rpm 60-80% load Ambient temperature 30–35℃ Speed ​​80–100 km / h Ck After statistical analysis of the healthy samples, the following results were obtained: median fan duty cycle: 60%; median water pump speed: 68%; median valve opening: 35%; fan start probability: 0.82; these are the baseline of actuator behavior.

[0136] Actuator behavior baseline is important because a typical characteristic of the early stages of coolant leakage is that the temperature may not be obviously abnormal yet, but the actuator is already "compensating" for it.

[0137] In other words: a healthy system: under this condition, the fan only needs to be running at 60%; a leaky system: under this condition, the fan may already be running at 78%; at this point: the temperature may appear normal; but the actuator behavior has deviated from the healthy baseline. This is exactly what "compensation behavior" aims to achieve.

[0138] The current actuator status and the actuator behavior baseline ratio will then be used: current fan duty cycle - fan baseline; current pump intensity - pump baseline; this forms the "actuator compensation deviation" or "control behavior deviation".

[0139] Therefore: Actuator behavior baseline = how the actuator should normally work under healthy conditions.

[0140] 3) Cooling capacity reference baseline refers to the normal cooling capacity level corresponding to the combined action of multiple actuators in a healthy system under a certain operating condition.

[0141] Under these operating conditions, the overall cooling capacity of the health system is more abstract than the previous two. This is because the temperature response baseline focuses on the "result," the actuator behavior baseline focuses on the "control action," while the cooling capacity reference baseline focuses on the "overall system capacity."

[0142] Step 205: Construct a multidimensional degradation bias sequence.

[0143] In this step, observation samples for each time point to be analyzed are obtained from the backtracking analysis samples obtained in step 202. It should be noted that the observation samples for each time point to be analyzed do not refer to the data of the entire backtracking window, but rather a set of real-time observations at a certain time within the backtracking window. Then, these are compared with the dynamic health reference baseline under the corresponding working condition obtained in step 204 time-by-time. Based on the comparison results, the deviation value of each key signal at the corresponding time point is calculated to obtain the one-dimensional degradation deviation value at the corresponding time point. At the same time point to be analyzed, the one-dimensional degradation deviation values ​​corresponding to all key signals are aggregated to obtain the multi-dimensional degradation deviation vector at the corresponding time point. The multi-dimensional degradation deviation vectors of each time point to be analyzed are stacked sequentially according to the time order of each time point to be analyzed to obtain a time-seriesd multi-dimensional degradation deviation sequence.

[0144] In this embodiment, for each key signal x_i, the deviation value Δ_i(t) is calculated at the time t to be analyzed. The deviation value is preferably calculated using one of the following methods: Difference-type deviation: Δ_i(t) = x_i(t) - B_i(t); Ratio-type deviation: Δ_i(t) = x_i(t) / B_i(t); Standardized residual bias: Δ_i(t) = [x_i(t) - B_i(t)] / σ_i(C_k). Where B_i(t) is the dynamic health reference baseline of the signal under the operating condition layer C_k, and σ_i(C_k) is the normal fluctuation range of the healthy sample under the operating condition layer.

[0145] In this embodiment, the key signal refers to the core observation that can reflect the degradation or failure evolution of the target equipment.

[0146] After calculating the one-dimensional degradation deviation value of each key signal at the corresponding time, at the same time point to be analyzed, the one-dimensional degradation deviation values ​​corresponding to all key signals are aggregated, and the multi-dimensional degradation deviation vectors at the corresponding time points are stacked sequentially according to the time order of each time point to be analyzed, resulting in a time-series multi-dimensional degradation deviation sequence. The multi-dimensional degradation deviation vectors at each time point to be analyzed are stacked according to the following formula: Weighted superposition: D(t) = Σw_i·Δ_i(t); Maximum anomaly dominance: D(t) = max{Δ_i(t)}; Principal component projection or dimensionality reduction fusion. Where w_i represents the importance weight of different signals.

[0147] In the engine coolant leakage prediction scenario, key signals may include, but are not limited to, the following signals: Temperature-related signals, such as engine temperature, coolant temperature, rate of temperature rise, and rate of temperature fall.

[0148] Operating condition related signals, such as engine speed, engine load, and vehicle speed, are usually used more for operating condition matching, but in some cases they can also be used for deviation analysis.

[0149] Actuator-related signals, such as fan duty cycle, water pump operating intensity or speed, valve opening, heat exchanger actuator status, etc. The system integrates characteristic signals, such as cooling capacity indicators, actuator comprehensive compensation strength, thermal behavior residuals, and temperature-actuator coupling deviations.

[0150] A sequence is usually a series of deviation values ​​arranged in chronological order, also known as a time series.

[0151] Furthermore, the biased sequence can be smoothed, for example by moving average, exponential smoothing, or median filtering, to suppress single-point noise.

[0152] The multidimensional degradation deviation sequence obtained in step 205 is directly used in step 206 for persistence determination, trend consistency determination, and multi-signal synergy determination; and in step 208 it is used for stage boundary identification.

[0153] For example, in the engine coolant leak prediction scenario, the multidimensional degradation deviation sequence obtained in step 205 can be further mapped into temperature rise deviation, temperature drop deviation, and actuator compensation deviation, and used as inputs to the leak determination index CLI. It should be noted that this is not simply about "being able to be used for prediction," but rather about essentially substituting the output deviation amount into the risk index construction process for this scenario.

[0154] In this step, step 205 yields a multidimensional deviation sequence for general degradation identification, including temperature-related deviations, actuator-related deviations, and overall capability-related deviations. In the specific application scenario of engine coolant leakage prediction, based on the cooling system failure mechanism, the multidimensional deviation sequence can be further converted into several judgment quantities with clear physical meanings, including: temperature rise deviation, used to characterize the degree of abnormality of the current system's heating behavior relative to the healthy temperature response baseline; temperature drop deviation, used to characterize the degree of decrease in the current system's heat dissipation or cooling capacity relative to the healthy baseline; and actuator compensation deviation, used to characterize the degree of additional compensation of the current actuator output (fan, water pump, valve, etc.) relative to the healthy actuator behavior baseline. Subsequently, the temperature rise deviation, temperature drop deviation, and actuator compensation deviation can be used as inputs to construct the leakage determination index (CLI), which is used to assess the risk and development stage of coolant leakage. Specifically, this includes: Assuming there is a certain time t, step 205 has already calculated these deviations: Coolant temperature deviation = +2.5℃; Engine temperature deviation = +1.8℃; Fan duty cycle deviation = +15%; Pump strength deviation = +12%; Cooling capacity deviation = -0.10. In this scenario, the temperature rise deviation can be further summarized as follows: the temperature rise deviation is extracted from the temperature-related deviation and the temperature rise rate deviation.

[0155] ΔHeating = The degree to which the current temperature response is more aggressive than the healthy baseline.

[0156] For example, we get: ΔHeating = 0.07 Temperature drop deviation, extracted from cooling capacity deviation and cooling rate deviation: ΔCooling = the degree to which the current heat dissipation effect is worse than the healthy baseline. For example, we get: ΔCooling=0.05.

[0157] Actuator compensation deviation is achieved by a combination of deviations from the fan, water pump, valve, etc.

[0158] The current level of additional compensation for the actuator. For example, we can obtain: Then substitute CLI(t) = α 0.07+β 0.05+γ 0.18 will give you CLI.

[0159] Step 206: Perform multi-scale persistence and trend consistency determination on the multidimensional degradation deviation sequence to identify multiple candidate degradation starting points.

[0160] In this step, for the multidimensional degradation deviation sequence obtained in step 205, sliding windows are constructed on short-term, medium-term, and long-term time scales. Within each window, it is determined whether the following conditions are met: whether the deviation continuously exceeds the normal fluctuation range; whether the deviation changes in the same direction; and whether the mean, maximum, or cumulative value of the deviation exceeds a preset threshold. If a deviation exceeds the limit only on a short time scale but cannot be maintained on medium and long time scales, it is judged as a short-term disturbance; if a deviation persists on multiple time scales and shows a continuous increase, cumulative deterioration or development trend in the same direction, then the time point or interval is identified as a candidate degradation starting point.

[0161] In addition, a synergistic analysis is performed on the deviation synchronization of multiple key signals. If multiple signals deviate simultaneously within a similar time range, a synergistic deviation enhancement score is generated; if only a single signal deviates in isolation, the confidence of the candidate starting point is reduced.

[0162] Step 206 outputs a set of candidate degradation initiation points, not the final start time. Step 207 further sorts these candidate degradation initiation point sets, removes false anomalies, and selects the true degradation initiation point (i.e., the true degradation start time).

[0163] For example, in the engine coolant leakage prediction scenario, step 206 can identify early candidate moments where "the temperature has not yet significantly exceeded the limit, but the actuator compensation deviation and thermal behavior deviation have continued to exist." These candidate moments will be further screened as the starting point of the compensation period.

[0164] Step 207: Sort the candidate degradation starting points, remove pseudo-anomalies, and invert the final start time for the candidate degradation starting point set.

[0165] Specifically, for each candidate degradation starting point t_j obtained in step 206, a corresponding candidate comprehensive score Score(t_j) is constructed according to multiple set indicators.

[0166] The comprehensive score includes at least the following components (i.e., the set multi-dimensional evaluation indicators): Persistence component: The longer the abnormality lasts, the higher the score. Amplitude component: The larger the deviation amplitude, the higher the score. Growth slope component: The more stable and obvious the deviation growth, the higher the score. Anchor point association component: The higher the score, the more likely it is to evolve to the vicinity of the faulty anchor point in time. Evolutionary continuity component: Whether a candidate starting point can be continuously connected to a significant degradation stage or a failure stage. Multi-signal consistency component: Whether multiple key signals jointly support the candidate starting point. Specifically, the comprehensive score is constructed according to the following formula: Score(t_j) = p1×persistence + p2×amplitude + p3×growth slope + p4×anchor correlation + p5×evolutionary continuity + p6×multi-signal consistency.

[0167] Where p1, p2, p3, p4, p5 and p6 are weights, and p1+p2+p3+p4+p5+p6=1.

[0168] Then, based on the comprehensive scoring results, pseudo-anomalies are eliminated, that is, candidate degenerate pseudo-anomalies are eliminated. Among them, pseudo-anomalies include at least: transient anomalies caused by environmental changes; and mode changes caused by operating condition switching. Short-term offsets caused by maintenance operations or calibration actions; false fluctuations caused by data interruptions or communication jitter. If a candidate degeneration starting point scores highly within a local time window but fails to form a continuous chain of deterioration, or highly overlaps with the aforementioned pseudo-abnormal events, it will be downweighted or directly removed.

[0169] Finally, the candidate degradation starting point with the highest comprehensive score after screening, which can form a complete evolution loop with the subsequent significant degradation stage and failure stage, is determined as the true degradation start time T1. The true degradation start time T1 is used to divide the equipment healthy operation stage and the performance degradation stage, and provides a time series benchmark for degradation law modeling, remaining service life prediction, and fault early warning threshold calibration.

[0170] In other words, T1, the output of step 207, is not saved as a single point in time, but is directly used as the starting point of the initial degradation stage in step 208, and is used to continue to identify the starting point T2 of the significant degradation stage and the starting point T3 of the fault stage, as well as the boundaries of subsequent stages.

[0171] For example, in the scenario of predicting engine coolant leakage, T1 can be used as the starting point label for the cooling system to enter the "compensation period" from the "healthy state". This starting point label is further used to define the stage boundary and early warning lead of the training samples.

[0172] Step 208: Automatically divide the degradation stage and output the results.

[0173] Starting from the actual degradation start time T1 obtained in step 207, and combining it with the deviation sequence D(t) obtained in step 205, the evolution trend of the deviation in subsequent time periods is further analyzed. That is, starting from T1, we observe how D(t) changes. For example, we observe whether the deviation continues to increase, whether the rate of deviation increase accelerates, whether multiple signals begin to deteriorate synchronously, and when the "significant degradation stage" and the "failure stage" begin. In other words, by using time points and a deviation evolution curve, we further identify the boundaries of subsequent stages.

[0174] In this embodiment, the initial degradation stage is defined when the deviation first continuously exceeds the normal fluctuation range and continues to rise slowly; or the starting point T2 of the significant degradation stage is defined when the deviation amplitude increases significantly, the growth rate accelerates, or the degree of synchronous deterioration of multiple signals increases; or the starting point T3 of the fault stage is defined when the deviation reaches the vicinity of the fault anchor point or corresponds to a clear fault event in time.

[0175] Among them, the identification of the stage boundary can preferably be based on one or more of the following rules: the deviation amplitude crosses the preset level threshold; the deviation growth rate undergoes a significant jump; the number of synchronous deviations of multiple signals reaches the preset threshold; and the time correlation between the stage boundary and the after-sales anchor point interval meets the constraints. Therefore, we obtain: the starting point T1 of the initial degradation stage; the starting point T2 of the significant degradation stage; and the starting point T3 of the failure stage. T3 can also be the starting point of the failure repair stage, as detailed below. Figure 3 The diagram shown is a schematic diagram of a device degradation evolution timeline provided in an embodiment of this application. Figure 3 The diagram also shows the healthy stage, the initial degradation stage, the significant degradation stage, and the failure stage.

[0176] Simultaneously, it outputs the duration of each stage, the average deviation within the stage, the growth rate, and the confidence level of the start time.

[0177] It should be noted that the output of step 208 is not simply a set of time data, but a structured fault evolution path result. Based on this, downstream methods can use it as follows: Using T1, T2, and T3 as stage label boundaries, historical samples are segmented into "healthy - initial degradation - significant degradation - failure" segments; After dividing the deviation sequence in step 205 into T1, T2, and T3, the features of each stage are extracted and used as input for downstream prediction or classification models. The time difference between T1 and the fault anchor point is used as the basis for estimating the early warning lead time. The rate of deviation growth within a phase is used as a quantitative indicator of how fast a fault develops.

[0178] For example, in the engine coolant leakage prediction scenario: T1 may correspond to the start point of the cooling system compensation period; T2 may correspond to the start point of the critical period; T3 may correspond to the failure period or the maintenance confirmation period.

[0179] Herein, the deviation sequence obtained in step 205 may further extract thermal behavior deviation features and compensation behavior features within the interval from T1 to T3, and substitute them into the leakage determination index CLI or the pattern library matching module in the engine coolant leakage prediction scenario; The time span from T1 to T3 can be used as the advance time window for coolant leakage warning. Herein, T1 indicates that the system starts to enter a recognizable initial degradation stage, which is more suitable as the "warning start point" or "enhanced monitoring start point" rather than directly as the maintenance trigger point.

[0180] Therefore, in the embodiments of the present application, the phase boundaries and the deviation sequence output in step 208 are further converted into the determination input and hierarchical output in the specific leakage risk modeling.

[0181] It should be noted that this embodiment only takes the engine coolant leakage prediction scenario as an example, and the actual application is not limited thereto. For example, the present application is also applicable to battery degradation, motor performance degradation, and other progressive fault scenarios, which is not limited in this embodiment.

[0182] By adopting the embodiments of the present application, with the actual degradation start time as the starting point, after the historical samples are divided according to T1, T2, and T3, labeled data is obtained, wherein T<T1: healthy stage; T1≤T<T2: initial degradation stage; T2≤T<T3: significant degradation stage; T≥T3: fault stage Thereafter, features are extracted from each stage, for example: temperature deviation mean, temperature deviation growth, fan compensation deviation mean, maximum water pump compensation deviation, growth rate of the comprehensive degradation deviation D(t), number of multi-signal synchronous deteriorations, etc.

[0183] After these extracted features are input into the classification model, the classification model can learn: "what kind of feature combination corresponds to the healthy stage, what kind of combination corresponds to the initial degradation stage, and what kind of combination corresponds to the significant degradation stage".

[0184] For example, when a new vehicle is available, the above extracted features are input into the classification model, and the classification model can output the current state of the current vehicle according to the features of the current time period: being currently in the healthy stage, having entered the initial degradation stage, having entered the significant degradation stage, or being close to the fault stage, etc. Further, based on the stage of the current device determined by the classification model, warning operations such as determining whether a warning is needed, determining the warning level, and determining whether to suggest entering a service station for maintenance are performed.

[0185] Furthermore, based on the stage classification, the classification stages can also be mapped to the corresponding risk levels.

[0186] For example, the healthy stage corresponds to low risk; the initial degradation stage corresponds to moderate risk; the significant degradation stage corresponds to high risk; and the failure stage corresponds to serious risk, all of which require maintenance. This classification result can be directly incorporated into after-sales or monitoring strategies. For example: Low risk: continue to observe, do not trigger alarms; Moderate risk: trigger a minor warning, it is recommended to shorten the monitoring cycle; High risk: trigger a critical warning, it is recommended to check the cooling system; Severe risk: trigger an in-site recommendation or restrict continued high-load operation. In other words, "after classification", the continuous deviation evolution information can be transformed into an executable level of operation and maintenance decision.

[0187] Furthermore, after classification, it is also possible to predict how far away a failure is.

[0188] If the classification model identifies that the current stage is the initial degradation stage or has already entered the significant degradation stage, then it can further combine historical statistical patterns to estimate: how long it will take on average from the current stage to the failure stage, what the current early warning lead time is, and whether the failure development speed is too fast or too slow. In other words, the classification results not only tell you "which segment you are in now", but also help you estimate "how far away is the next segment".

[0189] For example, if the current stage is judged to be the initial degradation stage, historical samples can be used to estimate that there are 20 days left until significant degradation. If the current stage is judged to be the significant degradation stage, historical samples can be used to estimate that there are 5 days left until failure. This can further support predictive maintenance.

[0190] Based on the classification results, different failure evolution modes can be distinguished because not all equipment degrades at the same speed or in the same manner. After segmenting the samples, further "mode classification" can be performed, such as: slow-heating degradation, actuator compensation-driven degradation, simultaneous deterioration of temperature and compensation, and rapid failure after intermittent fluctuations. With this classification, downstream methods not only know that "there is a risk," but also know how the risk is developing. Different maintenance strategies can then be adopted for different risks.

[0191] For example, if it is an actuator-compensated type, the water pump / fan / valve control chain should be checked first; if it is a rapidly deteriorating temperature type, coolant leakage or heat dissipation path problems should be considered first.

[0192] The working mechanism of this application embodiment can be summarized as "determining the fault endpoint using weak after-sales labels, determining the health reference using operating condition matching, and inferring the true degradation starting point using the continuous evolution chain." Specifically, it includes: First, since the start time of equipment performance degradation cannot be directly measured, this application does not rely directly on manual labeling. Instead, it extracts multiple candidate times related to the fault from after-sales records and generates a target fault anchor point through category matching, consistency verification, and time logic filtering. In this way, the method obtains a traceable fault endpoint reference.

[0193] Secondly, this application extracts multi-source time-series operational data of the equipment around this anchor point, and constructs a dynamic health baseline from individual historical health samples and group health samples under the same or similar operating conditions. In this way, the deviation calculated subsequently is no longer a comparison of a fixed threshold, but a relative deviation from the health status under comparable operating conditions.

[0194] Then, this application detects whether the deviation persists at multiple time scales and simultaneously determines whether the deviation shows a trend of continuous increase, cumulative deterioration, and synchronous deviation of multiple signals. If there is only a short-term over-limit, a single deviation after a sudden change in operating conditions, or an anomaly that cannot continue into the subsequent evolution stage, it is not considered as the starting point of degradation; if a candidate anomaly satisfies both persistence and trend consistency, and forms a closed loop with the subsequent significant degradation and failure stages in time, it is taken as the actual degradation start time.

[0195] Finally, this application automatically divides the initial degradation stage, significant degradation stage, and failure stage based on the evolution law of degradation deviation, and outputs a structured failure evolution path and confidence index. In essence, it maps the "degradation start time that cannot be directly observed" to the "deviation start point with a continuous deterioration law under operating condition constraints", and completes time inversion through weak post-sales labeling and time-series evolution closed loop.

[0196] The degradation start time, stage boundaries, deviation evolution trajectory, and confidence results output by this application can not only be used for the analysis and interpretation of the equipment degradation process itself, but also serve as upstream inputs for various downstream methods, such as fault prediction, health status classification, stage label generation, pattern library construction, and early warning model training. The engine coolant leakage prediction based on thermal behavior deviation and multi-source data analysis is just one application example of this application.

[0197] This application proposes a multi-level after-sales weak label anchor point construction and fault endpoint inversion mechanism. In other words, instead of simply using repair time, reporting time, or replacement time as the fault endpoint, this application first constructs multiple candidate times from after-sales records, and then performs a comprehensive scoring based on fault category matching, time logic, consistency verification, and repair action intensity to generate a target fault time anchor point or anchor point interval. This transforms the originally discrete, lagging, and noisy after-sales records into a structured endpoint reference that can be used for time inversion, thus providing a foundation for retrospective identification of the true degradation start time.

[0198] This application proposes a mechanism for constructing a dynamic health baseline and multidimensional degradation deviation under operating condition matching. Instead of directly comparing current observations with fixed thresholds or static means, it first constructs a hierarchical operating condition space based on load, speed, temperature, mode, and environmental conditions. Then, it extracts individual health samples and group health samples under the same or similar operating conditions to form a dynamic health reference baseline. Furthermore, it converts current state quantities, behavioral quantities, or fusion indicators into a multidimensional degradation deviation sequence. This ensures that the deviation truly reflects the "change in relative health status," rather than normal fluctuations caused by differences in operating conditions, mode switching, or environmental changes, thereby improving the authenticity and comparability of degradation identification.

[0199] This application proposes a degradation initiation time inversion and stage boundary identification mechanism based on a continuous evolution chain. Instead of directly identifying the first anomaly as the degradation starting point, it forms a candidate starting point set through multi-timescale persistence determination, trend consistency determination, and multi-signal synergy determination. Then, it combines the anomaly duration, deviation amplitude, growth slope, anchor point correlation degree, and subsequent evolution continuity for comprehensive scoring and pseudo-anomaly elimination, ultimately inverting the true degradation initiation time and automatically dividing it into initial degradation stage, significant degradation stage, and failure stage. In other words, it identifies not arbitrary anomaly moments, but the true degradation starting point and its complete evolution path that "can continuously evolve and ultimately lead to failure," thus providing directly callable boundary results for downstream prediction, early warning, and stage modeling.

[0200] This application utilizes after-sales time records to generate fault anchor points, and organizes and aligns multi-source equipment runtime sequence data; it constructs a backtracking window with fault nodes as endpoints, and divides the operating condition space according to operating parameters; it generates a dynamic adaptation baseline by combining individual and group health samples; it calculates and fuses signal deviations to obtain a degradation sequence; it screens degradation candidate nodes through multi-scale, trend, and multi-signal collaborative judgment; after scoring and eliminating false anomalies, it locks the true degradation start time; finally, it divides multi-level degradation stages and outputs the evolution results. This application does not require a large amount of manually labeled data, can adapt to complex and changing operating conditions, effectively identifies early and subtle degradation, reduces misjudgment problems, and finely dissects the entire equipment degradation process. It can be widely used in fault tracing, health assessment, and intelligent operation and maintenance of various electromechanical equipment.

[0201] It should be noted that all numerical values ​​in the embodiments of this application are for illustrative purposes only and are not limited to these in practical applications. For the method embodiments, for the sake of simplicity, they are all described as a series of actions; however, those skilled in the art should understand that this disclosure is not limited to the described order of actions, because according to this application, some steps can be performed in other orders or simultaneously. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are all preferred embodiments, and the actions involved are not necessarily essential to this application.

[0202] Please also see Figure 4 This is a block diagram of a device for determining the start time of device performance degradation, provided in an embodiment of this application. The device includes: a first acquisition module 401, a first determination module 402, a second determination module 403, a layering module 404, a third determination module 405, a fourth determination module 406, and a first filtering module 407, wherein... The first acquisition module 401 is used to acquire the after-sales maintenance records of the target device; The first determining module 402 is used to determine the fault time anchor point of the target equipment based on the after-sales maintenance record; The second determining module 403 is used to determine the backtracking analysis sample within the preset backtracking window based on the fault time anchor point; The stratification module 404 is used to stratify the backtracking analysis sample according to the operating condition variables of the target equipment's operating status to obtain the corresponding operating condition stratification space. The third determining module 405 is used to determine a multidimensional degradation deviation sequence based on the observation samples at each time to be analyzed within the backtracking window and the dynamic health reference baseline under the corresponding working condition matching, wherein the dynamic health reference baseline is constructed based on the working condition hierarchical space and the reference samples of the corresponding historical health stage. The fourth determining module 406 is used to determine multiple candidate degradation starting points of the target device based on the multidimensional degradation deviation sequence; The first screening module 407 is used to comprehensively score each candidate degradation starting point according to multi-dimensional evaluation indicators, and select the candidate degradation starting point with the highest comprehensive score and the ability to form a complete evolutionary closed loop as the actual degradation start time.

[0203] Optionally, in another embodiment, based on the above embodiments, the first determining module includes: The first extraction module is used to extract multiple fault candidate times related to the fault from the after-sales records; The first scoring module is used to give a comprehensive score to each candidate time of a fault based on the fault category matching degree, time sequence, record consistency and maintenance action intensity. The first selection module is used to select the fault candidate time with the highest comprehensive score as the fault time anchor point.

[0204] Optionally, in another embodiment, based on the above embodiments, the second determining module includes: The setting module is used to preset a backtracking window forward, with the fault time anchor point as the backtracking endpoint; The second extraction module is used to extract multi-source time-series operation data of the target device within the preset backtracking window; The second filtering module is used to analyze and filter the multi-source time-series running data at each time point in the backtracking window to obtain the backtracking analysis sample after analysis and filtering.

[0205] Optionally, in another embodiment, based on the above embodiments, the layered module includes: The third extraction module is used to extract operating condition variables reflecting the operating status of the target equipment for each time point based on the backtracking analysis samples in the backtracking window. The processing module is used to perform working condition discretization processing on the working condition vector; A module is established to perform working condition stratification on the discretized working condition variables and to establish a working condition stratification space.

[0206] Optionally, in another embodiment, based on the above embodiments, the device may further include: a construction module, used to construct a dynamic health reference baseline based on the working condition hierarchical space obtained by the hierarchical module and the reference samples of the corresponding historical health stages.

[0207] The building module includes: The working condition layer determination module is used to determine the working condition layer to which each backtracking moment belongs within the backtracking window based on the working condition layering space. The fourth extraction module is used to extract samples of the historical health stages of the target equipment within each operating condition layer to obtain an individual health sample set; The second acquisition module is used to acquire a group reference sample of the same type of health equipment under the same or similar working conditions when the individual health sample set is insufficient, so as to obtain a group health sample set. The first calculation module is used to calculate the individual health sample baseline and the group health sample baseline of the individual health sample set and the group health sample set; The first fusion module is used to perform weighted fusion of the individual health sample baseline and the group health sample baseline to obtain a dynamic health reference baseline.

[0208] Optionally, in another embodiment, based on the above embodiments, the third determining module includes: The third acquisition module is used to extract the set of observation values ​​corresponding to each time to be analyzed from the observation samples of the target device for each time within the backtracking window; The comparison module is used to compare the set of observations at each time point to be analyzed with the dynamic health reference baseline under the corresponding working condition on a time-by-time basis. The second calculation module is used to calculate the deviation value of each key signal at the corresponding time based on the comparison results, and obtain the one-dimensional degradation deviation value of the corresponding key signal. The aggregation module is used to aggregate the single-dimensional degradation deviation values ​​corresponding to all key signals at the same time point to be analyzed, and obtain the multi-dimensional degradation deviation vector at the corresponding time point. The second fusion module is used to sequentially superimpose the multidimensional degradation deviation vector of each time step to be analyzed according to the time order of each time step to obtain a time-seriesd multidimensional degradation deviation sequence.

[0209] Optionally, in another embodiment, based on the above embodiments, the fourth determining module includes: A sliding window construction module is used to construct sliding windows on short time scales, medium time scales, and long time scales based on the multidimensional degradation bias sequence. The judgment module is used to determine whether the degradation deviation in the multidimensional degradation deviation sequence meets the set conditions within each sliding window; The identification module is used to identify all time points that the judgment module determines meet the set conditions as candidate degradation start points.

[0210] Optionally, in another embodiment, based on the above embodiments, the first screening module includes: The second scoring module is used to comprehensively score each candidate degradation starting point according to the set multi-dimensional evaluation indicators. The multi-dimensional evaluation indicators include at least the following two indicators: persistence component indicator, amplitude component indicator, growth slope component indicator, anchor point correlation component indicator, evolution continuity component indicator, and multi-signal consistency component indicator. The elimination module is used to eliminate candidate degradation initiation points of pseudo-anomalies based on the comprehensive scoring results; The second selection module is used to select the candidate degradation starting point with the highest comprehensive score after elimination and which can form a complete evolution loop with the subsequent significant degradation stage and failure stage, as the true degradation start time. The true degradation start time is used to divide the equipment healthy operation stage and the performance degradation stage, and to provide a time series benchmark for degradation law modeling, remaining service life prediction, and fault early warning threshold calibration.

[0211] Optionally, in another embodiment, based on the above embodiments, the apparatus further includes: The segmentation module is used to divide the deviation evolution trend in the subsequent time period of the multidimensional degradation deviation sequence into an initial degradation stage, a significant degradation stage, and a failure stage, starting from the actual degradation start time. The output module is used to output the initial degradation start time, stage boundary time, fault time, and fault evolution path results.

[0212] Optionally, embodiments of this application also provide an electronic device, including: It includes a processor, a memory; and a program or instructions stored in the memory and executable on the processor, wherein when the program or instructions are executed by the processor, they implement the steps of the device performance degradation start time determination method as described above.

[0213] Optionally, embodiments of this application also provide a vehicle including the aforementioned electronic device.

[0214] Optionally, embodiments of this application also provide a readable storage medium storing a program or instructions, which, when executed by a processor of an electronic device, implement the steps of the device performance degradation start time determination method described above.

[0215] Optionally, embodiments of this application also provide a computer program product, including a computer program or instructions, which, when executed by a processor of an electronic device, implement the steps of the device performance degradation start time determination method as described above.

[0216] Regarding the apparatus in the above embodiments, the specific manner in which each module performs its operation has been described in detail in the embodiments related to the method, and will not be elaborated upon here.

[0217] 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.

[0218] Figure 5 This application provides a frame of an electronic device 500. Figure 5 As shown in the figure, it includes a processor 501, a communication interface 502, a memory 503, and a communication bus 504. The processor 501, the communication interface 502, and the memory 503 communicate with each other through the communication bus 504. Memory 503 is used to store the processor-executable instructions; The processor 501, when executing executable instructions on the memory 503, implements the method described above.

[0219] In this embodiment, the communication bus can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. This communication bus can be divided into an address bus, a data bus, a control bus, etc. For ease of illustration, only one thick line is used to represent it in the figure, but this does not mean that there is only one bus or one type of bus.

[0220] The communication interface is used for communication between the aforementioned electronic devices and other devices.

[0221] The memory may include random access memory (RAM) or non-volatile memory, such as at least one disk storage device. Optionally, the memory may also be at least one storage device located remotely from the aforementioned processor.

[0222] The processors mentioned above can be general-purpose processors, including central processing units (CPUs), network processors (NPs), etc.; they can also be digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.

[0223] In another embodiment provided in this application, a vehicle is also provided, including the aforementioned electronic equipment.

[0224] In another embodiment provided in this application, a readable storage medium is also provided, on which a program or instruction is stored. When the program or instruction is executed by a processor of an electronic device, the processor is able to execute the various processes of the device performance degradation start time determination method embodiment described above, and achieve the same technical effect. To avoid repetition, it will not be described again here. For example, the readable storage medium includes computer-readable storage media, such as ROM, random access memory (RAM), CD-ROM, magnetic tape, floppy disk, and optical data storage devices.

[0225] In another embodiment provided in this application, a computer program product is also provided, including a computer program or instructions. When the computer program or instructions are executed by the processor of an electronic device, they implement the various processes of the device performance degradation start time determination method embodiment described above, and can achieve the same technical effect. To avoid repetition, they will not be described again here.

[0226] In the above embodiments, implementation can be achieved entirely or partially through software, hardware, firmware, or any combination thereof. When implemented using software, it can be implemented entirely or partially as a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that integrates one or more available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium (e.g., solid state disk (SSD)).

[0227] Figure 6 This is a block diagram of an apparatus 600 for determining the start time of device performance degradation, as provided in an embodiment of this application. For example, apparatus 600 can be provided as a server. See also... Figure 6The apparatus 600 includes a processing component 622, which further includes one or more processors, and memory resources represented by memory 632 for storing instructions, such as application programs, that can be executed by the processing component 622. The application programs stored in memory 632 may include one or more modules, each corresponding to a set of instructions. Furthermore, the processing component 622 is configured to execute instructions to perform the methods described above.

[0228] Device 600 may also include a power supply component 626 configured to perform power management of device 600, a wired or wireless network interface 650 configured to connect device 600 to a network, and an input / output (I / O) interface 658. Device 600 may operate on an operating system stored in memory 632, such as Windows Server™, Mac OS X™, Unix™, Linux™, FreeBSD™, or similar.

[0229] 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. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk), and includes several instructions to cause a terminal (which may be a mobile phone, computer, server, or network device, etc.) to execute the methods described in the various embodiments of this application.

[0230] The embodiments of this application have been described above with reference to the accompanying drawings. However, this application is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of this application without departing from the spirit and scope of the claims, and all of these forms are within the protection scope of this application.

Claims

1. A method for determining the start time of equipment performance degradation, characterized in that, include: Obtain the after-sales maintenance records of the target equipment; The failure time anchor point of the target equipment is determined based on the after-sales maintenance records; Based on the fault time anchor point, determine the backtracking analysis sample within the preset backtracking window; The backtracking analysis samples are stratified according to the operating condition variables of the target equipment's operating status to obtain the corresponding operating condition stratification space; A multidimensional degradation deviation sequence is determined based on the observation samples at each time point to be analyzed within the backtracking window and the dynamic health reference baseline under the corresponding operating conditions. The dynamic health reference baseline is constructed based on the operating condition hierarchical space and the reference samples of the corresponding historical health stages. Based on the multidimensional degradation deviation sequence, multiple candidate degradation initiation points of the target device are determined; Each candidate degradation starting point is comprehensively scored according to multi-dimensional evaluation indicators. The candidate degradation starting point with the highest comprehensive score and the ability to form a complete evolutionary loop is selected as the actual degradation start time.

2. The method for determining the start time of equipment performance degradation according to claim 1, characterized in that, Determining the fault time anchor point of the target equipment based on the after-sales maintenance record includes: Extract multiple candidate fault times related to the fault from the after-sales records; Based on the fault category matching degree, time sequence, record consistency and maintenance action intensity, a comprehensive score is given to each fault candidate time. The fault candidate time with the highest comprehensive score is selected as the fault time anchor point.

3. The method for determining the start time of equipment performance degradation according to claim 1, characterized in that, The step of determining the backtracking analysis samples within the preset backtracking window based on the fault time anchor point includes: Using the aforementioned fault time anchor point as the backtracking endpoint, a backtracking window is preset forward; Extract the multi-source time-series operation data of the target device within the preset backtracking window; The multi-source time-series data at each time point within the backtracking window are analyzed and filtered to obtain the backtracking analysis sample after analysis and filtering.

4. The method for determining the start time of equipment performance degradation according to claim 1, characterized in that, The backtracking analysis sample is stratified according to the operating condition variables of the target equipment's operating status to obtain an operating condition stratification space, including: Based on the backtracking analysis samples within the backtracking window, for each time point, operating condition variables reflecting the operating status of the target equipment are extracted; The operating condition vector is discretized. The discretized operating condition variables are stratified into operating condition layers to establish an operating condition stratification space.

5. The method for determining the start time of equipment performance degradation according to claim 1, characterized in that, Based on the aforementioned operating condition hierarchical space and the corresponding historical health stage reference samples, a dynamic health reference baseline is constructed according to the following steps: Based on the aforementioned working condition layering space, determine the working condition layer to which each backtracking moment belongs within the backtracking window; Extract samples of the historical health stages of the target equipment within each operating condition layer to obtain an individual health sample set; When the individual health sample set is insufficient, a group reference sample of the same type of health equipment under the same or similar working conditions is added to obtain a group health sample set; Calculate the individual health sample baseline and the group health sample baseline of the individual health sample set and the group health sample set; The individual health sample baseline and the group health sample baseline are weighted and fused to obtain a dynamic health reference baseline.

6. The method for determining the start time of equipment performance degradation according to claim 1, characterized in that, The determination of the multidimensional degradation bias sequence based on the observation samples at each time point to be analyzed within the backtracking window and the dynamic health reference baseline under the corresponding operating conditions includes: For each time point to be analyzed within the backtracking window, the set of observation values ​​for the corresponding time point is extracted from the observation samples of the target device; The set of observations for each time point to be analyzed is compared with the dynamic health reference baseline under the corresponding working condition on a time-by-time basis. Based on the comparison results, the deviation value of each key signal at the corresponding time is calculated to obtain the one-dimensional degradation deviation value of the corresponding key signal; At the same time point to be analyzed, the single-dimensional degradation deviation values ​​corresponding to all key signals are aggregated to obtain the multi-dimensional degradation deviation vector at the corresponding time point; By sequentially superimposing the multidimensional degradation deviation vectors of each time point to be analyzed in chronological order, a time-seriesd multidimensional degradation deviation sequence is obtained.

7. The method for determining the start time of equipment performance degradation according to claim 1, characterized in that, The determination of multiple candidate degradation initiation points for the target device based on the multidimensional degradation deviation sequence includes: Based on the multidimensional degradation bias sequence, sliding windows are constructed at short, medium, and long time scales, respectively. Within each sliding window, determine whether the degradation deviation in the multidimensional degradation deviation sequence meets the set conditions; All time points that meet the set conditions are identified as candidate degradation start points.

8. The method for determining the start time of equipment performance degradation according to claim 1, characterized in that, The process of comprehensively scoring each candidate degradation initiation point according to a set multi-dimensional evaluation index, and selecting the candidate degradation initiation point with the highest comprehensive score that can form a complete evolutionary closed loop as the true degradation start time, includes: For each candidate degradation starting point, a comprehensive score is given according to the set multi-dimensional evaluation indicators, wherein the multi-dimensional evaluation indicators include at least the following two indicators: persistence component indicator, amplitude component indicator, growth slope component indicator, anchor point correlation component indicator, evolution continuity component indicator, and multi-signal consistency component indicator. Candidate degradation initiation points with pseudo-anomalies were eliminated based on the comprehensive scoring results; The candidate degradation starting point with the highest comprehensive score after elimination and which can form a complete evolution loop with the subsequent significant degradation stage and failure stage is selected as the true degradation start time. The true degradation start time is used to divide the equipment healthy operation stage and the performance degradation stage, and to provide a time series benchmark for degradation law modeling, remaining service life prediction and fault early warning threshold calibration.

9. An electronic device, characterized in that, include: Including processor and memory; And a program or instructions stored on the memory and executable on the processor, wherein the program or instructions, when executed by the processor, implement the steps of the method for determining the device performance degradation start time as described in any one of claims 1 to 8.

10. A vehicle, characterized in that, Includes the electronic device as described in claim 9.

Images