A deep learning-based automobile part life prediction method

Abstract

This invention relates to the field of lifespan prediction technology and discloses a deep learning-based method for predicting the lifespan of automotive parts. The method includes: dividing the full lifespan data of a target automotive part into operating condition segments to obtain degradation data fragments; performing cross-domain feature decoupling on the degradation data fragments to obtain operating condition-independent degradation features; evaluating the similarity mapping of the operating condition-independent degradation features based on the health baseline state of the target automotive part to obtain a continuous health score; performing adaptive detection on the full lifespan data based on the continuous health score to obtain the degradation initiation time; performing trend extrapolation on the continuous health score based on the degradation initiation time to obtain the health change trajectory of the target automotive part; and performing reverse failure point determination on the health change trajectory to obtain the predicted remaining lifespan value of the automotive part. This invention can improve the efficiency of deep learning-based automotive part lifespan prediction.

Description

Technical Field

[0001] This invention relates to the field of life prediction technology, and in particular to a deep learning-based method for predicting the life of automotive parts. Background Technology

[0002] In existing technologies for predicting the lifespan of automotive parts, there is a lack of refined methods for classifying operating conditions in the processing of data throughout the entire life cycle of parts. The raw data is not effectively processed to remove outliers and smooth out outliers, nor is the data accurately identified and homogenized. This makes it impossible to accurately screen out data segments with significant degradation trends, which makes the subsequent feature extraction process severely affected by fluctuations in operating conditions. It is difficult to obtain effective data that can truly reflect the degradation state of parts, thus creating potential data-level errors for lifespan prediction.

[0003] Existing technologies for analyzing component degradation characteristics and predicting lifespan have several shortcomings. On the one hand, it is difficult to effectively decouple condition-independent degradation features, and the feature extraction process lacks precise residual correction and energy weighting control, resulting in insufficient accuracy of extracted degradation features. On the other hand, the similarity mapping for health status assessment lacks directional sensitivity measurement and weighted distance accumulation, the detection of degradation initiation lacks an adaptive statistical feature mutation judgment mechanism, and the trend extrapolation of health change trajectories does not undergo refined rate smoothing and connection correction. Furthermore, the failure point judgment does not combine curvature extrema for piecewise linear fitting, ultimately leading to low accuracy and reliability in predicting the remaining lifespan of components and an inability to accurately match the actual degradation patterns of components. Therefore, improving the efficiency of lifespan prediction for automotive components has become an urgent problem to be solved. Summary of the Invention

[0004] This invention provides a deep learning-based method for predicting the lifespan of automotive components to address the problems mentioned in the background section.

[0005] To achieve the above objectives, this invention provides a deep learning-based method for predicting the lifespan of automotive components, comprising: S01. Divide the working condition status of the full life cycle data of the target automotive parts to obtain the degradation data fragment of the target automotive parts. S02. Perform cross-domain feature decoupling on the degraded data segment to obtain the condition-independent degradation features of the degraded data segment; S03. Based on the health baseline state of the target automotive component, perform similarity mapping evaluation on the condition-independent degradation features to obtain the continuous health score of the target automotive component. S04. Based on the continuous health score, adaptive detection is performed on the full life cycle data to obtain the degradation start time of the target automotive component; S05. Based on the degradation start time, perform trend extrapolation on the continuous health score to obtain the health change trajectory of the target automotive component; S06. Based on a preset failure threshold, reverse the failure point determination of the health change trajectory to obtain the remaining life prediction value of the automotive component.

[0006] In a preferred embodiment, the step of dividing the full lifecycle data of the target automotive component into operating condition segments to obtain degradation data fragments of the target automotive component includes: Local outlier removal is performed on the full lifecycle data of the target automotive parts to obtain smoothed basic data of the full lifecycle data; The smoothed basic data is subjected to operational mode membership determination to obtain the operating condition labels of the full life cycle data; Based on the operating condition labels, modal boundary cutting and division are performed on the smoothed basic data to obtain the homogeneous operating condition data segments of the full life cycle data; The homogeneous operating condition data segment is screened for significant degradation trends to obtain the degradation data segment of the target automotive component.

[0007] In a preferred embodiment, the step of performing cross-domain feature decoupling on the degraded data segment to obtain the condition-independent degradation features of the degraded data segment includes: The degradation data segment is subjected to operating condition fluctuation frequency analysis to obtain the operating condition characteristic frequency band distribution of the degradation data segment; Based on the frequency band distribution of the operating conditions, frequency domain notch blocking is applied to the degraded data segment to obtain the pure degraded component of the degraded data segment. The intrinsic degradation curve of the pure degradation component is obtained by performing an autoregressive moving average fitting on the pure degradation component. Based on the intrinsic degradation curve, residual correction compensation is performed on the pure degradation component, and feature condensation is performed on the compensated component to obtain the condition-independent degradation features of the degradation data segment.

[0008] In a preferred embodiment, the step of performing residual correction compensation on the pure degradation component based on the intrinsic degradation curve, and performing feature extraction on the compensated component to obtain the condition-independent degradation features of the degradation data segment, includes: The point-by-point difference between the pure degradation component and the intrinsic degradation curve is measured to obtain the instantaneous residual sequence of the pure degradation component; Based on the instantaneous residual sequence, the residual energy weighting index of the pure degenerate component is calculated, wherein the calculation formula of the residual energy weighting index is: ; In the formula, The residual energy weighting index is... The length of the instantaneous residual sequence is... The th in the instantaneous residual sequence One element, The element index of the instantaneous residual sequence, It is an exponential function. The intrinsic degradation curve is the first one. The point and the first Local rate of change at each point This is the cumulative index of the intrinsic degradation curve. The intrinsic degradation curve is the first one. The point and the first Local rate of change at each point The intrinsic degradation curve is the first one. One element; Based on the residual energy weighting index, the instantaneous residual sequence is amplitude corrected, and based on the corrected residual sequence, the intrinsic degradation curve is superimposed and fused to obtain the compensated component of the pure degradation component. The core information of the compensated components is extracted to obtain the condition-independent degradation features of the degraded data fragment.

[0009] In a preferred embodiment, the step of performing a similarity mapping evaluation on the condition-independent degradation features based on the health baseline state of the target automotive component to obtain a continuous health score for the target automotive component includes: The health baseline state of the target automotive component is embedded in a multidimensional manifold space to obtain the baseline distribution manifold of the health baseline state; Based on the baseline distribution manifold, neighborhood projection matching is performed on the condition-independent degradation features to obtain the nearest neighbor mapping points of the condition-independent degradation features. The offset vector between the nearest neighbor mapping point and the condition-independent degradation feature is measured by direction sensitivity to obtain the deviation pointing vector of the condition-independent degradation feature. Based on the deviation pointing vector, the weighted distance accumulation of the nearest neighbor mapping point is performed to obtain the projection residual distance of the condition-independent degradation feature; Based on preset health assessment rules, the projected residual distance is normalized and mapped to obtain the continuous health score of the target automotive component.

[0010] In a preferred embodiment, the step of performing a direction sensitivity measurement on the offset vector between the nearest neighbor mapping point and the condition-independent degradation feature to obtain the deviation pointing vector of the condition-independent degradation feature includes: Spatial orientation determination is performed on the offset vector of the nearest neighbor mapping point that is unrelated to the degradation characteristics of the working condition to obtain the spatial azimuth parameter of the offset vector; The magnitude of the offset vector is measured to obtain the amplitude intensity parameter of the offset vector; Based on the baseline distribution manifold, the local manifold curvature of the offset vector is evaluated to obtain the orientation sensitivity factor of the offset vector, wherein the formula for calculating the orientation sensitivity factor is: ; In the formula, The direction-sensitive factor, The spatial azimuth parameter is... The principal degradation direction angle of the reference distribution manifold at the nearest neighbor mapping point, The angular dispersion measure of the nearest neighbor mapping point. For absolute value operations, It is an exponential function. The preset adjustment coefficient, The geodesic curvature of the reference distribution manifold at the nearest neighbor mapping point. The maximum value of the geodesic curvature in the reference distribution manifold; Based on the direction-sensitive factor, the amplitude intensity parameter is weighted and modulated to obtain the direction modulation amplitude of the offset vector; Vector reconstruction is performed on the spatial azimuth parameter and the direction modulation amplitude to obtain the deviation pointing vector of the condition-independent degradation feature.

[0011] In a preferred embodiment, the step of adaptively detecting the full lifecycle data based on the continuous health score to obtain the degradation start time of the target automotive component includes: The continuous health score is segmented by a sliding time window to obtain local time segments of the continuous health score; Statistical feature parameters are extracted from the local time series segments to obtain the concentration trend of the fluctuation amplitude of the local time series segments; Based on the central trend characterization of the fluctuation amplitude, the significant differences of the local time series segments are compared to obtain the statistical feature mutation location index of the continuous health score; Based on the statistical feature mutation location index, the time axis of the full life cycle data is traced backward to obtain the degradation start time of the target automotive component.

[0012] In a preferred embodiment, the step of extrapolating the continuous health score based on the degradation initiation time to obtain the health change trajectory of the target automotive component includes: Based on the degradation initiation time, the local rate of change of the continuous health score is evaluated to obtain the instantaneous degradation rate of the continuous health score; Fluctuation component smoothing suppression is applied to the instantaneous degradation rate to obtain the intrinsic trend rate component of the instantaneous degradation rate; Based on the intrinsic trend rate component, the continuous health score is recursively extrapolated in multiple steps to obtain the extrapolated predicted value of the continuous health score. The extrapolated predicted value is adjusted to be connected with the continuous health score to obtain the smoothing adjustment amount of the extrapolated predicted value; Based on the smoothing correction amount, the extrapolated predicted value is adjusted point by point, and the adjusted predicted value is smoothly integrated to obtain the health change trajectory of the target automotive component.

[0013] In a preferred embodiment, the step of adjusting the extrapolated predicted value and the continuous health score to obtain a smoothing adjustment amount for the extrapolated predicted value includes: The deviation distribution between the extrapolated predicted value and the continuous health score is obtained by comparing the difference in the neighborhood of the connection point. Based on the deviation distribution, morphological similarity matching is performed on the fluctuation mode of the continuous health score to obtain the intrinsic fluctuation phase of the continuous health score. Based on the intrinsic wave phase, the deviation distribution is phase aligned to obtain the phase synchronization correction component of the deviation distribution; The phase synchronization correction component is subjected to amplitude constraint scaling to obtain the smooth correction amount of the extrapolated prediction value.

[0014] In a preferred embodiment, the step of reversely determining the failure point based on the health change trajectory according to a preset failure threshold to obtain the predicted remaining life of the automotive component includes: By performing curvature extreme value time localization on the health change trajectory, the key turning points of the health change trajectory are obtained; Based on the key turning point, the health change trajectory is approximated by piecewise linear fitting to obtain the rate of decline of the health change trajectory. Based on the descent rate and the preset failure threshold, the health change trajectory is extrapolated and extended to obtain the coordinates of the failure crossing point of the health change trajectory. The remaining life prediction value of the automotive component is obtained by measuring the time axial interval of the coordinates of the failure crossing point.

[0015] Compared with the prior art, the present invention has the following beneficial effects: 1. This invention performs refined operational condition classification on the entire lifecycle data of automotive parts, and combines frequency domain notch blocking and residual correction compensation with residual energy weighting to achieve cross-domain feature decoupling of degradation data and accurately extract operational condition-independent degradation features. At the same time, it embeds the health baseline state into a multi-dimensional manifold space, and completes similarity mapping evaluation through direction sensitivity measurement and weighted distance accumulation to output a continuous health score. This makes the quantitative assessment of the health status of parts more consistent with the actual degradation law, greatly improves the accuracy of degradation feature extraction and health status assessment, and provides high-quality features and data support for life prediction.

[0016] 2. This invention achieves adaptive and accurate detection of the degradation initiation moment through sliding window segmentation and statistical feature mutation determination. Based on this moment, the health score is extrapolated after degradation rate smoothing and connection correction to accurately fit the health change trajectory. Furthermore, key turning points are located by curvature extreme values, and the failure point is determined in reverse by piecewise linear fitting. This makes the health trajectory extrapolation and failure point determination more consistent with the actual degradation trend of the parts, effectively improving the accuracy and reliability of remaining life prediction. At the same time, it simplifies the data processing and feature analysis links in the prediction process, improving the overall efficiency of automotive parts life prediction. Attached Figure Description

[0017] Figure 1 This is a flowchart illustrating a deep learning-based method for predicting the lifespan of automotive parts, as provided in an embodiment of the present invention. The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0018] It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0019] This application provides a deep learning-based method for predicting the lifespan of automotive components. The execution entity of this deep learning-based method includes, but is not limited to, at least one of the following electronic devices that can be configured to execute the method provided in this application: a server, a terminal, etc. In other words, the deep learning-based method for predicting the lifespan of automotive components can be executed by software or hardware installed on a terminal device or a server device. The server includes, but is not limited to, a single server, a server cluster, a cloud server, or a cloud server cluster. The server can be an independent server or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content delivery networks (CDNs), and big data and artificial intelligence platforms.

[0020] Reference Figure 1 The diagram shown is a flowchart illustrating a deep learning-based method for predicting the lifespan of automotive parts according to an embodiment of the present invention. In this embodiment, the deep learning-based method for predicting the lifespan of automotive parts includes: S01. Divide the working condition status of the full life cycle data of the target automotive parts to obtain the degradation data fragment of the target automotive parts. In this embodiment of the invention, the step of dividing the full lifecycle data of the target automotive component into operating condition segments to obtain degradation data fragments of the target automotive component includes: Local outlier removal is performed on the full lifecycle data of the target automotive parts to obtain smoothed basic data of the full lifecycle data; The smoothed basic data is subjected to operational mode membership determination to obtain the operating condition labels of the full life cycle data; Based on the operating condition labels, modal boundary cutting and division are performed on the smoothed basic data to obtain the homogeneous operating condition data segments of the full life cycle data; The homogeneous operating condition data segment is screened for significant degradation trends to obtain the degradation data segment of the target automotive component.

[0021] The entire lifecycle data of the target automotive parts is continuously analyzed in time series. The continuity of each data point with the data points before and after it in the time series dimension is checked point by point. Isolated data points that have no reasonable time series correlation with the numerical changes of the data points before and after them are identified as local outliers. For all identified local outliers, the corresponding numerical values ​​are completed by linear interpolation of the values ​​of the valid data points before and after them. After the identification and completion of all local outliers, the continuous time series data without isolated anomalies is the smoothing base data of the entire lifecycle data.

[0022] Based on the actual operating attributes of the target automotive parts, the characteristics of various operating modes covered during the operation of the parts are sorted out. At the same time, the feature representation dimensions corresponding to different operating modes are determined. The smoothed basic data is split into corresponding dimensions according to the sorted feature representation dimensions. The split data of each dimension is matched with the feature representation dimensions of various operating modes one by one. Based on the feature matching results, the operating mode type corresponding to each continuous data segment in the smoothed basic data is determined. Each continuous data segment corresponding to different operating mode types is marked with a unique mode identifier. The mode identifier marked for each data segment is the operating condition label of the whole life cycle data.

[0023] Extract all location nodes labeled with operating condition tags from the smoothed base data, accurately locate the specific locations where the operating condition tags in the smoothed base data change, and use these locations as cutting boundaries to perform time series segmentation on the smoothed base data. All data points in each independent data segment formed after the segmentation are labeled with the same operating condition tag. Then, perform complete time series data integration on each independent data segment formed after the segmentation. Each independent data segment obtained after integration is the homogeneous operating condition data sub-segment of the whole life cycle data.

[0024] For each homogeneous operating condition data segment, a complete analysis of the numerical change trend is conducted on a time series basis. The continuous numerical change trend from the starting data point to the ending data point within each homogeneous operating condition data segment is clearly identified. Combined with the degradation attributes of the target automotive parts, the numerical change characteristics that match these attributes are identified. Homogeneous operating condition data segments whose numerical change trends exhibit these characteristics are selected and retained, while those that do not exhibit these characteristics are directly eliminated. All the selected and retained homogeneous operating condition data segments are then sequentially spliced ​​together according to the original time series order. The complete time series data formed after splicing is the degradation data segment of the target automotive parts.

[0025] The beneficial effects include identifying and completing local outliers in the full lifecycle data of target automotive parts, effectively eliminating the interference of isolated abnormal data on the original data, forming continuous and smoothed basic data without anomalies, and ensuring the integrity and temporal continuity of the data itself. Combining the operational attributes of parts with condition labeling enables accurate differentiation of data operational modes. The segmentation based on condition labels maintains the uniformity of condition attributes within homogeneous condition data segments, avoiding mutual interference between different condition data. Screening for significant degradation trends based on the degradation attributes of parts accurately retains data segments reflecting the degradation state of parts. The degradation data fragments obtained through orderly splicing can truly reflect the actual degradation state of parts, providing accurate and effective data support for subsequent data processing operations related to automotive part lifespan prediction, and improving the targeting and effectiveness of the data processing stage.

[0026] S02. Perform cross-domain feature decoupling on the degraded data segment to obtain the condition-independent degradation features of the degraded data segment; In this embodiment of the invention, the step of performing cross-domain feature decoupling on the degraded data segment to obtain the condition-independent degradation features of the degraded data segment includes: The degradation data segment is subjected to operating condition fluctuation frequency analysis to obtain the operating condition characteristic frequency band distribution of the degradation data segment; Based on the frequency band distribution of the operating conditions, frequency domain notch blocking is applied to the degraded data segment to obtain the pure degraded component of the degraded data segment. The intrinsic degradation curve of the pure degradation component is obtained by performing an autoregressive moving average fitting on the pure degradation component. Based on the intrinsic degradation curve, residual correction compensation is performed on the pure degradation component, and feature condensation is performed on the compensated component to obtain the condition-independent degradation features of the degradation data segment.

[0027] The process of performing residual correction compensation on the pure degradation component based on the intrinsic degradation curve, and then performing feature extraction on the compensated component to obtain the condition-independent degradation features of the degradation data segment includes: The point-by-point difference between the pure degradation component and the intrinsic degradation curve is measured to obtain the instantaneous residual sequence of the pure degradation component; Based on the instantaneous residual sequence, the residual energy weighting index of the pure degenerate component is calculated, wherein the calculation formula of the residual energy weighting index is: ; In the formula, The residual energy weighting index is... The length of the instantaneous residual sequence is... The th in the instantaneous residual sequence One element, The element index of the instantaneous residual sequence, It is an exponential function. The intrinsic degradation curve is the first one. The point and the first Local rate of change at each point This is the cumulative index of the intrinsic degradation curve. The intrinsic degradation curve is the first one. The point and the first Local rate of change at each point The intrinsic degradation curve is the first one. One element; Based on the residual energy weighting index, the instantaneous residual sequence is amplitude corrected, and based on the corrected residual sequence, the intrinsic degradation curve is superimposed and fused to obtain the compensated component of the pure degradation component. The core information of the compensated components is extracted to obtain the condition-independent degradation features of the degraded data fragment.

[0028] A complete frequency characteristic analysis is conducted on the degraded data segment according to the time series, and the numerical changes of the degraded data segment are transformed into the characteristic manifestation in the frequency domain. The numerical fluctuation intensity corresponding to different frequency intervals is identified. Based on the fluctuation performance of the degraded data segment in different frequency intervals, the frequency interval range related to the operating condition fluctuation is defined. The defined frequency interval range related to the operating condition fluctuation is the operating condition characteristic frequency band distribution of the degraded data segment.

[0029] Based on the obtained operating condition characteristic frequency band distribution, signal blocking processing is performed on the frequency range of the corresponding operating condition characteristic frequency band distribution in the frequency domain. All fluctuation signals in this frequency range are completely blocked, while the signal content of the other frequency ranges in the degraded data segment is retained. The signal after frequency domain blocking processing is restored to the numerical form of a time series. The restored time series value is the pure degraded component of the degraded data segment.

[0030] The pure degradation component is subjected to continuous numerical trend fitting according to the time series. With time as the horizontal axis and the value of the pure degradation component as the vertical axis, a continuous curve fitting operation is performed along the trend of the value change of the pure degradation component. The fitted continuous curve is the intrinsic degradation curve of the pure degradation component.

[0031] The pure degradation component and the intrinsic degradation curve are compared one by one at the same data point position on the time axis. The difference between the pure degradation component value and the intrinsic degradation curve value at each same position is identified. The differences at all positions are arranged in order according to the time series. The resulting difference sequence is the instantaneous residual sequence of the pure degradation component.

[0032] An overall energy characteristic analysis is performed on all values ​​of the instantaneous residual sequence. Combined with the numerical change characteristics of the intrinsic degradation curve, the values ​​of the instantaneous residual sequence are weighted accordingly. By comprehensively considering the overall numerical distribution of the instantaneous residual sequence and the continuous change trend of the intrinsic degradation curve, a comprehensive index that reflects the correlation between the energy characteristics of the instantaneous residual sequence and the intrinsic degradation curve is obtained. This comprehensive index is the residual energy weighting index of the pure degradation component.

[0033] The amplitude of each value in the instantaneous residual sequence is adjusted and corrected according to the residual energy weighting index. The magnitude of each value in the instantaneous residual sequence is changed according to the adjustment rule corresponding to the index to obtain the corrected residual sequence. The values ​​of the corrected residual sequence are then superimposed one by one onto the values ​​at the corresponding positions of the intrinsic degradation curve according to their time axis positions to complete the numerical fusion processing of the entire sequence. The new curve data obtained after the fusion processing is the compensated component of the pure degradation component.

[0034] The core information of all data content of the compensated component is extracted and sorted out. Meaningless numerical fluctuations and redundant information in the compensated component are removed, and the core numerical characteristics and continuous trend information that can truly reflect the degradation state of the target automotive parts are retained. The extracted and retained core information is systematically integrated and refined, and the resulting core information set is the condition-independent degradation feature of the degradation data segment.

[0035] All elements involved in the residual energy weighting index are products of previous data processing. The instantaneous residual sequence is obtained by comparing the differences between the pure degradation component and the intrinsic degradation curve point by point along the same position on the time axis and arranging them in an orderly manner. Its length is determined by the total number of difference data points contained in the sequence, and each element in the sequence is the difference result at the corresponding position. The intrinsic degradation curve is a continuous curve obtained by fitting the pure degradation component to the continuous numerical trend of the time series. Its local rate of change is obtained by calculating the difference between the values ​​of two adjacent points on the curve. The local rate of change at each position corresponds to the difference between the values ​​of adjacent points on the curve. Each element of the intrinsic degradation curve is the value at the corresponding time position on the curve.

[0036] The residual energy weighting index is used to comprehensively reflect the correlation between the energy distribution characteristics of the instantaneous residual sequence and the changes in the intrinsic degradation curve. It reflects the energy attribute of the residual by squaring the elements of the instantaneous residual sequence, and makes targeted adjustments to the residual energy by combining the distribution characteristics of the local rate of change of the intrinsic degradation curve. The adjusted residual energy is then accumulated, and finally the ratio is calculated with the sum of the squares of all elements of the intrinsic degradation curve. This achieves weighted quantification of the residual energy and accurately characterizes the degree of energy contribution of the instantaneous residual sequence relative to the intrinsic degradation curve.

[0037] As the element values ​​of the instantaneous residual sequence increase, the residual energy value after squaring will increase synchronously. If the local rate of change distribution of the intrinsic degradation curve is more concentrated, the adjustment range of the residual energy is more stable, and the adjusted cumulative result will increase with the increase of residual energy, and the residual energy weighting index will increase accordingly. When the sum of the element values ​​of the intrinsic degradation curve increases, its squared cumulative result will increase synchronously. If the adjusted residual energy cumulative result remains unchanged, the residual energy weighting index will decrease accordingly.

[0038] The beneficial effects include frequency characteristic analysis of degraded data segments and delineation of operating condition characteristic frequency band distribution; precise removal of signal interference caused by operating condition fluctuations through frequency domain notch filtering; and the resulting pure degradation components accurately reflecting the basic degradation state of components. The pure degradation components are fitted with intrinsic degradation curves, and residuals are measured point-by-point to form an instantaneous residual sequence. The residuals are corrected using a residual energy weighting index and fused with the intrinsic degradation curve. The resulting compensated components correct data biases, making degradation-related data more closely match the actual degradation patterns of components. After refining the core information of the compensated components and removing redundant and meaningless fluctuations, the extracted operating condition-independent degradation features are completely free from operating condition interference, accurately preserving the core degradation features of components. This provides accurate and effective feature support for subsequent health status assessments, improving the targeting and effectiveness of feature extraction.

[0039] The source and acquisition method of each element in the residual energy weighting index are clearly defined, ensuring that the data source for index calculation is traceable and fully consistent with the pre-processed products, thus guaranteeing the accuracy of the index calculation from the root. This index can accurately quantify the correlation between the energy distribution of the instantaneous residual sequence and the changes in the intrinsic degradation curve. Through targeted adjustments, it achieves scientific weighting of residual energy, clearly characterizing the energy contribution of the residual to the intrinsic degradation curve. Simultaneously, the clearly defined index change trend can accurately reflect the impact of changes in the values ​​of the residual and the intrinsic degradation curve on the index, providing a clear and reliable quantitative basis for subsequent amplitude correction of the instantaneous residual sequence, ensuring the accuracy of residual correction compensation, and thus improving the effectiveness of extracting condition-independent degradation features.

[0040] S03. Based on the health baseline state of the target automotive component, perform similarity mapping evaluation on the condition-independent degradation features to obtain the continuous health score of the target automotive component. In this embodiment of the invention, the step of performing a similarity mapping evaluation on the condition-independent degradation features based on the health baseline state of the target automotive component to obtain a continuous health score for the target automotive component includes: The health baseline state of the target automotive component is embedded in a multidimensional manifold space to obtain the baseline distribution manifold of the health baseline state; Based on the baseline distribution manifold, neighborhood projection matching is performed on the condition-independent degradation features to obtain the nearest neighbor mapping points of the condition-independent degradation features. The offset vector between the nearest neighbor mapping point and the condition-independent degradation feature is measured by direction sensitivity to obtain the deviation pointing vector of the condition-independent degradation feature. Based on the deviation pointing vector, the weighted distance accumulation of the nearest neighbor mapping point is performed to obtain the projection residual distance of the condition-independent degradation feature; Based on preset health assessment rules, the projected residual distance is normalized and mapped to obtain the continuous health score of the target automotive component.

[0041] The step of performing a direction sensitivity measurement on the offset vector of the nearest neighbor mapping point and the condition-independent degradation feature to obtain the deviation pointing vector of the condition-independent degradation feature includes: Spatial orientation determination is performed on the offset vector of the nearest neighbor mapping point that is unrelated to the degradation characteristics of the working condition to obtain the spatial azimuth parameter of the offset vector; The magnitude of the offset vector is measured to obtain the amplitude intensity parameter of the offset vector; Based on the baseline distribution manifold, the local manifold curvature of the offset vector is evaluated to obtain the orientation sensitivity factor of the offset vector, wherein the formula for calculating the orientation sensitivity factor is: ; In the formula, The direction-sensitive factor, The spatial azimuth parameter is... The principal degradation direction angle of the reference distribution manifold at the nearest neighbor mapping point, The angular dispersion measure of the nearest neighbor mapping point. For absolute value operations, It is an exponential function. The preset adjustment coefficient, The geodesic curvature of the reference distribution manifold at the nearest neighbor mapping point. The maximum value of the geodesic curvature in the reference distribution manifold; Based on the direction-sensitive factor, the amplitude intensity parameter is weighted and modulated to obtain the direction modulation amplitude of the offset vector; Vector reconstruction is performed on the spatial azimuth parameter and the direction modulation amplitude to obtain the deviation pointing vector of the condition-independent degradation feature.

[0042] The health baseline status of target automotive parts is expanded and distributed in a multidimensional space. Each characteristic index of the health baseline status is taken as a dimension of the multidimensional space, and the characteristic data of the health baseline status are projected onto the corresponding coordinate position in the multidimensional space. The distribution pattern of all projection points in the multidimensional space is continuously fitted to form a spatial form that can completely represent the distribution law of the health baseline status characteristics. This spatial form is the baseline distribution manifold of the health baseline status.

[0043] The feature data of the condition-independent degradation feature is fully projected onto the multidimensional space of the baseline distribution manifold. The specific location of the feature data in the multidimensional space is accurately located. With this location point as the center, a fixed range of neighborhood area is defined within the coverage of the baseline distribution manifold. All distribution points on the baseline distribution manifold within the neighborhood area are comprehensively searched. The distribution point that is spatially closest to the location point of the condition-independent degradation feature is selected. This distribution point is the nearest neighbor mapping point of the condition-independent degradation feature.

[0044] Starting from the nearest neighbor mapping point and ending at the location of the condition-independent degradation feature in the multidimensional space, a corresponding spatial vector is constructed in the multidimensional space. This spatial vector is the offset vector between the nearest neighbor mapping point and the condition-independent degradation feature. Based on the preset coordinate system and reference direction of the multidimensional space, the angle attribute of this offset vector relative to the preset reference direction in the multidimensional space is determined. This angle attribute is the spatial azimuth parameter of the offset vector.

[0045] Based on the unified distance calculation rules of multidimensional space, the straight-line distance from the starting point to the ending point of the offset vector is accurately measured, and the specific value of this distance is used as the amplitude of the offset vector. This value is the amplitude intensity parameter of the offset vector.

[0046] The specific location on the reference distribution manifold corresponding to the starting point of the offset vector is determined. The curvature characteristics of the reference distribution manifold within a fixed range around this location are comprehensively analyzed. The degree and direction of curvature of the reference distribution manifold at this location are carefully analyzed. All curvature characteristics obtained from the analysis are comprehensively quantified to form a comprehensive index that can fully characterize the curvature attribute at this location. This comprehensive index is the direction sensitivity factor of the offset vector.

[0047] The specific value of the amplitude intensity parameter is fused with the quantitative index of the direction sensitivity factor. The value of the amplitude intensity parameter is adjusted strictly according to the index attribute of the direction sensitivity factor. The new value obtained after complete adjustment is the direction modulation amplitude of the offset vector.

[0048] The spatial pointing attribute represented by the spatial azimuth parameter of the offset vector is fully preserved. The amplitude of the original offset vector is replaced with the specific value of the directional modulation amplitude. A new spatial vector is reconstructed in the original multidimensional space where the reference distribution manifold is located. The reconstructed spatial vector is the deviation pointing vector of the condition-independent degradation feature.

[0049] Using the nearest neighbor mapping point as the starting point for distance accumulation, the spatial distance represented by the deviation pointing vector is continuously accumulated along the spatial direction represented by the deviation pointing vector. At the same time, the accumulated distance value is weighted in a targeted manner by combining the vector attributes of the deviation pointing vector. The comprehensive distance value obtained after weighting is the projection residual distance of the condition-independent degradation feature.

[0050] A fixed health assessment rule is pre-defined. This rule contains a one-to-one mapping relationship between the projected residual distance value and the health score. The mapped health score is within a fixed numerical range. The specific value of the projected residual distance is substituted into the health assessment rule. According to the corresponding mapping relationship in the rule, the projected residual distance is completely converted into a continuous value within the corresponding numerical range. This continuous value is the continuous health score of the target automotive component.

[0051] The spatial azimuth parameter is derived from the spatial orientation determination of the offset vector, which is the angle attribute of the offset vector determined based on the multi-dimensional spatial preset coordinate system and the reference direction. The principal degradation direction angle is the core direction angle obtained by analyzing the degradation characteristics of the reference distribution manifold at the nearest neighbor mapping point after locating the nearest neighbor mapping point. The angle dispersion measure is a quantitative result formed by comprehensively sorting out the differences in the direction angles of various points on the reference distribution manifold around the nearest neighbor mapping point. The geodesic curvature is the curvature attribute obtained by analyzing the degree of curvature of the reference distribution manifold at the location of the nearest neighbor mapping point after locating the location of the nearest neighbor mapping point on the reference distribution manifold. The maximum value of the geodesic curvature is the curvature result with the highest value selected from all locations on the reference distribution manifold. The adjustment coefficient is a fixed value pre-set according to the actual needs of health assessment before carrying out the direction-sensitive factor processing.

[0052] The orientation sensitivity factor is used to comprehensively measure the degree of fit between the spatial orientation of the offset vector and the principal degradation direction of the reference distribution manifold at the nearest neighbor mapping point, as well as the influence of the curvature characteristics of the reference distribution manifold at that point on the offset metric. By comparing the difference between the spatial azimuth parameter and the principal degradation direction angle, and adjusting it in combination with the angular dispersion metric, and further correcting it based on the correlation between geodesic curvature and the maximum geodesic curvature, a quantitative result that can accurately reflect the importance of the offset vector orientation is finally formed, providing a core basis for the amplitude modulation of the offset vector.

[0053] When the difference between the spatial azimuth parameter and the principal degenerate direction angle decreases, the direction sensitivity factor will increase synchronously; when the geodesic curvature value at the nearest neighbor mapping point decreases, the direction sensitivity factor will increase accordingly; when the adjustment coefficient remains fixed and the maximum value of the geodesic curvature increases, the direction sensitivity factor will increase synchronously, and vice versa.

[0054] The beneficial effect is that embedding the health baseline state into a multi-dimensional manifold space to form a baseline distribution manifold provides a precise reference system for the similarity assessment of condition-independent degradation features. By locking the nearest neighbor mapping point through neighborhood projection matching and modulating the amplitude using a direction-sensitive factor, the deviation pointing vector is reconstructed, ensuring that the measurement of spatial offset fully conforms to the baseline distribution pattern. The projection residual distance calculation based on this vector achieves precise weighted quantification of the degree of feature deviation. The continuous health score obtained after normalization according to preset rules can continuously and accurately reflect the health status of components, providing a highly reliable health quantification basis for subsequent degradation initiation detection and lifespan prediction.

[0055] By clearly defining the sources and acquisition methods of each parameter of the direction-sensitive factor, the data sources for factor calculations can be traced back to the products of previous data processing, ensuring the accuracy and rationality of factor calculations from a fundamental level. This factor comprehensively considers the degree of fit between the spatial orientation of the offset vector and the main degradation direction of the baseline distribution manifold, while also making targeted corrections based on the manifold bending characteristics. It accurately quantifies the importance of the offset vector direction, providing a scientific and core basis for the amplitude modulation of the offset vector. Clearly defined factor change trends clarify the specific impact of changes in various related parameters on the factor, making the adjustment direction of amplitude modulation clearer and improving the accuracy of the offset orientation vector construction. This, in turn, provides reliable support for subsequent projection residual distance calculations and continuous health score assessments.

[0056] S04. Based on the continuous health score, adaptive detection is performed on the full life cycle data to obtain the degradation start time of the target automotive component; In this embodiment of the invention, the step of adaptively detecting the full lifecycle data based on the continuous health score to obtain the degradation start time of the target automotive component includes: The continuous health score is segmented by a sliding time window to obtain local time segments of the continuous health score; Statistical feature parameters are extracted from the local time series segments to obtain the concentration trend of the fluctuation amplitude of the local time series segments; Based on the central trend characterization of the fluctuation amplitude, the significant differences of the local time series segments are compared to obtain the statistical feature mutation location index of the continuous health score; Based on the statistical feature mutation location index, the time axis of the full life cycle data is traced backward to obtain the degradation start time of the target automotive component.

[0057] A fixed-length time window is set for the continuous health score. This time window is aligned with the beginning of the time series data of the continuous health score. All continuous health score data within the coverage area of ​​the time window is extracted to form the first data segment. Then, the time window is slid along the time axis of the continuous health score with a single time unit as the sliding step. After each slide, all continuous health score data within the coverage area of ​​the time window is extracted to form an independent data segment, until the time window slides to the end of the time series data of the continuous health score. All the independent data segments extracted in this way are the local time series sub-segments of the continuous health score.

[0058] Each local time series segment is independently analyzed for numerical characteristics. First, the values ​​of all continuous health scores within the segment are extracted to determine the fluctuation range of the values ​​within the segment. Then, the central reference value of all values ​​within the segment is calculated, and the deviation of each value from the central reference value is statistically analyzed. The fluctuation range and the deviation of all values ​​are systematically integrated to form a feature set that can fully reflect the overall state of the fluctuation amplitude of continuous health scores within the local time series segment. This feature set is the central tendency representation of the fluctuation amplitude of the local time series segment.

[0059] A comprehensive feature comparison is performed on the central tendency representations of the fluctuation amplitudes corresponding to two adjacent local time series segments on the time axis. Each feature content of the fluctuation range and numerical deviation in the two representations is checked one by one. All feature differences between the two representations are sorted out and integrated as a whole. The central tendency representations of the fluctuation amplitudes of all adjacent local time series segments are compared sequentially along the time axis of the continuous health score. The position where the feature difference changes abruptly is located, and the corresponding position number in the continuous health score time series data is marked at this position. This number is the statistical feature change position index of the continuous health score.

[0060] The statistical feature mutation location index is precisely matched with the time axis of the whole life cycle data to determine the specific time position of the index on the time axis of the whole life cycle data. Starting from the time position, the time axis of the whole life cycle data is traced point by point in the direction of earlier time. At each time point traced, the fluctuation amplitude central tendency of the local time series segment to which the continuous health score belongs is checked until the time point when the first fluctuation amplitude central tendency recovers to a stable state is traced. This time point is the degradation start time of the target automotive part.

[0061] The beneficial effects include: by using a sliding segmentation of a fixed-length time window, continuous health scores are decomposed into multiple local time-series segments, enabling refined segmented analysis of the temporal changes in health scores and providing clear analytical units for subsequent feature extraction. Extracting the central tendency representation of fluctuation amplitude for each local time-series segment allows for precise quantification of the fluctuation state and central tendency of health scores within each segment, fully presenting the local variation characteristics of health scores. By comparing the representational differences between adjacent segments to locate the index of statistical feature mutation locations, abrupt change nodes in health scores can be accurately captured. Then, by tracing back along the timeline, the time point of the first stable state can be identified, ensuring the accuracy and reliability of the determination of the degradation initiation time. This provides a solid time benchmark for subsequent extrapolation of health change trajectories and lifespan prediction, improving the targeting and reliability of the entire lifespan prediction process.

[0062] S05. Based on the degradation start time, perform trend extrapolation on the continuous health score to obtain the health change trajectory of the target automotive component; In this embodiment of the invention, the step of extrapolating the continuous health score based on the degradation initiation time to obtain the health change trajectory of the target automotive component includes: Based on the degradation initiation time, the local rate of change of the continuous health score is evaluated to obtain the instantaneous degradation rate of the continuous health score; Fluctuation component smoothing suppression is applied to the instantaneous degradation rate to obtain the intrinsic trend rate component of the instantaneous degradation rate; Based on the intrinsic trend rate component, the continuous health score is recursively extrapolated in multiple steps to obtain the extrapolated predicted value of the continuous health score. The extrapolated predicted value is adjusted to be connected with the continuous health score to obtain the smoothing adjustment amount of the extrapolated predicted value; Based on the smoothing correction amount, the extrapolated predicted value is adjusted point by point, and the adjusted predicted value is smoothly integrated to obtain the health change trajectory of the target automotive component.

[0063] The step of linking and correcting the extrapolated predicted value with the continuous health score to obtain a smoothing correction amount for the extrapolated predicted value includes: The deviation distribution between the extrapolated predicted value and the continuous health score is obtained by comparing the difference in the neighborhood of the connection point. Based on the deviation distribution, morphological similarity matching is performed on the fluctuation mode of the continuous health score to obtain the intrinsic fluctuation phase of the continuous health score. Based on the intrinsic wave phase, the deviation distribution is phase aligned to obtain the phase synchronization correction component of the deviation distribution; The phase synchronization correction component is subjected to amplitude constraint scaling to obtain the smooth correction amount of the extrapolated prediction value.

[0064] The specific location of the onset of degradation in the continuous health score time series data is determined. Starting from this location, two adjacent continuous health score data points are selected sequentially along the time axis. The correspondence between the numerical changes and time changes between the two points is analyzed. This analysis operation is completed by combining all adjacent data points on the time axis. The analysis results corresponding to each location are arranged in an orderly manner according to the time series. The resulting sequence is the instantaneous degradation rate of the continuous health score.

[0065] By analyzing the full sequence of instantaneous degradation rate values, irregular numerical fluctuations are identified in the time series. The values ​​of these fluctuations are then corrected point by point using the mean of adjacent values. During the correction process, the overall trend of instantaneous degradation rate is fully preserved. After completing the correction of all fluctuations, the resulting continuous numerical sequence is the intrinsic trend rate component of the instantaneous degradation rate.

[0066] The last data point of the continuous health score time series data is located. The health score value at this point is used as the starting point for recursion. According to the trend of numerical change represented by the intrinsic trend rate component, the health score value corresponding to each subsequent time unit is calculated sequentially. The change law of the intrinsic trend rate component is strictly followed during the calculation process. All the calculated health score values ​​of subsequent time units are arranged in an orderly manner according to the time series. The resulting numerical sequence is the extrapolated predicted value of the continuous health score.

[0067] Locate the connection point between the last data point of the continuous health score and the first data point of the extrapolated predicted value. Delineate a fixed time neighborhood range centered on this connection point. Extract all corresponding time position values ​​of the continuous health score and the extrapolated predicted value within the neighborhood. Calculate the difference between the two sets of values ​​one by one according to their time positions. Arrange all the calculated differences in an ordered time series. The resulting difference series is the deviation distribution between the extrapolated predicted value and the continuous health score.

[0068] By analyzing the numerical variation patterns of continuous health scores over a full time period, recurring fluctuation patterns are extracted as basic fluctuation modes. The numerical variation patterns of the deviation distribution are compared with the morphological details of each basic fluctuation mode. The basic fluctuation mode with the highest morphological fit is selected as the matching result. The combination of the fluctuation time characteristics and morphological characteristics corresponding to this matching result is the intrinsic fluctuation phase of the continuous health score.

[0069] Using the temporal and morphological characteristics of the intrinsic wave phase as a unified benchmark, the numerical arrangement time axis of the deviation distribution is adjusted so that the numerical change pattern of the deviation distribution is completely consistent with the pattern of the intrinsic wave phase. During the adjustment process, the original numerical magnitude characteristics of the deviation distribution are fully preserved. The new difference sequence obtained after the adjustment is the phase synchronization correction component of the deviation distribution.

[0070] A fixed amplitude variation range is set for the phase synchronization correction component. Each value in the phase synchronization correction component is adjusted to this fixed amplitude range by the same proportion. During the adjustment process, the relative magnitude relationship between the values ​​remains unchanged. The new sequence obtained after the amplitude adjustment operation of all values ​​is completed is the smoothing correction amount of the extrapolated prediction value.

[0071] Each value in the smoothing correction is superimposed onto the corresponding value of the extrapolated prediction value according to the corresponding position on the time axis, completing the point-by-point adjustment operation of the extrapolated prediction value. The adjusted extrapolated prediction value is seamlessly connected with the original continuous health score according to the time series. The connected overall numerical series is continuously morphologically fitted to keep the numerical changes in the series continuous and smooth. The complete time series curve obtained after fitting is the health change trajectory of the target automotive parts.

[0072] The beneficial effect is that it assesses the instantaneous degradation rate of the continuous health score from the moment of degradation onset, and then smooths and suppresses its fluctuation components to obtain the intrinsic trend rate component. This effectively eliminates irregular fluctuation interference and retains the core trend of degradation rate change, providing an accurate rate basis for health score extrapolation. After obtaining the extrapolated predicted value based on this component, the deviation distribution is analyzed around the connection point. Through intrinsic fluctuation phase matching, phase alignment, and amplitude constraint scaling, a smoothing correction amount is obtained, making the correction amount highly consistent with the inherent fluctuation law of the health score. By adjusting the extrapolated predicted value through the smoothing correction amount and seamlessly fitting it with the original health score, the resulting health change trajectory is continuous and smooth, accurately reflecting the actual degradation trend of the component's health status. This provides a complete and reliable time series reference for subsequent failure point reverse determination, improving the accuracy and rationality of health trajectory extrapolation.

[0073] S06. Based on a preset failure threshold, reverse the failure point determination of the health change trajectory to obtain the predicted remaining life of the automotive component. In this embodiment of the invention, the step of reversely determining the failure point based on the health change trajectory according to a preset failure threshold to obtain the predicted remaining life of the automotive component includes: By performing curvature extreme value time localization on the health change trajectory, the key turning points of the health change trajectory are obtained; Based on the key turning point, the health change trajectory is approximated by piecewise linear fitting to obtain the rate of decline of the health change trajectory. Based on the descent rate and the preset failure threshold, the health change trajectory is extrapolated and extended to obtain the coordinates of the failure crossing point of the health change trajectory. The remaining life prediction value of the automotive component is obtained by measuring the time axial interval of the coordinates of the failure crossing point.

[0074] By analyzing the full-time curve of the health change trajectory, the curvature of each position on the curve is analyzed point by point along the time axis. The curvature characteristics corresponding to each position are identified, and the curvature characteristics of all positions are fully recorded in time sequence. From all the recorded curvature characteristics, the curve positions where the curvature characteristics reach the extreme values ​​are selected. The specific time points corresponding to these extreme value positions on the time axis of the health change trajectory are accurately marked. These marked time points are the key turning points of the health change trajectory.

[0075] All key turning points are used as segment nodes of the time series curve of the health change trajectory. According to the distribution order of the nodes on the time axis, the time series curve of the entire health change trajectory is divided into multiple continuous curve segments. Linear fitting is performed on each independent curve segment after segmentation. A straight line that fits the curve segment is drawn along the numerical change trend of each curve segment. The drawn straight line is kept in close contact with all data points of the corresponding curve segment. Then, the correspondence between the numerical change and the time change of each fitted straight line is analyzed one by one to determine the rate of change represented by each fitted straight line. The rate characteristics that reflect the continuous decline of health score in all fitted straight lines are integrated to form the rate result that reflects the overall decline law of the health change trajectory, which is the decline rate of the health change trajectory.

[0076] Define the health score value corresponding to the preset failure threshold, and take the last data point of the health change trajectory time series curve as the starting position for extrapolation. Strictly follow the numerical change law represented by the rate of decline, and draw a trend line of health status change continuously along the time axis from this starting position. During the extension process, always follow the core change characteristics of the rate of decline, until the drawn trend line intersects with the horizontal line containing the health score value corresponding to the failure threshold. Accurately determine the x-coordinate and y-coordinate values ​​of this intersection point in the coordinate system of the health change trajectory, where the x-coordinate is the value of the time dimension and the y-coordinate is the health score value corresponding to the failure threshold. The coordinate point containing both time and health score values ​​is the coordinate of the failure crossing point of the health change trajectory.

[0077] The specific time dimension values ​​are accurately extracted from the coordinates of the failure crossing point. At the same time, the current time point is determined when the remaining life prediction operation of the automotive parts is carried out. Using a unified time measurement unit, the time interval between the time value of the failure crossing point and the time value of the current time point is calculated. The number of complete time units contained between the two time points is counted. Then, the remaining time part in the calculated time interval is checked. The number of complete time units and the remaining time part are systematically integrated to form a continuous and complete time measurement result. This time measurement result is the remaining life prediction value of the automotive parts.

[0078] The beneficial effects include: by analyzing the curvature characteristics of the health change trajectory point by point and locating the key turning points corresponding to extreme values, the core nodes of health status changes are accurately captured, providing a precise basis for subsequent segmented analysis of the trajectory. The descent rate obtained by segmenting linearly fitting with key turning points as nodes closely matches the actual descent pattern of the health change trajectory, providing an accurate rate reference for health status extrapolation. The coordinates of the failure crossing point obtained by extrapolating along the descent rate to the failure threshold accurately pinpoint the correspondence between the failure occurrence time and the health score. Then, by calculating the interval between the failure crossing point and the current time using a unified time unit, the remaining life prediction value is obtained. This ensures that the prediction results are continuous and closely match the actual degradation state of the components, significantly improving the accuracy of remaining life prediction and providing a reliable time basis for the operation and maintenance management of automotive components.

[0079] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0080] This application embodiment can acquire and process relevant data based on artificial intelligence technology. Artificial intelligence is the theory, method, technology, and application system that uses digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to obtain optimal results.

[0081] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A deep learning-based method for predicting the lifespan of automotive parts, characterized in that, The method includes: S01. Divide the working condition status of the full life cycle data of the target automotive parts to obtain the degradation data fragment of the target automotive parts. S02. Perform cross-domain feature decoupling on the degraded data segment to obtain the condition-independent degradation features of the degraded data segment; S03. Based on the health baseline state of the target automotive component, perform similarity mapping evaluation on the condition-independent degradation features to obtain the continuous health score of the target automotive component. S04. Based on the continuous health score, adaptive detection is performed on the full life cycle data to obtain the degradation start time of the target automotive component; S05. Based on the degradation start time, perform trend extrapolation on the continuous health score to obtain the health change trajectory of the target automotive component; S06. Based on a preset failure threshold, reverse the failure point determination of the health change trajectory to obtain the remaining life prediction value of the automotive component.

2. The method for predicting the lifespan of automotive parts based on deep learning as described in claim 1, characterized in that, The process of dividing the full lifecycle data of the target automotive parts into operating condition segments to obtain degradation data fragments of the target automotive parts includes: Local outlier removal is performed on the full lifecycle data of the target automotive parts to obtain smoothed basic data of the full lifecycle data; The smoothed basic data is subjected to operational mode membership determination to obtain the operating condition labels of the full life cycle data; Based on the operating condition labels, modal boundary cutting and division are performed on the smoothed basic data to obtain the homogeneous operating condition data segments of the full life cycle data; The homogeneous operating condition data segment is screened for significant degradation trends to obtain the degradation data segment of the target automotive component.

3. The method for predicting the lifespan of automotive parts based on deep learning as described in claim 1, characterized in that, The process of performing cross-domain feature decoupling on the degraded data segment to obtain the condition-independent degradation features of the degraded data segment includes: The degradation data segment is subjected to operating condition fluctuation frequency analysis to obtain the operating condition characteristic frequency band distribution of the degradation data segment; Based on the frequency band distribution of the operating conditions, frequency domain notch blocking is applied to the degraded data segment to obtain the pure degraded component of the degraded data segment. The intrinsic degradation curve of the pure degradation component is obtained by performing an autoregressive moving average fitting on the pure degradation component. Based on the intrinsic degradation curve, residual correction compensation is performed on the pure degradation component, and feature condensation is performed on the compensated component to obtain the condition-independent degradation features of the degradation data segment.

4. The method for predicting the lifespan of automotive parts based on deep learning as described in claim 3, characterized in that, The process of performing residual correction compensation on the pure degradation component based on the intrinsic degradation curve, and then performing feature extraction on the compensated component to obtain the condition-independent degradation features of the degradation data segment includes: The point-by-point difference between the pure degradation component and the intrinsic degradation curve is measured to obtain the instantaneous residual sequence of the pure degradation component; Based on the instantaneous residual sequence, the residual energy weighting index of the pure degenerate component is calculated, wherein the calculation formula of the residual energy weighting index is: ； In the formula, The residual energy weighting index is... The length of the instantaneous residual sequence is... The th in the instantaneous residual sequence One element, The element index of the instantaneous residual sequence, It is an exponential function. The intrinsic degradation curve is the first one. The point and the first Local rate of change at each point This is the cumulative index of the intrinsic degradation curve. The intrinsic degradation curve is the first one. The point and the first Local rate of change at each point The intrinsic degradation curve is the first one. One element; Based on the residual energy weighting index, the instantaneous residual sequence is amplitude corrected, and based on the corrected residual sequence, the intrinsic degradation curve is superimposed and fused to obtain the compensated component of the pure degradation component. The core information of the compensated components is extracted to obtain the condition-independent degradation features of the degraded data fragment.

5. The method for predicting the lifespan of automotive parts based on deep learning as described in claim 1, characterized in that, Based on the health baseline state of the target automotive component, a similarity mapping evaluation is performed on the condition-independent degradation features to obtain a continuous health score for the target automotive component, including: The health baseline state of the target automotive component is embedded in a multidimensional manifold space to obtain the baseline distribution manifold of the health baseline state; Based on the baseline distribution manifold, neighborhood projection matching is performed on the condition-independent degradation features to obtain the nearest neighbor mapping points of the condition-independent degradation features. The offset vector between the nearest neighbor mapping point and the condition-independent degradation feature is measured by direction sensitivity to obtain the deviation pointing vector of the condition-independent degradation feature. Based on the deviation pointing vector, the weighted distance accumulation of the nearest neighbor mapping point is performed to obtain the projection residual distance of the condition-independent degradation feature; Based on preset health assessment rules, the projected residual distance is normalized and mapped to obtain the continuous health score of the target automotive component.

6. The method for predicting the lifespan of automotive parts based on deep learning as described in claim 5, characterized in that, The step of performing a direction sensitivity measurement on the offset vector of the nearest neighbor mapping point and the condition-independent degradation feature to obtain the deviation pointing vector of the condition-independent degradation feature includes: Spatial orientation determination is performed on the offset vector of the nearest neighbor mapping point that is unrelated to the degradation characteristics of the working condition to obtain the spatial azimuth parameter of the offset vector; The magnitude of the offset vector is measured to obtain the amplitude intensity parameter of the offset vector; Based on the baseline distribution manifold, the local manifold curvature of the offset vector is evaluated to obtain the orientation sensitivity factor of the offset vector, wherein the formula for calculating the orientation sensitivity factor is: ； In the formula, The direction-sensitive factor, The spatial azimuth parameter is... The principal degradation direction angle of the reference distribution manifold at the nearest neighbor mapping point, The angular dispersion measure of the nearest neighbor mapping point. For absolute value operations, It is an exponential function. The preset adjustment coefficient, The geodesic curvature of the reference distribution manifold at the nearest neighbor mapping point. The maximum value of the geodesic curvature in the reference distribution manifold; Based on the direction-sensitive factor, the amplitude intensity parameter is weighted and modulated to obtain the direction modulation amplitude of the offset vector; Vector reconstruction is performed on the spatial azimuth parameter and the direction modulation amplitude to obtain the deviation pointing vector of the condition-independent degradation feature.

7. The method for predicting the lifespan of automotive parts based on deep learning as described in claim 1, characterized in that, The step of adaptively detecting the full lifecycle data based on the continuous health score to obtain the degradation start time of the target automotive component includes: The continuous health score is segmented by a sliding time window to obtain local time segments of the continuous health score; Statistical feature parameters are extracted from the local time series segments to obtain the concentration trend of the fluctuation amplitude of the local time series segments; Based on the central trend characterization of the fluctuation amplitude, the significant differences of the local time series segments are compared to obtain the statistical feature mutation location index of the continuous health score; Based on the statistical feature mutation location index, the time axis of the full life cycle data is traced backward to obtain the degradation start time of the target automotive component.

8. The method for predicting the lifespan of automotive parts based on deep learning as described in claim 1, characterized in that, The step of extrapolating the continuous health score based on the degradation initiation time to obtain the health change trajectory of the target automotive component includes: Based on the degradation initiation time, the local rate of change of the continuous health score is evaluated to obtain the instantaneous degradation rate of the continuous health score; Fluctuation component smoothing suppression is applied to the instantaneous degradation rate to obtain the intrinsic trend rate component of the instantaneous degradation rate; Based on the intrinsic trend rate component, the continuous health score is recursively extrapolated in multiple steps to obtain the extrapolated predicted value of the continuous health score. The extrapolated predicted value is adjusted to be connected with the continuous health score to obtain the smoothing adjustment amount of the extrapolated predicted value; Based on the smoothing correction amount, the extrapolated predicted value is adjusted point by point, and the adjusted predicted value is smoothly integrated to obtain the health change trajectory of the target automotive component.

9. The method for predicting the lifespan of automotive parts based on deep learning as described in claim 8, characterized in that, The step of linking and correcting the extrapolated predicted value with the continuous health score to obtain a smoothing correction amount for the extrapolated predicted value includes: The deviation distribution between the extrapolated predicted value and the continuous health score is obtained by comparing the difference in the neighborhood of the connection point. Based on the deviation distribution, morphological similarity matching is performed on the fluctuation mode of the continuous health score to obtain the intrinsic fluctuation phase of the continuous health score. Based on the intrinsic wave phase, the deviation distribution is phase aligned to obtain the phase synchronization correction component of the deviation distribution; The phase synchronization correction component is subjected to amplitude constraint scaling to obtain the smooth correction amount of the extrapolated prediction value.

10. The method for predicting the lifespan of automotive parts based on deep learning as described in claim 1, characterized in that, The method of reversely determining the failure point based on the preset failure threshold and obtaining the remaining life prediction value of the automotive component includes: By performing curvature extreme value time localization on the health change trajectory, the key turning points of the health change trajectory are obtained; Based on the key turning point, the health change trajectory is approximated by piecewise linear fitting to obtain the rate of decline of the health change trajectory. Based on the descent rate and the preset failure threshold, the health change trajectory is extrapolated and extended to obtain the coordinates of the failure crossing point of the health change trajectory. The remaining life prediction value of the automotive component is obtained by measuring the time axial interval of the coordinates of the failure crossing point.

Images