Special vehicle engine digital maintenance system based on big data analysis

Abstract

The application discloses a special vehicle engine digital maintenance system based on big data analysis, comprising: a data set generation module for forming an engine time series data set; a feature generation module for forming a window feature sequence; a prototype construction module for forming a working condition prototype set and an ordered working condition label sequence; a similarity generation module for forming a directed transition similarity matrix; a cost generation module for forming an adaptive segmentation cost sequence; an initial segmentation module for forming an initial working condition state segment set and an initial state segment boundary by using an improved SegmenTier algorithm; an evidence interval generation module for forming a precursor evidence interval; and a re-segmentation output module for outputting a working condition recognition result, a precursor segment positioning result and a maintenance triggering result. The application improves the working condition segmentation accuracy and the maintenance triggering reliability.

Description

Technical Field

[0001] This invention relates to the field of digital engine maintenance technology, and in particular to a digital maintenance system for special vehicle engines based on big data analysis. Background Technology

[0002] With the increasing demand for special vehicles in extreme environments such as off-road, Gobi, desert, plateau, hot, and cold regions, digital maintenance technologies focusing on engine operating status monitoring, fault precursor identification, and maintenance decision optimization have received widespread attention. Existing engine maintenance systems primarily rely on manual experience-based interpretation, fixed threshold alarms, or single fault diagnosis results for condition assessment. However, these systems commonly suffer from the following problems in practical applications: The collected operational monitoring data, status event data, and maintenance feedback data are scattered, have inconsistent sampling rhythms, and are frequently subject to abnormal disturbances. This results in weak correlation and poor continuity of data at different time points, making it difficult to form a stable and reliable foundation for time series analysis. For multi-condition and non-stationary engine operation processes, existing methods often only focus on numerical anomalies or static category classifications at a single moment, making it difficult to accurately characterize directional changes, state continuity characteristics, and precursor clustering characteristics during the condition transition process. This leads to coarse initial state segment divisions and inaccurate boundary positioning. During maintenance implementation, existing solutions typically lack effective utilization of maintenance feedback records, making it difficult to retrospectively correct preceding state segments based on subsequent maintenance events. This results in delayed precursor interval identification and insufficient basis for maintenance triggering, affecting the accuracy and practicality of digital engine maintenance results.

[0003] Therefore, how to provide a digital maintenance system for special vehicle engines based on big data analysis is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0004] One objective of this invention is to propose a digital maintenance system for special vehicle engines based on big data analysis. This invention employs an improved SegmenTier algorithm, combined with directed transition similarity calculation, adaptive segmentation cost generation, and evidence interval resegmentation processing, to recursively divide engine operating condition segments, locate precursor intervals, and determine maintenance triggers. It has the advantages of high accuracy in state segmentation, strong precursor recognition capability, and high reliability in maintenance triggering.

[0005] The digital maintenance system for special vehicle engines based on big data analysis according to an embodiment of the present invention includes: The dataset generation module is used to perform time alignment, anomaly correction, and numerical normalization on engine data to form an engine time-series dataset. The feature generation module is used to perform time window decomposition and feature calculation on the engine time series dataset to form a window feature sequence; The prototype building module is used to cluster and merge window feature sequences to form a set of working condition prototypes and an ordered working condition label sequence. The similarity generation module is used to calculate the directed transition similarity matrix corresponding to the set of working condition prototypes based on the evolution direction of adjacent time windows; The cost generation module is used to calculate the adaptive segmented cost sequence based on the degree of state continuity and the degree of precursor aggregation at the time position; The initial segmentation module is used to recursively segment the ordered working condition label sequence using the improved SegmenTier algorithm, integrate the directed transition similarity matrix into the cumulative score within the segment, and couple the adaptive segmentation cost sequence to the boundary determination process to form the initial working condition state segment set and the initial state segment boundary. The evidence interval generation module is used to locate the post-maintenance verification evidence anchor point based on the maintenance feedback record and backtrack to form the precursor evidence interval; The re-segmentation output module is used to update the initial working condition state segment set and the initial state segment boundary within the precursor evidence interval, and output the working condition identification result, precursor segment location result, and maintenance trigger result.

[0006] Optionally, the dataset generation module specifically comprises: Collect engine operation monitoring data, status event data, and maintenance feedback data; extract data identifiers, timestamps, parameter fields, and numerical fields to form a raw data record set; and sort and map the raw data record set according to the timestamp to form a time-aligned record set. For missing locations in the time-aligned record set, interpolation and neighborhood completion are performed to form a completed record set. Adjacent differences and local fluctuation values ​​are calculated for parameter fields and numerical fields in the completed record set. Abnormal records are identified by combining preset value ranges, difference thresholds, and fluctuation thresholds to form an abnormal marked record set. Replacement and smoothing corrections are performed on the abnormal locations according to the abnormality marker record set to form an abnormality correction record set. Numerical normalization is then performed on the numerical fields in the abnormality correction record set to form a normalized record set. Finally, the records in the normalized record set are arranged in chronological order to form an engine time series dataset.

[0007] Optionally, the feature generation module specifically comprises: Sliding decomposition is performed on the engine time series dataset according to the preset time window length and time window step. A time window segment set is established according to the start time position, end time position and window identifier corresponding to each time window, and parameter field sequence and numerical field sequence are extracted in each time window segment. For the parameter field sequence and numerical field sequence in each time window segment, calculate the numerical change, adjacent difference, continuous change slope, local fluctuation value and abnormal distribution value according to the time position, and combine the numerical change, adjacent difference, continuous change slope, local fluctuation value and abnormal distribution value according to the field order to form the time window feature record; The time window feature records are arranged in chronological order to form a window feature sequence.

[0008] Optionally, the prototype building module specifically comprises: Arrange the time window feature records in the window feature sequence according to the time position, extract the window identifier and the feature value corresponding to the window identifier, and combine them according to the field order to form a time window feature record set; Within the time window feature record set, each time window feature record is compared one by one. The feature difference value is calculated according to the difference between the feature values, and the feature difference value is written into the comparison relationship between the time window feature records to form a feature comparison record set. Based on the feature comparison record set, clustering and merging are performed according to the feature difference value. Time window feature records whose feature difference value meets the merging condition are classified into the same working condition category, and the time window feature records are aggregated according to the working condition category to form a working condition category record set. Around the working condition category in the working condition category record set, the feature values ​​belonging to the same working condition category are aggregated to form a working condition prototype set, and the working condition categories corresponding to the time window feature records are arranged in order according to the time position to form an ordered working condition label sequence.

[0009] Optionally, the similarity generation module specifically comprises: Arrange the time window feature records in the window feature sequence according to the time position, extract the previous time window feature records and the next time window feature records corresponding to adjacent time positions, and calculate the change in feature values ​​of the previous time window feature records and the next time window feature records at the same field position to form an evolution direction record set. The previous time window feature records and the next time window feature records in the evolution direction record set are respectively mapped to the previous working condition category and the next working condition category in the ordered working condition label sequence. A working condition transfer record set is established according to the correspondence between the previous working condition category and the next working condition category, and the change in feature value is written into the working condition transfer record in the working condition transfer record set. For each work condition transfer record in the work condition transfer record set, similarity calculation is performed based on the consistency of the direction of the change in feature value at the field position. The results are then written into matrix cells according to the arrangement of the previous work condition category and the next work condition category in the work condition prototype set, forming a directed transfer similarity matrix corresponding to the work condition prototype set.

[0010] Optionally, the cost generation module specifically comprises: Arrange the time window feature records in the window feature sequence according to the time position, extract the time window feature records corresponding to each time position and the corresponding working condition category in the ordered working condition label sequence, perform continuous statistics on the same working condition category at consecutive time positions, and accumulate the feature value change amplitude between time window feature records in the continuous interval. Calculate the state continuity degree corresponding to each time position based on the working condition category continuity length and feature value change amplitude. Around the time window feature records in the window feature sequence, the abnormal distribution values ​​are statistically analyzed in continuous intervals according to the time position. The cumulative results of abnormal distribution values ​​within the continuous interval and the increase or decrease of abnormal distribution values ​​between adjacent time positions are jointly calculated. The degree of precursor aggregation corresponding to each time position is calculated based on the concentration and change intensity of abnormal distribution values ​​within the continuous interval. The state continuation degree and precursor aggregation degree corresponding to each time position are paired according to the time position, and the state continuation degree and precursor aggregation degree at the same time position are combined and calculated to form the segmented cost value corresponding to each time position. Then, the segmented cost values ​​are arranged in the order of time position to form an adaptive segmented cost sequence.

[0011] Optionally, the initial segmentation module specifically comprises: Arrange the ordered working condition label sequence in chronological order, construct candidate segment intervals based on the continuous relationship between time positions, and record the start time position, end time position, and working condition category within the interval according to the candidate segment intervals to form a segmented candidate record set; Within the segmented candidate record set, the improved SegmenTier algorithm is used to perform intra-segment cumulative scoring. The directed transition similarity between adjacent working condition categories is extracted sequentially along the time position in the candidate segment interval. The directed transition similarity between each time position in the same candidate segment interval is accumulated item by item to form an intra-segment scoring record set. Around the candidate segment intervals in the score record set within the segment, the improved SegmenTier algorithm is used to perform boundary determination. The segmentation cost value corresponding to the boundary position of the candidate segment interval is introduced into the segmentation calculation, and the cumulative score result within the segment is jointly compared with the segmentation cost value corresponding to the boundary position to form a boundary determination record set. Based on the boundary determination record set, the candidate segment intervals are recursively updated. The candidate segment intervals that meet the retention conditions are confirmed as the initial working condition segments. The candidate segment intervals that meet the splitting conditions are split into new intervals according to the boundary positions, forming a segmented state record set. The initial state segment boundary is determined based on the recursive results in the segmented state record set, and the ordered working condition label sequence is divided into intervals according to the initial state segment boundary to form the initial working condition state segment set and the initial state segment boundary corresponding to the initial working condition state segment set.

[0012] Optionally, the evidence interval generation module specifically comprises: Arrange the maintenance feedback records in chronological order, and extract the record identifier, timestamp, maintenance result marker, and associated status information to form a maintenance feedback record set. Based on the maintenance feedback records in the maintenance feedback record set, abnormal maintenance events are identified according to the maintenance result markers, and the time location corresponding to the abnormal maintenance events is determined as the post-verification maintenance evidence anchor point, forming an evidence anchor point record set; Starting from the posterior dimension guarantee anchor point in the evidence anchor point record set, backtrack along the time position to extract the corresponding initial working condition state segment and the boundary of the initial state segment within the backtracking range, forming the evidence backtracking record set. Based on the posterior maintenance guarantee anchor point, initial operating condition segment, and initial state segment boundary in the evidence retrospective record set, the start position and end position of the precursor interval corresponding to the posterior maintenance guarantee anchor point are determined, forming the precursor evidence interval.

[0013] Optionally, the re-segmentation output module specifically comprises: The precursor evidence intervals are mapped to the corresponding interval ranges in the initial operating condition segment set. The initial operating condition segments and the boundaries of the initial operating condition segments within the interval ranges are extracted to form a re-segmentation candidate record set. Around the initial working condition state segment in the candidate record set for resegmentation, the cumulative result of the directed transition similarity between adjacent working condition categories within the interval is recalculated according to the time position, and the boundary determination is re-executed in combination with the segmentation cost value corresponding to each time position within the interval to form a resegmentation determination record set. Based on the boundary determination results in the re-segmentation determination record set, the initial working condition state segment within the precursor evidence interval is subjected to interval preservation, interval segmentation, and interval merging to form the updated working condition state segment set and the updated state segment boundary. The operating condition segments in the set are updated in chronological order, and the operating condition identification results, precursor segment location results, and maintenance trigger results are output.

[0014] The beneficial effects of this invention are: This invention achieves continuous and comparable engine time-series characterization results by uniformly aligning, correcting anomalies, and normalizing engine operation monitoring data, status event data, and maintenance feedback data, combined with time window decomposition, feature calculation, operating condition prototype construction, and directed transition similarity matrix generation. Compared to existing methods relying on fixed threshold alarms, single diagnostic judgments, or static category classification, this invention utilizes an improved SegmenTier algorithm to integrate the directed transition similarity matrix into the segment cumulative score, coupling the adaptive segmentation cost sequence to the boundary judgment process. This allows the segmentation of operating condition states to move beyond simple numerical fluctuation judgments, simultaneously reflecting the direction of operating condition transition, the degree of state continuity, and the degree of precursor aggregation. This improves the accuracy of the initial set of operating condition segments and the initial state segment boundaries, enhancing the stability and reliability of operating condition identification results under complex engine operating conditions.

[0015] Furthermore, this invention locates post-maintenance verification data anchor points based on maintenance feedback records and traces back to form a precursor evidence interval. Within the precursor evidence interval, it performs resegmentation processing on the initial operating condition segment set and the initial state segment boundaries, enabling maintenance results to participate in precursor interval identification and boundary correction. Compared to existing solutions where maintenance feedback records are only used for post-event archiving or simple maintenance confirmation, this invention establishes a clear correlation between maintenance feedback and preceding operating states, improving the consistency between precursor segment location results and the actual anomaly evolution process. This enhances the relevance and reliability of maintenance triggering results, reduces the risk of misjudgment due to coarse operating condition segment division and delayed precursor identification, and provides more accurate and practical technical support for digital maintenance of special vehicle engines. Attached Figure Description

[0016] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is a structural diagram of the digital maintenance system for special vehicle engines based on big data analysis proposed in this invention. Figure 2 This is a schematic diagram of the initial segmentation process of the digital maintenance system for special vehicle engines based on big data analysis proposed in this invention. Figure 3 This is a schematic diagram of the re-segmentation process of the digital maintenance system for special vehicle engines based on big data analysis proposed in this invention. Detailed Implementation

[0017] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0018] refer to Figures 1-3 A digital maintenance system for special vehicle engines based on big data analytics includes: The dataset generation module is used to perform time alignment, anomaly correction, and numerical normalization on engine data to form an engine time-series dataset. The feature generation module is used to perform time window decomposition and feature calculation on the engine time series dataset to form a window feature sequence; The prototype building module is used to cluster and merge window feature sequences to form a set of working condition prototypes and an ordered working condition label sequence. The similarity generation module is used to calculate the directed transition similarity matrix corresponding to the set of working condition prototypes based on the evolution direction of adjacent time windows; The cost generation module is used to calculate the adaptive segmented cost sequence based on the degree of state continuity and the degree of precursor aggregation at the time position; The initial segmentation module is used to recursively segment the ordered working condition label sequence using the improved SegmenTier algorithm, integrate the directed transition similarity matrix into the cumulative score within the segment, and couple the adaptive segmentation cost sequence to the boundary determination process to form the initial working condition state segment set and the initial state segment boundary. The evidence interval generation module is used to locate the post-maintenance verification evidence anchor point based on the maintenance feedback record and backtrack to form the precursor evidence interval; The re-segmentation output module is used to update the initial working condition state segment set and the initial state segment boundary within the precursor evidence interval, and output the working condition identification result, precursor segment location result, and maintenance trigger result.

[0019] In this embodiment, the dataset generation module specifically comprises: Collect engine operation monitoring data, status event data, and maintenance feedback data. Extract data identifiers, timestamps, parameter fields, and numerical fields according to the data source. Then, associate and organize the parameter fields and numerical fields corresponding to the same data identifier within the same time slice to form a raw data record set. Next, sort the raw data record set according to the timestamp and map the sorted timestamps to a unified time base so that records from different sources correspond to the same time position under a unified time scale, forming a time-aligned record set. For missing positions in the time-aligned record set, interpolation is performed to complete the missing positions based on the valid records before and after the missing positions. Neighborhood completion is also performed by combining the related records of adjacent time positions to form a complete record set. Then, the difference between adjacent time positions and the fluctuation value within the local time range are calculated for the parameter fields and numerical fields in the complete record set. The difference and local fluctuation value are compared with the preset value range, difference threshold, and fluctuation threshold, respectively, to identify abnormal records corresponding to numerical mutations, continuous drift, and local abnormal fluctuations, forming an abnormal marked record set. Replacement correction and smoothing correction are performed on the abnormal positions according to the abnormal mark record set. Replacement correction is used to replace the abnormal value with the adjacent valid record, and smoothing correction is used to smooth the continuous values ​​before and after the abnormal position to form an abnormal correction record set. Then, the numerical fields in the abnormal correction record set are subjected to unit unification and numerical regularization to form a regularized record set. Finally, the records in the regularized record set are arranged in chronological order to form the engine time series dataset.

[0020] In this embodiment, the feature generation module specifically comprises: Sliding decomposition is performed on the engine time series dataset according to the preset time window length and preset time window step. The preset time window length is determined based on the sampling time interval corresponding to a single record in the engine time series dataset and the number of consecutive records to be covered in a single time window. The preset time window step is determined based on the number of overlapping records to be retained between adjacent time windows. A time window segment set is established according to the start time position, end time position and window identifier corresponding to each time window, and parameter field sequence and numerical field sequence are extracted in each time window segment. For the parameter field sequence and numerical field sequence in each time window segment, the numerical change, adjacent difference, continuous change slope, local fluctuation value, and abnormal distribution value are calculated according to the time position. The numerical change is used to characterize the change amplitude between the values ​​in the previous segment and the values ​​in the next segment within the time window segment. The adjacent difference is used to characterize the degree of numerical change between adjacent time positions. The continuous change slope is used to characterize the continuous change trend of the numerical field along the time position. The local fluctuation value is used to characterize the dispersion of the values ​​around the change trend within the time window segment. The abnormal distribution value is used to characterize the distribution density of abnormal values ​​along the time position within the time window segment. The numerical change, adjacent difference, continuous change slope, local fluctuation value, and abnormal distribution value are combined in the field order to form the time window feature record. The time window feature records are arranged in chronological order to form a window feature sequence.

[0021] In this embodiment, the prototype construction module specifically includes: Arrange the time window feature records in the window feature sequence according to the time position order, extract the window identifier and the feature value corresponding to the window identifier, and combine them according to the field order to form a time window feature record set. The time position order is used to maintain the sequential correspondence of the time window feature records in the engine time series dataset, and the field order is used to ensure that each time window feature record corresponds to the same type of feature value in the same field position. Within the time window feature record set, each time window feature record is compared one by one. The feature difference value is calculated according to the feature value difference at the corresponding field position, and the feature difference value is written into the comparison relationship between the time window feature records to form a feature comparison record set. In the comparison process, the difference value is calculated for the feature values ​​at each field position of any two time window feature records, and then the difference values ​​corresponding to each field position are summarized into the feature difference value used to characterize the overall degree of difference. Based on the feature comparison record set, clustering and merging are performed according to the feature difference value. Time window feature records whose feature difference value meets the merging condition are classified into the same working condition category. The time window feature records are then aggregated according to the working condition category to form a working condition category record set. The merging process first determines the correspondence of time window feature records that meet the merging condition, and then merges the time window feature records with merging relationship into the same working condition category. The aggregation process summarizes and aggregates the records according to the working condition category. Around the working condition category record set, the feature values ​​belonging to the same working condition category are aggregated to form a working condition prototype set. The working condition categories corresponding to the time window feature records are arranged in order according to the time position to form an ordered working condition label sequence. The aggregation process summarizes and calculates the feature values ​​of each time window feature record in the same field position within the same working condition category, so that each working condition category corresponds to a set of working condition prototype results that can represent the overall characteristics of the category.

[0022] In this embodiment, the similarity generation module specifically comprises: Arrange the time window feature records in the window feature sequence according to the time position, extract the previous time window feature records and the next time window feature records corresponding to adjacent time positions, and calculate the change of feature values ​​of the previous time window feature records and the next time window feature records at the same field position to form an evolution direction record set. In the calculation process, the change results of the previous time window feature records to the next time window feature records are obtained for the feature values ​​at the same field position, so that each record in the evolution direction record set corresponds to a set of feature value changes along the time position. The previous time window feature records and the next time window feature records in the evolution direction record set are respectively mapped to the previous working condition category and the next working condition category in the ordered working condition label sequence. The working condition transfer record set is established according to the correspondence between the previous working condition category and the next working condition category. The feature value change is written into the working condition transfer record in the working condition transfer record set. The establishment process first determines the previous working condition category and the next working condition category based on the arrangement position of the time window feature records in the ordered working condition label sequence. Then, the previous working condition category, the next working condition category and the corresponding feature value change are combined into the same working condition transfer record. For each work condition transfer record in the work condition transfer record set, similarity calculation is performed based on the directional consistency of the feature value changes at the field positions. The results are then written into matrix cells according to the arrangement of the previous and next work condition categories in the work condition prototype set, forming a directed transfer similarity matrix corresponding to the work condition prototype set. The similarity calculation process compares the directional consistency of the feature value changes in the work condition transfer records field by field, and then summarizes the directional consistency results corresponding to each field position into the similarity results of the transfer relationship of the corresponding work condition category. The row positions in the matrix cells are used to identify the previous work condition category, and the column positions in the matrix cells are used to identify the next work condition category.

[0023] In this embodiment, the cost generation module specifically comprises: The time window feature records in the window feature sequence are arranged in chronological order, and the time window feature records corresponding to each time position and the corresponding work condition categories in the ordered work condition label sequence are extracted. The same work condition category at consecutive time positions is statistically analyzed, and the feature value change amplitude between time window feature records within the consecutive interval is accumulated. The state continuity degree corresponding to each time position is calculated based on the work condition category continuity length and the feature value change amplitude. The calculation of the state continuity degree is processed using textual rules. First, the number of time positions where the work condition category remains unchanged on the time axis represents the work condition category continuity length. Then, the absolute value difference of the feature values ​​corresponding to adjacent time positions within the continuity interval is accumulated item by item to represent the feature value change amplitude. The longer the work condition category continuity length, the higher the probability that the current time position is in a continuous state. The smaller the feature value change amplitude, the higher the probability that the current time position is in a stable transition state. The state continuity degree is determined by the work condition category continuity length and the feature value change amplitude. Time positions with a longer work condition category continuity length and a smaller feature value change amplitude correspond to a higher state continuity degree, while time positions with a shorter work condition category continuity length and a larger feature value change amplitude correspond to a lower state continuity degree. Around the time window feature records in the window feature sequence, the abnormal distribution values ​​are statistically analyzed in continuous intervals according to time position. The cumulative results of abnormal distribution values ​​within the continuous interval and the increase or decrease of abnormal distribution values ​​between adjacent time positions are jointly calculated. The degree of precursor aggregation corresponding to each time position is calculated based on the concentration and intensity of change of abnormal distribution values ​​within the continuous interval. The calculation of the degree of precursor aggregation is processed using textual rules. First, the time window feature records in which abnormal distribution values ​​appear continuously at adjacent time positions are classified into the same continuous interval. Then, the abnormal distribution values ​​corresponding to each time position within the continuous interval are accumulated one by one to characterize the concentration of abnormal signals within the continuous interval. Then, the abnormal distribution values ​​between adjacent time positions are compared before and after to characterize the enhancement of abnormal signals along the time position. The larger the cumulative result of abnormal distribution values ​​within the continuous interval, the more obvious the aggregation of abnormal signals. The more obvious the increase of abnormal distribution values ​​between adjacent time positions, the more obvious the expansion of abnormal signals. The degree of precursor aggregation is determined by the cumulative result of abnormal distribution values ​​and the increase or decrease of abnormal distribution values. Time positions with larger cumulative results of abnormal distribution values ​​and continuous increase of abnormal distribution values ​​correspond to a higher degree of precursor aggregation. The state continuation degree and precursor aggregation degree corresponding to each time position are paired according to time position, and the state continuation degree and precursor aggregation degree at the same time position are combined for calculation to form the segmented cost value corresponding to each time position. Then, the segmented cost values ​​are arranged in order of time position to form an adaptive segmented cost sequence. The calculation of segmented cost values ​​is processed using textual rules. The state continuation degree is used to characterize the strength of maintaining the continuous extension of the current working condition category, and the precursor aggregation degree is used to characterize the strength of triggering boundary segmentation at the current time position. When the state continuation degree is high, the segmented cost value is increased, and when the precursor aggregation degree is high, the segmented cost value is decreased, so that the stable continuation position corresponds to the higher segmented cost value, and the precursor aggregation position corresponds to the lower segmented cost value. After being arranged in order of time position, an adaptive segmented cost sequence is formed.

[0024] In this embodiment, the initial segmentation module is specifically as follows: The ordered working condition label sequence is arranged in chronological order. Candidate segment intervals are constructed based on the continuous relationship between time positions. The start time position, end time position, and working condition category arrangement results within the interval are recorded for each candidate segment interval to form a segmented candidate record set. The construction process takes each time position in the ordered working condition label sequence as the starting point of the interval and extends the interval continuously along the time position to the end point, so that each candidate segment interval corresponds to a continuous time range. The working condition category arrangement results within the interval are used to record the sequential distribution of each working condition category from the start time position to the end time position. Within the segmented candidate record set, the improved SegmenTier algorithm is used to perform intra-segment cumulative scoring. The directed transition similarity between adjacent working condition categories is extracted sequentially along the time position in the candidate segment interval. The directed transition similarity between each time position in the same candidate segment interval is accumulated item by item to form an intra-segment scoring record set. The intra-segment cumulative scoring process uses adjacent time positions as the calculation unit. The directed transition similarity from the working condition category corresponding to the previous time position to the working condition category corresponding to the next time position is sequentially accumulated. The cumulative result is used to characterize the degree of transition consistency when the working condition categories within the candidate segment interval are continuously extended along the time position. The higher the cumulative result, the more coherent the working condition transition within the candidate segment interval. Around the candidate segment intervals in the intra-segment scoring record set, an improved SegmenTier algorithm is used to perform boundary determination. The segmentation cost value corresponding to the boundary position of the candidate segment interval is introduced into the segmentation calculation. The cumulative intra-segment scoring result is jointly compared with the segmentation cost value corresponding to the boundary position to form a boundary determination record set. In the boundary determination process, the segmentation cost value corresponding to the end position of each candidate segment interval is located. Then, the cumulative intra-segment scoring result and the segmentation cost value are placed in the same determination relationship. The cumulative intra-segment scoring result is used to characterize the strength of maintaining the continuous extension of the current candidate segment interval, and the segmentation cost value is used to characterize the strength of performing interval segmentation at the boundary position. Based on the boundary judgment record set, the candidate segment intervals are recursively updated. The candidate segment intervals that meet the retention conditions are confirmed as the initial working condition segments. The candidate segment intervals that meet the splitting conditions are split into new intervals according to the boundary positions to form a segmented state record set. The recursive update process is gradually advanced along the time position. The boundary judgment result of the current candidate segment interval is used to perform a retention judgment or a splitting judgment. The retention judgment is used to directly write the current candidate segment interval into the segmented state record set. The splitting judgment is used to divide the current candidate segment interval into the front segment interval and the back segment interval at the boundary position. The split intervals are then included in the recursive update process. The initial state segment boundary is determined based on the recursive results in the segmented state record set. The ordered working condition label sequence is then divided into intervals according to the initial state segment boundary, forming an initial working condition state segment set and the initial state segment boundary corresponding to the initial working condition state segment set. The determination process uses the interval start time position and interval end time position in the segmented state record set as the boundary positioning basis. The interval division process uses the initial state segment boundary as the dividing point to divide the ordered working condition label sequence into interconnected continuous intervals, so that each continuous interval corresponds to an initial working condition state segment.

[0025] This invention introduces an improved SegmenTier algorithm into the initial segmentation module, which has at least three improvements and corresponding beneficial effects compared to the unimproved SegmenTier algorithm. First, this invention does not use the intra-segment scoring method that focuses on static category consistency in the unimproved SegmenTier algorithm. Instead, it introduces directed transition similarity into the intra-segment cumulative scoring process. This allows the scoring results of candidate segment intervals to not only reflect the degree of clustering of working condition categories within the interval, but also the degree of transition coherence when adjacent working condition categories extend continuously along the time position. This solves the problem that segmenting based solely on static category similarity easily ignores the direction of working condition migration and easily categorizes state changes with opposite directions but similar categories into the same interval, thus improving the accuracy of working condition transition identification in complex operating scenarios. Secondly, this invention does not adopt the uniform and fixed segmentation penalty approach of the unmodified SegmentTier algorithm. Instead, it couples the adaptive segmentation cost generated based on the degree of state continuity and the degree of precursor aggregation to the boundary determination process. This allows different time positions to correspond to different segmentation intensities, raising the boundary segmentation threshold in intervals with continuous and stable states, and lowering it in intervals with significant precursor aggregation and prominent anomaly enhancement. This suppresses invalid segmentation within stable operating conditions, retains truly diagnostically significant abnormal boundaries, and improves the rationality of the initial set of operating condition state segments and their boundaries. Thirdly, this invention jointly compares the cumulative score within a segment with the segmentation cost at the boundary position during the recursive update process. Instead of using a single score result to directly determine retention or segmentation, it integrates the continuity of operating condition transitions and the boundary segmentation intensity into the same recursive determination framework. This ensures that interval retention, interval splitting, and boundary determination are all based on dual constraints, thereby enhancing the adaptability of the segmentation results to the operating condition evolution process. Based on the above improvements, this invention can effectively reduce the problems of over-segmentation, missegmentation, and omission of abnormal precursors in the processing of non-stationary data under multiple engine operating conditions by the unimproved SegmenTier algorithm. It makes the initial segmentation results more consistent with the engine state evolution law, provides a more stable basis for the generation of precursor evidence intervals and re-segmentation output, and ultimately improves the accuracy and reliability of operating condition identification results, precursor segment location results, and maintenance triggering results.

[0026] In this embodiment, the evidence interval generation module specifically comprises: The maintenance feedback records are arranged in chronological order. The record identifier, timestamp, maintenance result marker, and associated status information are extracted to form a maintenance feedback record set. The arrangement process is completed according to the chronological relationship of the timestamps. The record identifier is used to distinguish different maintenance feedback records. The maintenance result marker is used to characterize the result status after maintenance processing. The associated status information is used to record the status association content corresponding to the maintenance feedback record. Based on the maintenance feedback records in the maintenance feedback record set, abnormal maintenance events are identified according to the maintenance result markers, and the time position corresponding to the abnormal maintenance event is determined as the post-verification maintenance evidence anchor point, forming an evidence anchor point record set. The identification process checks each maintenance result marker, and the maintenance feedback record that represents the abnormal handling result, abnormal confirmation result, and abnormal repair result is determined as an abnormal maintenance event. The determination process writes the time position corresponding to the abnormal maintenance event into the evidence anchor point record set to mark the starting position of evidence backtracking. Starting from the posterior dimension guarantee anchor point in the evidence anchor point record set, backtracking along the time position, extracting the corresponding initial working condition state segment and the boundary of the initial state segment within the backtracking range, forming an evidence backtracking record set. The backtracking process starts from the time position corresponding to the posterior dimension guarantee anchor point, and searches segment by segment in reverse according to the time position. The initial working condition state segment and the boundary of the initial state segment falling within the backtracking range are written into the evidence backtracking record set in sequence, so that the evidence backtracking record set maintains the time correlation between the posterior dimension guarantee anchor point and the previous initial working condition state segment. Based on the posterior safeguard anchor point, initial operating condition segment, and initial state segment boundary in the evidence backtracking record set, the start position and end position of the precursor interval corresponding to the posterior safeguard anchor point are determined to form a precursor evidence interval. The determination process uses the start position of the initial operating condition segment that earliest satisfies the correlation condition within the backtracking range as the start position of the precursor interval, and the time position corresponding to the posterior safeguard anchor point as the end position of the precursor interval. The start position and end position of the precursor interval are then combined to form the precursor evidence interval.

[0027] In this embodiment, the re-segmentation output module specifically comprises: The precursor evidence interval is mapped to the corresponding interval range in the initial operating condition segment set. The initial operating condition segments and the boundaries of the initial state segments within the interval range are extracted to form a resegmentation candidate record set. The mapping process uses the start and end positions of the precursor evidence interval as the positioning basis. The interval range that has a positional overlap relationship with the precursor evidence interval is determined in the initial operating condition segment set. The extraction process writes the initial operating condition segments that fall within the interval range and the boundaries of the initial state segments corresponding to the initial state segments into the resegmentation candidate record set. Around the initial working condition state segment in the candidate record set for resegmentation, the cumulative result of directed transition similarity between adjacent working condition categories within the interval is recalculated according to time position. Combined with the segmentation cost value corresponding to each time position within the interval, the boundary determination is re-executed to form a resegmentation determination record set. The recalculation process is carried out segment by segment in the order of time position within the interval. The directed transition similarity from the working condition category corresponding to the previous time position to the working condition category corresponding to the next time position is continuously accumulated, so that each initial working condition state segment corresponds to a new set of cumulative scores. The boundary determination process is re-executed to place the cumulative scores and the segmentation cost value corresponding to the boundary position within the interval in the same determination relationship, which is used to re-determine whether the boundary position is maintained, moved, or split. Based on the boundary determination results in the re-segmentation determination record set, the initial working condition state segments within the prodromal evidence interval are subjected to interval retention, interval segmentation, and interval merging to form an updated working condition state segment set and updated state segment boundaries. The interval retention process is used to maintain the initial working condition state segments whose boundary determination results meet the retention conditions unchanged. The interval segmentation process is used to divide a single interval into a front interval and a back interval at the position where the boundary determination results meet the splitting conditions. The interval merging process is used to combine adjacent intervals whose boundary determination results meet the connection conditions into a continuous interval, so that the updated working condition state segment set and the updated state segment boundaries correspond to the new segmentation results within the prodromal evidence interval. The working condition segments in the set of updated working condition segments are arranged in chronological order, and the working condition identification results, precursor segment location results, and maintenance trigger results are output. The arrangement process outputs the working condition segments in the set of updated working condition segments in chronological order from start to end position. The working condition identification results are used to characterize the distribution of working condition categories corresponding to each time position. The precursor segment location results are used to characterize the position of the precursor interval determined by resegmentation within the precursor evidence interval. The maintenance trigger results are used to characterize the maintenance trigger position corresponding to the precursor segment location results.

[0028] Example 1: To verify the feasibility of this invention in practice, it was applied to a digital maintenance scenario for engines of special vehicles undertaking heavy-load traction, frequent start-stop operations, and long-term idling. In this scenario, the engine's operating state changes frequently, thermal state changes significantly, and some abnormal signals have short durations and low amplitudes. Traditional maintenance methods relying on fixed thresholds and human experience are prone to two types of problems: one is misjudging local fluctuations as state boundaries within stable operating conditions, resulting in an excessive number of operating state segments and scattered maintenance judgments; the other is failing to identify the location of early warning signs in the early stages of anomaly formation, causing maintenance to be triggered too late. To address this issue, engine operation monitoring data, status event data, and maintenance feedback data are first collected. Time alignment, anomaly correction, and numerical normalization are then performed on the collected results to form an engine time-series dataset. Next, time window decomposition and feature calculation are performed on the engine time-series dataset to form a window feature sequence. Subsequently, clustering and merging are performed on the window feature sequence to form a set of operating condition prototypes and an ordered operating condition label sequence. Based on this, the directed transition similarity matrix corresponding to the set of operating condition prototypes is calculated according to the evolution direction of adjacent time windows. An adaptive segmentation cost sequence is calculated based on the degree of state continuity and the degree of precursor clustering at the time position. An improved SegmenTier algorithm is used to generate the initial set of operating condition state segments and the boundaries of the initial state segments. When an abnormal maintenance event occurs in the maintenance feedback record, the posterior maintenance evidence anchor point is located, and a precursor evidence interval is formed by backtracking. Within the precursor evidence interval, the initial set of operating condition state segments and the boundaries of the initial state segments are re-segmented. Finally, the operating condition identification result, precursor segment location result, and maintenance trigger result are output.

[0029] In actual implementation, a total of 12,480 original records were collected, and after time alignment, completion, and correction, 11,864 valid time-series records were formed. After sliding decomposition using a fixed time window length and a fixed time window step size, 426 time window feature records were formed, and six types of working condition prototypes were constructed. The comparative scheme uses a fixed threshold segmentation method, determining boundaries only based on local fluctuations and exceeding limits. The present invention introduces directed transition similarity when accumulating scores within segments, introduces adaptive segmentation costs when determining boundaries, and performs local resegmentation using precursor evidence intervals after an abnormal maintenance event occurs. This processing reduces invalid segmentation within stable working conditions, preserves continuous clustering features before anomaly formation, and allows precursor segments to be identified separately from complex fluctuations. For ease of explanation, the main results of the traditional method and the present invention are summarized in Table 1: Table 1 Comparison of Implementation Results

[0030] As shown in Table 1, under the same original record quantity, the number of effective time-series records generated by this invention is higher than that of the traditional fixed threshold segmentation method, indicating that the dataset generation process can improve the continuity and availability of the original data. Regarding the segmentation results, the number of initial state segments generated by this invention decreased from 41 to 29, indicating that the improved SegmentTier algorithm can suppress invalid boundaries within stable operating conditions. After resegmentation, the number of state segments was adjusted to 32, indicating that the generation of precursor evidence intervals and local resegmentation processing can supplement and recover boundary positions that truly have maintenance value. In terms of performance indicators, the state segment boundary accuracy increased from 79.3% to 92.1%, the precursor segment recognition rate increased from 58.6% to 87.4%, the maintenance triggering efficiency increased from 63.8% to 89.6%, and the number of false alarm shutdowns decreased from 5 to 1. This indicates that this invention can more accurately reflect the engine operating condition evolution process, more timely locate the precursor segment position, and provide maintenance triggering results with higher reliability. Therefore, this invention not only solves the problems of incorrect state boundary segmentation and delayed precursor identification in traditional solutions, but also enables maintenance feedback to be used in reverse for boundary correction, making the digital maintenance results closer to the actual operating state, and has high engineering application value.

[0031] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A digital maintenance system for special vehicle engines based on big data analysis, characterized in that, include: The dataset generation module is used to perform time alignment, anomaly correction, and numerical normalization on engine data to form an engine time-series dataset. The feature generation module is used to perform time window decomposition and feature calculation on the engine time series dataset to form a window feature sequence; The prototype building module is used to cluster and merge window feature sequences to form a set of working condition prototypes and an ordered working condition label sequence. The similarity generation module is used to calculate the directed transition similarity matrix corresponding to the set of working condition prototypes based on the evolution direction of adjacent time windows; The cost generation module is used to calculate the adaptive segmented cost sequence based on the degree of state continuity and the degree of precursor aggregation at the time position; The initial segmentation module is used to recursively segment the ordered working condition label sequence using the improved SegmenTier algorithm, integrate the directed transition similarity matrix into the cumulative score within the segment, and couple the adaptive segmentation cost sequence to the boundary determination process to form the initial working condition state segment set and the initial state segment boundary. The evidence interval generation module is used to locate post-maintenance verification evidence anchor points based on maintenance feedback records and to trace back to form a prodromal evidence interval. Specifically: Arrange the maintenance feedback records in chronological order, and extract the record identifier, timestamp, maintenance result marker, and associated status information to form a maintenance feedback record set. Based on the maintenance feedback records in the maintenance feedback record set, abnormal maintenance events are identified according to the maintenance result markers, and the time location corresponding to the abnormal maintenance events is determined as the post-verification maintenance evidence anchor point, forming an evidence anchor point record set; Starting from the posterior dimension guarantee anchor point in the evidence anchor point record set, backtrack along the time position to extract the corresponding initial working condition state segment and the boundary of the initial state segment within the backtracking range, forming the evidence backtracking record set. Based on the posterior maintenance guarantee anchor point, initial working condition segment and initial state segment boundary in the evidence retrospective record set, determine the start position and end position of the precursor interval corresponding to the posterior maintenance guarantee anchor point, and form the precursor evidence interval. The re-segmentation output module is used to update the initial operating condition segment set and initial state segment boundaries within the precursor evidence interval, and output the operating condition identification result, precursor segment location result, and maintenance trigger result, specifically: The precursor evidence intervals are mapped to the corresponding interval ranges in the initial operating condition segment set. The initial operating condition segments and the boundaries of the initial operating condition segments within the interval ranges are extracted to form a re-segmentation candidate record set. Around the initial working condition state segment in the candidate record set for resegmentation, the cumulative result of the directed transition similarity between adjacent working condition categories within the interval is recalculated according to the time position, and the boundary determination is re-executed in combination with the segmentation cost value corresponding to each time position within the interval to form a resegmentation determination record set. Based on the boundary determination results in the re-segmentation determination record set, the initial working condition state segment within the precursor evidence interval is subjected to interval preservation, interval segmentation, and interval merging to form the updated working condition state segment set and the updated state segment boundary. The operating condition segments in the set are updated in chronological order, and the operating condition identification results, precursor segment location results, and maintenance trigger results are output.

2. The digital maintenance system for special vehicle engines based on big data analysis according to claim 1, characterized in that, The dataset generation module specifically includes: Collect engine operation monitoring data, status event data, and maintenance feedback data; extract data identifiers, timestamps, parameter fields, and numerical fields to form a raw data record set; and sort and map the raw data record set according to the timestamp to form a time-aligned record set. For missing locations in the time-aligned record set, interpolation and neighborhood completion are performed to form a completed record set. Adjacent differences and local fluctuation values ​​are calculated for parameter fields and numerical fields in the completed record set. Abnormal records are identified by combining preset value ranges, difference thresholds, and fluctuation thresholds to form an abnormal marked record set. Replacement and smoothing corrections are performed on the abnormal locations according to the abnormality marker record set to form an abnormality correction record set. Numerical normalization is then performed on the numerical fields in the abnormality correction record set to form a normalized record set. Finally, the records in the normalized record set are arranged in chronological order to form an engine time series dataset.

3. The digital maintenance system for special vehicle engines based on big data analysis according to claim 1, characterized in that, The feature generation module is specifically as follows: Sliding decomposition is performed on the engine time series dataset according to the preset time window length and time window step. A time window segment set is established according to the start time position, end time position and window identifier corresponding to each time window, and parameter field sequence and numerical field sequence are extracted in each time window segment. For the parameter field sequence and numerical field sequence in each time window segment, calculate the numerical change, adjacent difference, continuous change slope, local fluctuation value and abnormal distribution value according to the time position, and combine the numerical change, adjacent difference, continuous change slope, local fluctuation value and abnormal distribution value according to the field order to form the time window feature record; The time window feature records are arranged in chronological order to form a window feature sequence.

4. The digital maintenance system for special vehicle engines based on big data analysis according to claim 1, characterized in that, The prototype construction module is specifically as follows: Arrange the time window feature records in the window feature sequence according to the time position, extract the window identifier and the feature value corresponding to the window identifier, and combine them according to the field order to form a time window feature record set; Within the time window feature record set, each time window feature record is compared one by one. The feature difference value is calculated according to the difference between the feature values, and the feature difference value is written into the comparison relationship between the time window feature records to form a feature comparison record set. Based on the feature comparison record set, clustering and merging are performed according to the feature difference value. Time window feature records whose feature difference value meets the merging condition are classified into the same working condition category, and the time window feature records are aggregated according to the working condition category to form a working condition category record set. Around the working condition category in the working condition category record set, the feature values ​​belonging to the same working condition category are aggregated to form a working condition prototype set, and the working condition categories corresponding to the time window feature records are arranged in order according to the time position to form an ordered working condition label sequence.

5. The digital maintenance system for special vehicle engines based on big data analysis according to claim 1, characterized in that, The similarity generation module specifically includes: Arrange the time window feature records in the window feature sequence according to the time position, extract the previous time window feature records and the next time window feature records corresponding to adjacent time positions, and calculate the change in feature values ​​of the previous time window feature records and the next time window feature records at the same field position to form an evolution direction record set. The previous time window feature records and the next time window feature records in the evolution direction record set are respectively mapped to the previous working condition category and the next working condition category in the ordered working condition label sequence. A working condition transfer record set is established according to the correspondence between the previous working condition category and the next working condition category, and the change in feature value is written into the working condition transfer record in the working condition transfer record set. For each work condition transfer record in the work condition transfer record set, similarity calculation is performed based on the consistency of the direction of the change in feature value at the field position. The results are then written into matrix cells according to the arrangement of the previous work condition category and the next work condition category in the work condition prototype set, forming a directed transfer similarity matrix corresponding to the work condition prototype set.

6. The digital maintenance system for special vehicle engines based on big data analysis according to claim 1, characterized in that, The cost generation module is specifically: Arrange the time window feature records in the window feature sequence according to the time position, extract the time window feature records corresponding to each time position and the corresponding working condition category in the ordered working condition label sequence, perform continuous statistics on the same working condition category at consecutive time positions, and accumulate the feature value change amplitude between time window feature records in the continuous interval. Calculate the state continuity degree corresponding to each time position based on the working condition category continuity length and feature value change amplitude. Around the time window feature records in the window feature sequence, the abnormal distribution values ​​are statistically analyzed in continuous intervals according to the time position. The cumulative results of abnormal distribution values ​​within the continuous interval and the increase or decrease of abnormal distribution values ​​between adjacent time positions are jointly calculated. The degree of precursor aggregation corresponding to each time position is calculated based on the concentration and change intensity of abnormal distribution values ​​within the continuous interval. The state continuation degree and precursor aggregation degree corresponding to each time position are paired according to the time position, and the state continuation degree and precursor aggregation degree at the same time position are combined to form the segmented cost value corresponding to each time position. Then, the segmented cost values ​​are arranged in the order of time position to form an adaptive segmented cost sequence.

7. The digital maintenance system for special vehicle engines based on big data analysis according to claim 1, characterized in that, The initial segmentation module is specifically as follows: Arrange the ordered working condition label sequence in chronological order, construct candidate segment intervals based on the continuous relationship between time positions, and record the start time position, end time position, and working condition category within the interval according to the candidate segment intervals to form a segmented candidate record set; Within the segmented candidate record set, the improved SegmenTier algorithm is used to perform intra-segment cumulative scoring. The directed transition similarity between adjacent working condition categories is extracted sequentially along the time position in the candidate segment interval. The directed transition similarity between each time position in the same candidate segment interval is accumulated item by item to form an intra-segment scoring record set. Around the candidate segment intervals in the score record set within the segment, the improved SegmenTier algorithm is used to perform boundary determination. The segmentation cost value corresponding to the boundary position of the candidate segment interval is introduced into the segmentation calculation, and the cumulative score result within the segment is jointly compared with the segmentation cost value corresponding to the boundary position to form a boundary determination record set. Based on the boundary determination record set, the candidate segment intervals are recursively updated. The candidate segment intervals that meet the retention conditions are confirmed as the initial working condition segments. The candidate segment intervals that meet the splitting conditions are split into new intervals according to the boundary positions, forming a segmented state record set. The initial state segment boundary is determined based on the recursive results in the segmented state record set, and the ordered working condition label sequence is divided into intervals according to the initial state segment boundary to form the initial working condition state segment set and the initial state segment boundary corresponding to the initial working condition state segment set.

Images