Point switch health assessment method based on dynamic segmentation and improved DTW

Abstract

The invention discloses a dynamic segmentation and improved DTW (dynamic time warping)-based switch machine health assessment method, which realizes adaptive curve segmentation according to the dynamic characteristics of a switch machine, does not need to manually set a fixed threshold value for segmentation, and avoids the problem of inaccurate segmentation caused by the change of working conditions. Moreover, by combining with the improved DTW algorithm, in the abnormal curve screening process, not only is the difference sensitivity of key action point locations considered, but also the matching path search range is constrained through the self-adaptive structured window, and the curve characteristics can be more comprehensively and deeply mined, so that the abnormal curve is more accurately screened out, and the monitoring accuracy and reliability are improved. Compared with a traditional method, the method has higher abnormal curve screening precision, better working condition adaptability and higher anti-jamming capability, the health state of the switch machine can still be stably and accurately monitored in the face of a complex operation environment and variable fault modes, and a more reliable basis is provided for fault diagnosis and maintenance of the switch machine.

Description

Technical Field

[0001] This invention relates to the field of rail transit equipment monitoring technology, and in particular to a method for health assessment of switch machines based on dynamic segmentation and improved DTW. Background Technology

[0002] Switch machines change the direction of turnout operation and lock the turnout switch rails and frog rails, serving as crucial signaling infrastructure reflecting the turnout's position. Railway transportation relies heavily on signaling systems, and as a key component of these systems, the normal operation of switch machines directly impacts train safety and transportation efficiency. Since turnout switch machines operate outdoors year-round, they are significantly affected by weather, temperature, and vibration; therefore, accurate monitoring of their health status is essential.

[0003] Existing switch machine condition monitoring mainly relies on fixed thresholds to determine operational status, which is difficult to adapt to dynamic changes under complex operating conditions. Furthermore, traditional algorithms lack sufficient sensitivity to differences in key stages such as unlocking, switching, and locking when processing switch machine operation curves, resulting in low accuracy in anomaly curve screening and a lack of interpretability in fault location. In recent years, the rapid development of artificial intelligence technology has spurred more efficient and accurate health monitoring methods, becoming a core direction for future research in this field. Therefore, developing an advanced switch machine fault diagnosis method based on artificial intelligence technology to overcome the limitations of existing technologies is of great strategic significance for improving the intelligence level of switch machine fault diagnosis and ensuring the safe operation of rail transit.

[0004] Many researchers have attempted to integrate artificial intelligence technology into the health monitoring research of turnout switch machines, and have achieved good results. Reference 1 (Wang Mengmeng, Liu Bohong. Research on the HealthState Evaluation of Point Machine Based on FCM and HSMM[C] / / 2021 International Conference on Communications, Information System and Computer Engineering (CISCE): IEEE, 2021: 235-238.) proposes a health state assessment method for switch machines based on fuzzy c-means clustering (FCM) and hidden semi-Markov model (HSMM). This method decomposes the power data collected by the switch machine using the complementary set empirical mode decomposition (CEEMD) method, and then uses the FCM clustering method to mine and classify the degradation state, which can quickly identify the health state of the switch machine. Reference 2 (Liu Chuanzhu, Huang Yongjie, Liu Ruiqi, et al. Health status classification of turnout switch based on ZD6 switch machine current data [J]. Railway Operation Technology, 2024, 30(04): 1-5.) collected the working current data of ZD6 switch machine under different loads in the range of 100%-120% of rated traction force to construct a dataset, and used CNN-SVM (convolutional neural network-support vector machine) network to classify the health status. Reference 3 (Zheng Yunshui, Chen Weimin. Health Indicator Construction and Life Prediction of the PointSwitch Machine[J]. Journal of Failure Analysis and Prevention, 2022, 22(3):1031-1039.) proposes a switch machine remaining life prediction model based on multi-feature information fusion. It combines deep belief network (DBN) and locally linear embedding (LLE) to construct health indicators, and then uses a hidden Markov model (HMM) to predict remaining life. However, the model has a large number of parameters, making it difficult to deploy and apply in practice.Reference 4 (Chen Yonggang, Dai Qianjun, Du Tao, et al. Switch machine condition assessment based on fuzzy evaluation of deterioration degree and optimal weight [J]. Journal of Railway Science and Engineering, 2019, 16(05): 1315-1322.) addresses the random and fuzzy characteristics of mechanical failures of switch machines by constructing an equipment health assessment model based on deterioration degree. The deterioration degree is calculated from three dimensions: monitoring data, maintenance score, and equipment service life. The subjective AHP and objective entropy weight method are integrated to determine the health status weights, and the weight allocation is optimized by the minimum variance method. Finally, fuzzy theory is used to comprehensively assess the health status, effectively addressing the fuzziness and complexity of health influencing factors.

[0005] Currently, while AI-based fault diagnosis solutions possess the ability to autonomously extract knowledge and patterns from data, and perform well in terms of classification stability and system scalability, they remain weak in areas such as data processing capabilities and intelligent state assessment. The key existing problems are concentrated in the following dimensions: (1) Fixed threshold segmentation has poor adaptability: Traditional methods rely on manually setting thresholds and cannot dynamically adjust the segmentation strategy according to the equipment's operating status.

[0006] (2) Insufficient feature extraction in key stages: The differences in dynamic characteristics of different action stages are not fully considered, resulting in low sensitivity of anomaly detection.

[0007] (3) Lack of interpretability in fault location: Only abnormal curves are output, without providing specific stage information of the fault occurrence.

[0008] In view of this, the present invention is hereby proposed. Summary of the Invention

[0009] The purpose of this invention is to provide a switch machine health assessment method based on dynamic segmentation and improved DTW, which solves the problems of poor segmentation adaptability, incomplete feature extraction and ambiguous fault location in the prior art, and achieves high-precision abnormal curve screening and interpretable fault stage location.

[0010] The objective of this invention is achieved through the following technical solution: A switch machine health assessment method based on dynamic segmentation and improved DTW includes: Based on the dynamic characteristics of the switch machine's unlocking, switching and locking stages, a mapping relationship between the action state and the power curve shape of the switch machine is established. The stage boundary point is dynamically identified through the change point detection algorithm to realize the adaptive segmentation of the switch machine's power curve. An improved DTW algorithm is used to calculate the similarity between each segment of the switch machine power curve and a given standard curve, and convert it into a score for the corresponding stage. The scores of all stages are combined and weighted to obtain a comprehensive score. Fault analysis is performed by combining the scores of each stage and the comprehensive score to obtain the status monitoring results of each stage corresponding to each curve segment and the entire stage, thus completing the health assessment. In the improved DTW algorithm, an adaptive regularization window constraint is introduced to limit the search range of the matching path, and each segment of the switch machine power curve and the given standard curve are standardized separately before calculating the similarity.

[0011] As can be seen from the technical solutions provided by the present invention, (1) adaptive curve segmentation can be achieved based on the dynamic characteristics of the switch machine itself, without relying on manually setting fixed thresholds for segmentation, avoiding the problem of inaccurate segmentation caused by changes in working conditions, and reducing the time and effort invested in manually adjusting the thresholds. By establishing a mapping relationship between the action state and the power curve shape, the stages are automatically divided according to the equipment operating state, improving the scientificity and accuracy of segmentation. (2) By using dynamic segmentation combined with the improved DTW method, in the process of abnormal curve screening, both the difference sensitivity of key action points and the matching path search range are considered. This multi-dimensional processing method overcomes the limitation of the traditional DTW algorithm in analyzing switch machine action curves from only a single perspective, and can more comprehensively and deeply mine curve features, thereby more accurately screening abnormal curves and improving the accuracy and reliability of monitoring. (3) The abnormal curve screening architecture based on dynamic segmentation and improved DTW has stronger capabilities in fault feature localization and anomaly judgment. By enhancing the sensitivity to differences in key action points through segmented weighting coefficients and flexibly adjusting the adaptive regularization window, the model can better adapt to changes in the switch machine's action curves under different operating conditions. Compared to traditional methods, this architecture has higher accuracy in anomaly curve screening, better adaptability to operating conditions, and stronger anti-interference capabilities. Even in the face of complex operating environments and changing fault modes, it can still stably and accurately monitor the health status of the switch machine, providing a more reliable basis for fault diagnosis and maintenance. Attached Figure Description

[0012] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0013] Figure 1 This is a flowchart of a switch machine health assessment method based on dynamic segmentation and improved DTW, provided for an embodiment of the present invention.

[0014] Figure 2 This is a schematic diagram of the overall process of a switch machine health assessment method based on dynamic segmentation and improved DTW, provided for an embodiment of the present invention.

[0015] Figure 3 A flowchart for preliminary determination of the integrity of the motion curve provided in an embodiment of the present invention.

[0016] Figure 4 The power operation curve of the turnout switch machine is provided for an embodiment of the present invention.

[0017] Figure 5 This is a schematic diagram illustrating the search range limitation provided in an embodiment of the present invention.

[0018] Figure 6 This is a partial dataset of standard curves provided for embodiments of the present invention.

[0019] Figure 7 The diagram shows the experimental results of dynamic segmentation of the power curve provided in the embodiment of the present invention.

[0020] Figure 8 This is a schematic diagram of the minimum alignment path between the comparison curve and the standard curve provided in an embodiment of the present invention.

[0021] Figure 9 The curve segmentation evaluation results provided in the embodiment of the present invention are for the position-to-reverse position conversion.

[0022] Figure 10 The segmented evaluation results are provided for when an abnormality occurs during the locking phase, as provided in the embodiments of the present invention. Detailed Implementation

[0023] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the protection scope of the present invention.

[0024] First, the following explanations are provided for the terms that may be used in this article: The terms "comprising," "including," "containing," "having," or other similar semantic descriptions should be interpreted as non-exclusive inclusion. For example, including a technical feature element (such as raw material, component, ingredient, carrier, dosage form, material, size, part, component, mechanism, device, step, process, method, reaction conditions, processing conditions, parameter, algorithm, signal, data, product or article of manufacture, etc.) should be interpreted as including not only the expressly listed technical feature element, but also other technical feature elements that are not expressly listed and are well-known in the art.

[0025] The term "composed of" excludes any technical features not expressly listed. When used in a claim, it closes the claim to exclude all technical features other than those expressly listed, except for associated conventional impurities. If the term appears only in a clause of a claim, it limits the claim to the elements expressly listed in that clause; elements recited in other clauses are not excluded from the overall claim.

[0026] The following is a detailed description of a switch machine health assessment method based on dynamic segmentation and improved DTW provided by this invention. Contents not described in detail in the embodiments of this invention are prior art known to those skilled in the art. For conditions not specified in the embodiments of this invention, conventional conditions or conditions recommended by the manufacturer should be followed. Reagents or instruments used in the embodiments of this invention, unless otherwise specified, are all commercially available conventional products.

[0027] like Figure 1 The diagram shows a flowchart of a switch machine health assessment method based on dynamic segmentation and improved DTW (Dynamic Time Warping) provided by an embodiment of the present invention. The method mainly includes the following steps: Step 1: Based on the dynamic characteristics of the switch machine's unlocking, switching and locking stages, establish a mapping relationship between the action state and the power curve shape of the switch machine. Dynamically identify the stage boundary points through the change point detection algorithm to achieve adaptive segmentation of the switch machine's power curve.

[0028] In this embodiment of the invention, before implementing adaptive segmentation of the switch machine power curve, an integrity check is performed on the switch machine power curve. If the integrity check passes, adaptive segmentation of the switch machine power curve is performed (i.e., step 1 is executed); if the integrity check fails, fault analysis is performed based on the integrity check results.

[0029] In this embodiment of the invention, the integrity check of the switch machine power curve includes: Part 1: Unlocking Phase Detection. Check if the action curve exists. If it exists, proceed to the next part. If the curve is 0 or does not exist, it means that the integrity check has not been passed. Part Two: Conversion Phase Detection. The curve status is continuously monitored. If a 0 curve or signal interruption occurs during this process, it indicates an abnormality in the conversion phase and failure to pass the integrity check; otherwise, proceed to the next part. Part 3: Locking phase detection, verifying whether the curve is released normally; if a 0 curve appears at this time, or the signal disappears, or the signal drop exceeds the set threshold, it is determined that the locking phase is abnormal and the integrity check has not been passed; otherwise, continue to the next part; Part 4: Final state detection, confirming that the curve has completely ended; if the curve still has signal output at this time, it is determined that the transition or latch-up phase is abnormal and the integrity check has not passed; otherwise, the integrity check has passed.

[0030] In this embodiment of the invention, the adaptive segmentation of the switch machine power curve includes: (1) Calculate the slope corresponding to each data point in the switch machine power curve; and set the window length to calculate the dispersion of the data within the window corresponding to each data point.

[0031] (2) Based on the dynamic characteristics of the unlocking, switching and locking stages of the switch machine, establish the mapping relationship between the action state and the power curve shape of the switch machine, and perform change point detection according to the slope change of each data point and the degree of dispersion, dynamically identify the stage boundary point, and segment the power curve of the switch machine according to the stage boundary point.

[0032] In this embodiment of the invention, the slope corresponding to each data point in the power curve of the switch machine is calculated; and, setting a window length, the calculation of the dispersion of the data within the window corresponding to each data point includes:

[0033] For the i-th data point Its slope Calculated using the following formula: ; in, For the (i+1)th data point, This represents the time interval between two adjacent data points.

[0034] Set the window length to L, for the i-th data point The degree of dispersion of data within the window Calculated using the following formula: ; in, Indicates that the index is The data points, where L is an odd number; The average value of the data within the window is expressed as: .

[0035] In this embodiment of the invention, the step of detecting change points and dynamically identifying stage boundary points based on the changes in slope and the degree of dispersion corresponding to each data point includes: Within the first set time range Within, set the first discrete threshold. Iterate through each moment within the first set time range. When the slope changes from negative to positive and the dispersion is less than a threshold, mark the corresponding moment as the boundary between the unlocking phase and the transition phase. , represented as: ; in, The degree of dispersion of the data within the window corresponding to the i-th data point. Let i be the time corresponding to the i-th data point. , , The slopes are, in order, the slopes corresponding to the (i-2)th data point, the (i-1)th data point, and the ith data point. The end time of the first set time range; Second set time range Within, set a second discrete threshold. Iterate through the moments within the first set time range. When the slope is less than 0 and the dispersion is greater than the threshold, the corresponding time is marked as the boundary between the transition phase and the latching-and-release phase. , represented as: ; in, , These are the start and end times of the second set time range, respectively. .

[0036] The switch machine power curve data is a time series, and the timestamp (moment) corresponding to each data point is recorded as follows: Therefore, the process is divided into segments based on the above dividing points, and the unlocking phase corresponds to the following time points: The corresponding time for the transition phase: The corresponding time for the locking phase: This allows us to divide the curves corresponding to each stage.

[0037] After completing the curve segmentation, a power threshold is introduced to determine whether the power value corresponding to each data point in the curve during the conversion stage exceeds the power threshold. If it does, the switch machine power curve data is judged as an abnormal curve.

[0038] Step 2: Using the improved DTW algorithm, calculate the similarity between each segment of the switch machine power curve and the given standard curve, and convert it into a score for the corresponding stage. Combine the scores of all stages and obtain a comprehensive score through weighting. Combine the scores of each stage and the comprehensive score to perform fault analysis, obtain the status monitoring results of each stage corresponding to each curve and the entire stage, and complete the health assessment.

[0039] In this embodiment of the invention, the switch machine operation curve data collected by the station signal centralized monitoring system is used as the basis. The data is classified according to the switch machine model and equipment number to construct a standard dataset, and a standard curve is generated based on the dataset.

[0040] In this embodiment of the invention, the improved DTW algorithm introduces an adaptive regularization window constraint on the matching path search range, and standardizes each segment of the switch machine power curve with a given standard curve before calculating the similarity. Specifically, the calculation of the similarity between each segment of the switch machine power curve and the given standard curve using the improved DTW algorithm includes: (1) Standardize each curve segment and each data point in the standard curve separately.

[0041] (2) Calculate each data point after standardization of each curve segment. , and each data point in the standard curve after standardization The Euclidean distance between them, combined with distance parameters, is used to calculate the nonlinear distance. And construct the cumulative distance matrix between each curve segment and the standard curve.

[0042] (3) Introduce an adaptive regularization window r to constrain the search range of the matching path. Search for regularized paths in the cumulative distance matrix between each curve segment and the standard curve. The points (u,o) in the regular path satisfy the following conditions: Where u is the index of the data point in each curve segment, and o is the index of the data point in the standard curve.

[0043] (4) For the regular path W1 corresponding to the single-segment curve Q1 and the standard curve C1, define the minimum cumulative distance corresponding to the point (u,o). for: ; in, This represents the minimum cumulative distance between points (u-1,o-1), (u-1,o), and (u,o-1).

[0044] (5) Finally, the minimum cumulative distance between the single-segment curve Q1 and the standard curve C1 is: ,in, For point The corresponding minimum cumulative distance, U is the index of the last data point of the single-segment curve Q1, and O is the index of the last data point of the standard curve C1.

[0045] (6) The minimum cumulative distance between a single curve segment and the standard curve is taken as the similarity between the single curve segment and the standard curve. The similarity between each curve segment and the given standard curve is calculated in the same way.

[0046] In this embodiment of the invention, the adaptive power curve of the switch machine is divided into three segments, and the similarity between each segment and a given standard curve is denoted as . The similarity score is converted to a percentage rating using the following formula: ; Where y is the index number of the stage. The corresponding stage is scored out of 100. This represents the similarity distance threshold for the corresponding stage. This represents the ideal matching distance for the corresponding stage.

[0047] In addition, the weighting coefficients are denoted as follows: , , And satisfy The comprehensive score is calculated using the following formula. : .

[0048] Set scoring thresholds for each stage and a total scoring threshold. Determine the working status of the corresponding stage based on the relationship between the score of the corresponding stage and the corresponding scoring threshold. Determine the working status of the entire stage based on the relationship between the comprehensive score and the total scoring threshold. Finally, determine the status monitoring results for the entire stage.

[0049] In this embodiment of the invention, the method further includes: when the comprehensive score is greater than the total score threshold, storing the corresponding switch machine power curve into a standard dataset, so that the standard curve can be updated in subsequent work.

[0050] The above-mentioned solutions provided by the embodiments of the present invention have the following main advantages: (1) Improve the accuracy and adaptability of segmentation: It can establish a mapping relationship between the action state and the power curve shape based on the dynamic characteristics of the switch machine's unlocking-conversion-locking transmission stage, and realize adaptive curve segmentation based on the equipment's operating state. It overcomes the shortcomings of traditional fixed threshold segmentation, which cannot adapt to different working conditions. It can more accurately reflect the state of the switch machine in different working stages, improve the accuracy and scientific nature of segmentation, and make subsequent analysis more in line with the actual operating conditions of the equipment.

[0051] (2) Enhanced abnormal curve screening capability: By introducing segmented weight coefficients to strengthen the sensitivity of key action points to differences, and combining adaptive regularization window constraints to match the path search range. Compared with the traditional DTW algorithm, it is more sensitive to the feature capture of key stages in the switch machine action curve, and can more accurately screen out curves containing abnormal morphological features, reduce misjudgment and omission, and improve the accuracy and reliability of abnormal curve screening.

[0052] (3) Provide interpretable fault location clues: The high-suspicion curve set output by the screening not only identifies abnormal curves, but its dynamic segmentation marking results also provide interpretable fault stage location clues for subsequent fault classification. Maintenance personnel can quickly locate the approximate stage of the fault based on these clues, clarify the direction of fault investigation, shorten the fault diagnosis time, improve maintenance efficiency, and ensure the safety and efficiency of railway transportation.

[0053] (4) Real-time monitoring and early warning: The constructed model can be used to monitor the switch machine power curve in real time, score the curve status, and promptly detect and warn of abnormalities in the curve. Before a fault occurs, maintenance personnel can be notified in advance to take measures to avoid train delays or safety accidents caused by switch machine failures, ensure the stable operation of the railway system, and reduce maintenance costs and operational risks.

[0054] To more clearly demonstrate the technical solution and its effects provided by the present invention, the method provided by the embodiments of the present invention will be described in detail below with reference to specific examples.

[0055] I. Overall Overview of the Plan

[0056] With the development of intelligent sensing technology and data-driven methods, the Dynamic Time Warping (DTW) algorithm has shown advantages in time series data matching. However, traditional DTW has problems such as insufficient sensitivity to key stages and rigid segmentation strategies in switch machine curve analysis.

[0057] To address this, this invention proposes a switch machine health assessment method based on dynamic segmentation and improved DTW (Dynamic Dynamic Waveform Design). By constructing a fusion architecture of an adaptive dynamic segmentation model and an improved DTW algorithm, it achieves high-precision anomaly screening and fault stage localization of the switch machine's power curve. This method overcomes the limitations of traditional fixed-threshold segmentation, dynamically dividing action stages based on the switch machine's dynamic characteristics. It enhances the sensitivity to differences in key features through segmentation weight coefficients and adaptive regularization windows, providing a more intelligent and interpretable solution for switch machine health monitoring.

[0058] This invention does not limit the type of switch machine and is applicable to mainstream switch machine types such as ZDJ9 and S700K. Its core technology can be extended to the time-series data health monitoring of other railway equipment, such as track circuits, and has broad engineering application prospects and market promotion value.

[0059] The dynamic segmentation and improved DTW evaluation method provided by this invention lies in two major technological innovations: 1. Dynamic segmentation model: Based on the dynamic characteristics of the switch machine's unlocking-conversion-locking stages, a mapping relationship between the action state and the power curve shape is established. The stage boundary point is dynamically identified through the change point detection algorithm to achieve adaptive segmentation and solve the problem of working condition adaptability of traditional fixed threshold segmentation.

[0060] 2. Improved DTW algorithm model: Introduce segmented weight coefficients to enhance the sensitivity of key action points to differences, and combine adaptive regularization window constraints to match the search range of the path. While retaining the advantages of DTW time series matching, improve the screening accuracy of abnormal curves, and provide interpretable stage location clues for fault diagnosis through segmented marking.

[0061] like Figure 2 The diagram shown is a schematic of the overall process of the switch machine health assessment method based on dynamic segmentation and improved DTW.

[0062] The technical solution of this invention will be generally introduced from the following three aspects: (1) Construction of dynamic segmentation model.

[0063] This method mainly completes curve integrity check (obtaining the integrity status of input data) and curve segmentation (dynamically dividing curves with integrity). It dynamically divides the unlocking, conversion, and locking stages through a change point detection algorithm and checks whether the power value is within the threshold range during the conversion process.

[0064] (2) Improve the construction of DTW model.

[0065] This section primarily focuses on calculating curve similarity, converting it into a score, and then weighting the scores in segments to obtain a comprehensive score. It also assists in fault analysis for abnormal curves. Simultaneously, a standard dataset is established. If the switch machine is determined to be operating normally, the curve is added to the standard dataset and dynamically updated to avoid deviations in the standard curve caused by seasonal or temperature changes, which could lead to abnormal hazard identification.

[0066] (3) Data processing and experimental setup.

[0067] Before being input into the model, data needs to undergo preprocessing such as standardization and noise reduction to eliminate the influence of dimensions and improve signal quality. To control variables in the analysis of switch machine health status, power curves of a ZYJ7 switch machine at a certain station under different operating conditions were collected to construct a multi-condition dataset. The curves after comprehensive scoring were visualized and the model mechanism was analyzed to verify the feasibility and effectiveness of the method of this invention.

[0068] II. Dynamic Segmentation Model.

[0069] In this embodiment of the invention, the dynamic segmentation model mainly includes: curve integrity check and curve segmentation.

[0070] 1. Curve integrity check.

[0071] Based on the power curve characteristics of the turnout switch machine, design as follows Figure 3 The flowchart illustrates the curve integrity check process. This flowchart describes the phased detection logic for judging the integrity of the ZYJ7 switch machine's motion curve. The specific process is as follows: Part 1: Unlocking Phase Detection (0.6-1 second). During this time period, the presence of the motion curve must be checked. If the curve shows a "0 curve" (i.e., no valid signal output) or is completely absent, it is determined that the unlocking phase is abnormal. This abnormality may involve problems such as power failure, motor failure to start, or sensor signal loss.

[0072] Part Two: Switching Phase Detection (1-7 seconds). During the first 7 seconds of switch machine operation, the curve status must be continuously monitored. If a "0 curve" or signal interruption occurs during this process, it indicates an abnormality in the switching locking phase, which may be caused by mechanical jamming, locking device failure, or poor circuit contact.

[0073] Part 3: Locking Phase Detection (7-8 seconds). When the action reaches 7-8 seconds, it is necessary to verify whether the curve is releasing normally. If a "0 curve" appears at this time or the signal suddenly drops / disappears, it is determined that the locking phase is abnormal, which may be related to insufficient relay release time, capacitor failure, or abnormal drive circuit.

[0074] Part Four: Final State Detection (25 seconds). At 25 seconds after the start of the action, it is necessary to confirm that the curve has completely ended. If the curve still has a signal output at this time (not returning to zero), it is determined to be an abnormality in the transition or latch-up phase, and it is necessary to check whether the equipment has failed to reset or has a jamming problem.

[0075] Subsequent processing: If no abnormalities are found in any of the above stages, proceed to the next stage of "dynamic curve segmentation". If the curve is incomplete in any stage, proceed to the fault analysis process.

[0076] This process systematically checks the integrity of the curve through a phased detection method, providing good data samples for subsequent curve segmentation.

[0077] 2. Curve segmentation.

[0078] In this embodiment of the invention, the dynamic partitioning method for implementing these three stages is described in detail by combining slope analysis, dispersion judgment and change point detection.

[0079] Suppose the switch machine power curve data is a time series. ,in Let be the i-th data point, representing the power value at time i, where i = 1, 2, ..., n, and n is the total number of data points. The timestamp corresponding to the time series is... ,and (Sampling at equal time intervals).

[0080] Power curve as shown Figure 4 As shown. To analyze the trend of the power curve, it is necessary to calculate the slope at each data point. For the i-th data point, its slope is... It can be approximated by the ratio of the power difference to the time difference between two adjacent points: ; in, .

[0081] To measure the fluctuation of the power curve, a degree of dispersion is defined. For the i-th data point, a window of length L (where L is an odd number) is taken centered on that point, and the degree of dispersion of the data within that window is calculated. Here, standard deviation is used to represent: , ; in It is the average value of the data within the window. When i is close to the ends of the sequence, the actual window length may be less than L, in which case it is calculated based on the actual available data.

[0082] Method for determining the dividing point between the unlocking and conversion phases: During the unlocking phase, power rises rapidly and fluctuates significantly. After unlocking is completed and the conversion phase begins, the power curve tends to stabilize. The specific method for determining the dividing point is as follows: Within the time range Within, set discrete thresholds Iterate through every moment within this time range. When the slope changes from negative to positive and the dispersion is less than the threshold, this moment is considered a positive moment. Marked as the dividing point between the unlocking phase and the transition phase .Right now: .

[0083] Method for determining the dividing point between the transition phase and the locking phase: When the equipment completes the transition and enters the locking phase, the curve will show a slow initial rise followed by a significant downward jump, forming a small step in the release phase. Within the time range... Internally set discrete threshold When there is a certain moment If the slope is less than 0 and the dispersion is greater than a threshold, then this moment... Marked as the dividing point between the transition phase and the locking phase. .Right now: .

[0084] By following the steps above, two dividing points are determined. and This dynamically divides the switch machine power curve into three stages: Unlocking Phase: Transition phase: Lockdown phase: During the conversion phase, a power threshold is set to protect against excessively high power values ​​during the conversion process. If the value exceeds the threshold, it is also judged as an abnormal curve.

[0085] It should be noted that the values ​​provided in this section are for illustrative purposes only and are not intended to constitute a limitation. In practical applications, users can adjust them according to their actual circumstances.

[0086] III. Introduction to the Improved DTW Model.

[0087] As mentioned earlier, the improved DTW model includes curve similarity calculation, score calculation, and curve fault-assisted analysis.

[0088] 1. Curve similarity calculation.

[0089] In the Dynamic Time Warping (DTW) algorithm, Euclidean distance serves as a fundamental metric for measuring the similarity between two time series data points. It is used to calculate the numerical difference between the two points and is also the cornerstone for constructing the DTW cumulative distance. The definition is as follows: ; in, For sequence The i-th data point, For sequence The j-th data point.

[0090] When comparing two time series, Q and C (one corresponding to switch machine power curve data and the other to a standard curve), alignment is needed to determine the optimal matching path. This method involves constructing a... The cumulative distance matrix D is used to achieve this: .

[0091] A regularized path W is a path in the cumulative distance matrix D, defined as: ; in: It is a point on the path, corresponding to the first point of Q. The data point and C's first There are 10 data points; k is the path length.

[0092] When performing path search, the following three principles should be followed for organization: Boundary conditions: Each regular path must strictly start from the lower left starting point (1,1) of the matrix and eventually reach the upper right ending point (n,m). That is, the first node W1 of the path is fixed at (1,1), and the last node W... k It is fixed at (n, m).

[0093] Continuity condition: The sequence of nodes on the path must remain continuous. That is, for any adjacent nodes in the path, the subsequent node must satisfy a specific spatial adjacency relationship with the preceding node. Specifically, the path can be extended by horizontal movement (equivalent to data insertion), vertical movement (equivalent to data deletion), or diagonal movement (equivalent to data matching), but there cannot be any cross-point jump matching. All movements must be strictly aligned with adjacent nodes.

[0094] Monotonicity condition: In matrix space, a regular path must strictly follow a monotonically increasing direction of movement, and any form of path regression is prohibited. This means that during the path's movement, both its row and column coordinates must maintain a non-decreasing trend.

[0095] Given the above constraints, there may be multiple regularized paths that meet the requirements. To obtain the optimal solution, a dynamic programming algorithm is needed to traverse and search all feasible paths, ultimately determining a path scheme that minimizes the regularization cost. ; In the above formula, the parameter K plays a role in balancing the impact of differences in path length. Since longer paths encompass more "point pairs," the accumulated distance value increases accordingly. The distance per unit path length is obtained by dividing the total distance by K. During path traversal, the distances calculated for each previously visited point are accumulated. When the path reaches the endpoint (n, m), this accumulated distance reflects the similarity between sequences Q and C.

[0096] set up Let be the minimum cumulative distance between the first i points of sequence Q and the first j points of sequence C, specifically defined as follows: ; in, It is the Euclidean distance between the current point pairs. Indicates alignment along the diagonal (matching) and The minimum cumulative distance of the previous pair of points, express Corresponding to The minimum cumulative distance to the previous point, express The previous point corresponds to The minimum cumulative distance, and the initial conditions are: .

[0097] The final minimum cumulative distance between the two sequences is: ; The method described above can be used to calculate the minimum similarity distance between each action curve and the standard curve.

[0098] To better facilitate engineering applications, an improved DWT algorithm is proposed based on the above method, which is optimized in the following two aspects.

[0099] (1) Local Constraint Window Optimization. The above algorithm requires constructing a global alignment matrix, with a computational complexity of O(n log n). In large-scale data applications, this method is inefficient. This model introduces a local constraint window to limit the search range to the area near the main diagonal or a diamond-shaped region, such as... Figure 5 As shown. Considering that the actual motion curve may have a large deviation, the alignment path is selected to be limited to the range r near the main diagonal, that is, for the point (i,j) in the regularized path W, the following condition is met. By imposing such restrictions, the computational complexity is significantly reduced to... The window width r can be dynamically adjusted based on the statistical characteristics of the standard curve. A suitable initial value is determined based on the time axis fluctuation of a large amount of normal switch machine power curve data. During the subsequent monitoring process, it is further fine-tuned according to the changing trend of real-time data to balance computational efficiency and matching accuracy.

[0100] (2) Nonlinear distance metric. Due to the nonlinear characteristics of the switch machine degradation process, Euclidean distance is insufficient to effectively capture these features. This model adopts the idea of ​​nonlinear distance, first standardizing each data point in sequences Q and C: ; in, The average value of the sequence. denoted as the standard deviation of the sequence.

[0101] Define the formula for nonlinear distance: ; in, The Euclidean distance between two points after standardization. It is a distance parameter, usually defined as the mean of the distance between two sequences: .

[0102] Nonlinear distance amplifies minute differences through an exponential function, more sensitively reflecting the differences in curves under different health states of the turnout. This is achieved in the cumulative distance matrix. During the construction process, this nonlinear distance is used to replace the traditional Euclidean distance in the dynamic programming formula of DTW, thereby more sensitively reflecting the differences in curves under different health states of the turnout: .

[0103] The final nonlinear DTW distance is: .

[0104] The above uses two time series as examples to illustrate the calculation process of curve similarity. In this embodiment of the invention, after the aforementioned curve segmentation process, the switch machine power curve is adaptively divided into three curve segments. Each curve segment is calculated to have a similarity to the standard curve based on the above principle, and the corresponding similarity is denoted as... .

[0105] 2. Scoring calculation and curve fault auxiliary analysis.

[0106] The switch machine power curve, after dynamic segmentation, is divided into three stages: unlocking, switching, and locking. Different stages have varying importance in determining the switch machine's health status. This invention assigns different weights to these three stages, denoted as... , , And satisfy The total similarity distance is calculated using the following formula, combining the calculated curve similarity: ; in, This represents the total similarity distance.

[0107] Based on historical fault data and expert experience, it has been found that the probability of faults occurring during the transition phase is relatively high and that it has a significant impact on the overall performance of the switch machine. Therefore, the weight of the transition phase can be appropriately increased. This segmented weighting method highlights the importance of key stages in the evaluation, making the total similarity distance more accurately reflect the health status of the switch machine.

[0108] However, considering that the total similarity distance cannot intuitively reflect the fault problem, this invention converts the similarity corresponding to each stage into a percentage score, thereby performing fault analysis for each stage separately. At the same time, by calculating a weighted comprehensive score, a holistic fault analysis of the entire stage is performed.

[0109] In this embodiment of the invention, to intuitively express the matching status at each stage, it is proposed to convert curve similarity into a percentage score. The specific method is as follows: Set similarity distance thresholds for each stage. Distance to ideal match ,in It is the maximum allowed distance. Let represent the ideal distance for a perfect match, and y be the index number of the stage. The similarity is converted to a percentage score using linear interpolation. : .

[0110] Among them, ideal matching distance This is the minimum distance for an ideal match, typically 0, because completely overlapping curves theoretically have no similarity distance. Similarity distance threshold. To determine the maximum acceptable distance, a threshold is set based on an understanding of the equipment's operating status and the range of curve fluctuations. When the similarity distance exceeds... When the score is 0, it means that the target curve is completely dissimilar to the standard curve at this stage, i.e., if Then directly order Similarly, in the curve integrity check section, curves that fail the integrity check are directly assigned a score of 0. Further fault analysis is performed on each stage, setting three scoring thresholds. If the first part... A score below the first scoring threshold may indicate a problem during the startup phase, such as excessive motor starting resistance; if the second part... If the score is below the second threshold, it may be due to jamming or wear of mechanical parts during the conversion process; if the third part... If the score is below the third scoring threshold, it may be related to a decrease in the performance of relevant components during the locking and releasing phase.

[0111] In addition, a weighted comprehensive score is calculated. Meanwhile, a total score threshold is introduced. When the comprehensive score is lower than the total score threshold, it indicates that there is a large difference between the currently monitored power curve and the normal curve, and the switch machine may be faulty; otherwise, it indicates that the switch machine is working normally as a whole, and the corresponding power curve can be stored in the standard dataset.

[0112] The above methods provide clear directions and clues for the fault diagnosis of switch machines, helping maintenance personnel to quickly locate the fault point, improve maintenance efficiency, and ensure the safe and stable operation of the railway system.

[0113] IV. Data Collection, Processing, and Experiment Setup.

[0114] Using the action data of a switch machine at a certain station as raw data, relevant preprocessing operations were performed to construct a standard curve library, and the accuracy of the dynamic segmentation method and the improved DTW algorithm was verified.

[0115] 1. Data processing and the establishment of a standard curve library.

[0116] Based on the switch machine operation curve data collected by the station signal centralized monitoring system, the data is classified according to switch machine model and equipment number to construct a standard dataset. For example, for newly built stations, the first 10 operation curves within the normal operating range are selected to establish an initial standard curve set; for stations undergoing equipment upgrades, the 10 most recent representative normal operation curves are selected from historical data to construct the standard dataset. Figure 6 As shown. Based on this, a standard curve is generated from the above dataset, and the subsequently collected action curves are compared with the standard curve. If the comparison score is within the set threshold range, the curve is included in the dataset. At the same time, the curve with the earliest action time in the dataset is removed in chronological order, thereby realizing the dynamic updating of the standard curve library. In addition, the maximum allowable value of the conversion process needs to be set, and an alarm mechanism is triggered when the data exceeds the critical value or exceeds the limit.

[0117] It should be noted that the data provided here are for illustrative purposes only and are not intended to constitute a limitation. In practical applications, users can adjust the data according to their specific circumstances.

[0118] 2. Curve segmentation verification.

[0119] Multiple sets of operation curves from the same turnout switch machine were randomly selected for segmentation accuracy verification. The experiments verified the integrity of the curves and accurately divided them. The experimental results are as follows: Figure 7 As shown, points p1 and p2 are dynamic dividing points. This demonstrates the dynamic division result of the curve from the initial position to the reverse position of turnout #2 at a certain station. It can be seen that, based on the curve shape, it is accurately divided into three parts.

[0120] It should be noted that the present invention mainly provides some data examples for ease of understanding. In practical applications, users can make corresponding adjustments according to the actual situation. The present invention does not impose any specific limitations.

[0121] 3. Improve the verification of the DTW algorithm model.

[0122] For example, using the test data of a certain turnout No. 2 at a certain station, Figure 8 The diagram shows the optimal matching path between the test curve and the standard curve. This path is determined based on the principle of minimizing cumulative error, which is the improved DTW distance. The geometric characteristics of this path—especially its proximity to the diagonal—directly reflect the similarity between the two curves. Regions where the path deviates from the diagonal indicate differences between the standard and test curves at these time points. The diagram demonstrates how the two curves can be matched through nonlinear stretching or compression, which can be used to analyze local correspondences within a sequence. By observing the path's trajectory, the synchronicity of the curves can be assessed, and regions of significant similarity or difference can be located.

[0123] Figure 9The display shows the power curves of the No. 2 switch machine at a certain station during normal operation, specifically the power curves during reverse positioning and reverse positioning. These curves were dynamically segmented, and each segment was compared to a standard curve for similarity and scoring. A comprehensive evaluation result was then derived. Based on the test results, the scores for the normal operation curves were all above 94 points. Furthermore, the standard curves were overlaid as a background using colored blocks, allowing for a visual comparison of the curve differences and aiding in the judgment.

[0124] Figure 10 The evaluation results show the situation when abnormal conditions occur during the locking phase of the curve. During the locking phase, a jamming problem in the switch machine caused an abnormal increase in power, resulting in a low locking phase score, and the evaluation system also gave a low overall score. If a piece of equipment frequently exhibits low scores over a period of time, maintenance personnel need to pay special attention to the status of that equipment and conduct further fault analysis on these low-scoring curves.

[0125] Fifty-four action curve data points generated by turnout #2 at the station within one week were selected to verify the evaluation results. Of these 54 curves, two were identified as abnormal, scoring 85.48 and 88.66 points respectively, while the remaining 52 curves all scored above 94 points. This verification result fully demonstrates the feasibility and effectiveness of the adopted evaluation method.

[0126] It should be noted that the parameters provided here are for illustrative purposes only and are not intended to be limiting. In practical applications, users can adjust them according to their specific circumstances.

[0127] Through the above description of the embodiments, those skilled in the art can clearly understand that the above embodiments can be implemented by software, or by using software plus necessary general-purpose hardware platforms. Based on this understanding, the technical solutions of the above embodiments can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (such as a CD-ROM, USB flash drive, mobile hard drive, etc.), including several instructions to cause a computer device (such as a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments of the present invention.

[0128] The above description is merely a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims. The information disclosed in the background section is intended only to enhance the understanding of the overall background technology of the present invention and should not be construed as an admission or implication in any way that such information constitutes prior art known to those skilled in the art.

Claims

1. A switch machine health assessment method based on dynamic segmentation and improved DTW, characterized in that, include: Based on the dynamic characteristics of the switch machine's unlocking, switching and locking stages, a mapping relationship between the action state and the power curve shape of the switch machine is established. The stage boundary point is dynamically identified through the change point detection algorithm to realize the adaptive segmentation of the switch machine's power curve. An improved DTW algorithm is used to calculate the similarity between each segment of the switch machine power curve and a given standard curve, and convert it into a score for the corresponding stage. The scores of all stages are combined and weighted to obtain a comprehensive score. Fault analysis is performed by combining the scores of each stage and the comprehensive score to obtain the status monitoring results of each segment of the curve and the entire stage, thus completing the health assessment. In the improved DTW algorithm, an adaptive regularization window is introduced to constrain the matching path search range, and each segment of the switch machine power curve and the given standard curve are standardized before calculating the similarity.

2. The switch machine health assessment method based on dynamic segmentation and improved DTW according to claim 1, characterized in that, Also includes: Before implementing adaptive segmentation of the switch machine power curve, an integrity check is performed on the switch machine power curve. If the integrity check passes, adaptive segmentation of the switch machine power curve is then performed. If the integrity check fails, the fault analysis will be performed based on the integrity check results.

3. The switch machine health assessment method based on dynamic segmentation and improved DTW according to claim 2, characterized in that, The integrity check of the switch machine power curve includes: Part 1: Unlocking Phase Detection. Check if the action curve exists. If it exists, proceed to the next part. If the curve is 0 or does not exist, it means that the integrity check has not been passed. Part Two: Conversion Phase Detection. The curve status is continuously monitored. If a 0 curve or signal interruption occurs during this process, it indicates an abnormality in the conversion phase and failure to pass the integrity check; otherwise, proceed to the next part. Part 3: Locking phase detection, verifying whether the curve is released normally; if a 0 curve appears at this time, or the signal disappears, or the signal drop exceeds the set threshold, it is determined that the locking phase is abnormal and the integrity check has not been passed; otherwise, continue to the next part; Part 4: Final state detection, confirming that the curve has completely ended; if the curve still has signal output at this time, it is determined that the transition or latch-up phase is abnormal and the integrity check has not passed; otherwise, the integrity check has passed.

4. The switch machine health assessment method based on dynamic segmentation and improved DTW according to claim 1, characterized in that, The process of establishing a mapping relationship between the action state and the power curve shape of the switch machine based on the dynamic characteristics of the unlocking, switching, and locking stages, and dynamically identifying the stage boundary points through a change-point detection algorithm to achieve adaptive segmentation of the switch machine power curve includes: Calculate the slope corresponding to each data point in the switch machine power curve; and, set the window length to calculate the dispersion of the data within the window corresponding to each data point. Based on the dynamic characteristics of the switch machine's unlocking, switching, and locking stages, a mapping relationship between the action state and the switch machine's power curve shape is established. According to the changes in the slope corresponding to each data point and the degree of dispersion, change point detection is performed to dynamically identify the stage boundary points. The switch machine's power curve is then segmented according to the stage boundary points.

5. The switch machine health assessment method based on dynamic segmentation and improved DTW according to claim 4, characterized in that, The slope corresponding to each data point in the calculated switch machine power curve; Furthermore, by setting the window length, the dispersion of the data within the window corresponding to each data point is calculated, including: For the i-th data point Its slope Calculated using the following formula: ； in, For the (i+1)th data point, The time interval between two adjacent data points; Set the window length to L, for the i-th data point The degree of dispersion of data within the window Calculated using the following formula: ； in, Indicates that the index is The data points, where L is an odd number; The average value of the data within the window is expressed as: .

6. A switch machine health assessment method based on dynamic segmentation and improved DTW according to claim 4 or 5, characterized in that, The step of detecting change points and dynamically identifying stage boundary points based on the slope changes and dispersion of each data point includes: Within the first set time range Within, set the first discrete threshold. Iterate through each moment within the first set time range. When the slope changes from negative to positive and the dispersion is less than a threshold, mark the corresponding moment as the boundary between the unlocking phase and the transition phase. , is represented as: ； in, The degree of dispersion of the data within the window corresponding to the i-th data point. Let i be the time corresponding to the i-th data point. , , The slopes are, in order, the slopes corresponding to the (i-2)th data point, the (i-1)th data point, and the ith data point. The end time of the first set time range; Second set time range Within, set a second discrete threshold. Iterate through the moments within the first set time range. When the slope is less than 0 and the dispersion is greater than the threshold, the corresponding time is marked as the boundary between the transition phase and the latching-and-release phase. , is represented as: ； in, , These are the start and end times of the second set time range, respectively. .

7. The switch machine health assessment method based on dynamic segmentation and improved DTW according to claim 1, characterized in that, After implementing adaptive segmentation of the switch machine power curve, the following steps are also included: A power threshold is introduced to determine whether the power value corresponding to each data point in the curve during the conversion stage exceeds the power threshold. If it does, the switch machine power curve data is judged as an abnormal curve.

8. The switch machine health assessment method based on dynamic segmentation and improved DTW according to claim 1, characterized in that, The improved DTW algorithm is used to calculate the similarity between each segment of the switch machine power curve and a given standard curve, including: Each curve segment is standardized and compared with each data point in the standard curve; Calculate each data point after standardization for each curve segment. , and each data point in the standard curve after standardization The Euclidean distance between them, combined with distance parameters, is used to calculate the nonlinear distance. And construct the cumulative distance matrix between each curve segment and the standard curve; An adaptive regularization window r is introduced to constrain the search range of the matching path. Regularized paths are then searched using the cumulative distance matrix between each curve segment and the standard curve. Points (u, o) within the regularized paths satisfy the following conditions: Where u is the index of the data point in each curve segment, and o is the index of the data point in the standard curve; For a single curve Q1 and a standard curve C1, the regular path W1 is defined as the minimum cumulative distance corresponding to the point (u,o). for: ； in, Let represent the minimum cumulative distance between points (u-1,o-1), (u-1,o), and (u,o-1). Finally, the minimum cumulative distance between the single-segment curve Q1 and the standard curve C1 is: ,in, For point The corresponding minimum cumulative distance, U is the index of the last data point of the single-segment curve Q1, and O is the index of the last data point of the standard curve C1; The minimum cumulative distance between a single curve segment and the standard curve is used as the similarity between the single curve segment and the standard curve. The similarity between each curve segment and the given standard curve is calculated in the same way.

9. The switch machine health assessment method based on dynamic segmentation and improved DTW according to claim 1, characterized in that, The scores converted to the corresponding stage include: The adaptive power curve of the switch machine is divided into three segments, and the similarity between each segment and the given standard curve is denoted as . The similarity score is converted to a percentage rating using the following formula: ； Where y is the index number of the stage. The corresponding stage is scored out of 100. This represents the similarity distance threshold for the corresponding stage. This represents the ideal matching distance for the corresponding stage.

10. A switch machine health assessment method based on dynamic segmentation and improved DTW according to claim 1 or 9, characterized in that, A comprehensive score is obtained by weighting the scores from all stages. Fault analysis is then performed by combining the scores from each stage with the comprehensive score to obtain the status monitoring results for each stage corresponding to each curve segment and for the entire stage, including: The adaptive power curve of the switch machine is divided into three segments, and the scores corresponding to the three stages are denoted as follows: , , The weighting coefficients are denoted as follows: , , And satisfy The comprehensive score is calculated using the following formula. : ； Set scoring thresholds for each stage and a total scoring threshold. Determine the working status of the corresponding stage based on the relationship between the score of the corresponding stage and the corresponding scoring threshold. Determine the working status of the entire stage based on the relationship between the comprehensive score and the total scoring threshold, and determine the status monitoring results of the entire stage.

Images