Rail intelligent sensor detection data processing method, terminal and rail detection system
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- CHENGDU CHENGXIN YIHE SYSTEM INTEGRATION DEVELOPMENT CO LTD
- Filing Date
- 2026-04-23
- Publication Date
- 2026-07-21
Smart Images
Figure CN122430458A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of track condition detection technology, and in particular to a method for processing track intelligent sensor detection data, a terminal, and a track detection system. Background Technology
[0002] During long-term service, rail tracks are prone to changes in condition, such as crack initiation, local loosening, peeling damage, or abnormal structural response, in the rail body, joint adjacent areas, turnout adjacent areas, and areas with dense local auxiliary structures. To promptly identify these changes, existing technologies typically employ sensors deployed at different locations along the track to collect track vibration signals or acoustic detection signals. These signals are then combined with threshold determination, peak value extraction, short-time energy analysis, cross-correlation alignment, or single propagation path matching methods to detect the track condition.
[0003] Existing track acoustic wave detection schemes can achieve a certain degree of state recognition in ordinary uniform sections. This is because, in such scenarios, the propagation path of sound waves along the track is relatively simple, and local reflections are weak. Although the detection signals collected by track smart sensors at different locations have propagation delays and amplitude attenuation, they can still be approximated as translational responses of the same event at different locations. In this case, using fixed time windows, first peak extraction, strongest peak extraction, or local cross-correlation calculations can usually complete basic state judgment.
[0004] However, in abrupt structural changes such as in turnout adjacency zones, rail joint proximity zones, and areas with dense local ancillary structures, the propagation behavior of sound waves in the track undergoes significant alterations. Specifically, sound waves in these regions are prone to boundary reflections, mode conversions, multipath superposition, and local standing wave enhancement. In these situations, the responses of the same real track anomaly event on different track smart sensors no longer satisfy a simple single-peak propagation relationship. Instead, they may manifest as multiple delayed peaks, multiple superimposed energy clusters, and combinations of local peaks whose strength relationships do not align with their propagation sequence. Furthermore, the amplitude of certain reflection peaks formed by structural boundaries on a single track smart sensor may even exceed the peak value corresponding to the actual anomaly response.
[0005] In the aforementioned scenarios, if the traditional fixed time window, single-peak extraction, single propagation path estimation, and fixed threshold judgment methods are still used, the following problems are likely to occur: On the one hand, the system may misjudge strong reflection responses caused by structural abrupt changes as true orbital anomaly responses; on the other hand, true anomaly responses may not be stably extracted due to aliasing with multi-path reflections, resulting in missed detections. Especially when multiple orbital smart sensors deployed at different locations participate in detection simultaneously, traditional methods often can only process the local results of a single sensor or a pair of sensors, lacking the overall utilization of the consistency relationship, propagation chain relationship, and reflection delay relationship between multi-location detection responses. Therefore, it is difficult to obtain stable and reliable detection results in special scenarios.
[0006] Therefore, how to perform more targeted data processing on the complex acoustic detection responses collected by multiple track-intelligent sensors deployed at different locations, using only a small amount of key data such as acoustic detection sequences and location data from track-intelligent sensors, so that the real track anomaly response and structural reflection response can be distinguished, and further output reliable track state detection results, remains a technical problem that urgently needs to be solved in this field. Summary of the Invention
[0007] This invention provides a method, terminal, and track detection system for processing data detected by an intelligent track sensor, which at least solves the problem that existing track acoustic wave detection technology has difficulty distinguishing between real track anomaly response and structural reflection response in scenarios of structural abrupt changes, resulting in high false alarm and false negative rates.
[0008] To achieve the above objectives, the present invention provides a method for processing data detected by a smart track sensor, the method comprising: Acquire acoustic wave detection sequences and position data of several track smart sensors deployed at different locations within the current detection window, and construct a candidate event cluster set based on the acoustic wave detection sequences; Based on the candidate event cluster set and location data, candidate relative time differences are generated between smart sensors on different orbits. The global time difference correction result is obtained to obtain the corrected acoustic wave detection sequence. Based on the corrected acoustic wave detection sequence and location data, the main propagation arrival chain and the reflection propagation arrival chain are constructed, and the corresponding chain index results are determined. Based on the corrected acoustic wave detection sequence, main propagation arrival chain, reflection propagation arrival chain, and chain index results, the main propagation observation relationship and reflection propagation observation relationship are established, and joint decomposition processing is performed on the corrected acoustic wave detection sequence to obtain the true anomaly source field and structure reflection field. The focusing anomaly index is calculated based on the real anomaly source field and the structural reflection field, and the track state detection result based on the end-to-end collaborative analysis between track intelligent sensors is output according to the focusing anomaly index.
[0009] Optionally, acquire the acoustic wave detection sequences and position data of several track-based smart sensors deployed at different locations within the current detection window, and construct a candidate event cluster set based on the acoustic wave detection sequences, specifically including: Acquire the acoustic wave detection sequences of several track smart sensors deployed at different locations within the current detection window, and establish a position mapping relationship with each track smart sensor. Construct an envelope response for each acoustic detection sequence; Based on the envelope response, local energy continuous segments are extracted as candidate event clusters, and the corresponding start time, end time, peak time, peak amplitude and local cluster sequence are recorded for each candidate event cluster. The candidate event clusters corresponding to each track smart sensor are organized according to the positional order of the track smart sensor to form the candidate event cluster set.
[0010] Optionally, candidate relative time differences between smart sensors on different orbits are generated based on the candidate event cluster set and location data, and the global time difference correction result is obtained to get the corrected acoustic wave detection sequence, specifically including: Candidate event clusters from different orbital smart sensors are paired and compared, and normalized correlation values between clusters are calculated. The time shift that maximizes the normalized correlation value is determined as the candidate relative time difference for the corresponding orbital smart sensor pair, and the normalized correlation value is used as the correlation weight for the corresponding candidate relative time difference. The global time difference correction result is obtained based on the candidate relative time difference and related weights of each track intelligent sensor pair; Based on the global time difference correction result, the acoustic wave detection sequence of each track intelligent sensor is time-corrected to obtain the corrected acoustic wave detection sequence.
[0011] Optionally, based on the corrected acoustic wave detection sequence and location data, a main propagation arrival chain and a reflection propagation arrival chain are constructed, and the corresponding chain index results are determined, specifically including: The effective propagation speed is determined based on the global time difference correction result and the location data. Based on the corrected acoustic detection sequence, candidate peak values corresponding to each track smart sensor are extracted, and the effective propagation speed is used as a propagation constraint to connect the candidate peak values of different track smart sensors across sensors. The peak connection result with the lowest connection cost and the largest number of track smart sensors is determined as the main propagation arrival chain, and the peak connection results among the remaining peaks that are distributed with delay relative to the main propagation arrival chain are determined as the reflection propagation arrival chain. Based on the main propagation arrival chain and the reflection propagation arrival chain, chain delay features, chain stability features, and chain energy ratio features are extracted to form the chain index result.
[0012] Optionally, based on the corrected acoustic wave detection sequence, main propagation arrival chain, reflection propagation arrival chain, and chain index results, the main propagation observation relationship and reflection propagation observation relationship are established, specifically including: The detection section is discretized along the track length direction to form multiple track discrete nodes; The main propagation arrival delay is determined based on the main propagation arrival chain, and the reflection propagation arrival delay is determined based on the reflection propagation arrival chain and the chain index result. For each orbital intelligent sensor and each orbital discrete node, a master propagation response term and a reflection propagation response term are constructed respectively, and an observation relationship is established for joint decomposition processing of the real anomaly source field and the structural reflection field.
[0013] Optionally, a joint decomposition process is performed on the corrected acoustic detection sequence to obtain the true anomaly source field and structure reflection field, specifically including: A joint solution objective is constructed based on the aforementioned observation relationship; Based on the joint solution objective, the real anomaly source field and the structural reflection field are updated alternately, so that the response that satisfies the main propagation response relationship and has spatial focusing characteristics is assigned to the real anomaly source field, and the response that satisfies the reflection propagation response relationship is assigned to the structural reflection field. When the joint solution objective reaches the convergence condition, the true anomaly source field and the structural reflection field are output.
[0014] Optionally, the focusing anomaly index is calculated based on the actual anomaly source field and the structural reflection field, specifically including: Calculate the real anomaly focusing energy corresponding to each discrete node of the track based on the real anomaly source field; The reflection ratio of each track discrete node is calculated based on the structural reflection field, and the focusing anomaly index is calculated by combining the multi-sensor consistency factor of each track discrete node. The focused anomaly indices corresponding to each discrete node of the track are organized in order of track position to form an anomaly index sequence for outputting track status detection results.
[0015] Optionally, the track state detection results based on end-to-end collaborative analysis between track smart sensors are output according to the focused anomaly index, specifically including: The median and median absolute deviation are calculated based on the abnormal index sequence, and a state determination threshold is generated. Track discrete nodes with a focusing anomaly index lower than the warning threshold are identified as normal nodes; track discrete nodes with a focusing anomaly index greater than or equal to the warning threshold but lower than the anomaly threshold are identified as warning nodes; and track discrete nodes with a focusing anomaly index greater than or equal to the anomaly threshold are identified as anomaly nodes. Adjacent abnormal nodes and adjacent early warning nodes are continuously merged to form corresponding abnormal sections or early warning sections, and the track status detection results are output.
[0016] Furthermore, to achieve the above objectives, the present invention also provides a track intelligent sensor detection terminal, the terminal being used to execute the track intelligent sensor detection data processing method as described in any of the preceding claims, specifically including: processor; Memory, used to store computer programs; When the computer program is executed by the processor, it performs the following operations: Acquire acoustic wave detection sequences and position data of several track smart sensors deployed at different locations within the current detection window, and construct a candidate event cluster set based on the acoustic wave detection sequences; Based on the candidate event cluster set and location data, candidate relative time differences are generated between smart sensors on different orbits. The global time difference correction result is obtained to obtain the corrected acoustic wave detection sequence. Based on the corrected acoustic wave detection sequence and location data, the main propagation arrival chain and the reflection propagation arrival chain are constructed, and the corresponding chain index results are determined. Based on the corrected acoustic wave detection sequence, main propagation arrival chain, reflection propagation arrival chain, and chain index results, the main propagation observation relationship and reflection propagation observation relationship are established, and joint decomposition processing is performed on the corrected acoustic wave detection sequence to obtain the true anomaly source field and structure reflection field. The focusing anomaly index is calculated based on the real anomaly source field and the structural reflection field, and the track state detection result based on the end-to-end collaborative analysis between track intelligent sensors is output according to the focusing anomaly index.
[0017] Furthermore, to achieve the above objectives, the present invention also provides a track detection system, the system comprising: Several smart track sensors deployed at different locations are configured to collect acoustic wave detection sequences within the current detection window; Data transmission interface, used to transmit acoustic wave detection sequences and position data of each track smart sensor; The track intelligent sensor detection terminal described above is used to perform data processing on the acoustic detection sequence and the position data to output track state detection results; The result output unit is used to display or alarm the track status detection results.
[0018] The beneficial effects of this invention are as follows: It proposes a method, terminal, and track detection system for processing data from intelligent track sensors. The method first organizes the acoustic wave detection sequences collected by multiple intelligent track sensors deployed at different locations into candidate event clusters. Then, it uses the candidate relative time difference relationships between multiple sensors to obtain the global time difference correction result. Using the corrected acoustic wave detection sequences and location data, it constructs the main propagation arrival chain and the reflection propagation arrival chain, and introduces both types of propagation relationships into the subsequent joint decomposition processing. This allows the real track anomaly response and structural reflection response, which were originally mixed in the same detection window, to be separated at the data processing level. Furthermore, by comprehensively utilizing the real anomaly source field, structural reflection field, and multi-sensor consistency relationships, it forms a focused anomaly index and outputs track state detection results based on end-to-end collaborative analysis between intelligent track sensors. This improves the accuracy, stability, and interpretability of track state detection under structural abrupt change scenarios, and reduces the risk of false alarms and missed alarms caused by strong reflections, multi-path superposition, and misjudgment of local peaks. Attached Figure Description
[0019] Figure 1 This is a flowchart illustrating the data processing method for track intelligent sensor detection according to an embodiment of the present invention. Detailed Implementation
[0020] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.
[0021] This invention provides a method for processing data detected by a smart track sensor, referring to... Figure 1 , Figure 1 This is a flowchart illustrating the data processing method for track intelligent sensor detection according to an embodiment of the present invention.
[0022] In this embodiment, a method for processing data detected by a smart track sensor includes: S1: Obtain the acoustic wave detection sequences and position data of several track smart sensors deployed at different locations within the current detection window, and construct a candidate event cluster set based on the acoustic wave detection sequences.
[0023] Specifically, the acoustic wave detection sequences corresponding to several track-based smart sensors deployed at different locations within the current detection window are obtained, and a positional mapping relationship is established with each track-based smart sensor. An envelope response is constructed for each acoustic wave detection sequence. Based on the envelope response, local energy continuous segments are extracted as candidate event clusters, and the corresponding start time, end time, peak time, peak amplitude, and local cluster sequence are recorded for each candidate event cluster. The candidate event clusters corresponding to each track-based smart sensor are organized according to the positional order of the track-based smart sensors to form the candidate event cluster set.
[0024] In this embodiment of the invention, step S1 is used to transform the responses with complex morphology, obvious oscillations, and local multi-peak overlap in the original continuous acoustic wave detection sequence into data units that can be compared, matched, and organized subsequently. The candidate event cluster set is not directly equivalent to the actual orbital anomaly response; rather, it preserves local energy events that may participate in subsequent propagation consistency analysis to avoid premature rejection of the original sequence at the initial processing stage. It should be noted that in structural abrupt change regions, if a single peak value or fixed threshold is used for hard judgment of events in the initial stage, it is easy to mistakenly delete the actual anomaly response or mistakenly retain the structural reflection response. Therefore, in this embodiment of the invention, the candidate event cluster organization method is first adopted to provide a more sufficient data foundation for subsequent steps.
[0025] In practical applications, the first step is to acquire the acoustic detection sequence corresponding to each track-based intelligent sensor within the current detection window, and establish a position mapping relationship between each track-based intelligent sensor and its corresponding location. It should be noted that the track-based intelligent sensors are configured with both acoustic wave sensing elements and position sensing elements. This position mapping relationship is used to determine the exact location on the track from which each acoustic wave detection sequence originates, ensuring that propagation delay, propagation speed, connection cost, and arrival chain information encountered in subsequent steps can be traced back to the specific spatial location on the track. Furthermore, in one executable implementation, the track-based intelligent sensors can be sequentially numbered along the track length, establishing a one-to-one correspondence between each acoustic wave detection sequence and the location of each track-based intelligent sensor. The start time and length of the current detection window are also recorded to uniformly constrain the time range of this data processing.
[0026] Subsequently, an envelope response is constructed for each acoustic wave detection sequence. Specifically, in this embodiment of the invention, the envelope response satisfies: ; in, This represents the envelope response corresponding to the i-th orbital smart sensor. This represents the acoustic wave detection sequence corresponding to the i-th orbital smart sensor. The expression represents the Hilbert transform, where j represents the imaginary unit and t represents time. Its purpose is to transform the original oscillating waveform into a more intuitive energy profile curve, clearly demonstrating the duration and energy concentration of local events along the time axis. It should be noted that the envelope response is not a simple smoothing of the original data, but rather provides a unified metric for identifying local energy segments from continuous time-domain signals, particularly suitable for special detection scenarios involving multipath superposition and local destructive / constructive interactions.
[0027] After obtaining the envelope response, continuous segments of local energy are extracted as candidate event clusters based on the envelope response. For each candidate event cluster, the corresponding start time, end time, peak time, peak amplitude, and local cluster sequence are recorded. Specifically, continuous segments in the envelope response that are consistently higher than the local background energy level and satisfy the minimum duration condition can be identified and retained as candidate event cluster objects. Furthermore, in one executable implementation, the start time of the cluster can be used to describe the entry position of the local event, the end time to describe the exit position, and the peak time and peak amplitude to describe the most significant response position and intensity of the local event. The local cluster sequence then serves as the direct input for subsequent normalized correlation calculations. In this way, the local energy events originally scattered throughout the original acoustic wave detection sequence are organized into multiple structured data units.
[0028] After completing the above processing, the candidate event clusters corresponding to each track smart sensor are organized according to the positional order of the track smart sensors to form a candidate event cluster set. This set will be directly used in step S2 to calculate the candidate relative time difference between different track smart sensors. In one specific embodiment, five track smart sensors can be deployed along the length of the rail in the turnout adjacency area. The above processing is performed on the acoustic wave detection sequence within a preset detection window after the train passes. After processing, each track smart sensor can obtain several candidate event clusters. The number of these candidate event clusters may not be the same among different track smart sensors, but they are all retained for subsequent comparison processes, thereby providing a sufficient candidate basis for identifying the real propagation relationship in complex scenarios.
[0029] S2: Generate candidate relative time differences between smart sensors on different orbits based on candidate event cluster sets and location data, and obtain global time difference correction results to obtain the corrected acoustic wave detection sequence.
[0030] Specifically, candidate event clusters from different orbital smart sensors are paired and compared, and normalized correlation values between clusters are calculated. The time shift that maximizes the normalized correlation value is determined as the candidate relative time difference for the corresponding orbital smart sensor pair, and the normalized correlation value is used as the correlation weight for the corresponding candidate relative time difference. A global time difference correction result is obtained based on the candidate relative time difference and correlation weight for each orbital smart sensor pair. The acoustic wave detection sequence of each orbital smart sensor is time-corrected according to the global time difference correction result to obtain the corrected acoustic wave detection sequence.
[0031] In this embodiment of the invention, step S2 is used to establish a unified and stable time reference among the multi-location detection responses. It should be noted that in the structural abrupt change scenario corresponding to the current technical problem, the local peak values collected by different orbital intelligent sensors often do not have a one-to-one correspondence. Therefore, if the arrival time of a certain local peak is directly used as the basis for global alignment, it is easy to cause all subsequent analyses to deviate from the true propagation relationship due to the selection of the wrong peak. In this embodiment of the invention, instead of directly making a rigid correspondence to the original peak values, it compares the morphological consistency of the candidate event cluster set formed in step S1 at the cluster level, and further obtains the global time difference correction result within the overall scope of the multi-sensor system.
[0032] In this embodiment of the invention, candidate event clusters from different orbital intelligent sensors are first paired and compared, and the normalized correlation value between the clusters is calculated. Specifically, the normalized correlation value satisfies: ; in, This indicates that the candidate event clusters of the i-th orbital smart sensor and the candidate event clusters of the j-th orbital smart sensor have a time shift of . Normalized correlation value at time, and These represent the two paired local wave cluster sequences. This represents a very small positive number. Therefore, even when the amplitude may be affected by propagation attenuation, installation coupling, or local boundary conditions, the consistency of the overall shape of the wave cluster is still used to assess whether two local events might originate from the same propagation process. Through this processing, even if the local response amplitudes on two orbital smart sensors differ significantly, as long as their shapes exhibit high consistency under appropriate time shifts, they can still be retained as valid time difference candidates.
[0033] After obtaining the normalized correlation value, the time shift that maximizes the normalized correlation value is determined as the candidate relative time difference for the corresponding orbital smart sensor pair, and the normalized correlation value is used as the correlation weight for the corresponding candidate relative time difference. Further, the global time difference correction result is obtained based on the candidate relative time differences and correlation weights of each orbital smart sensor pair. Specifically, the global time difference correction result in this embodiment satisfies: ; in, This represents the correlation weight between the i-th and j-th track smart sensors. and Let represent the global time difference correction values corresponding to the i-th and j-th orbital smart sensors, respectively. This represents the candidate relative time difference between the i-th and j-th orbital smart sensors. It should be noted that the above process does not rely solely on the local results of a single pair of orbital smart sensors to determine the global alignment relationship. Instead, it integrates the candidate relative time differences between all pairs of orbital smart sensors to ensure that the overall solution simultaneously satisfies multiple sets of observation relationships as much as possible, thereby suppressing the disruption to the overall timing relationship caused by local mispairing.
[0034] Based on the global time difference correction results, the acoustic detection sequences of each orbital smart sensor are time-corrected to obtain the corrected acoustic detection sequences. The corrected acoustic detection sequences output in step S2 will then serve as the direct input for step S3. In other words, step S2 not only outputs a set of time correction values, but more importantly, it places the responses collected by multiple orbital smart sensors under the same time reference, ensuring a unified temporal basis for the subsequent construction of the main propagation arrival chain and the reflection propagation arrival chain. In one specific implementation, if some orbital smart sensor pairs exhibit locally high correlation values due to structural reflection mismatches, these locally high correlation values will not be infinitely amplified during the global solution process. Instead, they will be constrained by the consistency relationships of other orbital smart sensor pairs, thus making the final correction result more closely resemble the true propagation relationship supported by multiple locations.
[0035] S3: Construct the main propagation arrival chain and the reflection propagation arrival chain based on the corrected acoustic wave detection sequence and location data, and determine the corresponding chain index results.
[0036] Specifically, the effective propagation velocity is calculated based on the global time difference correction result and the position data; candidate peaks corresponding to each orbital smart sensor are extracted based on the corrected acoustic detection sequence, and cross-sensor connections are made between candidate peaks of different orbital smart sensors using the effective propagation velocity as a propagation constraint; the peak connection result with the lowest connection cost and covering the largest number of orbital smart sensors is determined as the main propagation arrival chain, and the peak connection result of the remaining peaks that are distributed with delay relative to the main propagation arrival chain is determined as the reflection propagation arrival chain; chain delay characteristics, chain stability characteristics, and chain energy ratio characteristics are extracted based on the main propagation arrival chain and the reflection propagation arrival chain to form the chain index result.
[0037] In this embodiment of the invention, step S3 is used to extract a more physically meaningful propagation chain structure from the corrected multi-location detection response. It should be noted that, in the current scenario, not all local peaks on different orbital smart sensors belong to the same propagation process. Generally, a set of peak connections that better conforms to the propagation laws of orbital space, covers more orbital smart sensors, and has lower connection costs is more likely to correspond to the main propagation process induced by the actual anomaly. Conversely, another set of peak connections that exhibits an overall delay relative to this main propagation process but still possesses a certain degree of spatial consistency is more likely to correspond to a reflection propagation process caused by structural boundaries or auxiliary structures. This step is designed for this purpose.
[0038] In this embodiment of the invention, the effective propagation velocity is first determined based on the global time difference correction result and location data. Specifically, the effective propagation velocity satisfies: ; in, Indicates the effective transmission speed. and These represent the position data of the i-th and j-th track smart sensors, respectively. and Let represent the global time difference correction values corresponding to the i-th and j-th orbital smart sensors, respectively. This represents a very small positive number. The effective propagation velocity here is not required to be exactly the same as the theoretical wave velocity in an ideal homogeneous medium, but rather it is used to characterize the equivalent propagation velocity within the current detection window, under the combined constraints of the current orbital structure conditions and the current multi-location measurement results. Using the median instead of the mean helps to reduce the impact of a few outliers on the propagation velocity estimation.
[0039] Subsequently, candidate peak values corresponding to each orbital smart sensor are extracted based on the corrected acoustic detection sequence, and cross-sensor connections are made between candidate peak values from different orbital smart sensors using the effective propagation speed as a propagation constraint. Specifically, the connection cost between candidate peak values satisfies: ; Where J represents the connection cost between candidate peaks. and These represent the arrival times of the two candidate peaks, respectively. and These represent the position data of the orbital smart sensor corresponding to the two candidate peaks. and These represent the peak amplitudes of the two candidate peaks, and This represents the weighting coefficient. The first part of the formula reflects whether the relationship between time variation and spatial distance conforms to the effective propagation law under the current window, while the second part reflects whether the local response intensity varies within a reasonable range. By considering both factors simultaneously, errors caused by connecting solely based on temporal sequence or peak magnitude can be avoided.
[0040] After constructing the connection cost, the peak connection result with the lowest connection cost and covering the largest number of orbital smart sensors is determined as the main propagation arrival chain. The peak connection results among the remaining peaks that exhibit a delayed distribution relative to the main propagation arrival chain are determined as reflection propagation arrival chains. Further, chain delay features, chain stability features, and chain energy proportion features are extracted based on the main propagation arrival chain and the reflection propagation arrival chain to form a chain index result. The chain delay feature reflects the overall time offset relationship of a certain reflection propagation arrival chain relative to the main propagation arrival chain; the chain stability feature reflects the consistency of the propagation slope of the same propagation chain among different orbital smart sensors; and the chain energy proportion feature reflects the proportion of energy accounted for by a certain reflection propagation process in the overall observation response. It should be noted that the chain index result is not independently generated data, but rather an important constraint basis used in step S4 to establish the main propagation response term and the reflection propagation response term.
[0041] In one specific implementation, if the current detection section includes a turnout adjacent structure, the main propagation arrival chain typically corresponds to a set of peaks that approximately satisfy a uniform propagation slope on most track smart sensors. The reflection propagation arrival chain, however, may appear as another set of peak connections with a delay of several sampling intervals relative to the main propagation arrival chain. By explicitly distinguishing these two types of arrival chains in step S3, subsequent steps can avoid treating all responses as signals from the same source, thus laying the foundation for separating the actual abnormal response from the structural reflection response at the data level.
[0042] S4: Based on the corrected acoustic wave detection sequence, main propagation arrival chain, reflection propagation arrival chain, and chain index results, establish the main propagation observation relationship and the reflection propagation observation relationship, and perform joint decomposition processing on the corrected acoustic wave detection sequence to obtain the true anomaly source field and structure reflection field.
[0043] Specifically, the detection section is discretized along the track length to form multiple track discrete nodes; the main propagation arrival delay is determined based on the main propagation arrival chain, and the reflection propagation arrival delay is determined based on the reflection propagation arrival chain and the chain index result; the main propagation response term and the reflection propagation response term are constructed for each track smart sensor and each track discrete node, and an observation relationship is established for the joint decomposition processing of the real anomaly source field and the structural reflection field.
[0044] Based on this, a joint solution objective is constructed based on the observation relationship; the real anomaly source field and the structural reflection field are alternately updated based on the joint solution objective, so that the response that satisfies the master propagation response relationship and has spatial focusing characteristics is assigned to the real anomaly source field, and the response that satisfies the reflection propagation response relationship is assigned to the structural reflection field; when the joint solution objective reaches the convergence condition, the real anomaly source field and the structural reflection field are output.
[0045] In this embodiment of the invention, step S4 is used to enable the main propagation response and reflection propagation response, which were originally mixed together within the same detection window, to be jointly processed under a unified observation relationship, and further output the real anomaly source field and structural reflection field with physical interpretation significance. First, the response relationships corresponding to the main propagation and reflection propagation are established respectively. Then, the two types of response relationships are used to jointly decompose the corrected acoustic wave detection sequence, so that the response components that are more in line with the main propagation law and have local focusing characteristics are assigned to the real anomaly source field, while the response components that are in line with the delayed reflection law are assigned to the structural reflection field.
[0046] In this embodiment of the invention, the detection segment is first discretized along the track length direction to form multiple discrete track nodes. This is done to transform the continuous track segment into a finite set of nodes that are easy to calculate and associate, so that each subsequent discrete node can carry the corresponding description of the real anomaly response and reflection response. Further, the main propagation arrival delay is determined based on the main propagation arrival chain obtained in step S3, and the reflection propagation arrival delay is determined based on the reflection propagation arrival chain and chain index results.
[0047] Subsequently, the master propagation response term and reflection propagation response term were constructed for each orbital smart sensor and each orbital discrete node, and the following observation relationship was established: ; in, This represents the corrected acoustic wave detection sequence corresponding to the i-th orbital smart sensor. This represents the main propagation response coefficient from the nth discrete node on the track to the ith smart sensor on the track. This represents the true anomaly source field corresponding to the nth discrete orbit node. This represents the main propagation arrival delay from the nth discrete node on the track to the ith smart sensor on the track. This represents the reflection propagation response coefficient from the nth discrete node on the track to the ith smart sensor on the track. This represents the structural reflection field corresponding to the nth discrete node of the orbit. This represents the reflection propagation arrival delay from the nth discrete node on the track to the ith smart sensor on the track. Let N represent the residual error term, and N represent the total number of discrete nodes in the orbit. This observation relationship implies that the response observed by each orbital smart sensor is no longer interpreted solely as the result of a single propagation source, but rather as the superposition of the main propagation response caused by the real anomaly and the reflection propagation response caused by the structural boundary.
[0048] After establishing the aforementioned observation relationships, a joint solution objective is further constructed, and the real anomaly source field and the structural reflection field are jointly decomposed. Specifically, the joint solution objective satisfies: ; Where y represents the observation data composed of the corrected acoustic detection sequences from each orbital smart sensor. This represents the main propagation response matrix, composed of the main propagation response coefficients. Let represent the reflection propagation response matrix composed of each reflection propagation response coefficient, s represent the unsolved quantity composed of the real anomaly source fields corresponding to each discrete node of the orbit, and h represent the unsolved quantity composed of the structural reflection fields corresponding to each discrete node of the orbit. , and Let denote the regularization coefficient, and p denote the fractional sparsity exponent satisfying 0. <p<1, This represents the adjacency weight between discrete nodes on the track. In the above objective, the first term is used to ensure the overall reconstruction result remains consistent with the actual observation data; the second term is used to make the true anomaly source field spatially more inclined to form a focused response at a few local nodes; the third term is used to maintain a reasonable continuity of the true anomaly response between adjacent discrete nodes on the track; and the fourth term is used to make the structural reflection field absorb response components that match the reflection propagation relationship as much as possible. It should be noted that the above settings are all centered around the technical problem this invention aims to solve: how to stably separate the true anomaly response from the mixed response in scenarios of strong reflection and multipath superposition.
[0049] In this embodiment of the invention, the real anomaly source field and the structural reflection field are alternately updated based on the joint solution objective. Responses satisfying the master propagation response relationship and exhibiting spatial focusing characteristics are assigned to the real anomaly source field, while responses satisfying the reflection propagation response relationship are assigned to the structural reflection field. When the joint solution objective reaches convergence, the real anomaly source field and the structural reflection field are output. Furthermore, in one executable implementation, the structural reflection field can be fixed first to update the real anomaly source field, and then the real anomaly source field can be fixed again to update the structural reflection field. This process is repeated iteratively until the change in the objective function between two adjacent iterations is less than a preset threshold. In this way, the real anomaly source field and structural reflection field output in step S4 not only maintain consistency with the propagation chain results in step S3, but also provide a direct and interpretable data foundation for subsequent orbital state determination.
[0050] S5: Calculate the focusing anomaly index based on the real anomaly source field and the structural reflection field, and output the track state detection result based on the end-to-end collaborative analysis between track intelligent sensors according to the focusing anomaly index.
[0051] Specifically, the true anomalous focusing energy corresponding to each discrete node of the track is calculated based on the true anomalous source field; the reflection ratio corresponding to each discrete node of the track is calculated based on the structural reflection field, and the focusing anomaly index is calculated in combination with the multi-sensor consistency factor corresponding to each discrete node of the track; the focusing anomaly index corresponding to each discrete node of the track is organized according to the track position order to form an anomaly index sequence for outputting track status detection results.
[0052] Based on this, the median and median absolute deviation are calculated based on the anomaly index sequence, and a state determination threshold is generated. Track discrete nodes with a focusing anomaly index lower than the warning threshold are identified as normal nodes, track discrete nodes with a focusing anomaly index greater than or equal to the warning threshold but lower than the anomaly threshold are identified as warning nodes, and track discrete nodes with a focusing anomaly index greater than or equal to the anomaly threshold are identified as anomaly nodes. Adjacent anomaly nodes and adjacent warning nodes are continuously merged to form corresponding anomaly segments or warning segments, and the track state detection results are output.
[0053] In this embodiment of the invention, step S5 is used to further convert the real anomaly source field and structural reflection field output in step S4 into orbital state results. It should be noted that simply obtaining the real anomaly source field and structural reflection field is insufficient to directly form a final conclusion, because the energy distribution, reflection ratio, and multi-sensor support at different discrete nodes still vary. Therefore, this embodiment of the invention introduces a focusing anomaly index to comprehensively express multiple factors that can reflect the reliability of the real anomaly, thereby forming a determineable, sortable, and outputtable orbital state detection result.
[0054] In this embodiment of the invention, the true anomaly focusing energy corresponding to each discrete node of the orbit is first calculated based on the true anomaly source field. Specifically, the true anomaly focusing energy satisfies: ; in, This represents the true anomaly focusing energy corresponding to the nth discrete orbital node. This represents the true anomaly source field corresponding to the nth discrete orbit node. This expression quantifies the local energy intensity of each discrete orbit node within the true anomaly source field. It's important to note that the true anomaly focused energy reflects the portion of energy still attributed to the true anomaly response after joint decomposition processing, rather than the total energy in the original observation. Therefore, this differs fundamentally from the traditional method of directly extracting peak values or energy from the original waveform.
[0055] After obtaining the true anomalous focusing energy, the reflection ratio corresponding to each discrete node of the orbit is calculated based on the structural reflection field, and the focusing anomaly index is calculated by combining the multi-sensor consistency factor corresponding to each discrete node of the orbit. Specifically, the focusing anomaly index satisfies: ; in, This represents the focusing anomaly index corresponding to the nth discrete orbit node. This represents the multi-sensor consistency factor corresponding to the nth discrete orbit node. This represents the reflection ratio corresponding to the nth discrete orbital node. The meaning of this expression is: the higher the true anomaly focusing energy, the more likely the node is to have a true anomaly; the higher the multi-sensor consistency factor, the more likely the anomaly interpretation of the node is supported by multiple orbital smart sensors; the higher the reflection ratio, the heavier the structural reflection component in the node's observation response, and its anomaly confidence should be appropriately lowered. By combining these three factors, nodes with simply high energy can be distinguished from truly reliable anomaly nodes.
[0056] After generating the focusing anomaly indices corresponding to the discrete nodes of each orbit, they are organized according to the orbital position order to form an anomaly index sequence. Further, the median and median absolute deviation are calculated based on the anomaly index sequence, and a state determination threshold is generated. Specifically, the state determination threshold satisfies: ; ; ; ; in, The index represents the median of the focus anomaly index, and MAD represents the median absolute deviation of the focus anomaly index. Indicates the warning threshold. Indicates the abnormal threshold. and Represents the threshold coefficient and Greater than The purpose of using the median and median absolute deviation is to enhance the robustness of the threshold generation process to a small number of extremely high reflectivity nodes, making the state determination within the current detection window more adaptable to skewed distributions in special scenarios.
[0057] After obtaining the status determination threshold, discrete track nodes with a focusing anomaly index below the warning threshold are identified as normal nodes; discrete track nodes with a focusing anomaly index greater than or equal to the warning threshold but lower than the anomaly threshold are identified as warning nodes; and discrete track nodes with a focusing anomaly index greater than or equal to the anomaly threshold are identified as anomaly nodes. Subsequently, adjacent anomaly nodes and adjacent warning nodes are continuously merged to form corresponding anomaly segments or warning segments, and the track status detection results are output. It should be noted that the continuous merging setting ensures that the final output is no longer limited to a single discrete point, but rather better reflects the spatial extension characteristics of track defects, thereby improving the usability of the final result for operation and maintenance applications.
[0058] In one specific embodiment, multiple track smart sensors can be deployed along the length of the rail in the turnout adjacency area, and acoustic wave detection sequences can be collected within a detection window after a train passes. A candidate event cluster set is formed through step S1, a corrected acoustic wave detection sequence is obtained through step S2, the main propagation arrival chain and the reflection propagation arrival chain are distinguished through step S3, the actual anomaly source field and structural reflection field are obtained through step S4, and finally, the corresponding anomaly section and warning section are output through step S5. It should be noted that the above embodiments are only used to illustrate the data processing flow and operating logic of the present invention and should not be construed as limiting the scope of protection of the present invention. Under different track structures, different numbers of track smart sensors, different sampling rates, and different detection window lengths, the specific parameter settings can be adjusted according to actual application needs. As long as the technical concept does not deviate from the core concept of the present invention, it should be considered to fall within the scope of protection of the present invention.
[0059] In a preferred embodiment, the present invention also provides a track intelligent sensor detection terminal, which is used to execute the track intelligent sensor detection data processing method as described in any of the preceding claims, specifically including: a processor; a memory for storing a computer program; wherein, when the computer program is executed by the processor, it is used to perform the following operations: Acquire acoustic wave detection sequences and position data of several track smart sensors deployed at different locations within the current detection window, and construct a candidate event cluster set based on the acoustic wave detection sequences; Based on the candidate event cluster set and location data, candidate relative time differences are generated between smart sensors on different orbits. The global time difference correction result is obtained to obtain the corrected acoustic wave detection sequence. Based on the corrected acoustic wave detection sequence and location data, the main propagation arrival chain and the reflection propagation arrival chain are constructed, and the corresponding chain index results are determined. Based on the corrected acoustic wave detection sequence, main propagation arrival chain, reflection propagation arrival chain, and chain index results, the main propagation observation relationship and reflection propagation observation relationship are established, and joint decomposition processing is performed on the corrected acoustic wave detection sequence to obtain the true anomaly source field and structure reflection field. The focusing anomaly index is calculated based on the real anomaly source field and the structural reflection field, and the track state detection result based on the end-to-end collaborative analysis between track intelligent sensors is output according to the focusing anomaly index.
[0060] Other embodiments or specific implementations of the intelligent track sensor detection terminal of the present invention can be referred to the above-described method embodiments, and will not be repeated here.
[0061] In a preferred embodiment, the present invention also provides a track detection system, which includes: Several smart track sensors deployed at different locations are configured to collect acoustic wave detection sequences within the current detection window; Data transmission interface, used to transmit acoustic wave detection sequences and position data of each track smart sensor; The track intelligent sensor detection terminal described above is used to perform data processing on the acoustic detection sequence and the position data to output track state detection results; The result output unit is used to display or alarm the track status detection results.
[0062] Other embodiments or specific implementations of the track detection system of the present invention can be referred to the above-described method embodiments, and will not be repeated here.
[0063] It is understood that in the description of this specification, references to terms such as "one embodiment," "another embodiment," "other embodiments," or "first embodiment to Nth embodiment," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the present invention. In this specification, illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.
[0064] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or system that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or system. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or system that includes that element.
[0065] The above are merely preferred embodiments of the present invention and do not limit the scope of the patent. Any equivalent structural or procedural transformations made based on the description and drawings of the present invention, or direct or indirect applications in other related technical fields, are similarly included within the scope of patent protection of the present invention.
Claims
1. A track intelligent sensor detection data processing method, characterized in that, The method includes: Acquire acoustic wave detection sequences and position data of several track smart sensors deployed at different locations within the current detection window, and construct a candidate event cluster set based on the acoustic wave detection sequences; Based on the candidate event cluster set and location data, candidate relative time differences are generated between smart sensors on different orbits. The global time difference correction result is obtained to obtain the corrected acoustic wave detection sequence. Based on the corrected acoustic wave detection sequence and location data, the main propagation arrival chain and the reflection propagation arrival chain are constructed, and the corresponding chain index results are determined. Based on the corrected acoustic wave detection sequence, main propagation arrival chain, reflection propagation arrival chain, and chain index results, the main propagation observation relationship and reflection propagation observation relationship are established, and joint decomposition processing is performed on the corrected acoustic wave detection sequence to obtain the true anomaly source field and structure reflection field. The focusing anomaly index is calculated based on the real anomaly source field and the structural reflection field, and the track state detection result based on the end-to-end collaborative analysis between track intelligent sensors is output according to the focusing anomaly index.
2. The method for processing data detected by a smart track sensor as described in claim 1, characterized in that, Acquire acoustic wave detection sequences and position data from several track-based smart sensors deployed at different locations within the current detection window. Construct a candidate event cluster set based on the acoustic wave detection sequences, specifically including: Acquire the acoustic wave detection sequences of several track smart sensors deployed at different locations within the current detection window, and establish a position mapping relationship with each track smart sensor. Construct an envelope response for each acoustic detection sequence; Based on the envelope response, local energy continuous segments are extracted as candidate event clusters, and the corresponding start time, end time, peak time, peak amplitude and local cluster sequence are recorded for each candidate event cluster. The candidate event clusters corresponding to each track smart sensor are organized according to the positional order of the track smart sensor to form the candidate event cluster set.
3. The method for processing track intelligent sensor detection data as described in claim 1, characterized in that, Candidate relative time differences between smart sensors on different orbits are generated based on candidate event cluster sets and location data. Global time difference correction results are then obtained to acquire the corrected acoustic detection sequence, specifically including: Candidate event clusters from different orbital smart sensors are paired and compared, and normalized correlation values between clusters are calculated. The time shift that maximizes the normalized correlation value is determined as the candidate relative time difference for the corresponding orbital smart sensor pair, and the normalized correlation value is used as the correlation weight for the corresponding candidate relative time difference. The global time difference correction result is obtained based on the candidate relative time difference and related weights of each track intelligent sensor pair; Based on the global time difference correction result, the acoustic wave detection sequence of each track intelligent sensor is time-corrected to obtain the corrected acoustic wave detection sequence.
4. The method for processing track intelligent sensor detection data as described in claim 1, characterized in that, Based on the corrected acoustic wave detection sequence and location data, the main propagation arrival chain and the reflection propagation arrival chain are constructed, and the corresponding chain index results are determined, specifically including: The effective propagation speed is determined based on the global time difference correction result and the location data. Based on the corrected acoustic detection sequence, candidate peak values corresponding to each track smart sensor are extracted, and the effective propagation speed is used as a propagation constraint to connect the candidate peak values of different track smart sensors across sensors. The peak connection result with the lowest connection cost and the largest number of track smart sensors is determined as the main propagation arrival chain, and the peak connection results among the remaining peaks that are distributed with delay relative to the main propagation arrival chain are determined as the reflection propagation arrival chain. Based on the main propagation arrival chain and the reflection propagation arrival chain, chain delay features, chain stability features, and chain energy ratio features are extracted to form the chain index result.
5. The method for processing track intelligent sensor detection data as described in claim 1, characterized in that, Based on the corrected acoustic wave detection sequence, main propagation arrival chain, reflection propagation arrival chain, and chain index results, the main propagation observation relationship and the reflection propagation observation relationship are established, specifically including: The detection section is discretized along the track length direction to form multiple track discrete nodes; The main propagation arrival delay is determined based on the main propagation arrival chain, and the reflection propagation arrival delay is determined based on the reflection propagation arrival chain and the chain index result. For each orbital intelligent sensor and each orbital discrete node, a master propagation response term and a reflection propagation response term are constructed respectively, and an observation relationship is established for joint decomposition processing of the real anomaly source field and the structural reflection field.
6. The method for processing track intelligent sensor detection data as described in claim 5, characterized in that, The corrected acoustic detection sequence is subjected to joint decomposition processing to obtain the true anomaly source field and structure reflection field, specifically including: A joint solution objective is constructed based on the aforementioned observation relationship; Based on the joint solution objective, the real anomaly source field and the structural reflection field are updated alternately, so that the response that satisfies the main propagation response relationship and has spatial focusing characteristics is assigned to the real anomaly source field, and the response that satisfies the reflection propagation response relationship is assigned to the structural reflection field. When the joint solution objective reaches the convergence condition, the true anomaly source field and the structural reflection field are output.
7. The method for processing data detected by a smart track sensor as described in claim 1, characterized in that, The focusing anomaly index is calculated based on the real anomaly source field and the structural reflection field, specifically including: Calculate the real anomaly focusing energy corresponding to each discrete node of the track based on the real anomaly source field; The reflection ratio of each track discrete node is calculated based on the structural reflection field, and the focusing anomaly index is calculated by combining the multi-sensor consistency factor of each track discrete node. The focused anomaly indices corresponding to each discrete node of the track are organized in order of track position to form an anomaly index sequence for outputting track status detection results.
8. The method for processing data detected by a smart track sensor as described in claim 7, characterized in that, Based on the focused anomaly index, the output is the track state detection result based on end-to-end collaborative analysis between track smart sensors, specifically including: The median and median absolute deviation are calculated based on the abnormal index sequence, and a state determination threshold is generated. Track discrete nodes with a focusing anomaly index lower than the warning threshold are identified as normal nodes; track discrete nodes with a focusing anomaly index greater than or equal to the warning threshold but lower than the anomaly threshold are identified as warning nodes; and track discrete nodes with a focusing anomaly index greater than or equal to the anomaly threshold are identified as anomaly nodes. Adjacent abnormal nodes and adjacent early warning nodes are continuously merged to form corresponding abnormal sections or early warning sections, and the track status detection results are output.
9. A smart track sensor detection terminal, characterized in that, The terminal is used to execute the track intelligent sensor detection data processing method according to any one of claims 1-8, specifically including: processor; Memory, used to store computer programs; When the computer program is executed by the processor, it performs the following operations: Acquire acoustic wave detection sequences and position data of several track smart sensors deployed at different locations within the current detection window, and construct a candidate event cluster set based on the acoustic wave detection sequences; Based on the candidate event cluster set and location data, candidate relative time differences are generated between smart sensors on different orbits. The global time difference correction result is obtained to obtain the corrected acoustic wave detection sequence. Based on the corrected acoustic wave detection sequence and location data, the main propagation arrival chain and the reflection propagation arrival chain are constructed, and the corresponding chain index results are determined. Based on the corrected acoustic wave detection sequence, main propagation arrival chain, reflection propagation arrival chain, and chain index results, the main propagation observation relationship and reflection propagation observation relationship are established, and joint decomposition processing is performed on the corrected acoustic wave detection sequence to obtain the true anomaly source field and structure reflection field. The focusing anomaly index is calculated based on the real anomaly source field and the structural reflection field, and the track state detection result based on the end-to-end collaborative analysis between track intelligent sensors is output according to the focusing anomaly index.
10. A track detection system, characterized in that, The system includes: Several smart track sensors deployed at different locations are configured to collect acoustic wave detection sequences within the current detection window; Data transmission interface, used to transmit acoustic wave detection sequences and position data of each track smart sensor; The track intelligent sensor detection terminal as described in claim 9 is used to perform data processing on the acoustic detection sequence and the position data to output track state detection results; The result output unit is used to display or alarm the track status detection results.