A Machine Vision-Based Non-Destructive Testing Method and System for Track Fastener Condition

Abstract

This invention discloses a non-destructive testing method and system for track fasteners based on machine vision, belonging to the field of track inspection technology. By organizing the sleeper-beam nodes of the track network and creating inspection objects, a detection cloud center is built; associated sleeper-beam sets are extracted and the inspection objects and data clusters are labeled; a set of detection behavior features and anomaly feature sequences are generated; and the correlation of detection anomalies is analyzed using a sliding time window, achieving accurate inspection and maintenance of track fasteners. This invention solves the problems of low efficiency, high missed detection rate, and lack of correlation analysis in traditional inspection methods, enabling efficient and accurate inspection of track fastener conditions, reducing maintenance costs, and improving track operation safety. It is applicable to fastener inspection scenarios for various track lines, including high-speed rail and conventional rail.

Description

Technical Field

[0001] This invention relates to the field of track inspection technology, specifically to a machine vision-based non-destructive testing method and system for track fasteners. Background Technology

[0002] Track fasteners are critical components of the track system, and their condition directly affects the safety and stability of train operation. Currently, the inspection of track fasteners mainly relies on two methods: manual inspection and traditional machine vision inspection, but both have significant drawbacks.

[0003] Manual inspection method: Staff need to walk or ride along the track to inspect, which is not only labor-intensive and inefficient (the average daily inspection mileage is less than 20 kilometers), but also easily affected by the environment (such as rain, snow, night) and human factors (such as fatigue, visual error), resulting in a high rate of missed inspections and failing to meet the high efficiency inspection needs of large-scale track lines.

[0004] Traditional machine vision inspection methods, while achieving partial automation, mostly adopt a "single object independent inspection" mode. They fail to systematically organize the sleeper beam nodes in the track line, making it impossible to establish the correlation between adjacent fasteners, resulting in the inability to trace abnormal states. At the same time, they lack dynamic analysis of inspection data, only able to determine the "pass / fail" of a single inspection, unable to capture the evolution trend of abnormal states, and the inspection results cannot directly guide targeted maintenance, requiring secondary investigation by staff, which increases maintenance costs and time.

[0005] With the continuous increase in the mileage of high-speed rail and conventional rail lines, the demand for "efficiency, accuracy and relevance" in the inspection of track fasteners is becoming increasingly urgent, and existing inspection technologies can no longer meet the actual needs of track operation and maintenance. Summary of the Invention

[0006] The purpose of this invention is to provide a machine vision-based non-destructive testing method and system for track fasteners to solve the problems mentioned in the background art.

[0007] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0008] This machine vision-based non-destructive testing system for track fasteners includes: a data storage center, a machine vision non-destructive testing cloud center, and modules for managing sleeper beam nodes and inspection objects, as well as modules for associating sleeper beam sets and data cluster marking, generating inspection behavior features and anomaly sequences, and analyzing and maintaining inspection anomaly associations. These modules work together to perform non-destructive testing on the condition of track fasteners. The data storage center stores sleeper beam node information, inspection object data, and inspection logs, while the machine vision non-destructive testing cloud center records inspection behavior and supports data interaction.

[0009] The sleeper beam node and inspection object management module is used to sort out the sleeper beam nodes in the track network, treat a single sleeper beam as an independent sleeper beam node, including main rail sleeper beams and sub-rail sleeper beams, create inspection objects corresponding to the track fastener modules between adjacent sleeper beam nodes, and build a machine vision non-destructive testing cloud center to record and store the inspection logs of the inspection objects, and allocate inspection data clusters for each inspection object in the data storage center to record the inspection data using the inspection item type;

[0010] The detection anomaly correlation analysis and maintenance module is used to extract adjacent sub-rail sleeper beam nodes in the track network with the main rail sleeper beam node as the center, forming a set of associated rail sleeper beams; to mark the detection objects between the main rail sleeper beam node and the sub-rail sleeper beam node, and to clarify the detection data cluster corresponding to each detection object, which is used to record the abnormal detection behavior of the detection objects under different detection batches;

[0011] The detection behavior feature and abnormal sequence generation module is used to extract the abnormal state set of detection behavior from the detection data cluster corresponding to the detection object in the same detection batch to form a detection behavior mark; take the associated track sleeper beam set as the detection behavior feature sample, count all detection behavior marks in the same detection batch to generate a detection behavior feature set; and generate a detection abnormal feature sequence of the detection object according to the order of the detection batch by evaluating the detection abnormality coefficient of the detection object between adjacent detection batches.

[0012] The detection anomaly correlation analysis and maintenance module is used to set the inspection cycle range, construct a two-dimensional coordinate system of detection anomaly features with the detection batch index as the horizontal axis and the detection anomaly coefficient as the vertical axis, map each anomaly coefficient in the detection anomaly feature sequence to a coordinate point in the coordinate system, construct a sliding time window in combination with the inspection cycle range, analyze the detection anomaly feature sequence segments of different detection objects within the sliding time window to quantify the correlation degree of detection anomalies between detection objects, and instruct staff to perform manual correlation maintenance on the corresponding detection objects.

[0013] As a preferred embodiment of the present invention, the sleeper beam node and detection object management module includes a sleeper beam node sorting unit, a detection object creation unit, a cloud center construction unit, and a data cluster allocation unit;

[0014] The sleeper beam node sorting unit is used to sort out the sleeper beams in the track network, divide a single sleeper beam into an independent sleeper beam node, and distinguish between the main rail sleeper beam node and the sub-rail sleeper beam node;

[0015] The detection object creation unit is used to create detection objects between adjacent sleeper beam nodes, ensuring that each detection object corresponds one-to-one with a track fastener module;

[0016] The cloud center building unit is used to build a machine vision non-destructive testing cloud center, configure the inspection log recording function, and support the storage and retrieval of different inspection item types;

[0017] The data cluster allocation unit is used to allocate a dedicated data cluster to each detection object within the data storage center, ensuring that the detection behavior of the detection object is completely recorded through the data cluster.

[0018] As a preferred embodiment of the present invention, the associated bolster set and data cluster marking module includes an associated bolster set extraction unit, a detection object marking unit, and a data cluster marking unit;

[0019] The associated sleeper beam set extraction unit, with the main rail sleeper beam node as the center, filters and extracts adjacent sub-rail sleeper beam nodes in the track network to form an associated track sleeper beam set;

[0020] The detection object marking unit is used to uniquely identify the detection object between the main rail sleeper beam node and the sub-rail sleeper beam node, ensuring accurate positioning of the detection object;

[0021] The data cluster labeling unit is used to associate and label each detection object with the corresponding data cluster, clarifying the correspondence between the detection batch and the abnormal status of the detection behavior recorded by the data cluster.

[0022] As a preferred embodiment of the present invention, the detection behavior feature and abnormal sequence generation module includes an abnormal state set extraction unit, a detection behavior feature set generation unit, an abnormal coefficient calculation unit, and an abnormal feature sequence generation unit.

[0023] The abnormal state set extraction unit is used to extract the abnormal state set of detection behavior from the data cluster corresponding to the detection object in the same detection batch and form a detection behavior label.

[0024] The detection behavior feature set generation unit uses the associated track sleeper beam set as feature samples to statistically analyze all detection behavior markers within the same detection batch and generate a detection behavior feature set.

[0025] The anomaly coefficient calculation unit is used to calculate the detection anomaly coefficient of the same detection object between two adjacent detection batches, and to quantify the changing trend of detection anomalies.

[0026] The abnormal feature sequence generation unit integrates the detection abnormality coefficients of the detected objects into a detection abnormality feature sequence according to the order of the detection batches from first to last, which intuitively reflects the abnormal evolution process of the detected objects.

[0027] As a preferred embodiment of the present invention, the detection anomaly correlation analysis and maintenance module includes an inspection cycle setting unit, a coordinate system construction unit, a sliding time window generation unit, a correlation degree calculation unit, and a manual maintenance instruction unit.

[0028] The inspection cycle setting unit is used to set a reasonable inspection cycle range according to track maintenance needs, and to clarify the start and end intervals of the inspection batch;

[0029] The coordinate system construction unit constructs a two-dimensional coordinate system for detecting anomalies, with the detection batch index as the horizontal axis and the detection anomaly coefficient as the vertical axis, to achieve a visual mapping of the anomaly coefficient.

[0030] The sliding time window generation unit determines the minimum and maximum values ​​of the abnormal coefficients in the detected abnormal feature sequence, and constructs a sliding time window in combination with the inspection cycle range to limit the spatiotemporal range of abnormal analysis.

[0031] The correlation calculation unit is used to analyze the abnormal feature sequence segments of different detected objects within the sliding time window and calculate the correlation degree of detected anomalies between the detected objects.

[0032] The manual maintenance instruction unit has a preset threshold for detecting abnormal correlation. When the correlation between detected objects reaches or exceeds the threshold, it automatically sends a manual correlation maintenance instruction signal to the staff.

[0033] A machine vision-based non-destructive testing method for track fasteners, comprising the following steps:

[0034] Step S1: Organize the sleeper beam nodes in the track network, treat each sleeper beam as an independent sleeper beam node, including main rail sleeper beams and sub-rail sleeper beams, create inspection objects corresponding to the track fastener modules between adjacent sleeper beam nodes; and build a machine vision non-destructive testing cloud center to record and store the inspection logs of the inspection objects, and allocate inspection data clusters for each inspection object in the data storage center to record the inspection data using the inspection item type;

[0035] Step S2: Taking the main rail sleeper beam node as the center, extract the adjacent sub-rail sleeper beam nodes in the track network to form an associated rail sleeper beam set; mark the detection objects between the main rail sleeper beam node and the sub-rail sleeper beam node, and clarify the detection data cluster corresponding to each detection object, which is used to record the abnormal detection behavior of the detection objects under different detection batches;

[0036] Step S3: In the same inspection batch, extract the abnormal state set of inspection behavior from the inspection data cluster corresponding to the inspection object to form an inspection behavior label; take the associated track sleeper beam set as the inspection behavior feature sample, count all inspection behavior labels in the same inspection batch to generate an inspection behavior feature set; by evaluating the inspection abnormality coefficient of the inspection object between adjacent inspection batches, generate the inspection abnormality feature sequence of the inspection object in the order of inspection batches.

[0037] Step S4: Set the inspection cycle range, construct a two-dimensional coordinate system of detection anomaly features with the detection batch index as the horizontal axis and the detection anomaly coefficient as the vertical axis, map each anomaly coefficient in the detection anomaly feature sequence to a coordinate point in the coordinate system, construct a sliding time window in combination with the inspection cycle range, analyze the detection anomaly feature sequence segments of different detection objects within the sliding time window to quantify the correlation between detection anomalies between detection objects, and use this to instruct staff to manually maintain the correlation of the corresponding detection objects.

[0038] As a preferred embodiment of the present invention, the specific implementation process of step S1 includes:

[0039] Based on the track network, each sleeper beam in the track network is identified. The data storage center treats each sleeper beam as a sleeper beam node. The sleeper beam includes the main rail sleeper beam and the sub-rail sleeper beam. A detection object is created between two adjacent sleeper beam nodes, and one detection object corresponds to a track fastener module.

[0040] A machine vision nondestructive testing cloud center is established to record and store the inspection logs of the inspection objects. The inspection logs record different inspection item types, and inspection data clusters are allocated to the inspection objects in the data storage center to record the inspection behavior using the inspection item types.

[0041] As a preferred embodiment of the present invention, the specific implementation process of step S2 includes:

[0042] Taking the k-th main sleeper beam node as the center node, extract the sub-sleeper beam nodes adjacent to the k-th main sleeper beam node in the track network to form an associated track sleeper beam set, denoted as . ,in, This represents the l-th sleeper beam node. This represents the total number of sub-rail sleeper beam nodes adjacent to the k-th main rail sleeper beam node;

[0043] At the k-th main rail sleeper node Joint with sleeper beam The detection objects created between them are denoted as and the detection object The assigned detection data clusters are denoted as ,in, This indicates the inspection targets during the m-th batch of inspections of the rail transit network. The set of abnormal detection behaviors, and , This indicates the test objects in the m-th batch. The nth type of detection item that provides feedback on abnormal detection, where M represents the batch number and N represents the type number of the detection item.

[0044] As a preferred embodiment of the present invention, the specific implementation process of step S3 includes:

[0045] During the same m-th batch of testing, in the test data cluster Extract the abnormal state set of the detected behavior To constitute a detection behavior marker, denoted as ,and ,in, The logical AND symbol is used to enable the detection of data clusters. With the set of abnormal detection behaviors Generate association logic;

[0046] Related track sleeper beam set To detect behavioral feature samples, all detection behavioral markers generated in the same m-th batch of detection are statistically analyzed to generate a detection behavioral feature set. And quantify the detection objects between the m-th batch and the (m+1)-th batch. Detection anomaly coefficient In the formula, Indicates in the detection data cluster Extract the abnormal state set of the detected behavior The detection behavior is marked by the constructed behavior. Indicates the detection behavior marker The total number of detection item types included. Indicates the detection behavior marker With detection behavior markers The total number of detection item types included after the intersection operation;

[0047] The anomaly coefficients are analyzed in chronological order of the testing batches to generate the testing objects. The detected abnormal feature sequence is denoted as .

[0048] As a preferred embodiment of the present invention, the specific implementation process of step S4 includes:

[0049] Set inspection cycle range ,in, and The numbers are, in order, the batch number of the test, and ;

[0050] Establish a two-dimensional coordinate system for detecting abnormal features, and use an index... The x-axis is used as the independent variable to detect outlier coefficients. The ordinate is the dependent variable, and the coordinate points are obtained. And detect abnormal feature sequences Each detected anomaly coefficient is mapped to a coordinate point;

[0051] To detect anomalous feature sequences As the target for anomaly tracing, in detecting anomalous feature sequences The minimum and maximum values ​​of the anomaly detection coefficient are selected from the data, and are denoted as follows: and ;

[0052] Within the inspection cycle range The horizontal scale of the sliding time window is used to detect the range of anomaly coefficients formed by the minimum and maximum values ​​of the anomaly coefficients. The vertical scale of the sliding time window forms a sequence for detecting anomalous features. The sliding time window for anomaly tracing targets is denoted as... ;

[0053] In the sliding time window Each detected abnormal feature sequence fragment within will originate from the detected abnormal feature sequence. The detected abnormal feature sequence fragment is denoted as ;

[0054] If the detection object The corresponding generated anomaly feature sequence There are abnormal feature sequence fragments in the detection. In the sliding time window Inside, the assessment of the test object With the test object Within the inspection cycle Internal detection anomaly correlation In the formula, and The nodes are, in order, the main rail sleeper beam node and the branch rail sleeper beam node, and... , = or ≠ , and In the sliding time window, respectively Internal detection of abnormal feature sequence fragments The minimum and maximum values ​​of the anomaly detection coefficient in the data;

[0055] A preset threshold for detecting abnormal correlations is set; if an abnormal correlation is detected... If the correlation value is greater than or equal to the detection anomaly correlation threshold, it indicates that the problem is within the inspection cycle range. Internal detection object and the objects of detection If an abnormal correlation exists between the detected samples, staff will be instructed to investigate the samples. and the objects of detection Perform manual association maintenance.

[0056] Compared with the prior art, the beneficial effects achieved by the present invention are:

[0057] The sleeper beams in the track network are divided into independent nodes (main track sleeper beams + sub-track sleeper beams). Detection objects corresponding to fastener modules are created between adjacent nodes to achieve precise positioning of the detection objects. At the same time, a cloud center and data cluster are built to ensure that the detection behavior (including abnormal states) of each detection object can be recorded throughout the process. Through node-based management and cloud center storage, the detection process is automated and the data is centrally processed.

[0058] The associated sleeper beam set is extracted with the main rail sleeper beam node as the core. Fastener inspection objects in the same associated set are regarded as an "inspection unit" to avoid isolated inspection. By evaluating the inspection anomaly coefficient of adjacent batches, the abnormal changes of the inspection objects are quantified, and an anomaly feature sequence is generated to realize the transformation of "inspection data - anomaly trend". Through the analysis of associated sleeper beam set and anomaly coefficient, the anomaly feature sequence can capture progressive anomalies and realize "early detection and early maintenance".

[0059] By constructing a sliding time window based on the inspection cycle, abnormal feature segments of different inspection objects are analyzed within the time window, and the correlation is calculated. If the abnormal trends of two inspection objects are highly correlated (correlation meets the standard), it indicates that their abnormalities may be caused by the same reason (such as track foundation settlement). This guides staff to carry out targeted maintenance, accurately locate the root cause of the abnormality, and eliminates the need for staff to conduct a full-line inspection, avoiding blind inspection. Maintenance is only required for inspection objects with a high correlation, thus reducing maintenance costs. Attached Figure Description

[0060] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used together with the embodiments of the invention to explain the invention and do not constitute a limitation thereof.

[0061] Figure 1 This is a schematic diagram illustrating the steps of the non-destructive testing method for track fasteners based on machine vision according to the present invention. Detailed Implementation

[0062] 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 scope of protection of the present invention.

[0063] In this first embodiment, a machine vision-based non-destructive testing system for track fasteners is provided. This system includes: a data storage center, a machine vision non-destructive testing cloud center, and modules for managing sleeper beam nodes and inspection objects, as well as modules for associating sleeper beam sets and data cluster marking, generating inspection behavior features and abnormal sequences, and analyzing and maintaining inspection anomaly associations. These modules work together to achieve non-destructive testing of the track fasteners. The data storage center stores sleeper beam node information, inspection object data, and inspection logs, while the machine vision non-destructive testing cloud center records inspection behavior and supports data interaction.

[0064] The sleeper beam node and inspection object management module is used to organize the sleeper beam nodes in the track network, treat a single sleeper beam as an independent sleeper beam node, including main rail sleeper beams and sub-rail sleeper beams, and create inspection objects corresponding to the track fastener modules between adjacent sleeper beam nodes; and build a machine vision non-destructive testing cloud center to record and store the inspection logs of the inspection objects, and allocate inspection data clusters for each inspection object in the data storage center to record the inspection data using the inspection item type;

[0065] The sleeper beam node and inspection object management module includes a sleeper beam node sorting unit, an inspection object creation unit, a cloud center construction unit, and a data cluster allocation unit.

[0066] The sleeper beam node sorting unit is used to sort out the sleeper beams in the track network, divide a single sleeper beam into an independent sleeper beam node, and distinguish between the main rail sleeper beam node and the sub-rail sleeper beam node;

[0067] The detection object creation unit is used to create detection objects between adjacent sleeper beam nodes, ensuring that each detection object corresponds one-to-one with a track fastener module;

[0068] The cloud center building unit is used to build a machine vision non-destructive testing cloud center, configure the inspection log recording function, and support the storage and retrieval of different inspection item types;

[0069] The data cluster allocation unit is used to allocate a dedicated data cluster to each detection object within the data storage center, ensuring that the detection behavior of the detection object is completely recorded through the data cluster;

[0070] The detection anomaly correlation analysis and maintenance module is used to extract adjacent sub-rail sleeper beam nodes in the track network with the main rail sleeper beam node as the center, forming a set of associated rail sleeper beams; it marks the detection objects between the main rail sleeper beam node and the sub-rail sleeper beam node, and clarifies the detection data cluster corresponding to each detection object, which is used to record the abnormal detection behavior of the detection objects under different detection batches;

[0071] The associated bolster set and data cluster labeling module includes an associated bolster set extraction unit, a detection object labeling unit, and a data cluster labeling unit.

[0072] The associated sleeper beam set extraction unit, with the main rail sleeper beam node as the center, filters and extracts adjacent sub-rail sleeper beam nodes in the track network to form an associated track sleeper beam set;

[0073] The detection object marking unit is used to uniquely identify the detection object between the main rail sleeper beam node and the sub-rail sleeper beam node, ensuring accurate positioning of the detection object;

[0074] The data cluster labeling unit is used to associate and label each detection object with the corresponding data cluster, clarifying the correspondence between the detection batch and the abnormal status of the detection behavior recorded by the data cluster;

[0075] The detection behavior feature and abnormal sequence generation module is used to extract the abnormal state set of detection behavior from the detection data cluster corresponding to the detection object in the same detection batch to form a detection behavior label; using the associated track sleeper beam set as the detection behavior feature sample, it statistically analyzes all detection behavior labels in the same detection batch to generate a detection behavior feature set; and by evaluating the detection abnormality coefficient of the detection object between adjacent detection batches, it generates a detection abnormality feature sequence of the detection object in the order of detection batches.

[0076] The detection behavior feature and abnormal sequence generation module includes an abnormal state set extraction unit, a detection behavior feature set generation unit, an abnormal coefficient calculation unit, and an abnormal feature sequence generation unit.

[0077] The abnormal state set extraction unit is used to extract the abnormal state set of detection behavior from the data cluster corresponding to the detection object in the same detection batch and form a detection behavior label.

[0078] The detection behavior feature set generation unit uses the associated track sleeper beam set as feature samples to statistically analyze all detection behavior markers within the same detection batch and generate a detection behavior feature set.

[0079] The anomaly coefficient calculation unit is used to calculate the detection anomaly coefficient of the same detection object between two adjacent detection batches, and to quantify the changing trend of detection anomalies.

[0080] The abnormal feature sequence generation unit integrates the detection abnormality coefficients of the detected objects into a detection abnormality feature sequence according to the order of the detection batches from first to last, which intuitively reflects the abnormal evolution process of the detected objects;

[0081] The detection anomaly correlation analysis and maintenance module is used to set the inspection cycle range, construct a two-dimensional coordinate system of detection anomaly features with the detection batch index as the horizontal axis and the detection anomaly coefficient as the vertical axis, map each anomaly coefficient in the detection anomaly feature sequence to the coordinate point in the coordinate system, construct a sliding time window in combination with the inspection cycle range, analyze the detection anomaly feature sequence segments of different detection objects within the sliding time window, so as to quantify the correlation degree of detection anomalies between detection objects, and instruct staff to perform manual correlation maintenance on the corresponding detection objects;

[0082] The detection anomaly correlation analysis and maintenance module includes an inspection cycle setting unit, a coordinate system construction unit, a sliding time window generation unit, a correlation degree calculation unit, and a manual maintenance instruction unit.

[0083] The inspection cycle setting unit is used to set a reasonable inspection cycle range according to track maintenance needs, and to clarify the start and end intervals of the inspection batch;

[0084] The coordinate system construction unit constructs a two-dimensional coordinate system for detecting anomalies, with the detection batch index as the horizontal axis and the detection anomaly coefficient as the vertical axis, to achieve a visual mapping of the anomaly coefficient.

[0085] The sliding time window generation unit determines the minimum and maximum values ​​of the abnormal coefficients in the detected abnormal feature sequence, and constructs a sliding time window in combination with the inspection cycle range to limit the spatiotemporal range of abnormal analysis.

[0086] The correlation calculation unit is used to analyze the abnormal feature sequence segments of different detected objects within the sliding time window and calculate the correlation degree of detected anomalies between the detected objects.

[0087] The manual maintenance instruction unit has a preset threshold for detecting abnormal correlation. When the correlation between detected objects reaches or exceeds the threshold, it automatically sends a manual correlation maintenance instruction signal to the staff.

[0088] Please see Figure 1 In this second embodiment, a non-destructive testing method for track fastener status based on machine vision is provided to be applicable to the first embodiment above. This embodiment takes a high-speed railway line (design speed 350km / h, track type ballastless track, line length 50km) as the testing scenario. The line has been in operation for 5 years and includes 20,000 main track sleepers and 15,000 branch track sleepers. The track fastener module between each adjacent sleeper node consists of 6 groups (including bolts, spring clips, and washers). It is necessary to detect anomalies such as "loosening, missing, and rusting" of fasteners.

[0089] The method includes the following steps:

[0090] Step S1: Organize the sleeper beam nodes in the track network, treat each sleeper beam as an independent sleeper beam node, including main rail sleeper beams and sub-rail sleeper beams, create inspection objects corresponding to the track fastener modules between adjacent sleeper beam nodes; and build a machine vision non-destructive testing cloud center to record and store the inspection logs of the inspection objects, and allocate inspection data clusters for each inspection object in the data storage center to record the inspection data using the inspection item type;

[0091] For example, based on the track network, each sleeper beam in the track network is identified. The data storage center treats a sleeper beam as a sleeper beam node. The sleeper beam includes the main rail sleeper beam and the sub-rail sleeper beam. A detection object is created between two adjacent sleeper beam nodes, and one detection object corresponds to a track fastener module.

[0092] A machine vision nondestructive testing cloud center is built to record and store the inspection logs of the inspection objects. The inspection logs record different inspection item types, and the inspection data clusters are allocated according to the inspection objects in the data storage center to record the inspection behavior using the inspection item types.

[0093] For example, the main rail sleeper beam spacing is 2.5 meters, and the sub-rail sleeper beam spacing is 1 meter; the inspection objects are named according to "main rail sleeper beam number - sub-rail sleeper beam number - fastener group number" (such as "Z1-F1-3" indicating the 3rd group of fasteners between the 1st main rail sleeper beam and the 1st sub-rail sleeper beam); the machine vision non-destructive testing cloud center adopts an edge computing architecture, the inspection log is updated once per minute, and the data cluster storage capacity is 1 year of inspection data for each inspection object (including 3 types of inspection items: "loose, missing, and corroded").

[0094] Step S2: Taking the main rail sleeper beam node as the center, extract the adjacent sub-rail sleeper beam nodes in the track network to form an associated rail sleeper beam set; mark the detection objects between the main rail sleeper beam node and the sub-rail sleeper beam node, and clarify the detection data cluster corresponding to each detection object, which is used to record the abnormal detection behavior of the detection objects under different detection batches;

[0095] For example, taking the k-th main sleeper beam node as the center node, the sub-sleeper beam nodes adjacent to the k-th main sleeper beam node are extracted from the track network to form an associated track sleeper beam set, denoted as . ,in, This represents the l-th sleeper beam node. This represents the total number of sub-rail sleeper beam nodes adjacent to the k-th main rail sleeper beam node;

[0096] At the k-th main rail sleeper node Joint with sleeper beam The detection objects created between them are denoted as and the detection object The assigned detection data clusters are denoted as ,in, This indicates the inspection targets during the m-th batch of inspections of the rail transit network. The set of abnormal detection behaviors, and , This indicates the test objects in the m-th batch. The nth type of detection item that provides feedback on abnormal detection, where M represents the batch number and N represents the type number of the detection item;

[0097] For example, taking a single main track sleeper beam node as the center, extract three sub-track sleeper beam nodes on each side to form an associated track sleeper beam set (each associated set contains 1 main track sleeper beam + 6 sub-track sleeper beams); the detection object is marked with a QR code, and the association response time between the data cluster and the detection object is less than 50ms.

[0098] Step S3: In the same inspection batch, extract the abnormal state set of inspection behavior from the inspection data cluster corresponding to the inspection object to form an inspection behavior label; take the associated track sleeper beam set as the inspection behavior feature sample, count all inspection behavior labels in the same inspection batch to generate an inspection behavior feature set; by evaluating the inspection abnormality coefficient of the inspection object between adjacent inspection batches, generate the inspection abnormality feature sequence of the inspection object in the order of inspection batches.

[0099] For example, during the same m-th batch of testing, in the test data cluster Extract the abnormal state set of the detected behavior To constitute a detection behavior marker, denoted as ,and ,in, The logical AND symbol is used to enable the detection of data clusters. With the set of abnormal detection behaviors Generate association logic;

[0100] Related track sleeper beam set To detect behavioral feature samples, all detection behavioral markers generated in the same m-th batch of detection are statistically analyzed to generate a detection behavioral feature set. And quantify the detection objects between the m-th batch and the (m+1)-th batch. Detection anomaly coefficient In the formula, Indicates in the detection data cluster Extract the abnormal state set of the detected behavior The detection behavior is marked by the constructed behavior. Indicates the detection behavior marker The total number of detection item types included. Indicates the detection behavior marker With detection behavior markers The total number of detection item types included after the intersection operation;

[0101] The anomaly coefficients are analyzed in chronological order of the testing batches to generate the testing objects. The detected abnormal feature sequence is denoted as ;

[0102] For example, if the testing batch is 1 time / day, and a total of 30 days of testing are conducted (i.e., 30 batches), in batches 10-15, the abnormal item change of a certain test object (Z500-F2-4) is "no abnormality → rust → rust + loosening", and the abnormality coefficient is 0 (batch 10-11), 1 (batch 11-12, common abnormal item is "rust"), and 0.8 (batch 12-13, common abnormal item is "rust"). The generated abnormal feature sequence accurately reflects the evolution process of the fastener from "normal → slight abnormality → moderate abnormality".

[0103] Step S4: Set the inspection cycle range, construct a two-dimensional coordinate system of detection anomaly features with the detection batch index as the horizontal axis and the detection anomaly coefficient as the vertical axis, map each anomaly coefficient in the detection anomaly feature sequence to the coordinate point in the coordinate system, construct a sliding time window in combination with the inspection cycle range, analyze the detection anomaly feature sequence segments of different detection objects within the sliding time window, so as to quantify the correlation between detection anomalies between detection objects, and use it to instruct staff to manually maintain the correlation of the corresponding detection objects;

[0104] For example, setting the inspection cycle range ,in, and The numbers are, in order, the batch number of the test, and ;

[0105] Establish a two-dimensional coordinate system for detecting abnormal features, and use an index... The x-axis is used as the independent variable to detect outlier coefficients. The ordinate is the dependent variable, and the coordinate points are obtained. And detect abnormal feature sequences Each detected anomaly coefficient is mapped to a coordinate point;

[0106] To detect anomalous feature sequences As the target for anomaly tracing, in detecting anomalous feature sequences The minimum and maximum values ​​of the anomaly detection coefficient are selected from the data, and are denoted as follows: and ;

[0107] Within the inspection cycle range The horizontal scale of the sliding time window is used to detect the range of anomaly coefficients formed by the minimum and maximum values ​​of the anomaly coefficients. The vertical scale of the sliding time window forms a sequence for detecting anomalous features. The sliding time window for anomaly tracing targets is denoted as... ;

[0108] In the sliding time window Each detected abnormal feature sequence fragment within will originate from the detected abnormal feature sequence. The detected abnormal feature sequence fragment is denoted as ;

[0109] If the detection object The corresponding generated anomaly feature sequence There are abnormal feature sequence fragments in the detection. In the sliding time window Inside, the assessment of the test object With the test object Within the inspection cycle Internal detection anomaly correlation In the formula, and The nodes are, in order, the main rail sleeper beam node and the branch rail sleeper beam node, and... , = or ≠ , and In the sliding time window, respectively Internal detection of abnormal feature sequence fragments The minimum and maximum values ​​of the anomaly detection coefficient in the data;

[0110] A preset threshold for detecting abnormal correlations is set; if an abnormal correlation is detected... If the correlation value is greater than or equal to the detection anomaly correlation threshold, it indicates that the problem is within the inspection cycle range. Internal detection object and the objects of detection If an abnormal correlation exists between the detected samples, staff will be instructed to investigate the samples. and the objects of detection Perform manual association maintenance;

[0111] For example, the inspection cycle ranges from 10 to 20 batches (a total of 10 days); the horizontal scale of the sliding time window is 10 batches, and the vertical scale is 0-1 (the range of anomaly coefficient values); the threshold for the correlation of detection anomalies is set to 0.6; within the sliding time window, the correlation between the detected objects Z500-F2-4 and Z500-F3-2 is 0.75 (both have anomaly trends of "rust → rust + loosening"), exceeding the threshold of 0.6; on-site inspection by staff revealed that there was slight settlement in the track foundation in this area, resulting in uneven stress on adjacent fasteners, which in turn led to the correlation anomaly; after maintenance, the anomaly coefficients of the two detected objects dropped to 0 in subsequent batches.

[0112] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0113] Finally, it should be noted that the above descriptions are merely preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for non-destructive testing of the state of rail fastenings based on machine vision, characterized in that, The method comprises the following steps: Step S1: combing the sleeper beam nodes in the track line network, taking a single sleeper beam as an independent sleeper beam node, including main track sleeper beams and split track sleeper beams, creating detection objects corresponding to track fastener modules between adjacent sleeper beam nodes; and building a machine vision non-destructive testing cloud center for recording and storing detection logs of detection objects, and assigning detection data clusters for each detection object in the data storage center for recording detection data using detection item types for detection; Step S2: taking the main track sleeper beam node as the center, extracting adjacent split track sleeper beam nodes in the track line network to form an associated track sleeper beam set; marking the detection objects between the main track sleeper beam node and the split track sleeper beam node, and clearly defining the detection data clusters corresponding to each detection object for recording detection behavior abnormalities of the detection objects under different detection batches; Step S3: In the same detection batch, extract the detection behavior abnormal state set from the detection data cluster corresponding to the detection object to form a detection behavior mark; take the associated track sleeper beam set as a detection behavior feature sample, count all detection behavior marks in the same detection batch, and generate a detection behavior feature set; by evaluating the detection abnormality coefficient of the detection object between adjacent detection batches, generate a detection abnormality feature sequence of the detection object in the order of detection batches; Step S4: set the inspection cycle range, build a detection abnormality feature two-dimensional coordinate system with detection batch index as horizontal coordinate and detection abnormality coefficient as vertical coordinate, map each abnormality coefficient in the detection abnormality feature sequence to a coordinate point in the coordinate system, combine the inspection cycle range to build a sliding time window, analyze the detection abnormality feature sequence fragments of different detection objects in the sliding time window, and quantify the detection abnormality correlation degree between detection objects to indicate the staff to manually associate and maintain the corresponding detection objects; The specific implementation process of step S3 comprises: At the same time, the detection data cluster is extracted from the detection behavior abnormal state set to constitute a detection behavior label, denoted as , and wherein, represents a logical and symbol, used to make the detection data cluster associated with the detection behavior abnormal state set produce a correlation logic; Associated track sleeper beam set To detect the behavior characteristic sample, the total detection behavior mark constituted in the same m batch detection is counted to generate the detection behavior characteristic set , and the detection object between the m batch detection and the m+1 batch detection is quantified The detection abnormal coefficient of the detection object , wherein, The detection object is constituted by the first Main track sleeper beam node And the first Sub-track sleeper beam node , The detection behavior abnormal state set is extracted in the detection data cluster To constitute the detection behavior mark , The detection item type total number contained in the detection behavior mark , The detection item type total number contained after the intersection operation of the detection behavior mark And the detection behavior mark ; The detection abnormal coefficient is arranged in the order of detection batch from early to late to generate a detection abnormal feature sequence of the detection object, denoted as . .

2. The method of claim 1, wherein, The specific implementation process of step S1 comprises: Based on the track line network, each sleeper beam in the track line network is combed, a data storage center regards a sleeper beam as a sleeper beam node, the sleeper beam includes main track sleeper beams and split track sleeper beams, and a detection object is created between two adjacent sleeper beam nodes, wherein one detection object corresponds to one track fastener module; Build a machine vision non-destructive testing cloud center to record and store detection logs of detection objects, which record different detection item types, and assign detection data clusters to detection objects in the data storage center to record detection behaviors using detection item types for detection.

3. The method of claim 2, wherein, The specific implementation process of step S2 comprises: Taking the kth main rail sleeper beam node as a center node, the sub-rail sleeper beam nodes adjacent to the kth main rail sleeper beam node are extracted in the track network to form an associated track sleeper beam set, denoted as wherein, the lth sub-rail sleeper beam node is denoted as denotes the total number of the sub-rail sleeper beam nodes adjacent to the kth main rail sleeper beam node. The detection object created between the kth main rail sleeper beam node and the sub rail sleeper beam node is denoted as , and the detection data cluster assigned to the detection object is denoted as , wherein represents the abnormal state set of the detection behavior of the detection object when the mth batch of detection is performed on the track network, and , represents the nth detection item type of the detection object that feeds back the detection abnormality when the mth batch of detection is performed, M represents the detection batch number, and N represents the detection item type number.

4. The method of claim 3, wherein, The specific implementation process of step S4 comprises: Setting a range of patrol cycles wherein, and are, in sequence, respectively, a detection batch number, and ; Establish a two-dimensional coordinate system for detecting abnormal features, and use an index... The x-axis is used as the independent variable to detect outlier coefficients. The ordinate is the dependent variable, and the coordinate points are obtained. And detect abnormal feature sequences Each detected anomaly coefficient is mapped to a coordinate point; Detecting abnormal feature sequence For abnormal source target, in detecting abnormal feature sequence The minimum and maximum values of the detecting abnormal coefficient are selected, and are sequentially recorded as And ; with the range of the patrol cycle as the lateral dimension of the sliding time window, with the range of the minimum and maximum of the anomaly coefficient as the anomaly coefficient range as the longitudinal dimension of the sliding time window, with the anomaly feature sequence as the detection target as the sliding time window of the anomaly traceability target, denoted as ; In the sliding time window , each detection abnormality feature sequence segment, the detection abnormality feature sequence segment derived from the detection abnormality feature sequence is recorded as ; If the detection object The corresponding generated anomaly feature sequence There are abnormal feature sequence fragments in the detection. In the sliding time window Inside, the assessment of the test object With the test object Within the inspection cycle Internal detection anomaly correlation In the formula, and The nodes are, in order, the main rail sleeper beam node and the branch rail sleeper beam node, and... , and In the sliding time window, respectively Internal detection of abnormal feature sequence fragments The minimum and maximum values ​​of the anomaly detection coefficient in the data; A preset detection anomaly correlation degree threshold is set. If the detection anomaly correlation degree is greater than or equal to the detection anomaly correlation degree threshold, it indicates that there is a detection anomaly correlation between the detection object and the detection object within the patrol cycle range, and the worker is instructed to manually maintain the correlation between the detection object and the detection object. ​​​​​​ 5. A machine vision based non-destructive testing system for rail fastener condition, performing the machine vision based non-destructive testing method for rail fastener condition according to any one of claims 1 to 4, characterized in that, The system comprises: a data storage center, a machine vision non-destructive testing cloud center, and sleeper beam node and detection object management modules, associated sleeper beam set and data cluster marking modules, detection behavior feature and abnormal sequence generation modules, and detection abnormality correlation analysis and maintenance modules; each module cooperates to realize non-destructive detection of the state of the track fastener, wherein the data storage center is used to store sleeper beam node information, detection object data and detection logs, and the machine vision non-destructive testing cloud center is used to record detection behaviors and support data interaction; The sleeper beam node and the detection object management module are used to sort out sleeper beam nodes in a track network, take a single sleeper beam as an independent sleeper beam node, include main track sleeper beams and split track sleeper beams, create detection objects corresponding to track fastener modules between adjacent sleeper beam nodes, and build a machine vision nondestructive testing cloud center for recording and storing detection logs of the detection objects and allocating detection data clusters for recording detection using detection item types for each detection object in a data storage center; The detection anomaly correlation analysis and maintenance module is used to extract adjacent split track sleeper beam nodes in the track network as a center of the main track sleeper beam node to form a set of associated track sleeper beams, mark detection objects between the main track sleeper beam node and the split track sleeper beam node, and clearly indicate detection data clusters corresponding to each detection object for recording detection behavior anomalies of the detection objects under different detection batches; The detection behavior feature and anomaly sequence generation module is used to extract a set of detection behavior anomaly states from the detection data cluster corresponding to the detection object in the same detection batch to form a detection behavior mark, take the set of associated track sleeper beams as a detection behavior feature sample, count all detection behavior marks in the same detection batch to generate a set of detection behavior features, and generate a detection anomaly feature sequence of the detection object according to the detection batch order by evaluating detection anomaly coefficients of the detection object between adjacent detection batches; The detection anomaly correlation analysis and maintenance module is used to set a range of inspection cycles, build a two-dimensional coordinate system of detection anomaly features with a detection batch index as an abscissa and a detection anomaly coefficient as an ordinate, map each anomaly coefficient in the detection anomaly feature sequence to a coordinate point in the coordinate system, build a sliding time window in combination with the range of inspection cycles, analyze a detection anomaly feature sequence fragment of each detection object in the sliding time window, and quantify a detection anomaly correlation degree between the detection objects for indicating a worker to manually maintain a corresponding detection object.

Images