A Smart Management Method and System for the Entire Lifecycle of Inspection, Diagnosis and Treatment of Railway Tunnels

Abstract

This invention discloses a method and system for intelligent management and control of the entire lifecycle of railway tunnel inspection, diagnosis, and treatment, belonging to the field of railway infrastructure operation and maintenance management technology. This invention expands the traditional linear process of detection-diagnosis-treatment into a five-loop closed-loop model: S1 Detection, S2 Diagnosis, S3 Treatment, S4 Verification, and S5 Closure, embedding an SDCA+PDCA dual-cycle driving mechanism. Multi-source detection data are unified with a spatiotemporal benchmark, and data-feature-decision level three-layer fusion is carried out based on adaptive weights according to data quality levels, outputting diagnostic results with confidence labels. Based on the matching degree between the defect and the knowledge base template, standardized or exploratory channels are automatically categorized. After treatment, re-inspection, and verification meet the standards, knowledge is forcibly extracted and stored in the database in a structured manner. This invention connects the entire process of detection, diagnosis, treatment, verification, and knowledge storage, taking into account both efficient handling of conventional defects and iterative solidification of new defects, shortening the defect closure cycle, and improving the intelligent level of railway tunnel operation and maintenance.

Description

Technical Field

[0001] This invention relates to the field of railway infrastructure operation and maintenance management technology, and in particular to a method and system for intelligent management and control of the entire life cycle of railway tunnel inspection, diagnosis and treatment. Background Technology

[0002] As a key control node of railway lines, the structural health of tunnels directly affects traffic safety and transportation order. During long-term service, tunnel linings inevitably suffer from various defects such as cracks, water leakage, voids, spalling, and arch heave due to the combined effects of multiple factors such as geological environment, train cyclic load, and groundwater erosion. According to relevant statistics, about 30% of the existing tunnels on the entire railway have defects of varying degrees, of which about 5% pose a direct threat to structural safety and operational order. Tens of thousands of tunnel inspection data, diagnostic reports, and treatment work orders are generated every year. However, such large-scale inspection, diagnosis, and treatment activities currently lack a unified full life cycle management framework. How can inspection data drive diagnosis, how can diagnostic conclusions guide treatment, and how can treatment effects be fed back to optimize the next round of inspection plans? There is no institutionalized organic connection between the various links, and the overall operation and maintenance is still mainly based on "firefighting" passive response, lacking a systematic method for proactive management from a full life cycle perspective.

[0003] Railway tunnel inspection and maintenance must be completed within extremely limited "maintenance windows"—high-speed railway maintenance windows are typically only 240 minutes per day, while conventional railway maintenance windows are approximately 120-180 minutes per day. In long tunnels, additional time for entry and exit must be factored in. The structural contradictions of tight schedules, heavy workloads, poor accessibility, and high risk of omissions further amplify the negative effects of the fragmented "inspection, diagnosis, and treatment" process: the limited work windows require a high degree of coordination among all links, but existing detection systems, diagnostic analysis tools, and treatment and construction management systems are all independently built for single business links. There is a lack of unified data interfaces and process connection mechanisms between systems. Detection data cannot automatically trigger diagnostic analysis, diagnostic conclusions cannot directly drive treatment decisions, and treatment results cannot be systematically fed back to the next round of inspection plans. Information transmission and coordination between links rely on manual processes, making it difficult to guarantee overall operation and maintenance efficiency and quality.

[0004] Currently, the railway tunnel operation and maintenance field faces the following core problems. First, multi-source detection data is difficult to integrate—existing detection methods encompass various technologies such as measurement robots, 3D laser scanning, high-definition cameras, drones, InSAR, distributed fiber optic sensing, and microseismic monitoring. However, these data follow different format standards, have inconsistent spatial reference systems, and their time sampling frequencies range from seconds to months, being scattered and stored in their respective independent systems. The phenomenon of "data silos" is prominent, making cross-source correlation analysis and comprehensive judgment difficult. Second, the processes of "inspection, diagnosis, and treatment" are fragmented—detection systems, diagnostic analysis tools, and treatment and construction management systems are all independently built for single business processes. There is a lack of unified data interfaces and process connection mechanisms between systems. Detection data cannot automatically trigger diagnostic analysis, diagnostic conclusions cannot directly drive treatment decisions, and treatment results cannot be systematically fed back to the next round of inspection plans. Information transmission and coordination between each stage rely on manual processes. Third, digital twin applications remain at the visualization level—currently, they mainly take the form of 3D model displays and monitoring data dashboards. The models lack physical properties and mechanical behavior information that can be loaded with boundary conditions for extrapolation and calculation. The interaction between monitoring data and 3D models is primarily offline import and static display, lacking real-time dynamic mapping and two-way interaction capabilities. Fourth, disease treatment experience cannot be systematically accumulated and reused—diagnostic conclusions, treatment plan selection criteria, and treatment effect data generated in each inspection and treatment practice exist mostly in unstructured documents or individual memory forms. There is a lack of mechanisms to automatically extract, structure, and incorporate this information into a searchable knowledge base. This means that when similar diseases recur, re-judgment still relies on individual experience, and organizations cannot continuously learn from historical cases. Summary of the Invention

[0005] This invention aims to solve the following technical problems existing in the prior art: In the existing railway tunnel operation and maintenance system, multi-source detection data cannot be integrated and analyzed due to inconsistent formats and spatiotemporal benchmarks; the detection, diagnosis, and treatment links rely on independent systems and manual connections, resulting in process fragmentation; digital twin applications are limited to three-dimensional visualization and lack computable physical attributes and two-way data interaction capabilities; and the diagnostic conclusions and treatment experiences generated in each inspection, diagnosis, and treatment practice cannot be systematically extracted and reused. These problems together constitute a systemic dilemma of "data silos → process fragmentation → shallow twin → knowledge loss".

[0006] Therefore, the present invention provides a smart management and control method for the entire life cycle of railway tunnel inspection, diagnosis and treatment. The management and control method includes a five-ring closed-loop management method and an SDCA+PDCA dual-cycle driving mechanism embedded in the five-ring closed-loop management method. The five-ring closed-loop management method expands the three-stage linear process of "detection → diagnosis → treatment" into a five-ring closed-loop model, which includes: S1, Inspection Ring: Multi-source heterogeneous data of tunnel structure are acquired through various monitoring devices. All data are uniformly converted to a spatiotemporal reference with "tunnel mileage + lining ring number + clock position" as the spatial index and UTC timestamp as the time index. After automatic data quality classification (A / B / C / D four levels), the data is stored in the database. The acquisition strategy supports three modes: window point triggering, real-time streaming and periodic inspection. S2, Diagnostic Loop: For all data after spatiotemporal unification, data-level fusion (spatial registration, temporal alignment, denoising and standardization), feature-level fusion (diagnostic feature extraction, splicing and dimensionality reduction), and decision-level fusion (comprehensive judgment of multi-source diagnostic tendency and conflict detection) are performed sequentially. The contribution weight of each data in the fusion is adaptively adjusted according to its data quality level. The fusion result is attached with a quantitative confidence label. When the confidence is lower than the preset threshold, a low confidence alarm is generated and pushed for manual review. The fused feature vector is input into the diagnostic model chain of the model layer (including the apparent disease identification model, the hidden disease analysis model, and the health status classification model), and the outputs disease type, five-level health level, and four-level risk level. The output conclusions of the diagnostic model are continuously optimized with the accumulation of re-examination data. When the accumulated re-examination deviation rate exceeds the preset threshold, the incremental training or retraining of the model is automatically triggered, so that the judgment ability continues to evolve with the increase of usage. S3, Treatment Loop: Based on the diagnostic results of the diagnostic loop, the treatment decision engine intelligently retrieves historical cases from the knowledge base and related knowledge from the knowledge graph, automatically matches historical successful cases and applicable standardized treatment plans similar to the current disease characteristics, and generates a treatment plan recommendation list after sorting according to the preset comprehensive scoring rules. When there are significant differences in the recommendation results based on different knowledge sources, the system automatically generates a prudent assessment mark to prompt decision-makers to pay attention to potential risks. The recommendation list can only generate a digital treatment work order after "preliminary review by the diagnostic engineer → approval by the technical leader" and is issued to the on-site work team for execution. Modification operations during the approval process are recorded as feedback signals to optimize the recommendation model. S4, Verification Ring: After treatment, the post-treatment monitoring data is obtained through re-inspection and automatically compared with the pre-treatment diagnostic baseline. If the standard is met, it enters S5; if the standard is not met, it returns to S3. A maximum of 2 returns can be made at a time. If the number of returns exceeds the limit, the approval process will be upgraded and an expert review will be triggered. S5, Closure Loop: After the verification loop passes the verification, knowledge extraction is forcibly triggered. The disease feature codes, key parameters of the treatment plan, treatment effect tags and lessons learned tags of this case are extracted in a structured manner. After review, they are stored in the knowledge base and the knowledge graph is updated, completing the closed-loop accumulation from detection data to organizational knowledge.

[0007] Furthermore, the SDCA+PDCA dual-cycle driving mechanism is based on the five-ring closed-loop management method. After the S2 diagnostic ring outputs its conclusions and before the S3 treatment ring starts, it embeds a routing decision mechanism driven by both SDCA and PDCA cycles. Based on whether the conclusions of the S2 diagnostic ring can match the existing standardized process templates in the knowledge base, it decides whether to enter the SDCA cycle or the PDCA cycle, thereby driving the S3 treatment ring, S4 verification ring, and S5 cancellation ring to execute according to differentiated paths.

[0008] Furthermore, the SDCA cycle refers to entering the SDCA standardized processing channel when a disease matches an existing standard process template in the knowledge base. Here, S refers to calling the standard process template, which includes diagnostic criteria, treatment plan, and acceptance criteria. D refers to executing each step according to the template. C refers to checking the execution quality and treatment effect. A refers to fine-tuning the template parameters based on the check results.

[0009] Furthermore, the PDCA cycle refers to the automatic switch to the PDCA exploratory treatment channel when the disease cannot be matched with any standard process template or the similarity is lower than a preset threshold. The initial value of the preset threshold is 0.6. Here, P refers to organizing multi-disciplinary consultations, combining literature search and similar case search, to formulate research-based diagnostic plans and experimental treatment plans. D refers to the refined execution of the experimental plan, with encrypted monitoring and recording throughout the process. C refers to multi-dimensional effect evaluation (short-term / medium-term / long-term). A refers to refining the verified plan into a standard process template and formally incorporating it into the PDCA system.

[0010] Furthermore, when the similarity between the input disease and all standard templates is less than the preset threshold of 0.6, or the treatment effectiveness rate of the case is less than 80% after 20 SDCA cycles, the SDCA cycle is converted into a PDCA cycle. The conditions for converting the PDCA cycle to the SDCA cycle are: (a) the PDCA cycle-based research diagnostic and experimental treatment plans have been successfully applied in at least 3 independent tunnels or locations; (b) the mid-term re-examination of the independent tunnels or locations where the plans were applied has met the standards, with a mid-term period of at least 3 months; and (c) there is no statistically significant difference in the effectiveness of each application.

[0011] Furthermore, the acceptance criteria for the S4 verification ring are as follows: For cracks, the width of the crack after treatment should not exceed 0.1 mm or 30% of the width of the crack before treatment; For deformation-related defects, the deformation rate after treatment should not exceed 0.5 mm / month or 20% of the deformation rate before treatment. For water leaks, once the treatment is completed, the water level at the leak point should drop to a damp stain or remain unchanged for three consecutive months.

[0012] Furthermore, in S5, the lessons learned label includes the reasons for deviations in scheme selection, records of construction anomalies, and warnings of recurrence risk.

[0013] Furthermore, in S3, a two-dimensional risk matrix and a three-level priority ranking rule are used to provide a quantitative basis for treatment priority decision-making. Two-dimensional risk matrix is Risk Value , For failure probability, For the consequences of failure The three-tier priority ranking rule is: risk level first → disease nature first → section importance first.

[0014] This invention also discloses a system for implementing a smart management and control method for the entire lifecycle of railway tunnel inspection, diagnosis, and treatment. The system adopts a five-layer, two-vertical design, specifically including: The perception layer adopts the device abstract adapter pattern to uniformly connect to multiple source devices such as measurement robots, InSAR, UAVs, LiDAR, and fiber optic sensors, and supports three acquisition strategy scheduling. The data layer organizes and manages multi-source heterogeneous data using "mileage + ring number + clock position + UTC time" as a unified spatiotemporal benchmark. The data layer has built-in automatic quality classification, active learning intelligent annotation, and knowledge base or case library. The model layer deploys disease identification models, health diagnosis models, risk classification models, solution recommendation models, and trend prediction models in a containerized manner, supporting hot updates and continuous evolution driven by error signals. When the cumulative re-inspection deviation rate exceeds the threshold, incremental training or retraining is automatically triggered. At the business layer, based on the BPMN 2.0 workflow engine, the S1-S5 five-ring process is solidified into a configurable digital workflow. The business layer adopts a human-machine hybrid decision-making mode of mandatory expert approval for high-risk diseases and automatic flow within the authorized scope for low-risk, high-matching diseases. The entire process operation is recorded in the chain hash structure audit log. The presentation layer uses Cesium or Three.js as the rendering engine to realize a 3D digital twin cockpit, supporting four levels of zoom (full-line bird's-eye view → section browsing → single-ring slice → disease details) and time-axis driven historical evolution backtracking; The two vertically integrated systems are the security and access control system and the data standards and interface specifications system.

[0015] The advantages of this invention compared to the prior art are: (1) Regarding the seamless integration of the entire process, this invention utilizes the S1-S5 five-loop flow model to automatically drive diagnosis based on detection data, directly link diagnostic conclusions to treatment decisions, and enforce knowledge extraction and storage after automatic comparison and determination of treatment effectiveness. This ensures that each detection, diagnosis, and treatment process forms a traceable, reusable, and optimizable closed loop, solving the problems in existing technologies where detection, diagnosis, and treatment rely on manual coordination and lack institutionalized effect verification and knowledge archiving after treatment. Data from Example 3 shows that after incorporating the SDCA standard process, the average closed-loop cycle for similar diseases is shortened from 98 days in the PDCA exploration phase to 32 days.

[0016] (2) In terms of management mechanism, this invention achieves a balance between standardization and innovative exploration through the SDCA+PDCA dual-cycle mechanism. For high-frequency diseases whose patterns have been mastered, standardized templates are used to ensure the quality and efficiency of treatment. For new diseases that cannot be matched with existing templates, the invention automatically switches to the exploratory treatment channel and, after successful verification, refines them into new standards and incorporates them into the SDCA system. This enables knowledge to be continuously transformed from individual experience to organizational capabilities, solving the dual dilemma of relying on personal experience for the quality of conventional disease treatment and lacking methodological guidance for new diseases.

[0017] (3) In terms of diagnostic credibility, the present invention uses an automatic data quality classification and fusion weight linkage mechanism—high-quality data automatically obtains higher decision weight while low-quality data is automatically downgraded or excluded—to make the fusion result accompanied by a quantitative confidence label to provide a credible basis for diagnostic decision-making. When the confidence is lower than the threshold, a low confidence alarm and manual review are triggered, avoiding the risk of blindly outputting a definitive diagnostic conclusion when the data quality is insufficient. Attached Figure Description

[0018] Figure 1 This is a schematic diagram of the five-ring closed-loop circulation model of the present invention.

[0019] Figure 2 This is a schematic diagram of the SDCA+PDCA dual-cycle driving mechanism of the present invention.

[0020] Figure 3 This is a schematic diagram of the five-layer, two-vertical platform architecture of the present invention. Detailed Implementation

[0021] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, 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, 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.

[0022] Example Deployment and operation of the five-ring closed-loop management method on the platform This embodiment uses a 14.5-kilometer-long operational railway tunnel (double-track single-bore, designed speed 350 km / h) within the jurisdiction of a railway bureau as an application object to illustrate the specific deployment and end-to-end operation mode of the five-ring closed-loop management method and SDCA+PDCA dual-cycle mechanism of the present invention on a preferred platform architecture. It should be understood that the following architectural details are only used to illustrate the engineering implementation of the method and are not intended to limit the operating environment of the method.

[0023] Perception layer deployment The following sensing devices are deployed along the entire length of the tunnel: 12 measuring robots, one every 1.2 kilometers along the tunnel's longitudinal direction, automatically monitor using a spatiotemporal reference provided by a single BeiDou positioning terminal, triggering automatic measurements at a maintenance window every quarter; a 3D laser scanner mounted on a track inspection vehicle, completing a full tunnel scan within a monthly maintenance window; a high-definition industrial camera array deployed every 50 meters on the tunnel lining surface, acquiring high-definition images of the lining appearance in real time; a distributed fiber optic sensing system laid along the entire length of the tunnel, continuously monitoring strain and temperature; microseismic monitoring sensors deployed in key sections of the tunnel (fault fracture zones, tunnel entrances, etc.), acquiring microseismic signals in real time; and stress / displacement sensors embedded in key sections of the lining, continuously acquiring stress and deformation data. All devices are uniformly connected to the sensing layer through a device abstraction adapter, automatically scheduling maintenance window triggering devices according to the maintenance window plan.

[0024] Data layer operation The raw data collected by the perception layer are transformed according to a unified spatiotemporal reference. Taking the convergence deformation data of a lining ring obtained by the measurement robot as an example: the three-dimensional coordinates (X0, Y0, Z0) in the original coordinate system are first transformed to the CGCS2000 coordinate system. Then, based on the measured mileage K12+345 and ring number R1234 of the ring, the deformation is associated with the unified spatial index of "mileage K12+345-ring number R1234-3 o'clock position"; the timestamp is unified to UTC time (e.g., 2026-03-15T02:35:18Z). The automatic data quality assessment module comprehensively evaluates the quality level of this data point as Grade A based on the equipment calibration status (calibration was completed within 24 hours prior to this measurement, marked as Grade A), data completeness (this location covers three types of data: 3D laser scanning point cloud, high-definition image, and distributed fiber stress, with coverage = 3 / 4 = 0.75), and timeliness (the acquisition time is 12 hours from the present, the half-life is set to 7 days, and the attenuation factor is ≈0.95). The confidence level label is 0.85.

[0025] Model layer processing The detection data of a certain section of the tunnel (mileage K8+200~K8+350) in the first quarter of 2026, after data fusion, triggered the apparent defect identification channel. The DeepLabV3+ model identified 3 main cracks (maximum width 0.35mm, maximum length 2.1m) and 1 seepage area (area approximately 0.8m², level "dripping") in the image. In the hidden defect inversion channel, the ground-penetrating radar data confirmed by FDTD inversion that there was a cavity behind the lining (area approximately 0.6m², depth approximately 12cm from the inner surface of the lining).

[0026] The comprehensive evaluation model, combined with deformation monitoring data (the convergence value of this section increased by 0.8 mm compared to the previous quarter), determined the health status of this section to be Level C (moderate abnormality). The system automatically generated a diagnostic report and pushed it to the business layer.

[0027] Business layer flow The business layer workflow engine matches the standard process template SDCA-T-012 (standard process for sealing lining cracks and grouting voids) based on the defect type ("lining cracks + voids"). It automatically enters the SDCA channel. Because the crack width (0.35mm) exceeds the recommended threshold (0.3mm) for grouting treatment in the template, the system automatically marks "grouting pressure parameter needs to be adjusted upwards" on the standard template and pushes it to the diagnostic engineer for initial review. After confirmation by the diagnostic engineer and approval by the technical manager, the system generates a digital treatment work order (including precise defect location K8+260~K8+310 / 4~8 o'clock orientation, recommended process "epoxy resin grouting + surface sealing", grouting pressure 0.3~0.5MPa, key control parameters, and acceptance standard ≤0.1mm). This work order is pushed to the on-site work team via mobile device.

[0028] Presentation layer presentation In the 3D digital twin cockpit, the tunnel model of this section is highlighted in yellow (Grade C). Clicking on it displays a disease details card, which includes the original image, 3D laser scan point cloud deformation cloud map, radar profile, diagnostic conclusion, and treatment work order status. Managers can drill down from the overall bird's-eye view to the single-ring slice view to intuitively understand the spatial distribution and historical evolution of the disease.

[0029] The closed-loop operation of the verification and cancellation loops was implemented. Seven days after the treatment was completed, the system automatically triggered a measurement robot and a high-definition camera to re-inspect the section according to the preset re-inspection plan. After spatiotemporal unification, the re-inspection data was automatically compared with the pre-treatment diagnostic baseline: the crack width decreased from 0.35mm before treatment to 0.06mm (meeting the acceptance standard of ≤0.1mm), the void behind it disappeared in the radar profile after grouting, and the convergence value stabilized at the pre-treatment level. Once the system determines that the treatment effect meets the standards, it automatically enters the cancellation loop. It extracts the following structured data from this case: the disease characteristic code (CR-001+CA-003, longitudinal crack + back cavity), key parameters of the treatment plan (epoxy resin grouting pressure 0.4MPa, grouting volume 38L, surface sealing material type), treatment effect label (effective – crack closed and cavity filled), and lessons learned label (reason for adjusting grouting pressure by 0.1MPa compared to the standard template: crack width 0.35mm exceeds the template's recommended threshold of 0.3mm). After review and confirmation by the diagnostic engineer, this data is written into the knowledge base and the statistical parameters of the SDCA-T-012 template in the knowledge graph are updated. If the re-inspection fails to meet the standards, the system automatically returns to the S3 treatment loop with a list of non-compliant items, with a maximum of two returns per instance. Thus, the disease in this section has undergone a complete five-loop process: S1 detection → S2 diagnosis → S3 treatment → S4 verification → S5 cancellation. The only manual intervention occurred at the two nodes of initial review by the diagnostic engineer and approval by the technical manager. All other steps were automatically driven by the platform.

[0030] The specific process of three-level data fusion

[0031] Taking the data fusion process of a certain defective section of the tunnel as an example, the implementation steps of the three-level fusion strategy are explained in detail. At the same lining location at mileage K12+345 (ring number R1234), four sources of data were acquired within one week: three-dimensional laser scanning point cloud (acquired on 2026-04-03, showing a cross-sectional convergence value of 12.5mm), high-definition image (acquired on 2026-04-05, showing a crack width of 0.42mm), distributed fiber optic strain data (continuously acquired from 2026-04-01 to 04-07, showing a strain increase of 120με at this location), and InSAR surface deformation data (acquired on 2026-04-02, showing a surface settlement rate of 2.5mm / month above the tunnel).

[0032] Step 1: Data-level Fusion Spatial registration: The 3D laser scanning point cloud data was converted to the tunnel mileage-ring number coordinate system via ICP registration, and located at the K12+345 / R1234 / 3~5 o'clock position; the high-resolution image was used to obtain the camera pose through LiDAR SLAM, and the pixel coordinates were back-projected onto the surface of the tunnel 3D model using the collinearity equation. The actual spatial location of the crack matched the deformation location of the 3D laser scanning point cloud (deviation < 5cm); the distributed fiber strain data was automatically associated with the unified spatial index by the sensor number at this location; the InSAR data was converted using the CGCS2000 intermediate reference to confirm that the surface subsidence center corresponds to the tunnel mileage K12+340~K12+360.

[0033] Time alignment: 2026-04-05T00:00:00Z is used as the unified reference time. The 3D laser scanning point cloud and high-definition image data are marked with the acquisition time (deviation ≤ 2 days, no interpolation is performed due to the slow development of the disease). The distributed fiber strain data is the average of 24 hours before and after the reference time (152με). The InSAR data is marked as 2026-04-02 due to the revisit period constraint (deviation 3 days, within the allowable range).

[0034] Denoising and standardization: The convergence value of the 3D laser scanning point cloud data was corrected to 11.8mm after wavelet threshold denoising; the crack width of the high-definition image was corrected to 0.38mm after subpixel precision optimization; the strain data was corrected for trend terms after Kalman filtering and standardized to the [0,1] interval.

[0035] Step 2: Feature-level fusion Feature vectors were extracted from various data sources: 3D laser scanning point cloud → cross-sectional convergence features (11.8 mm, an increase of 2.1 mm / month compared to the previous period); high-resolution images → crack geometric features (width 0.38 mm, length 1.8 m, approximately circumferential orientation, density 0.3 cracks / m²); distributed fiber optic strain data → strain trend features (increasing by 120 με / month, upward trend); InSAR landmark deformation data → surface deformation features (2.5 mm / month subsidence, affecting an area of ​​approximately 150 m). These four features were concatenated into a joint feature vector under a unified spatial location index, with a dimension of 4 × 3 = 12. After PCA dimensionality reduction (retaining 96.2% variance information), an 8-dimensional feature vector was output for decision-level fusion.

[0036] Step 3: Decision-level integration The four data sources independently provide diagnostic tendencies: 3D laser scanning point cloud + distributed optical fiber (structural deformation) → tends towards "lining collapse risk" (confidence 0.78); high-resolution image (appearance) → tends towards "crack development" (confidence 0.82); InSAR (surface) → tends towards "goaf activation impact" (confidence 0.65). Using DS evidence theory for synthesis: the initial BPA weights for each source are determined according to data quality level (point cloud A level 1.0, image A level 1.0, optical fiber A level 1.0, InSAR B level 0.75). The fused comprehensive diagnostic conclusion is "additional deformation and cracking of the lining caused by goaf activation," with a comprehensive confidence level of 0.79 (0.14 higher than a single InSAR source), and an uncertainty interval of 0.12. This conclusion is simultaneously pushed to the diagnostic stage, and the diagnostic model adjusts its health level assessment for this section accordingly (upgraded from B to C). The fused confidence level of 0.79 is greater than the preset threshold of 0.6, so no low-confidence alarm is triggered.

[0037] An example of closed-loop management driven by SDCA+PDCA dual-cycle approach.

[0038] The treatment process for two different types of tunnel defects fully illustrates the operation and transition between the SDCA and PDCA dual cycles.

[0039] (I) SDCA Cycle Operation Example - Conventional Crack Treatment In February 2026, inspection data revealed a circumferential crack at tunnel kilometer marker K3+200 (ring number R0320), measuring 0.18 mm in width and 1.2 m in length. There was no water seepage or signs of accelerated development. The business layer workflow engine coded this defect feature and matched it against the knowledge base, achieving a similarity score of 0.89 with the standard process template SDCA-T-008 (Standard Process for Surface Sealing of Circumferential Cracks). The system automatically entered the SDCA channel. S (Standardization): Use the template and specify the process as "epoxy resin surface coating and sealing", and the acceptance standard is "the width of the crack after sealing is ≤0.05mm, and there is no expansion after re-inspection within 6 months".

[0040] D (Execution): The diagnostic engineer conducts an initial review and confirms the diagnosis (15 minutes), the technical supervisor approves it (20 minutes), and a digital treatment work order is generated and pushed to the on-site work team; the team completes the closed operation within the designated window, and the whole process takes 90 minutes.

[0041] C (Inspection): A follow-up inspection was conducted on the 7th day after treatment, and the crack width was 0.04mm, meeting the acceptance standard; a second follow-up inspection was conducted on the 30th day, and the width was still 0.04mm, with no expansion.

[0042] A (Handling): If the verification results meet the standards, the system automatically generates a treatment effectiveness report and archives it. The effective treatment tag (effective) for this case is fed back to the knowledge base, updating the "effectiveness" statistics of this standard process template (from 91.2% to 91.3%).

[0043] The entire closed-loop cycle: from the submission of test data (February 5) to the cancellation and archiving (March 6, including 30 days of waiting for re-inspection), a total of 29 days were completed, with the total manual intervention time being less than 2 hours, demonstrating the high efficiency of the SDCA channel.

[0044] (II) SDCA→PDCA Switching Example – Standard Process Failure In March 2026, a water seepage defect appeared at kilometer marker K8+260 (ring number R0826) of the tunnel. The seepage level was "water leakage" (approximately 0.5 L / min), pH=4.2 (acidic), accompanied by white precipitates. The business layer workflow engine searched the knowledge base, and the most matching standard process template was SDCA-W-003 (standard process for grouting and sealing of lining water seepage). The similarity was only 0.52, which is less than the threshold θ1=0.6, triggering the SDCA→PDCA switch. The system also generated a switch alarm notification: "The water seepage defect at K8+260 cannot match the existing standard process (similarity 0.52). It is recommended to start the PDCA exploration channel. Preliminary analysis of the cause: the acidic water seepage (pH 4.2) exceeds the acid and alkali resistance range of conventional grouting materials." (III) PDCA Cycle Implementation Example – Exploration of New Diseases After the system enters the PDCA channel: P (Plan): The system automatically organizes multi-disciplinary consultations (structural engineers + geological engineers + materials engineers), combines literature search (7 relevant cases of acidic water leakage treatment in the past 5 years) and case database search (3 cases with similar pH conditions, all of which are experimental schemes), and formulates an experimental scheme: using acid and alkali resistant special grouting materials (epoxy-acrylate composite system), controlling the grouting pressure in stages (0.2MPa→0.4MPa→0.6MPa), and increasing the frequency of monitoring (daily recording of leakage, pH value, and lining strain).

[0045] D (Execution): After the test plan was approved by the Technical Committee, it was implemented during the designated window in April 2026. The on-site grouting was carried out according to the segmented grouting plan: initial grouting at 0.2 MPa (lasting 15 min, observing changes in leakage) → medium-pressure grouting at 0.4 MPa (lasting 20 min, leakage decreasing from 0.5 L / min to 0.1 L / min) → final pressure sealing at 0.6 MPa (lasting 10 min, leakage decreasing to 0). The grouting pressure curve, material usage (a total of 42 L of epoxy-acrylate composite grout), and lining strain response were recorded throughout the process (maximum additional strain 85 με, below the warning value of 200 με). Post-treatment, intensified monitoring was initiated: leakage was 0 for 7 consecutive days, the pH of the surface exudate increased from 4.2 to 5.8, and the lining strain stabilized at 85±5 με within one week.

[0046] C (Inspection): First re-inspection on day 7 after treatment – ​​leakage was 0, crack surface was dry, and pH test paper showed the surface was neutral. Second re-inspection on day 30 – no leakage recurrence, strain value stable. Third re-inspection on day 90 (mid-term evaluation) – leakage remained 0, pH recovered to 6.5, and lining strain decreased to 62 με. All three re-inspections confirmed that the treatment effect met the standards.

[0047] A (Handling): After the mid-term assessment on day 90 is passed, the system automatically triggers the PDCA→SDCA conversion assessment. Upon verification that three conditions are met—mid-term re-inspection is passed, successful application under similar conditions in two other tunnels (a total of 3 independent applications), and the effectiveness of each application is 100% with no statistically significant difference in effect between different locations (Fisher precise verification, p=0.25>0.05)—the standard process template SDCA-W-005 (Standard Process for Acid-Resistant Grouting of Acidic Leakage Water) is officially released and incorporated into the SDCA system after approval by the Technical Committee. This template includes applicable conditions (pH≤5.0), recommended materials, segmented grouting parameters (0.2→0.4→0.6MPa), and acceptance criteria (leakage = 0 and pH≥6.0 within 90 days after grouting). Since its release, the subsequent 3 cases of acidic leakage water have all been treated through the SDCA channel, shortening the average closed-loop cycle from 98 days in the PDCA exploration phase to 32 days.

[0048] This embodiment fully demonstrates the entire evolutionary process of the PDCA channel, from exploratory processing to knowledge standardization, and then its integration into the SDCA system to form a highly efficient and reusable dual-cycle system.

[0049] All or part of the steps in the above method embodiments can be implemented by hardware related to computer program instructions. The computer program can be stored in a computer-readable storage medium, including ROM, RAM, disk, or optical disk, etc. When the program is executed by a processor, it realizes the automated operation of steps S1 to S5 and the SDCA+PDCA dual-loop driving mechanism in the method embodiments of the present invention.

[0050] It should be noted that the specific parameters, algorithm selections, and numerical ranges given in the above embodiments are merely preferred embodiments of the present invention and are not intended to limit the scope of protection of the present invention. In other embodiments, those skilled in the art can make equivalent substitutions and modifications according to actual application scenarios.

[0051] All equivalent changes and modifications made in accordance with the teachings of this invention are still within the scope of this invention. Those skilled in the art will readily conceive of other embodiments of this invention upon considering the specification and the disclosure of practical truth. This application is intended to cover any variations, uses, or adaptations of this invention that follow the general principles of this invention and include common knowledge or conventional techniques in the art not described herein.

Claims

1. A smart management and control method for the entire lifecycle of railway tunnel inspection, diagnosis, and treatment, the management and control method comprising a five-ring closed-loop management method and an SDCA+PDCA dual-cycle driving mechanism embedded in the five-ring closed-loop management method, characterized in that: The five-ring closed-loop management method expands the three-stage linear process of "detection → diagnosis → treatment" into a five-ring closed-loop model, which includes: S1. Detection ring: Multi-source heterogeneous data of tunnel structure are acquired through various monitoring devices. All data are uniformly converted to a spatiotemporal reference with "tunnel mileage + lining ring number + clock position" as the spatial index and UTC timestamp as the time index. After automatic data quality classification (four levels: A / B / C / D), the data is stored in the database. S2, Diagnostic Loop: Data-level fusion, feature-level fusion, and decision-level fusion are sequentially performed on all data after spatiotemporal unification. The contribution weight of each data in the fusion is adaptively adjusted according to its data quality level. The fusion result is attached with a quantitative confidence label. When the confidence is lower than the preset threshold, a low confidence alarm is generated and pushed for manual review. The fused feature vector is input into the diagnostic model chain of the model layer, and the disease type, five-level health level, and four-level risk level are output. S3, Treatment Loop: Based on the diagnostic results of the diagnostic loop, the treatment decision engine intelligently retrieves historical cases in the knowledge base and related knowledge in the knowledge graph, automatically matches historical successful cases and applicable standardized treatment plans similar to the current disease characteristics, and generates a recommended list of treatment plans after sorting them according to the preset comprehensive scoring rules. S4, Verification Loop: After treatment, the post-treatment monitoring data is obtained through re-inspection and automatically compared with the pre-treatment diagnostic baseline. If the standard is met, it proceeds to S5; otherwise, it returns to S3. S5, Closure Loop: After the verification loop passes the verification, knowledge extraction is forcibly triggered. The disease feature codes, key parameters of the treatment plan, treatment effect tags and lessons learned tags of this case are extracted in a structured manner. After review, they are stored in the knowledge base and the knowledge graph is updated, completing the closed-loop accumulation from detection data to organizational knowledge.

2. The intelligent management and control method for the entire life cycle of inspection, diagnosis and treatment of railway tunnels according to claim 1, characterized in that: The SDCA+PDCA dual-cycle driving mechanism is based on the five-ring closed-loop management method. It further embeds a routing decision mechanism driven by the dual wheels of SDCA and PDCA cycles after the output conclusion of the S2 diagnostic ring and before the start of the S3 treatment ring. It determines whether to enter the SDCA cycle or the PDCA cycle based on whether the conclusion of the S2 diagnostic ring can match the existing standardized process template in the knowledge base, thereby driving the S3 treatment ring, S4 verification ring and S5 cancellation ring to execute according to differentiated paths.

3. The intelligent management and control method for the entire life cycle of inspection, diagnosis and treatment of railway tunnels according to claim 2, characterized in that: The SDCA cycle refers to the process of entering the SDCA standardized processing channel when a disease matches an existing standard process template in the knowledge base. Here, S refers to calling the standard process template, which includes diagnostic criteria, treatment plan, and acceptance criteria. D refers to executing each step according to the template. C refers to checking the execution quality and treatment effect. A refers to fine-tuning the template parameters based on the check results.

4. The intelligent management and control method for the entire life cycle of inspection, diagnosis and treatment of railway tunnels according to claim 3, characterized in that: The PDCA cycle refers to automatically switching to the PDCA exploratory treatment channel when a disease cannot be matched with any standard process template or the similarity is lower than a preset threshold. Here, P refers to organizing multi-disciplinary consultations, combining literature search and similar case search, to formulate research-based diagnostic plans and experimental treatment plans; D refers to refining the execution of the experimental plan, with encrypted monitoring and recording throughout the process; C refers to multi-dimensional effect evaluation; and A refers to refining the validated plan into a standard process template and formally incorporating it into the SDCA system.

5. The intelligent management and control method for the entire life cycle of railway tunnel inspection, diagnosis and treatment according to claim 4, characterized in that: When the similarity between the input disease and all standard templates is less than the threshold, or the treatment effectiveness rate of the case is less than 80% after 20 SDCA cycles, the SDCA cycle is converted into a PDCA cycle. The conditions for converting the PDCA cycle to the SDCA cycle are: (a) the PDCA cycle-based research diagnostic and experimental treatment plans have been successfully applied in at least 3 independent tunnels or locations; (b) the mid-term re-examination of the independent tunnels or locations where the plans were applied has met the standards, with a mid-term period of at least 3 months; and (c) there is no statistically significant difference in the effectiveness of each application.

6. The intelligent management and control method for the entire life cycle of inspection, diagnosis and treatment of railway tunnels according to claim 1, characterized in that: The acceptance criteria for the S4 verification ring are as follows: For cracks, the width of the crack after treatment should not exceed 0.1 mm or 30% of the width of the crack before treatment; For deformation-related defects, the deformation rate after treatment should not exceed 0.5 mm / month or 20% of the deformation rate before treatment. For water leaks, once the treatment is completed, the water level at the leak point should drop to a damp stain or remain unchanged for three consecutive months.

7. The intelligent management and control method for the entire life cycle of inspection, diagnosis and treatment of railway tunnels according to claim 1, characterized in that: In S5, the lessons learned label includes the reasons for deviations in scheme selection, records of construction anomalies, and warnings of recurrence risk.

8. The intelligent management and control method for the entire life cycle of inspection, diagnosis and treatment of railway tunnels according to claim 1, characterized in that: In S3, a two-dimensional risk matrix and a three-level priority ranking rule are used to provide a quantitative basis for treatment priority decision-making. Two-dimensional risk matrix is Risk Value , For failure probability, For the consequences of failure The three-tier priority ranking rule is: risk level first → disease nature first → section importance first.

9. A smart management and control system for the entire lifecycle of inspection, diagnosis, and treatment of railway tunnels, used to execute the management and control method according to any one of claims 1-8, characterized in that, The system adopts a five-layer, two-vertical design, specifically including: The perception layer adopts the device abstract adapter pattern to uniformly connect to multiple source devices such as measurement robots, InSAR, UAVs, LiDAR, and fiber optic sensors, and supports three acquisition strategy scheduling. The data layer organizes and manages multi-source heterogeneous data using "mileage + ring number + clock position + UTC time" as a unified spatiotemporal benchmark. The data layer has built-in automatic quality classification, active learning intelligent annotation, and knowledge base or case library. The model layer deploys disease identification models, health diagnosis models, risk classification models, solution recommendation models, and trend prediction models in a containerized manner, supporting hot updates and continuous evolution driven by error signals. When the cumulative re-inspection deviation rate exceeds the threshold, incremental training or retraining is automatically triggered. At the business layer, based on the BPMN 2.0 workflow engine, the S1-S5 five-ring process is solidified into a configurable digital workflow. The business layer adopts a human-machine hybrid decision-making mode of mandatory expert approval for high-risk diseases and automatic flow within the authorized scope for low-risk, high-matching diseases. The entire process operation is recorded in the chain hash structure audit log. The presentation layer uses Cesium or Three.js as the rendering engine to realize a 3D digital twin cockpit, supporting four levels of zoom (full-line bird's-eye view → section browsing → single-ring slice → disease details) and time-axis driven historical evolution backtracking; The two vertically integrated systems are the security and access control system and the data standards and interface specifications system.

Images