Intelligent anti-clogging double-layer replaceable side slope drainage pipe and early warning method

Abstract

The application discloses an intelligent anti-silt double-layer replaceable side slope drainage pipe and a warning method; the drainage pipe comprises a permanent buried structure, an inner-layer pipe replaceable section, a vortex anti-silt structure, a positioning flow-limiting butt joint structure, a monitoring module and a cloud intelligent management platform; the warning method comprises data acquisition, data preprocessing, feature extraction, silt diagnosis, residual life prediction and warning decision; the permanent buried structure is left in the slope body as permanent support, and the inner-layer pipe replaceable section can be pulled out and replaced from the slope surface, thus solving the risk of hole collapse during pipe replacement; the vortex anti-silt structure generates a rotating flow field by means of spiral flow guide ribs, and silt is prevented; by collecting temperature, strain and vibration data, and then extracting high-dimensional feature vectors such as temperature gradient, strain ratio and vibration frequency spectrum, the multi-task neural network model is input, the silt type and the residual life are identified, and four-level warning is performed; the replaceable maintenance and intelligent monitoring of the drainage pipe are realized, the slope instability is effectively prevented, and the whole-cycle operation and maintenance cost is reduced.

Description

Technical Field

[0001] This invention relates to the field of slope engineering drainage technology, and in particular to an intelligent anti-siltation double-layer replaceable slope drainage pipe and an early warning method. Background Technology

[0002] Groundwater seepage is one of the core factors affecting slope stability. Increased pore water pressure within the slope significantly reduces the effective stress and shear strength of the soil and rock mass, thereby inducing geological disasters such as landslides and collapses. Therefore, constructing an efficient and reliable slope drainage system to promptly divert groundwater and reduce seepage pressure is a fundamental engineering measure to ensure the long-term stability of slopes, and slope drainage pipes are the key components in this system for achieving deep water conduction.

[0003] However, traditional slope drainage pipes commonly face severe clogging and failure problems during long-term service. Groundwater continuously carries fine-grained silt, clay minerals, insoluble salt ions, and microbial metabolic products during seepage. These substances continuously deposit, adhere to, and crystallize on the inner walls of the drainage pipes, filter pores, and the outer filter layer, causing the cross-sectional area to gradually narrow, the flow resistance to increase exponentially, and the drainage efficiency to continuously decline, ultimately leading to the complete loss of drainage function. Once the drainage system fails, the pore water pressure inside the slope will re-accumulate, rapidly depleting the safety reserves invested in the initial project and easily triggering catastrophic accidents.

[0004] To address the aforementioned issues, while existing technologies have evolved from ordinary PVC perforated pipes to various composite filter pipes, systemic defects remain insurmountable in terms of operation and maintenance mechanisms and technical architecture. These defects manifest in the following four aspects: 1) Lagging status perception and lack of real-time capability: Traditional operation and maintenance heavily rely on periodic manual inspections and passive fault diagnosis, failing to capture early, subtle signs of siltation and development, making it difficult to meet the all-weather online monitoring needs of large-scale slope groups; 2) Vague siltation diagnosis and lack of quantitative basis: Existing technologies cannot distinguish the composition, distribution location, and severity of silt, resulting in a lack of precise data support for maintenance decisions; 3) Lack of proactive intervention capability, missing the optimal treatment window: Traditional drainage pipes... In the early stages of performance degradation, the system is completely passive and lacks any self-regulation or anti-clogging function. By the time a decrease in flow is detected, the clogging has often progressed to the middle or late stages, missing the opportunity to restore performance at low cost. 4) The maintenance methods are crude, with the risk of collapse after replacement: For severely clogged drainage pipes, the conventional treatment is to excavate and replace the entire pipe or use high-pressure water jet dredging. Excavating and replacing the entire pipe involves a huge amount of work, and deeply buried pipes are prone to collapse of the borehole wall or even inducing slope instability when pulled out. High-pressure water jet dredging can easily damage the filter layer structure and shorten the service life of the new pipe. In addition, the traditional integrated pipe design means that any local damage requires the entire pipe to be scrapped, resulting in high maintenance costs throughout the entire life cycle.

[0005] In recent years, some studies have attempted to introduce replaceable maintenance technologies. For example, the patent application CN202311101108.5, entitled "A Drainage Pipe with Replaceable Filter Cartridge and Its Filter Cartridge Replacement Method," proposed a replaceable filter cartridge and used finite element analysis to determine whether the drainage pipe needs to be replaced and when to replace it. However, the results of existing finite element analyses have not fully considered different types of clogging, and the effect on chemical crystallization clogging is minimal.

[0006] In conclusion, the field of slope engineering urgently needs a new drainage pipe technology that integrates permanent support, rapid replacement, proactive silt prevention, and intelligent diagnostics. This technology must, while ensuring the safety of the slope structure is not disturbed by maintenance work, achieve real-time perception of drainage status, accurate identification of siltation types, scientific prediction of remaining lifespan, and intelligent decision-making regarding maintenance plans. Summary of the Invention

[0007] To achieve the above-mentioned objectives, this invention provides an intelligent anti-clogging double-layer replaceable slope drainage pipe and an early warning method, aiming to solve the technical problems in the prior art, such as difficulty in replacing the pipe along its entire length, inability to identify the type of clogging, and inability to accurately predict the remaining lifespan.

[0008] A smart anti-siltation double-layer replaceable slope drainage pipe includes a permanently buried structure, a replaceable inner pipe section, a vortex anti-siltation structure, a positioning and flow-limiting docking structure, a monitoring module, and a cloud-based intelligent management platform.

[0009] The permanently buried structure includes an outer pipe, a bottom cover, a conical top cover, and an inner pipe fixing section. The outer pipe is drilled and inserted deep into the slope during construction, and is a mesh-like, rigid, and permeable pipe. The bottom cover is fixed to the port of the outer pipe that protrudes from the slope surface. It has internal threads machined on its inner side and regularly distributed drainage holes on its sidewalls to drain the collected water and serve as a mechanical interface for connecting to the replaceable section of the inner pipe. The conical top cover is fixed to the port of the outer pipe that is deeply buried in the slope. It is bullet-shaped to facilitate pushing during construction and installation and to prevent soil and rock from directly intruding into the pipe. The inner pipe fixing section is coaxially inserted into the end of the outer pipe that extends into the slope and is connected to the conical top cover at that end.

[0010] The replaceable section of the inner tube is coaxially inserted into the outer tube at one end near the slope surface. The end near the slope surface is connected to the bottom cover of the outlet, and the end extending into the slope body is connected to the fixed section of the inner tube through the positioning and flow-limiting docking structure. The replaceable section of the inner tube is a mesh-like rigid permeable pipe with evenly distributed small holes along the entire length of the pipe wall, and the pipe wall is wrapped with geotextile.

[0011] The vortex anti-siltation structure is fixedly installed on the inner wall of the inner tube fixed section and the inner tube replaceable section, and its inner wall is uniformly provided with spiral guide ribs along the circumference.

[0012] The positioning and flow-limiting docking structure is located at one end of the slope of the replaceable inner tube section, used to achieve coaxial docking and radial limiting between the replaceable inner tube section and the fixed inner tube section, and to block the unfiltered water bypass channel at the docking end face. The positioning and flow-limiting docking structure consists of a radial positioning ring, an end face water-blocking ring, and a self-aligning guide bearing. The radial positioning ring is fitted onto the outer wall of the end of the replaceable inner tube section, and is integrally vulcanized with elastic rubber. Several continuous arc-shaped positioning protrusions are provided along the axial direction on its outer circumference. The outer diameter of the positioning protrusions is slightly larger than the inner diameter of the outer tube. In the free state, the protrusions and the grid ribs of the inner wall of the outer tube form multi-point elastic abutment. The end face water-blocking ring is fixed to the end of the replaceable inner tube section, and has an annular protrusion on the side facing the fixed inner tube section. When the outlet cap is tightened, the annular protrusion presses against the end face of the fixed inner tube section. The self-aligning guide bearing is located inside the docking structure and is used to guide the coaxiality during installation.

[0013] The monitoring module employs a distributed fiber Bragg grating array, pre-embedded in pre-designed microgrooves within the wall of the replaceable inner tube section, spaced 0.5m to 1.0m along the tube's length. Each microgroove is a semi-circular groove shaped along the tube wall axially, with a depth of 0.5mm to 1.0mm and a width matching the diameter of the distributed fiber Bragg grating array, used to accommodate and fix the array. The distributed fiber Bragg grating array includes a temperature grating, a strain grating, and a vibration grating, wherein the temperature grating is used to sense the temperature of the fluid inside the tube. C 1 Strain gratings are used to sense the circumferential strain of the pipe wall caused by the pressure difference between the inside and outside. C 2 and pipe wall axial strain C 3 Vibration gratings are used to sense the vibration frequency of the pipe wall induced by the eddy current anti-siltation structure. C 4 Pipe wall vibration amplitude C 5 ; The cloud-based intelligent management platform is deployed on a cloud server and adopts a microservice architecture, including a data access module, a feature extraction module, an intelligent evaluation and prediction module, and an early warning and decision-making module. The data access module is responsible for receiving data collected by the distributed fiber Bragg grating array, including the temperature of the fluid inside the pipe. C 1 Circumferential strain of pipe wall C 2 Pipe wall axial strain C 3 Pipe wall vibration frequency C 4 and pipe wall vibration amplitude C5 And preprocess it to obtain standardized raw data. C o =( C 1 , C 2 , C 3 , C 4 , C 5 ); The feature extraction module is based on standardized raw data. C i Three types of high-dimensional feature vectors were calculated, including temperature features, strain features, and vibration features; among them, the temperature features included the rate of change of the temperature gradient along the pipe length. C 6 Rate of change of temperature over time C 7 Strain characteristics include the ratio of circumferential strain to axial strain. C 8 Vibration characteristics include the dominant frequency energy of the vibration signal. C 9 and spectral distribution C 10 Thus, high-dimensional feature vectors are obtained. C i =( C 6 , C 7 , C 8 , C 9 , C 10 ); The intelligent evaluation and prediction module embeds a trained multi-task neural network model, including a shared encoder, a diagnostic decoding head, and a lifetime prediction head; wherein, the shared encoder is configured to receive the high-dimensional feature vector. C i By mapping multiple convolutional and fully connected layers, robust deep semantic feature vectors are extracted. C l The diagnostic decoding head is configured to use the deep semantic feature vector. C l As input, output the blockage diagnosis results, including: probability distribution of blockage type. D 1 and the equivalent water area loss rate, which characterizes the severity. D 2Among them, the probability distribution of siltation types includes fine-grained soil siltation, chemical crystallization siltation, biofilm siltation, and mixed siltation; the lifetime prediction head is configured to use spliced ​​vectors [ C l , D 1 , D 2 , C 11 ] is the input, where C 11 The environmental condition vector includes cumulative runtime, cumulative rainfall, and forecasted rainfall; the lifetime prediction head uses a long short-term memory network to output the remaining effective days. D 3 and recommended maintenance window period D 4 ; The early warning decision module is used to make decisions based on the output of the intelligent assessment and prediction module. D 1 , D 2 , D 3 , D 4 Based on preset clogging level thresholds, the system automatically generates tiered early warning instructions and maintenance strategies, including: Normal state: When D 2 <20% and D 3 When the time exceeds 180 days, only data is recorded, and no alarm is triggered; Key monitoring and early warning: When 20% ≤ D 2 <40% or 60 days ≤ D 3 If the time is less than 180 days, increase the monitoring sampling frequency; Planned maintenance warning: When 40% ≤ D 2 <60% or 7 days ≤ D 3 When the maintenance period is less than 60 days, a planned replacement instruction will be pushed out: prompting maintenance personnel to complete the replacement of the replaceable section of the inner pipe within the recommended maintenance window, and simultaneously outputting the corresponding pipe section and operation instructions; Emergency Response Warning: When D 2 ≥60% or D 3 If the time limit is less than 7 days, an emergency shutdown and replacement command will be immediately triggered, and an audible and visual alarm will be sent through multiple channels to mark the drain pipe as "replace immediately".

[0014] An early warning method, applied to the aforementioned intelligent anti-clogging double-layer replaceable slope drainage pipe, includes the following steps: S1. Data Acquisition: The temperature of the fluid inside the pipe is collected by a distributed fiber optic grating array embedded in the microgrooves of the replaceable section wall of the inner pipe. C 1 Circumferential strain of pipe wall C 2 Pipe wall axial strain C 3 Pipe wall vibration frequency C 4 and pipe wall vibration amplitude C 5 Simultaneously, access environmental operating condition data. C 11 This includes cumulative runtime, cumulative rainfall, and forecasted rainfall; S2. Data Preprocessing: Preprocess the data collected in step S1 to obtain standardized raw data. C o =( C 1 , C 2 , C 3 , C 4 , C 5 Preprocessing includes the following steps: S201 Missing value repair: For individual data missing due to sensor momentary failure or transmission packet loss, linear interpolation based on time series or the mean of nearest neighbor method is used to fill in the missing data. S202. Outlier Removal: Setting a Sliding Time Window T Calculate in the sliding time window T mean of all data points m and standard deviation s According to statistics 3 s The principle is to identify and eliminate those that exceed [the limit]. m ±3 s Outliers within the range; S203, Standardization Processing: For each data value X Perform Z-score normalization transformation, and the transformed data values X norm =( X - m ) / s This transforms the data into a distribution with a mean of 0 and a standard deviation of 1, forming standardized raw data. C o =(C 1 , C 2 , C 3 , C 4 , C 5 ); S3. Feature Extraction: Based on standardized raw data C 0 Calculate the high-dimensional feature vector C i =( C 6 , C 7 , C 8 , C 9 , C 10 This includes the rate of change of the temperature gradient along the pipe length. C 6 Rate of change of temperature over time C 7 The ratio of circumferential strain to axial strain C 8 The dominant frequency energy of the vibration signal C 9 and spectral distribution C 10 ; S4. Blockage Diagnosis: High-dimensional feature vectors... C i The shared encoder of the pre-trained multi-task neural network model in the cloud-based intelligent management platform is used to map deep semantic feature vectors. C l Then C l The input is fed into the diagnostic decoder head, which outputs the blockage diagnosis results, including the probability distribution of blockage types. D 1 and equivalent water flow area loss rate D 2 ; S5. Remaining lifetime prediction: Constructing a splicing vector [ C l , D 1 , D 2 , C 11 Input the remaining effective days into the lifespan prediction header. D 3and recommended maintenance window period D 4 ; S6. Early Warning Decision: The early warning decision module, based on... D 2 , D 3 And preset thresholds, execute a four-level early warning strategy: Normal state: When D 2 <20% and D 3 When the time exceeds 180 days, only data is recorded, and no alarm is triggered; Key monitoring and early warning: When 20% ≤ D 2 <40% or 60 days ≤ D 3 If the time is less than 180 days, increase the monitoring sampling frequency; Planned maintenance warning: When 40% ≤ D 2 <60% or 7 days ≤ D 3 When the maintenance period is less than 60 days, a planned replacement instruction will be pushed out: prompting maintenance personnel to complete the replacement of the replaceable section of the inner pipe within the recommended maintenance window, and simultaneously outputting the corresponding pipe section and operation instructions; Emergency Response Warning: When D 2 ≥60% or D 3 If the time limit is less than 7 days, an emergency shutdown and replacement command will be immediately triggered, and an audible and visual alarm will be sent through multiple channels to mark the drain pipe as "replace immediately".

[0015] Preferably, the cross-sectional profile of the spiral guide rib is semi-circular or trapezoidal, and its protrusion height is... D 1 Set as inner diameter of inner tube D 2 3% to 5%; the pitch of the spiral guide ribs is set to 1 to 3 times the inner diameter.

[0016] Preferably, the multi-task neural network model adopts an end-to-end encoder-decoder architecture, including a shared encoder, a diagnostic decoding head, and a lifetime prediction head; The shared encoder is configured to receive the high-dimensional feature vector. C i And through hierarchical feature extraction, a deep semantic feature vector is mapped. C lThe shared encoder consists of an input layer, a first one-dimensional convolutional layer, a first max-pooling layer, a second one-dimensional convolutional layer, a second max-pooling layer, a global average pooling layer, and a first fully connected layer connected in series. The kernel size of both the first and second one-dimensional convolutional layers is set to 3, the stride is set to 1, and the activation function is a linear rectified function. The global average pooling layer is used to convert the two-dimensional feature map into a one-dimensional feature vector, and the first fully connected layer is used to map the feature vector to a high-dimensional latent space. The diagnostic decoding head is connected in series with the output of the shared encoder and configured to use deep semantic feature vectors. C l The input is a blockage diagnosis result; the diagnosis decoding head consists of a second fully connected layer and a Softmax classification layer, and the output dimension corresponds to the probability distribution of blockage type. D 1 Simultaneously, the diagnostic decoding head introduces a third fully connected layer and a linear activation layer in parallel, used for regression calculation of the equivalent water area loss rate characterizing the severity. D 2 ; The lifetime prediction head is configured to use concatenated vectors[ C l , D 1 , D 2 , C 11 The input is [image of input]. The lifespan prediction head consists of a Long Short-Term Memory (LSTM) network layer, a fourth fully connected layer, and a Dropout regularization layer. The number of hidden units in the LSM layer is set to 50 to capture temporal dependencies in the time series. The dropout rate of the Dropout regularization layer is set to 0.5 to prevent overfitting. The output layer of the lifespan prediction head contains two neurons, which output the remaining effective days. D 3 and recommended maintenance window period D 4 ; Preferably, in the training of the multi-task neural network model, the training dataset consists of a mixture of synthetic simulation dataset and field measured dataset, with the synthetic simulation dataset accounting for 70% and the field measured dataset accounting for 30%. The synthetic simulation dataset is generated by constructing a three-dimensional turbulence model using computational fluid dynamics software to simulate seepage states under different working conditions, specifically including: clear water condition, fine-grained soil clogging condition, chemical crystallization clogging condition, and biofilm clogging condition. For each working condition, by changing the inlet flow velocity, particle concentration, and crystallization layer thickness parameters, corresponding temperature field, strain field, and vibration spectrum data are generated in batches, and clogging type labels and equivalent flow area loss rate labels are automatically labeled. The field measured dataset is collected from laboratory physical model tests and deployed field test slopes. During the data acquisition process, the injection of sediment or chemical solution is manually controlled to simulate the clogging process, and a high-precision flow meter is used as the reference true value to synchronously calibrate the raw signals collected by the distributed fiber optic array. Optimization is performed using a multi-task joint loss function, the loss function is... L = a 1 · L 1 + a 2 · L 2 + a 3 · L 3 ,in L 1 Losses are categorized by type of siltation. L 2 For the regression loss of the water flow area loss rate, L 3 For the remaining lifetime regression loss, a 1 , a 2 , a 3 These are all hyperparameters that balance the weights of each task; The model training employs a phased training strategy, which includes the following steps: The first stage is the pre-training stage, which uses only the synthetic simulation dataset to train the shared encoder and diagnostic decoder head, freezing the lifetime prediction head parameters; the focus of the first stage is optimization. L 1 and L 2 This enables the model to acquire basic physical feature extraction capabilities; the optimizer used is the Adam algorithm, with an initial learning rate of 0.001, a batch size of 64, and 100 training epochs. The second stage is the fine-tuning stage, which involves loading pre-trained weights, unfreezing all network parameters, and performing end-to-end training using a mixed dataset; this stage enables the full multi-task joint loss function. L By adjusting hyperparameters a 1 , a 2 , a 3 To balance classification accuracy, an early stopping strategy is adopted during training, and training is terminated when the validation set loss no longer decreases for 10 consecutive rounds. The third stage is pre-deployment validation, which involves testing the trained model on an independent test set to ensure that its mean absolute percentage error is less than 5%, in order to meet the accuracy requirements of engineering applications.

[0017] In summary, compared with the prior art, the present invention has the following beneficial effects: 1) The dual-layer separation structure solves the pipe replacement problem: By physically separating the permanent buried structure from the functional drainage section, the risk of collapse caused by borehole instability during the replacement of the inner pipe is eliminated. The replaceable section and the fixed section of the inner pipe achieve rapid coaxial connection and radial positioning through a positioning and flow-limiting docking structure. When the inner pipe fails due to chemical crystallization or fine soil blockage, maintenance personnel only need to unscrew the bottom cover of the outlet to directly pull out the replaceable section of the inner pipe from the slope and insert a new pipe to complete the replacement. The entire maintenance process does not require slope excavation or removal of the outer pipe, and causes almost zero disturbance to the slope soil. This avoids the huge amount of engineering work and safety risks of traditional whole-system excavation and replacement, and also overcomes the defects of high-pressure water jet dredging that easily damages the filter layer structure and shortens the life of the new pipe. This design transforms the operation and maintenance of the drainage system throughout its entire life cycle from destructive overhaul to rapid core replacement, significantly reducing operation and maintenance costs and safety risks. 2) Embedded vortex anti-clogging structure enhances the anti-clogging capability of drainage pipes: By setting specific spiral guide ribs inside the pipe, the principle of fluid dynamics is used to force the water flow to generate a rotating flow field. This design produces a significant near-wall shear enhancement effect, increases the boundary layer velocity, and effectively inhibits the static deposition of fine particles and biofilms. At the same time, the low-frequency pressure pulsation induced under heavy rain conditions can actively remove initial deposits. Compared with traditional passive filtration, this structure delays the clogging process from the source and extends the effective service life of the drainage pipe. 3) Multi-physics field fusion sensing and AI diagnosis for accurate identification of clogging mechanisms: Unlike traditional single-parameter flow monitoring, this invention uses a distributed fiber optic grating array to simultaneously collect information from three fields: temperature, strain, and vibration. Combined with a cloud-deployed multi-task neural network model, it can not only calculate the equivalent water flow area loss rate but also accurately distinguish between fine-grained soil clogging, chemical crystallization clogging, biofilm clogging, and mixed clogging. This breakthrough solves the pain point of existing technologies being unable to identify clogging types, providing a scientific basis for targeted maintenance. 4) Improved prediction accuracy under small sample size by pre-trained model based on physical simulation: The model adopts a strategy of mixed training of computational fluid dynamics simulation data and field measured data. By constructing a three-dimensional turbulence model, a large amount of synthetic data covering different clogging conditions is generated for pre-training, and then fine-tuned using measured data. This method effectively solves the problem of poor generalization ability of AI model caused by the scarcity of fault samples in slope engineering, and ensures the reliability of remaining life prediction. 5) Four-level early warning and predictive maintenance to optimize the whole life cycle cost: By using the remaining effective days and maintenance window output by the intelligent assessment and prediction module, the operation and maintenance mode is transformed from reactive repair to predictive maintenance; the four-level early warning mechanism allows managers to intervene at low cost in the early stage of blockage, avoiding the passive situation of not repairing unless it breaks or collapsing after repair in traditional technology, and significantly reducing the operation and maintenance cost and safety risk of the whole life cycle. Attached Figure Description

[0018] Figure 1 This is a module connection diagram of an intelligent anti-clogging double-layer replaceable slope drainage pipe according to an embodiment of the present invention; Figure 2 This is a schematic diagram of the structure of an intelligent anti-clogging double-layer replaceable slope drainage pipe according to an embodiment of the present invention; Figure 3 This is a flowchart illustrating an early warning method according to an embodiment of the present invention; Attached diagram labels: 1-Permanently buried structure, 11-Outer pipe, 12-Outlet bottom cover, 13-Conical top cover, 14-Fixed section of inner pipe, 2-Replaceable section of inner pipe, 3-Edge flow anti-siltation structure, 4-Positioning and flow-limiting docking structure, 41-Radial positioning ring, 42-End face water-blocking ring, 43-Self-aligning guide bearing, 5-Monitoring module, 51-Temperature grating, 52-Strain grating, 53-Vibration grating, 6-Cloud intelligent management platform, 61-Data access module, 62-Feature extraction module, 63-Intelligent assessment and prediction module, 64-Early warning and decision-making module. Detailed Implementation

[0019] The specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. It should be understood that the specific embodiments given herein are for illustration and explanation only and are not intended to limit the present invention.

[0020] The first aspect of this application discloses as follows: Figure 1-Figure 2 The intelligent anti-siltation double-layer replaceable slope drainage pipe shown includes a permanently buried structure 1, an inner layer pipe replaceable section 2, a vortex anti-siltation structure 3, a positioning and flow limiting docking structure 4, a monitoring module 5, and a cloud-based intelligent management platform 6.

[0021] The permanent buried structure 1 includes an outer pipe 11, a bottom cover 12, a cone-shaped top cover 13, and an inner pipe fixing section 14.

[0022] The outer pipe 11 is drilled and implanted into the deep part of the slope in one go during construction, and adopts a mesh-like rigid permeable pipe. In specific implementation, the pipe wall of the outer pipe 11 is divided into a permeable area and an impermeable area along the circumference. The permeable area is distributed at the top of the pipe wall and covers a 240° arc segment, and the pipe wall has evenly distributed small holes. The impermeable area is distributed at the bottom of the pipe wall and covers a 120° arc segment, and the pipe wall has no small holes, serving as a water collection base. The pipe wall of the outer pipe 11 has circumferential curved ribs. The outer pipe 11 serves as the first coarse filtration barrier and undertakes the permanent support function of the slope hole wall to prevent soil collapse. The mesh-like rigid permeable pipe is a mesh-like pipe with uniform small holes formed by hot extrusion at high temperature using high-density polyethylene plastic as the main raw material.

[0023] The water outlet cover 12 is fixed to the port of the outer tube that is exposed on the slope. It has internal threads machined on its inner side and regularly distributed drainage holes on its side wall for draining the collected water and as a mechanical interface for connecting with the replaceable section 2 of the inner tube.

[0024] The cone-shaped top cover 13 is fixed to the port of the outer tube 11 that is deeply buried in the slope. Its shape is bullet-shaped, which facilitates the advancement during construction and installation and prevents the soil and rock from directly intruding into the tube.

[0025] The inner tube fixing section 14 is coaxially inserted into one end of the outer tube 11, which extends into the slope body, and is connected to the cone top cover 13 at that end.

[0026] The replaceable inner tube section 2 is coaxially inserted into the outer tube 11 at one end near the slope surface. The end near the slope surface is connected to the outlet cover 12, and the end extending into the slope body is connected to the fixed inner tube section through the positioning and flow-limiting docking structure 4. The replaceable inner tube section adopts a mesh-like rigid permeable pipe with evenly distributed small holes along the entire length of the pipe wall, and the pipe wall is wrapped with geotextile. As the main functional carrier for drainage and silt prevention, when the replaceable inner tube section 2 fails due to chemical crystallization or fine soil blockage, it can be directly pulled out from the slope for replacement, while the outer tube 11 is retained as permanent support to ensure that the slope soil does not collapse due to replacement operations.

[0027] The eddy current anti-siltation structure 3 is fixedly installed on the inner wall of the inner tube fixed section 14 and the inner tube replaceable section 2. At least three spiral-shaped guide ribs are evenly arranged circumferentially on its inner wall. In specific implementation, the cross-sectional profile of the spiral-shaped guide ribs is semi-circular or trapezoidal, and their protrusion height... D 1 Set as inner diameter of inner tube D 2 The pitch of the spiral guide ribs is set to 1 to 3 times the inner diameter; the vortex anti-siltation structure 3 is integrally injection molded with high-density polyethylene compatible with the pipe material. The geometric configuration of the vortex anti-siltation structure 3 is designed to guide the seepage water flowing through the pipe to generate a continuous rotating flow field. When the seepage water flows through the structure, the characteristics of the rotating flow field are used to achieve a near-wall shear enhancement effect, that is, the rotating flow forces the water to move spirally close to the pipe wall, increasing the tangential flow velocity and shear stress of the pipe wall boundary layer, thereby inhibiting the static deposition of fine particles of silt and biofilm and preventing the formation of mud skin; in addition, under heavy rain conditions, as the seepage flow increases, the vortex street detachment induced by the spiral guide ribs will generate low-frequency pressure pulsation in the pipe. This pulsation can loosen the initial soft deposits that have been attached to the pipe wall, making them easier to be carried away by the subsequent high-speed water flow.

[0028] The positioning and flow-limiting docking structure 4 is located at one end of the slope of the inner tube replaceable section, and is used to achieve coaxial docking and radial limiting between the inner tube replaceable section 2 and the inner tube fixed section 14, and to block the unfiltered water bypass channel at the docking end face; the positioning and flow-limiting docking structure 4 is composed of a radial positioning ring 41, an end face water-blocking ring 42, and a self-aligning guide bearing 43.

[0029] The radial positioning ring 41 is fitted onto the outer wall of the end of the replaceable section 2 of the inner tube. It is integrally vulcanized with elastic rubber and has several continuous arc-shaped positioning protrusions along the axial direction on its outer circumference. The outer diameter of the positioning protrusions is slightly larger than the inner diameter of the outer tube. In the free state, the protrusions and the grid ribs of the inner wall of the outer tube form multi-point elastic contact rather than surface contact sealing, which is used to maintain the coaxiality of the inner tube and prevent uneven wear.

[0030] The end face water-blocking ring 42 is fixed to the end of the replaceable section 2 of the inner tube, and has an annular protrusion on the side facing the fixed section 14 of the inner tube. When the bottom cover of the water outlet is tightened, the annular protrusion is pressed against the end face of the fixed section of the inner tube. The end face water-blocking ring is used to cover the annular joint at the junction of the two sections of the inner tube, physically preventing coarse particles in the annular gap water from directly invading the downstream pipe cavity through the joint, thereby achieving bypass prevention and flow restriction.

[0031] The self-aligning guide bearing 43 is located inside the docking structure and is used to guide the coaxiality during installation.

[0032] The monitoring module 5 employs a distributed fiber optic grating array, pre-embedded in pre-designed microgrooves within the wall of the replaceable inner tube section 2, spaced 0.5m to 1.0m along the tube length. The pre-designed microgrooves are semi-circular grooves formed along the tube wall axially, with a depth of 0.5mm to 1.0mm and a width matching the diameter of the distributed fiber optic grating array, used to accommodate and fix the array. The distributed fiber optic grating array includes a temperature grating 51, a strain grating 52, and a vibration grating 53, wherein the temperature grating 51 is used to sense the temperature of the fluid inside the tube. C 1 It identifies the exothermic or thermal conductivity changes accompanying chemical crystallization; the strain grating 52 is used to sense the circumferential strain of the pipe wall caused by the internal and external pressure difference. C 2 and pipe wall axial strain C 3 Vibration grating 53 is used to sense the pipe wall vibration frequency induced by the eddy current anti-siltation structure. C 4 Pipe wall vibration amplitude C 5 .

[0033] The cloud-based intelligent management platform 6 is deployed on a cloud server and adopts a microservice architecture, including a data access module 61, a feature extraction module 62, an intelligent evaluation and prediction module 63, and an early warning decision module 64.

[0034] Data access module 61 is responsible for receiving data collected by the distributed fiber Bragg grating array, including the temperature of the fluid inside the pipe. C 1 Circumferential strain of pipe wall C 2 Pipe wall axial strain C 3 Pipe wall vibration frequency C 4 and pipe wall vibration amplitude C 5 And preprocess it to obtain standardized raw data. C o =( C 1 , C 2 , C 3 , C 4 , C 5 ).

[0035] The feature extraction module 62 extracts data based on standardized raw data. C iThree types of high-dimensional feature vectors were calculated, including temperature features, strain features, and vibration features; among them, the temperature features included the rate of change of the temperature gradient along the pipe length. C 6 Rate of change of temperature over time C 7 This allows for the identification of localized heating regions caused by the exothermic reaction of chemical crystallization. Simultaneously, it analyzes the attenuation pattern of diurnal temperature fluctuations along the tube side, determining the thickness of the crystalline layer and its thermal barrier effect. Strain characteristics include the ratio of circumferential strain to axial strain. C 8 The increase in equivalent pipe wall stiffness caused by blockage accumulation is determined by the long-term trend of the strain sequence; vibration characteristics include the dominant frequency energy of the vibration signal. C 9 and spectral distribution C 10 Thus, high-dimensional feature vectors are obtained. C i =( C 6 , C 7 , C 8 , C 9 , C 10 ).

[0036] The intelligent evaluation and prediction module 63 embeds a trained multi-task neural network model, including a shared encoder, a diagnostic decoding head, and a lifetime prediction head; wherein, the shared encoder is configured to receive the high-dimensional feature vector. C i By mapping multiple convolutional and fully connected layers, robust deep semantic feature vectors are extracted. C l The diagnostic decoding head is configured to use the deep semantic feature vector. C l As input, output the blockage diagnosis results, including: probability distribution of blockage type. D 1 and the equivalent water area loss rate, which characterizes the severity. D 2 Among them, the probability distribution of siltation types includes fine-grained soil siltation, chemical crystallization siltation, biofilm siltation, and mixed siltation; the lifetime prediction head is configured to use spliced ​​vectors [ C l , D 1 , D 2 , C 11 ] is the input, whereC 11 The environmental condition vector includes cumulative runtime, cumulative rainfall, and forecasted rainfall; the lifetime prediction head uses a long short-term memory network to output the remaining effective days. D 3 and recommended maintenance window period D 4 .

[0037] The early warning decision module 64 is used to make decisions based on the output of the intelligent evaluation and prediction module 63. D 1 , D 2 , D 3 , D 4 Based on preset siltation level thresholds, it automatically generates graded early warning instructions and maintenance strategies.

[0038] The second aspect of the present invention discloses as follows Figure 3 The early warning method shown is applied to the aforementioned intelligent anti-clogging double-layer replaceable slope drainage pipe, and includes the following steps: S1. Data Acquisition: The temperature of the fluid inside the pipe is collected through a distributed fiber optic grating array embedded in the microgrooves of the replaceable section wall of the inner pipe. C 1 Circumferential strain of pipe wall C 2 Pipe wall axial strain C 3 Pipe wall vibration frequency C 4 and pipe wall vibration amplitude C 5 Simultaneously, access environmental operating condition data. C 11 This includes cumulative runtime, cumulative rainfall, and forecasted rainfall.

[0039] S2. Data Preprocessing: Preprocess the data collected in step S1 to obtain standardized raw data. C o =( C 1 , C 2 , C 3 , C 4 , C 5 Preprocessing includes the following steps: S201, Missing value repair: For individual data missing due to sensor momentary failure or transmission packet loss, linear interpolation based on time series or the mean of nearest neighbor method is used to fill in the missing data. S202, Outlier Removal: Setting a sliding time window T Calculate in the sliding time window T mean of all data points m and standard deviation s According to statistics 3 s The principle is to identify and eliminate those that exceed [the limit]. m ±3 s Outliers within the range; S203, Standardization Processing: For each data value X Perform Z-score normalization transformation, and the transformed data values X norm =( X - m ) / s This transforms the data into a distribution with a mean of 0 and a standard deviation of 1, forming standardized raw data. C o =( C 1 , C 2 , C 3 , C 4 , C 5 ).

[0040] S3. Feature Extraction: Based on standardized raw data C 0 Calculate the high-dimensional feature vector C i =( C 6 , C 7 , C 8 , C 9 , C 10 This includes the rate of change of the temperature gradient along the pipe length. C 6 Rate of change of temperature over time C 7 The ratio of circumferential strain to axial strain C 8 The dominant frequency energy of the vibration signal C 9 and spectral distributionC 10 .

[0041] S4. Blockage Diagnosis: High-dimensional feature vectors... C i The shared encoder of the pre-trained multi-task neural network model in the cloud-based intelligent management platform is used to map deep semantic feature vectors. C l Then C l The input is fed into the diagnostic decoder head, which outputs the blockage diagnosis results, including the probability distribution of blockage types. D 1 and equivalent water flow area loss rate D 2 .

[0042] S5. Remaining lifetime prediction: Constructing a splicing vector [ C l , D 1 , D 2 , C 11 Input the remaining effective days into the lifespan prediction header. D 3 and recommended maintenance window period D 4 .

[0043] S6. Early Warning Decision-Making: The early warning decision-making module, based on... D 2 , D 3 And preset thresholds, execute a four-level early warning strategy: Normal state: When D 2 <20% and D 3 When the time exceeds 180 days, only data is recorded, and no alarm is triggered; Key monitoring and early warning: When 20% ≤ D 2 <40% or 60 days ≤ D 3 If the time is less than 180 days, increase the monitoring sampling frequency; Planned maintenance warning: When 40% ≤ D 2 <60% or 7 days ≤ D 3 When the maintenance period is less than 60 days, a planned replacement instruction will be pushed out: prompting maintenance personnel to complete the replacement of the replaceable section of the inner pipe within the recommended maintenance window, and simultaneously outputting the corresponding pipe section and operation instructions; Emergency Response Warning: When D2 ≥60% or D 3 If the time limit is less than 7 days, an emergency shutdown and replacement command will be immediately triggered, and an audible and visual alarm will be sent through multiple channels to mark the drain pipe as "replace immediately".

[0044] In practice, the multi-task neural network model adopts an end-to-end encoder-decoder architecture, including a shared encoder, a diagnostic decoding head, and a lifetime prediction head. The shared encoder is configured to receive the high-dimensional feature vector. C i And through hierarchical feature extraction, a deep semantic feature vector is mapped. C l The shared encoder consists of an input layer, a first one-dimensional convolutional layer, a first max-pooling layer, a second one-dimensional convolutional layer, a second max-pooling layer, a global average pooling layer, and a first fully connected layer connected in series. The kernel size of both the first and second one-dimensional convolutional layers is set to 3, the stride is set to 1, and the activation function is a linear rectified function. The global average pooling layer is used to convert the two-dimensional feature map into a one-dimensional feature vector, and the first fully connected layer is used to map the feature vector to a high-dimensional latent space. The diagnostic decoding head is connected in series with the output of the shared encoder and configured to use deep semantic feature vectors. C l The input is a blockage diagnosis result; the diagnosis decoding head consists of a second fully connected layer and a Softmax classification layer, and the output dimension corresponds to the probability distribution of blockage type. D 1 Simultaneously, the diagnostic decoding head introduces a third fully connected layer and a linear activation layer in parallel, used for regression calculation of the equivalent water area loss rate characterizing the severity. D 2 ; The lifetime prediction head is configured to use concatenated vectors[ C l , D 1 , D 2 , C 11 The input is [image of input]. The lifespan prediction head consists of a Long Short-Term Memory (LSTM) network layer, a fourth fully connected layer, and a Dropout regularization layer. The number of hidden units in the LSM layer is set to 50 to capture temporal dependencies in the time series. The dropout rate of the Dropout regularization layer is set to 0.5 to prevent overfitting. The output layer of the lifespan prediction head contains two neurons, which output the remaining effective days. D 3 and recommended maintenance window period D 4 .

[0045] In practice, the training dataset for the multi-task neural network model consists of a mixture of synthetic simulation datasets and field measured datasets, with the synthetic simulation dataset accounting for 70% and the field measured dataset accounting for 30%. The synthetic simulation dataset is generated by constructing a three-dimensional turbulence model using computational fluid dynamics software to simulate seepage states under different working conditions, including: clear water condition, fine-grained soil clogging condition, chemical crystallization clogging condition, and biofilm clogging condition. For each working condition, by changing the inlet flow velocity, particle concentration, and crystallization layer thickness parameters, corresponding temperature field, strain field, and vibration spectrum data are generated in batches, and clogging type labels and equivalent water flow area loss rate labels are automatically added. The field measured dataset is collected from laboratory physical model experiments and deployed field test slopes. During data acquisition, the injection of sediment or chemical solution is manually controlled to simulate the clogging process, and a high-precision flow meter is used as the reference ground truth to synchronously calibrate the raw signals collected by the distributed fiber optic array. Optimization is performed using a multi-task joint loss function, the loss function is... L = a 1 · L 1 + a 2 · L 2 + a 3 · L 3 ,in L 1 Losses are categorized by type of siltation. L 2 For the regression loss of the water flow area loss rate, L 3 For the remaining lifetime regression loss, a 1 , a 2 , a 3 These are all hyperparameters that balance the weights of each task; The model training employs a phased training strategy, which includes the following steps: The first stage is the pre-training stage, which uses only the synthetic simulation dataset to train the shared encoder and diagnostic decoder head, freezing the lifetime prediction head parameters; the focus of the first stage is optimization. L 1 and L 2 This enables the model to acquire basic physical feature extraction capabilities; the optimizer used is the Adam algorithm, with an initial learning rate of 0.001, a batch size of 64, and 100 training epochs. The second stage is the fine-tuning stage, which involves loading pre-trained weights, unfreezing all network parameters, and performing end-to-end training using a mixed dataset; this stage enables the full multi-task joint loss function. L By adjusting hyperparameters a 1 , a 2 , a 3 To balance classification accuracy, an early stopping strategy is adopted during training, and training is terminated when the validation set loss no longer decreases for 10 consecutive rounds. The third stage is pre-deployment validation, which involves testing the trained model on an independent test set to ensure that its mean absolute percentage error is less than 5%, in order to meet the accuracy requirements of engineering applications.

[0046] The above describes one or more embodiments of the present invention in a relatively specific and detailed manner, but it should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these all fall within the protection scope of the present invention. Therefore, the protection scope of this patent should be determined by the appended claims.

Claims

1. A smart anti-clogging double-layer replaceable slope drainage pipe, characterized in that, This includes a permanent burial structure, a replaceable inner tube section, an eddy current anti-siltation structure, a positioning and flow-limiting docking structure, a monitoring module, and a cloud-based intelligent management platform. The permanently buried structure includes an outer pipe, a bottom cover for water outlet, a cone-shaped top cover, and an inner pipe fixing section. The outer pipe is drilled and inserted deep into the slope during construction, and is a mesh-like rigid permeable pipe. The bottom cover for water outlet is fixed to the port of the outer pipe that protrudes from the slope surface, and has internal threads machined on its inner side and regularly distributed drainage holes on its side wall. The cone-shaped top cover is fixed to the port of the outer pipe that is deeply buried in the slope, and is bullet-shaped. The inner pipe fixing section is coaxially inserted into the inner end of the outer pipe that extends into the slope, and is connected to the cone-shaped top cover at this end. The replaceable section of the inner tube is coaxially inserted into the outer tube at one end near the slope surface. The end near the slope surface is connected to the bottom cover of the outlet, and the end that extends into the slope body is connected to the fixed section of the inner tube through the positioning and flow-limiting docking structure. The replaceable section of the inner tube is a mesh-like rigid permeable pipe with evenly distributed small holes along the entire length of the pipe wall, and the pipe wall is wrapped with geotextile. The vortex anti-siltation structure is fixedly installed on the inner pipe wall of the inner pipe fixed section and the inner pipe replaceable section, and spiral guide ribs are evenly arranged along the circumference of the inner wall. The positioning and flow-limiting docking structure is located at one end of the inner tube replaceable section, extending into the slope. It consists of a radial positioning ring, an end face water-blocking ring, and a self-aligning guide bearing. The radial positioning ring is fitted onto the outer wall of the end of the inner tube replaceable section and is integrally vulcanized from elastic rubber. Several continuous arc-shaped positioning protrusions are provided along the axial direction on its outer circumference. The end face water-blocking ring is fixed to the end of the inner tube replaceable section, and an annular protrusion is provided on the side facing the fixed section of the inner tube. The self-aligning guide bearing is located inside the docking structure. The monitoring module employs a distributed fiber optic grating array, pre-embedded in pre-designed microgrooves within the wall of the replaceable inner tube section, arranged along the tube's length. These pre-designed microgrooves are semi-circular grooves formed along the tube wall's axial direction, with a width matching the diameter of the distributed fiber optic grating array. The distributed fiber optic grating array includes a temperature grating, a strain grating, and a vibration grating, wherein the temperature grating is used to sense the temperature of the fluid inside the tube. C 1 Strain gratings are used to sense the circumferential strain of the pipe wall caused by the pressure difference between the inside and outside. C 2 and pipe wall axial strain C 3 Vibration gratings are used to sense the vibration frequency of the pipe wall induced by the eddy current anti-siltation structure. C 4 Pipe wall vibration amplitude C 5 ; The cloud-based intelligent management platform is deployed on a cloud server and includes a data access module, a feature extraction module, an intelligent evaluation and prediction module, and an early warning and decision-making module. The data access module is responsible for receiving data collected by the distributed fiber Bragg grating array, including the temperature of the fluid inside the pipe. C 1 Circumferential strain of pipe wall C 2 Pipe wall axial strain C 3 Pipe wall vibration frequency C 4 and pipe wall vibration amplitude C 5 And preprocess it to obtain standardized raw data. C o =( C 1 , C 2 , C 3 , C 4 , C 5 ); The feature extraction module is based on standardized raw data. C i Three types of high-dimensional feature vectors were calculated, including temperature features, strain features, and vibration features; among them, the temperature features included the rate of change of the temperature gradient along the pipe length. C 6 Rate of change of temperature over time C 7 Strain characteristics include the ratio of circumferential strain to axial strain. C 8 Vibration characteristics include the dominant frequency energy of the vibration signal. C 9 and spectral distribution C 10 Thus, high-dimensional feature vectors are obtained. C i =( C 6 , C 7 , C 8 , C 9 , C 10 ); The intelligent evaluation and prediction module embeds a trained multi-task neural network model, including a shared encoder, a diagnostic decoding head, and a lifetime prediction head; wherein, the shared encoder is configured to receive the high-dimensional feature vector. C i By mapping multiple convolutional and fully connected layers, robust deep semantic feature vectors are extracted. C l The diagnostic decoding head is configured to use the deep semantic feature vector. C l As input, output the blockage diagnosis results, including: probability distribution of blockage type. D 1 and the equivalent water area loss rate, which characterizes the severity. D 2 Among them, the probability distribution of siltation types includes fine-grained soil siltation, chemical crystallization siltation, biofilm siltation, and mixed siltation; the lifetime prediction head is configured to use spliced ​​vectors [ C l , D 1 , D 2 , C 11 ] is the input, where C 11 The environmental condition vector includes cumulative runtime, cumulative rainfall, and forecasted rainfall; the lifetime prediction head uses a long short-term memory network to output the remaining effective days. D 3 and recommended maintenance window period D 4 ; The early warning decision module is used to generate graded early warning instructions and maintenance strategies based on the output of the intelligent assessment and prediction module and in combination with preset siltation level thresholds.

2. The intelligent anti-clogging double-layer replaceable slope drainage pipe according to claim 1, characterized in that, The cross-sectional profile of the spiral guide ribs is semi-circular or trapezoidal, and their protrusion height is... D 1 Set as inner diameter of inner tube D 2 3% to 5%; the pitch of the spiral guide ribs is set to 1 to 3 times the inner diameter.

3. The intelligent anti-clogging double-layer replaceable slope drainage pipe according to claim 1, characterized in that, The multi-task neural network model adopts an end-to-end encoder-decoder architecture, including a shared encoder, a diagnostic decoder head, and a lifetime prediction head; The shared encoder is configured to receive the high-dimensional feature vector. C i And through hierarchical feature extraction, a deep semantic feature vector is mapped. C l The shared encoder consists of an input layer, a first one-dimensional convolutional layer, a first max pooling layer, a second one-dimensional convolutional layer, a second max pooling layer, a global average pooling layer, and a first fully connected layer connected in series. The diagnostic decoding head is connected in series with the output of the shared encoder and configured to use deep semantic feature vectors. C l The input is a blockage diagnosis result; the diagnosis decoding head consists of a second fully connected layer and a Softmax classification layer, and the output dimension corresponds to the probability distribution of blockage type. D 1 Simultaneously, the diagnostic decoding head leads out a third fully connected layer and a linear activation layer in parallel. The lifetime prediction head is configured to use concatenated vectors[ C l , D 1 , D 2 , C 11 The input is [image of input]; the lifespan prediction head consists of a long short-term memory network layer, a fourth fully connected layer, and a Dropout regularization layer; the output layer of the lifespan prediction head contains two neurons, which output the remaining effective days. D 3 and recommended maintenance window period D 4 .

4. A smart anti-clogging double-layer replaceable slope drainage pipe according to any one of claims 1 to 3, characterized in that, In training the multi-task neural network model, the training dataset consists of a mixture of synthetic simulation datasets and field-tested datasets, with the synthetic simulation dataset accounting for 70% and the field-tested dataset accounting for 30%. A multi-task joint loss function is used for optimization. L = a 1 · L 1 + a 2 · L 2 + a 3 · L 3 ,in L 1 Losses are categorized by type of siltation. L 2 For the regression loss of the water flow area loss rate, L 3 For the remaining lifetime regression loss, a 1 , a 2 , a 3 All of these are hyperparameters used to balance the weights of each task; the model training adopts a phased training strategy.

5. An early warning method, characterized in that, The application of a smart anti-clogging double-layer replaceable slope drainage pipe as described in any one of claims 1 to 4 includes the following steps: S1. Data Acquisition: The temperature of the fluid inside the pipe is collected through a distributed fiber optic grating array embedded in the microgrooves of the replaceable section wall of the inner pipe. C 1 Circumferential strain of pipe wall C 2 Pipe wall axial strain C 3 Pipe wall vibration frequency C 4 and pipe wall vibration amplitude C 5 Simultaneously, access environmental operating condition data. C 11 This includes cumulative runtime, cumulative rainfall, and forecasted rainfall; S2. Data Preprocessing: Preprocess the data collected in step S1 to obtain standardized raw data. C o =( C 1 , C 2 , C 3 , C 4 , C 5 Preprocessing includes the following steps: S201, Missing value repair: For individual data missing due to sensor momentary failure or transmission packet loss, linear interpolation based on time series or the mean of nearest neighbor method is used to fill in the missing data. S202, Outlier Removal: Setting a sliding time window T Calculate in the sliding time window T mean of all data points μ and standard deviation σ According to statistics 3 σ The principle is to identify and eliminate those that exceed [the limit]. μ ±3 σ Outliers within the range; S203, Standardization Processing: For each data value X Perform Z-score normalization transformation, and the transformed data values X norm =( X - μ ) / σ This transforms the data into a distribution with a mean of 0 and a standard deviation of 1, forming standardized raw data. C o =( C 1 , C 2 , C 3 , C 4 , C 5 ); S3. Feature Extraction: Based on standardized raw data C 0 Calculate the high-dimensional feature vector C i =( C 6 , C 7 , C 8 , C 9 , C 10 This includes the rate of change of the temperature gradient along the pipe length. C 6 Rate of change of temperature over time C 7 The ratio of circumferential strain to axial strain C 8 The dominant frequency energy of the vibration signal C 9 and spectral distribution C 10 ; S4. Blockage Diagnosis: High-dimensional feature vectors... C i The shared encoder of the pre-trained multi-task neural network model in the cloud-based intelligent management platform is used to map deep semantic feature vectors. C l Then C l The input is fed into the diagnostic decoder head, which outputs the blockage diagnosis results, including the probability distribution of blockage types. D 1 and equivalent water flow area loss rate D 2 ; S5. Remaining lifetime prediction: Constructing a splicing vector [ C l , D 1 , D 2 , C 11 Input the remaining effective days into the lifespan prediction header and output the remaining effective days. D 3 and recommended maintenance window period D 4 ; S6. Early Warning Decision-Making: The early warning decision-making module, based on... D 2 , D 3 And preset thresholds, execute a four-level early warning strategy: Normal state: When D 2 <20% and D 3 When the time exceeds 180 days, only data is recorded, and no alarm is triggered; Key monitoring and early warning: When 20% ≤ D 2 <40% or 60 days ≤ D 3 If the time is less than 180 days, increase the monitoring sampling frequency; Planned maintenance warning: When 40% ≤ D 2 <60% or 7 days ≤ D 3 When the maintenance period is less than 60 days, a planned replacement instruction will be pushed out: prompting maintenance personnel to complete the replacement of the replaceable section of the inner pipe within the recommended maintenance window, and simultaneously outputting the corresponding pipe section and operation instructions; Emergency Response Warning: When D 2 ≥60% or D 3 If the time limit is less than 7 days, an emergency shutdown and replacement command will be immediately triggered, and an audible and visual alarm will be sent through multiple channels to mark the drain pipe as "replace immediately".

Images