Blockage and drainage abnormity intelligent detection method special for high-concentration self-flow filling pipeline

Abstract

The invention discloses an intelligent detection method for abnormal blockage and leakage special for a high-concentration self-flow filling pipeline, and relates to the technical field of mine filling, the method comprises the following steps: a plurality of monitoring stations are installed on the filling pipeline, and each station is composed of a pressure transmitter and an electromagnetic flowmeter; then monitoring, collecting and storing the data in a wired or wireless manner by using an SCADA system; performing regular learning training on the established deep learning network by using four-dimensional time sequence historical data formed by adjacent sites; finally, in the whole filling system, real-time window data samples of all adjacent monitoring stations are substituted into the deep learning intelligent prediction model, and the running states of pipelines between different stations are obtained. According to the technical scheme provided by the invention, automatic detection, early warning and positioning of blockage and leakage of the complex filling pipeline can be realized, so that maintenance personnel can reach an accident point as soon as possible to replace and process the pipeline, and support is provided for intelligent and safe filling.

Description

Technical Field

[0001] The present invention relates to the technical field of mine filling, and in particular to a special intelligent detection method for blockage and leakage anomalies of a high-concentration gravity filling pipeline. Background Art

[0002] Backfilling is a green underground mining method that can reduce environmental pollution and damage, provide effective ground pressure management, maximize resource recovery rates, and promote the sustainable development of the mining industry. After the filling slurry is evenly mixed at the surface filling station, it is gravity-flowed or pumped to the underground stope through boreholes and pipelines. During the service life of the filling pipe network system, leaks, blockages and other accidents are inevitable. Due to the long laying distance of the filling pipeline, it is difficult to obtain pipeline operation information in a timely manner, troubleshooting is difficult, and accident handling is inefficient. This may lead to tunnel pollution and additional cleanup costs. In severe cases, it may result in major losses such as borehole closures, delays in stope filling operations, and casualties. Therefore, real-time detection of pipeline conditions and rapid response are crucial to reducing mine losses and ensuring safety.

[0003] Currently, most pipelines in my country's mine backfill systems lack automated leak and blockage detection systems, relying primarily on manual inspections along the pipelines. Drawing on successful experiences in the water, natural gas, and oil industries, existing pipeline condition analysis and monitoring methods can be categorized into three types: external sensing, internal sensing, and remote sensing. However, mining backfill pipelines are extremely complex systems. Slurry, a solid-liquid two-phase flow, exhibits greater damping than media such as water, oil, and natural gas. Backfill pipelines exhibit complex characteristics such as turbulent vibration, impact wear, confined spaces, variable geometry, and extreme temperatures, all of which can significantly interfere with signals. Backfill pipelines are made of a wide variety of materials, ranging from ceramics, polyvinyl chloride, and high-density polyethylene plastics to induction-hardened or lightweight steel and manganese steel pipes with thermoplastic coatings. This places higher demands on sensor installation and signal acquisition. High-concentration, self-flowing backfill slurries, characterized by turbulence, partially filled pipes, and large height differences, make pipeline condition monitoring particularly challenging.

[0004] Regarding filling pipeline detection technology, after searching existing related patents, the following summary is provided. First, regarding "pipeline leakage", Patent CN202110000322.6 proposes a long-distance oil and gas pipeline leak detection and positioning method using an acoustic vector sensor, which is suitable for long-distance oil and gas pipeline leak detection and positioning methods; Patent CN202011191138.6 uses three types of electrical signals, namely light, electricity, and sound waves, to establish a pipeline leak detection method and system for detecting pipelines in the oil and gas storage and transportation field; Patent CN202210382198.9 proposes a method for pipeline leak detection, leakage flow rate estimation, and leakage positioning under flow conditions. This method mainly uses the mass conservation and momentum conservation equations to perform spatiotemporal modeling of the pipeline and obtain calculation results; Patent CN201810415336.2 proposes a heating pipeline leak detection system and method based on infrasound and reference points, realizing the calculation of the leak location through the transmission speed of the sound wave; Patent CN202211141247.6 proposes a pipeline leak detection method and equipment. Secondly, regarding the "leakage monitoring method," patent CN202110979611.5 proposes a pipeline leakage detection method based on the fusion of knowledge features and hybrid models. However, this method requires the establishment of a real-time transient pipeline mechanism model of the pipeline fluid, including the establishment of a continuity equation and a momentum equation for the pipeline fluid motion. Then, by extracting the equivalent friction coefficient in the pipeline, prediction is performed using the fusion of knowledge features and hybrid models. Patent CN202110151689.8 proposes a pipeline leakage detection method based on an accelerometer. Patent CN201910715183.8 establishes a time-domain-based pipeline leakage detection algorithm that gives a suitable threshold based on the reconstructed time-domain signal to determine the pipeline status. Finally, regarding "filling pipeline leakage detection," patent CN202110915552.5 proposes a method for rapidly calculating the leakage location of paste-filled long-distance horizontal conveying pipelines. This method, based mainly on the Buckingham formula, compares the resistance loss calculation models during normal filling and leakage to deduce the calculation of pipeline leakage location.

[0005] As can be seen, currently there are few pipeline operation status detection methods specifically tailored to the characteristics of mine filling technology, and in particular, no automated filling pipeline detection methods incorporating machine learning have been found. With the rapid rise of artificial intelligence, big data, and the Internet of Things (IoT) in recent years, they have received widespread attention across various industries. The intelligence level of filling pipeline detection and early warning platforms is still relatively low. It is necessary to utilize precision transmitters and big data technology to improve the automatic detection, early warning, and location intelligence of pipe blockages and leaks, promptly identify pipeline accidents, and issue accurate alarms to enable maintenance personnel to quickly reach the accident site for pipeline replacement and repair. Therefore, it is necessary to develop an accurate and intelligent detection method suitable for complex, high-concentration filling pipeline failures. Summary of the Invention

[0006] The purpose of the present invention is to provide an intelligent detection method for blockage and leakage anomalies specifically for high-concentration gravity-flowing filling pipelines, which is suitable for the automatic monitoring and early warning system of mine filling pipeline transportation processes. It can realize automatic detection, early warning and positioning of blockages and leakages in complex filling pipelines, so that maintenance personnel can reach the accident site as soon as possible to replace and handle the pipelines, providing support for intelligent and safe filling.

[0007] To achieve the above objectives, the present invention provides an intelligent detection method for abnormal blockage and leakage of high-concentration gravity-filled pipelines, the steps of which are as follows:

[0008] S1. Install multiple monitoring stations at intervals on the filling pipeline. Each station contains a pressure transmitter and an electromagnetic flowmeter, and use the SCADA system to collect pressure and flow data at each station.

[0009] S2. Build a historical database based on the collected data. Select the pressure and flow data of two adjacent monitoring stations to form a 4-dimensional time series historical data. Use a sliding window to segment the series to form historical data learning samples.

[0010] S3. Establish a long-short-term deep learning network model based on the long short-term memory network LSTM;

[0011] S4, using the historical data learning samples obtained in S2 to train the model of S3 to obtain an intelligent prediction model;

[0012] S5. During the operation of the filling system, the pressure and flow data of adjacent monitoring stations are collected in real time, and prediction samples are formed according to the same sliding window as S2. The prediction samples are input into the intelligent prediction model obtained in S4, and the judgment result of the current operation status of the pipeline is output, which includes normal, blocked or leaking.

[0013] Preferably, in S1, the SCADA system includes a host computer, a PLC master station, and a PLC acquisition substation. The host computer and the PLC master station are located in the surface control room. The PLC acquisition substation is connected to the pressure transmitter and electromagnetic flowmeter of the monitoring station through a shielded signal line. The PLC acquisition substation is connected to the PLC master station in a wired or wireless manner.

[0014] Preferably, the sampling frequency of the pressure and flow data of the SCADA system is 0.1 to 10 Hz.

[0015] Preferably, in S2, data is stored using a commercial or open source SQL relational database.

[0016] Preferably, in S2, the sliding window length is 5 seconds to 5 minutes.

[0017] Preferably, in S3, the long-short time deep learning network model consists of a data processing layer, an LSTM layer, a batch normalization layer, and a fully connected layer; wherein the data processing layer segments the 4-dimensional time series historical data input to provide standardized samples for LSTM layer learning; the LSTM layer consists of a first LSTM sublayer and a second LSTM sublayer, the first LSTM sublayer expands the 4-dimensional input to a high-dimensional feature space, the second LSTM sublayer fully learns the features in the high-dimensional space, and provides the features to the fully connected layer, each LSTM sublayer is followed by a batch normalization layer to normalize the data transferred between layers, and the fully connected layer maps the high-dimensional features into three categories of outputs, namely normal, blocked pipe, and leakage.

[0018] Preferably, in S4, the intelligent prediction model uses historical data accumulated during system operation to perform regular learning and updating.

[0019] According to the specific embodiments provided by the present invention, the present invention discloses the following technical effects: the intelligent prediction model adopts a deep learning network, which can make full use of production historical data and update learning regularly. The model prediction accuracy is high. It is not only easy to integrate into the automatic control system, but also can realize unmanned intelligent inspection of filling slurry pipeline transportation, reduce pipeline maintenance costs, and improve the intelligence level of filling.

[0020] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0022] Figure 1 This is a flow chart of an embodiment of an intelligent detection method for abnormal blockage and leakage in a high-concentration gravity-filled pipeline according to the present invention;

[0023] Figure 2 This is a schematic diagram of the distribution of a mine filling pipeline and monitoring stations according to an embodiment of the present invention;

[0024] Figure 3 This is a diagram of the long-short time deep learning network model architecture of an embodiment of the present invention. DETAILED DESCRIPTION

[0025] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0026] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0027] Example

[0028] Aiming at the shortcomings of existing technologies, we make full use of various machine learning and artificial intelligence algorithms to develop special algorithms for filling pipeline blockage and leakage detection, which can be integrated into the intelligent unmanned inspection system. This is an important way to improve filling automation and intelligence and reduce filling operation costs. The present invention provides a special intelligent detection method for blockage and leakage anomalies in high-concentration gravity filling pipelines, such as Figure 1 As shown, the method steps include:

[0029] S1. Taking a high-concentration gravity filling mine in China as an example, a high-concentration filling slurry with a concentration of 78% to 81% is prepared. The inner diameter of the filling pipe is 150 mm and the flow rate is 200-250 m3 / h. Multiple monitoring stations are installed on a certain middle section of the filling pipe network, such as Figure 2 As shown, each station consists of a Rosemount 3051 pressure transmitter (range 0~6 MPa) and a Yamatake MagneW 3000 intelligent electromagnetic flowmeter.

[0030] Optical fiber is used as the data transmission medium. The pressure and flow meter signals of each monitoring substation are connected to the PLC acquisition module through shielded signal cables. The PLC acquisition module is connected to the PLC master station in the surface control room through optical fiber. The PLC master station is connected to the host computer through an RJ-45 network cable. The SCADA system is used on the host computer to realize data acquisition.

[0031] S2. Build a historical database based on the collected data. Select the pressure and flow data of two adjacent monitoring stations to form a 4-dimensional time series historical data. Use a sliding window to segment the sequence to form historical data learning samples. The sampling frequency of the SCADA system is 2 Hz, and the data is stored in a Microsoft SQL database.

[0032] S3. Establish a long-short-term deep learning network model based on the long short-term memory network LSTM; the model framework is as follows Figure 3As shown in the figure, it mainly consists of a data processing layer, an LSTM layer, a batch normalization layer, and a fully connected layer. The data processing layer segments the 4-dimensional time series historical data input to provide standardized samples for LSTM layer learning. The LSTM layer consists of the first LSTM sublayer and the second LSTM sublayer. The first LSTM sublayer expands the 4-dimensional input to a high-dimensional feature space. The second LSTM sublayer fully learns the features in the high-dimensional space and provides the features to the fully connected layer. Each LSTM sublayer is followed by a batch normalization layer to normalize the data transferred between layers, which can effectively improve the model learning efficiency and reduce overfitting. The fully connected layer maps the high-dimensional features into three categories of output, namely normal, blocked pipe, and leakage. The model is written in Python. Figure 3 where N is the number of samples, p1 and p2 represent the pressure values ​​of two adjacent pressure gauges; q1 and q2 represent the monitoring values ​​of two adjacent flow meters; the superscript 1-120 represents the time segment; k, h and c represent the internal parameters of the network; and Z represents three output states.

[0033] S4. Use the historical data learning samples obtained in S2 to train the model in S3 to obtain an intelligent prediction model. In the model's data processing layer, the time window is set to 1 minute, that is, the pipeline operation status is judged based on data within 1 minute. Both positive and negative samples are production data in the database. To fully utilize the negative sample data of plugging and leaking, the time window step size of negative samples is set to 5 seconds, and the window step size of positive samples is set to 1 minute. The total number of samples generated is randomly divided into training, validation, and test sets in a ratio of 6:2:2. The validation set is used for model hyperparameter optimization, the training set is used for learning, and the test set is used for model verification.

[0034] The model's hyperparameters are composed of two parts: constant and variable. The constant parameters include the activation function of the LSTM sub-unit and the activation function type of the fully connected layer. The former maintains the standard sigmoid and tanh functions, while the latter adopts the softplus function suitable for multi-classification tasks. The variable parameters include the number of hidden layer units in the two LSTM sub-layers and the optimizer algorithm. The value range of the hyperparameters is shown in Table 1. The grid search algorithm is used to optimize the hyperparameters, and the parameters with the highest accuracy are regarded as the best result.

[0035] Table 1 Hyperparameter value range and optimization results

[0036] Hyperparameter optimization project Optimization parameter range Best results The number of hidden units in the first LSTM layer [40, 60, 120, 240, 360, 480] 240 Number of hidden units in the second LSTM sub-layer [4, 8, 16, 20, 40, 80, 120] 16 Optimizer ['Adam', 'RMSprop', 'Adagrad'] Adam

[0037] In order to prevent overfitting and underfitting during model training, we adopted techniques such as increasing the number of negative samples, early stopping technology, dynamically reducing the learning rate, and batch normalization. Cross entropy was used as the loss function in training, as shown in the following formula (1);

[0038] (1)

[0039] Where: p is the hot single-hot encoding of the true value, such as [1, 0, 0]; q is the predicted value encoding; n is 3 states; H is a bit.

[0040] S5. During the operation of the filling system, the pressure and flow data of adjacent monitoring stations are collected in real time, and prediction samples are formed according to the same sliding window as S2. The prediction samples are input into the intelligent prediction model obtained in S4, and the judgment result of the current operation status of the pipeline is output, which includes normal, blocked or leaking.

[0041] The model is evaluated using a combination of accuracy and Area under the Curve (AUC) curve. Based on the prediction results of the model for positive and negative samples, four possibilities can be obtained as shown in Table 2.

[0042] Table 2 Confusion Matrix of Prediction Results

[0043] Positive samples Negative samples True True Positive (TP) True Negative (TN) False False Positive (FP) False Negative (FN)

[0044] TP represents the number of positive samples accurately predicted as positive; TN represents the number of negative samples accurately predicted as negative; FP represents the number of negative samples incorrectly predicted as positive; FN represents the number of positive samples incorrectly predicted as negative.

[0045] (2)

[0046] (3)

[0047] (4)

[0048] Accuracy represents the accuracy rate; True Positive Rate (TPR) represents the ratio of correctly classified positive samples to the total number of positive samples; and False Positive Rate (FPR) represents the ratio of incorrectly classified negative samples to the total number of negative samples. AUC describes classification performance by measuring the area under the Receiver Operating Characteristic (ROC) curve (TPR vs. FPR), ranging from 0.5 to 1.0. A larger area under the ROC curve indicates better classification performance. The final model achieved a training accuracy of 98%, with an area under the ROC curve of 0.99.

[0049] The model was evaluated using production test samples. The overall prediction accuracy of the LSTM deep model was 98.31%, accurately identifying normal operating conditions. The false positive rate for abnormal pipeline blockages and leaks was only 3.21%. This demonstrates that this method has high accuracy for identifying complex filling pipeline conditions and a very low false positive rate for abnormal conditions.

[0050] The remaining technical features of the above embodiments can be flexibly selected by those skilled in the art to meet specific practical needs. However, it will be apparent to those skilled in the art that these specific details are not required to practice the present invention. In other instances, to avoid obscuring the present invention, well-known components, structures, or parts are not described in detail, and are therefore within the scope of protection of the technical solutions claimed in the claims.

[0051] Modifications and variations made by those skilled in the art without departing from the spirit and scope of the present invention are intended to be within the scope of the claims appended hereto. In the foregoing description, numerous specific details are set forth to provide a thorough understanding of the present invention. However, it will be apparent to those skilled in the art that these specific details are not necessarily required to practice the present invention. In other instances, well-known techniques, such as specific construction details, operating conditions, and other technical requirements, are not described in detail to avoid obscuring the present invention.

[0052] This document uses specific examples to illustrate the principles and implementation methods of the present invention. The above examples are only intended to help understand the method and core concept of the present invention. At the same time, those skilled in the art will find that the specific implementation methods and application scopes may vary based on the concept of the present invention. In summary, the contents of this specification should not be construed as limiting the present invention.

Claims

1. An intelligent detection method for abnormal blockage and leakage of high-concentration gravity-filled pipelines, characterized in that: Here are the steps: S1. Install multiple monitoring stations at intervals on the filling pipeline. Each station contains a pressure transmitter and an electromagnetic flowmeter, and use the SCADA system to collect pressure and flow data at each station. S2. Build a historical database based on the collected data. Select the pressure and flow data of two adjacent monitoring stations to form a 4-dimensional time series historical data. Use a sliding window to segment the series to form historical data learning samples. S3. Establish a long-short-term deep learning network model based on the long short-term memory network LSTM; S4, using the historical data learning samples obtained in S2 to train the model of S3 to obtain an intelligent prediction model; S5. During the operation of the filling system, the pressure and flow data of adjacent monitoring stations are collected in real time, and prediction samples are formed according to the same sliding window as S2. The prediction samples are input into the intelligent prediction model obtained in S4, and the judgment result of the current operation status of the pipeline is output, which includes normal, blocked or leaking.

2. The intelligent detection method for abnormal blockage and leakage of a high-concentration gravity-filled pipeline according to claim 1 is characterized by: In S1, the SCADA system includes a host computer, a PLC master station, and a PLC acquisition substation. The host computer and the PLC master station are located in the surface control room. The PLC acquisition substation is connected to the pressure transmitter and electromagnetic flowmeter of the monitoring station through a shielded signal line. The PLC acquisition substation is connected to the PLC master station in a wired or wireless manner.

3. The intelligent detection method for abnormal blockage and leakage in a high-concentration gravity-filled pipeline according to claim 2 is characterized by: The sampling frequency of the pressure and flow data of the SCADA system is 0.1 to 10 Hz.

4. The intelligent detection method for abnormal blockage and leakage in a high-concentration gravity-filled pipeline according to claim 1 is characterized by: In S2, data is stored using commercial or open source SQL relational databases.

5. The intelligent detection method for abnormal blockage and leakage in a high-concentration gravity-filled pipeline according to claim 1 is characterized by: In S2, the sliding window length is 5 seconds to 5 minutes.

6. The intelligent detection method for abnormal blockage and leakage in a high-concentration gravity-filled pipeline according to claim 1 is characterized by: In S3, the long-short time deep learning network model consists of a data processing layer, an LSTM layer, a batch normalization layer, and a fully connected layer; the data processing layer segments the 4-dimensional time series historical data input to provide standardized samples for LSTM layer learning; the LSTM layer consists of a first LSTM sublayer and a second LSTM sublayer, the first LSTM sublayer expands the 4-dimensional input to a high-dimensional feature space, the second LSTM sublayer fully learns the features in the high-dimensional space, and provides the features to the fully connected layer, each LSTM sublayer is followed by a batch normalization layer to normalize the data transferred between layers, and the fully connected layer maps the high-dimensional features into three categories of outputs, namely normal, blocked pipe, and leakage.

7. The intelligent detection method for abnormal blockage and leakage in a high-concentration gravity-filled pipeline according to claim 1 is characterized by: In S4, the intelligent prediction model uses historical data accumulated during system operation to perform regular learning and updating.

Images