High-concentration slurry pipeline blockage risk probability analysis and automatic slurry discharge control method

Abstract

The invention belongs to the technical field of automatic control, and particularly relates to a high-concentration slurry pipeline blockage risk probability analysis and automatic slurry discharge control method. The method comprises the following steps: 1, mounting an acoustic emission sensor array on the outer wall of a pipeline along the axial direction of the high-concentration slurry conveying pipeline, and arranging a plurality of acoustic emission sensors on each measurement section; 2, constructing a pipeline blockage state hidden Markov model, defining four hidden states including a normal flow state, a light deposition state, a serious deposition state and a blockage precursor state, and calculating a pipeline blockage risk probability according to a hidden state identification result; and 3, selecting a corresponding pulse slurry discharge control mode according to the hidden state recognition result. The operation safety and reliability of the high-concentration slurry pipeline conveying system are remarkably improved.

Description

Technical Field

[0001] This invention relates to industrial data processing, belonging to the field of automatic control technology, and specifically to a method for analyzing the risk of blockage in high-concentration slurry pipelines and for automatic slurry discharge control. Background Technology

[0002] High-concentration slurry pipeline transportation technology is widely used in fields such as mine backfilling, tailings disposal, coal hydraulic transportation, dredging projects, and chemical material transportation. Compared with traditional belt conveyors and vehicle transportation, pipeline transportation has significant advantages such as continuous operation, small footprint, low environmental pollution, and low operating costs, and has become an important method for long-distance transportation of solid materials. However, the most prominent problem faced by high-concentration slurry in pipeline transportation is pipeline blockage. Since the density of solid particles in slurry is usually greater than that of the carrying medium, the particles tend to settle to the bottom of the pipeline under the action of gravity. When the conveying velocity is lower than the critical settling velocity, the slurry concentration is too high, or the particle size distribution is unreasonable, solid particles will gradually deposit at the bottom of the pipeline, forming a sediment layer. The continuous growth of the sediment layer will lead to a reduction in the effective flow cross-section and an increase in conveying resistance, eventually causing pipeline blockage accidents. Pipeline blockage not only causes production interruptions and economic losses, but in severe cases, it may also lead to pipeline overpressure rupture, causing safety accidents and environmental pollution.

[0003] To address the problem of slurry pipeline blockage, existing technologies mainly focus on two aspects: blockage detection and prevention / control. In terms of blockage detection, differential pressure monitoring is the most widely used technique. This method involves placing pressure sensors along the pipeline to monitor changes in the pressure difference between adjacent measuring points. When deposits occur inside the pipeline, local resistance increases, and the pressure difference rises accordingly. Differential pressure monitoring is simple in principle and low in cost, but it suffers from insufficient sensitivity. Changes in pressure difference are a macroscopic manifestation of deposits developing to a certain extent; when a significant anomaly in pressure difference occurs, the deposits are often already quite severe, leaving a very limited window for intervention. Flow monitoring determines the blockage status by detecting changes in pipeline flow rate. Its limitations are similar to those of differential pressure monitoring, both being reactive rather than early warning systems. In terms of prevention / control, increasing the delivery velocity is the most direct method, but excessively high velocities exacerbate pipeline wear and increase energy costs. Periodic pipeline cleaning is another common method, using pigs to remove deposits from the pipes. However, cleaning operations require interrupting normal delivery, and there is a risk of pigs getting stuck in long-distance pipelines. Pulsed delivery technology has attracted attention in recent years. This technology generates periodic flow velocity fluctuations in pipelines, utilizing the shearing effect of the pulsating flow field to suppress particle sedimentation or remove existing deposits. Existing pulse delivery systems typically use pulse parameters with fixed frequency and amplitude, lacking the ability to sense the actual deposition state in the pipeline and unable to adaptively adjust the pulse strategy according to the degree of deposition. When the deposition is light, excessively strong pulses can cause unnecessary energy consumption and equipment wear; when the deposition is severe, insufficient pulse intensity is insufficient to effectively remove the deposits. Summary of the Invention

[0004] The main objective of this invention is to provide a method for analyzing the probability of blockage in high-concentration slurry pipelines and for automatic slurry discharge control, which significantly improves the operational safety and reliability of high-concentration slurry pipeline transportation systems.

[0005] To address the aforementioned technical problems, this invention provides: a method for analyzing the probability of blockage in high-concentration slurry pipelines and for automatic slurry discharge control, the method comprising: Step 1: Install an acoustic emission sensor array on the outer wall of the high-concentration slurry delivery pipeline along its axial direction, with multiple acoustic emission sensors arranged at each measurement section; perform time-frequency analysis on the acoustic emission signals to extract the frequency band energy vector, perform envelope detection processing on the acoustic emission signals, and count the acoustic emission events; calculate the directional energy difference vector based on the frequency band energy vectors of each acoustic emission sensor at the same measurement section, and form a section energy feature vector; calculate the event count range and the total number of events at the section based on the acoustic emission event count values; concatenate the section energy feature vector, event count vector, event count range, and total number of events at the section to form the pipeline flow state observation vector; Step 2: Construct a hidden Markov model for pipeline blockage, defining four hidden states: normal flow state, light deposition state, severe deposition state, and pre-blockage state. Establish observation feature judgment conditions for each hidden state; perform hidden state sequence inference based on pipeline flow observation vector and state transition permission matrix to obtain hidden state identification results, and calculate pipeline blockage risk probability based on hidden state identification results. Step 3: Select the corresponding pulse slurry discharge control mode based on the latent state identification result. When the latent state identification result is a light deposition state, start the high-frequency small pulse slurry discharge mode. When the latent state identification result is a severe deposition state, start the low-frequency large pulse slurry discharge mode. When the latent state identification result is a pre-blockage state, start the frequency sweep pulse slurry discharge mode to determine the resonance frequency of the deposition layer and continue to work at the resonance frequency of the deposition layer.

[0006] Furthermore, in step one, four acoustic emission sensors are arranged at each measurement section. The four acoustic emission sensors are located at the top, bottom, left and right positions of the pipe section, respectively. The axial spacing between adjacent measurement sections is five times the inner diameter of the pipe. The operating frequency range of the acoustic emission sensors is 100kHz to 1MHz, and the sampling frequency is 2MHz.

[0007] Furthermore, in step one, the specific process of time-frequency analysis is as follows: the continuously acquired acoustic emission signal is divided into analysis windows with a time length of 50ms. A short-time Fourier transform is performed on the acoustic emission signal in each analysis window to obtain a time-frequency spectrum. In the time-frequency spectrum, the frequency axis is divided into 8 equal-width frequency bands, and the energy value in each frequency band is counted to form an 8-dimensional frequency band energy vector. The specific process of envelope detection is as follows: the amplitude threshold is set to 3 times the root mean square value of the background noise. When the envelope signal exceeds the amplitude threshold, it is recorded as an acoustic emission event. The count value of acoustic emission events in a single analysis window is counted.

[0008] Furthermore, in step one, the directional energy difference vector includes a vertical energy difference vector and a horizontal energy difference vector; the vertical energy difference vector is the difference between the frequency band energy vector of the acoustic emission sensor located at the top position and the frequency band energy vector of the acoustic emission sensor located at the bottom position; the horizontal energy difference vector is the difference between the frequency band energy vector of the acoustic emission sensor located on the left position and the frequency band energy vector of the acoustic emission sensor located on the right position; the frequency band energy vectors of the acoustic emission sensors located at the top position, the bottom position, the left position, and the right position, the vertical energy difference vector, and the horizontal energy difference vector are sequentially concatenated to form a 48-dimensional cross-sectional energy feature vector; the 48-dimensional cross-sectional energy feature vector, the 4-dimensional event count vector, the event count range, and the total number of cross-sectional events are concatenated to form a 54-dimensional pipeline flow state observation vector.

[0009] Furthermore, in step two, the observation characteristic determination conditions include: For normal flow conditions, the observation characteristic determination conditions are: the event count range is less than 20% of the total number of events in the cross-section, and the absolute value of each component in the vertical energy difference vector is less than 15% of the total energy of the corresponding frequency band, and the absolute value of each component in the horizontal energy difference vector is less than 15% of the total energy of the corresponding frequency band; For slightly deposited conditions, the observation characteristic determination conditions are: the event count range is greater than or equal to 20% and less than 50% of the total number of events in the cross-section, or at least two components in the vertical energy difference vector have an absolute value greater than or equal to 15% of the total energy of the corresponding frequency band and less than the total energy of the corresponding frequency band. 40%; The observation characteristics of severe deposition state are determined as follows: at least one of the four acoustic emission sensors has an acoustic emission event count value less than 10% of the total number of cross-sectional events, and the sum of the energy of the four frequency bands in the high-frequency band of the corresponding frequency band energy vector of that sensor is less than 30% of the sum of the energy of the four frequency bands in the low-frequency band; The observation characteristics of blockage precursor state are determined as follows: the change in the total number of cross-sectional events between five consecutive adjacent analysis windows exceeds 100% of the total number of cross-sectional events in the previous analysis window, or more than half of the components in the 48-dimensional cross-sectional energy feature vector show a monotonically decreasing trend in three consecutive adjacent analysis windows and the cumulative decrease exceeds 60% of the initial value.

[0010] Furthermore, in step two, the state transition permission matrix is ​​a 4x4 matrix, with matrix elements taking values ​​of 0 or 1; the normal flow state allows transitions to the normal flow state and the slightly deposited state, the slightly deposited state allows transitions to the normal flow state, the slightly deposited state and the severely deposited state, the severely deposited state allows transitions to the slightly deposited state, the severely deposited state and the pre-blockage state, and the pre-blockage state allows transitions to the severely deposited state and the pre-blockage state. The matrix elements corresponding to the above-mentioned permitted transitions take values ​​of 1, and the remaining matrix elements take values ​​of 0.

[0011] Furthermore, in step two, the specific process of inferring the hidden state sequence is as follows: Based on the pipeline flow state observation vector, it is sequentially checked whether the observation feature judgment conditions of the four hidden states are met, and a current observation condition satisfaction vector is generated. The current observation condition satisfaction vector is a 4-dimensional vector. When the observation feature judgment condition of a certain hidden state is met, the corresponding position is set to 1, and when it is not met, the corresponding position is set to 0. The initial hidden state is set to the normal flow state. For each new analysis window, the state transition permission matrix is ​​queried based on the hidden state of the previous moment to obtain the set of hidden states that can be reached at the current moment. In the set of hidden states that can be reached, the hidden state with the corresponding position of the current observation condition satisfaction vector is selected as 1. When there are multiple hidden states that meet the conditions, the hidden state with the largest state number is selected as the hidden state identification result. When the current observation condition satisfaction vector corresponding to all hidden states in the set of allowed hidden states is 0, the hidden state of the previous moment remains unchanged.

[0012] Furthermore, in step two, the calculation process for the pipeline blockage risk probability is as follows: A risk level lookup table is established, with normal flow state corresponding to risk level value 0, slight deposition state corresponding to risk level value 1, severe deposition state corresponding to risk level value 2, and pre-blockage state corresponding to risk level value 3; the hidden state identification result sequence of the most recent 10 consecutive analysis windows is obtained, and the corresponding 10 risk level values ​​are obtained by querying the risk level lookup table; the maximum value among the 10 risk level values ​​is added to the arithmetic mean of the 10 risk level values, and then divided by 6 to obtain the pipeline blockage risk probability.

[0013] Furthermore, in step three, the pulse frequency of the high-frequency small-pulse slurry discharge mode is 15Hz and the duty cycle is 30%; the pulse frequency of the low-frequency large-pulse slurry discharge mode is 2Hz and the duty cycle is 50%; the sweep frequency pulse slurry discharge mode has a sweep frequency start frequency of 1Hz, a sweep frequency end frequency of 20Hz, and a frequency step value of 0.5Hz. Pulses are generated at each frequency and the system operates for 10 complete cycles. During this period, the average amplitude of the acoustic emission signal is recorded, and the frequency point with the largest average amplitude of the acoustic emission signal is determined as the resonant frequency of the deposition layer. Pulses are continuously generated at the resonant frequency of the deposition layer with a duty cycle set to 40%.

[0014] Furthermore, in step three, a pulse slurry discharge actuator is configured. The pulse slurry discharge actuator includes an accumulator, a pulse control valve, and a pulse generator controller. The pre-charge pressure of the accumulator is set to 1.5 times the normal operating pressure of the pipeline. The pulse control valve is an electromagnetic quick-opening valve with an opening response time of less than 30ms. During the execution of pulse slurry discharge control, the hidden state identification results and the pipeline blockage risk probability are continuously updated. When the pipeline blockage risk probability decreases to below 0.3 and the hidden state identification results return to the normal flow state and are maintained for 20 analysis windows, the current pulse slurry discharge control mode is exited and the normal continuous conveying condition is restored.

[0015] The method for analyzing the risk of blockage in high-concentration slurry pipelines and automatically controlling slurry discharge of the present invention has the following beneficial effects: The present invention provides a method for high-concentration slurry pipeline blockage risk probability analysis and automatic slurry discharge control, which achieves comprehensive perception of pipeline flow state through a multi-point arrangement of acoustic emission sensor arrays. Multiple acoustic emission sensors are deployed at each measurement cross-section to detect particle collision signals from different directions, accurately capturing the spatial distribution characteristics of the slurry within the pipeline cross-section. By calculating the energy differences between sensors at different locations, flow asymmetry caused by deposition can be sensitively identified, exhibiting higher spatial resolution and early warning sensitivity compared to traditional single-point detection methods. The joint extraction of frequency band energy vectors and acoustic emission event counts characterizes flow state features from both frequency domain characteristics and time domain statistics, forming a multi-dimensional pipeline flow state observation vector containing rich state information, laying a data foundation for subsequent accurate state identification.

[0016] This invention employs a Hidden Markov Model (HMM) to model pipeline blockage states, dividing the blockage evolution process into four latent states: normal flow, light deposition, severe deposition, and pre-blockage. It meticulously depicts the complete evolution path from normal transport to impending blockage. By setting a state transition permission matrix, the physical laws of the blockage process are explicitly encoded into the model, ensuring that the state inference results conform to the objective laws of deposition development and avoiding unreasonable state jumps caused by observation noise. The pipeline blockage risk probability calculated based on the latent state identification results comprehensively considers peak risk and average risk, enabling rapid response to sudden high-risk events and reflecting the persistence of the risk state, providing a reliable basis for graded early warning and precise decision-making.

[0017] The multi-mode adaptive pulse slurry discharge control strategy of this invention automatically matches the optimal slurry discharge mode based on the latent state identification results. For mild deposition, a high-frequency small-pulse mode is used to prevent further particle settling through continuous flow velocity disturbance; for severe deposition, a low-frequency large-pulse mode is used to strip the deposit layer from the pipe wall through strong shearing; for pre-blockage states, a frequency-sweeping pulse mode is used to maximize slurry discharge efficiency by searching for the resonant frequency of the deposit layer. This graded response strategy ensures slurry discharge efficiency while avoiding energy waste and equipment wear caused by excessive intervention. The entire system forms a complete closed loop of perception, identification, decision-making, and execution. During the slurry discharge process, the state identification results and risk probabilities are continuously updated, enabling real-time monitoring of the slurry discharge effect and dynamic adjustment of the control strategy, significantly improving the operational safety and reliability of high-concentration slurry pipeline transportation systems. Attached Figure Description

[0018] Figure 1 A schematic diagram of the distribution of pipeline flow state observation vectors in the feature space and the hidden state discrimination boundary provided in an embodiment of the present invention; Figure 2 A schematic diagram illustrating the time-series evolution of the pipeline blockage latent state identification results provided in an embodiment of the present invention; Figure 3 This is a schematic diagram illustrating the evolution curve of the probability of pipeline blockage risk provided in an embodiment of the present invention. Detailed Implementation

[0019] The method of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments.

[0020] Risk probability analysis of high-concentration slurry pipeline blockage and automatic slurry discharge control method, the method includes: Step 1: Install an acoustic emission sensor array on the outer wall of the high-concentration slurry delivery pipeline along its axial direction, with multiple acoustic emission sensors arranged at each measurement section; perform time-frequency analysis on the acoustic emission signals to extract the frequency band energy vector, perform envelope detection processing on the acoustic emission signals, and count the acoustic emission events; calculate the directional energy difference vector based on the frequency band energy vectors of each acoustic emission sensor at the same measurement section, and form a section energy feature vector; calculate the event count range and the total number of events at the section based on the acoustic emission event count values; concatenate the section energy feature vector, event count vector, event count range, and total number of events at the section to form the pipeline flow state observation vector; Step 2: Construct a hidden Markov model for pipeline blockage, defining four hidden states: normal flow state, light deposition state, severe deposition state, and pre-blockage state. Establish observation feature judgment conditions for each hidden state; perform hidden state sequence inference based on pipeline flow observation vector and state transition permission matrix to obtain hidden state identification results, and calculate pipeline blockage risk probability based on hidden state identification results. Step 3: Select the corresponding pulse slurry discharge control mode based on the latent state identification result. When the latent state identification result is a light deposition state, start the high-frequency small pulse slurry discharge mode. When the latent state identification result is a severe deposition state, start the low-frequency large pulse slurry discharge mode. When the latent state identification result is a pre-blockage state, start the frequency sweep pulse slurry discharge mode to determine the resonance frequency of the deposition layer and continue to work at the resonance frequency of the deposition layer.

[0021] When high-concentration slurry is transported in a pipeline, solid particles continuously collide with the pipe wall. The elastic waves generated by these collisions propagate along the pipe wall in the form of acoustic emission signals. When the slurry is in a normal, uniformly suspended state, particle collisions exhibit a statistically uniform distribution across the pipe wall. However, when slurry deposition begins, particle movement slows down in the bottom region of the pipeline, and the collision frequency and intensity decrease accordingly. This spatial non-uniformity can be captured by multi-point acoustic emission detection. Based on this physical mechanism, this embodiment employs an acoustic emission sensor array to monitor the pipeline flow in real time.

[0022] An acoustic emission sensor array is arranged axially on the outer wall of the high-concentration slurry delivery pipeline. During installation, the number and location of measurement sections are determined based on the total length of the pipeline and the distribution of areas at risk of blockage. For a slurry delivery pipeline with a total length of 500 meters, when the pipeline inner diameter is 200 mm, the axial spacing between adjacent measurement sections is set to five times the pipeline inner diameter, i.e., 1000 mm. This spacing takes into account the attenuation characteristics of the acoustic emission signal in the pipe wall: too small a spacing will cause mutual interference between signals from adjacent sections, while too large a spacing may miss local deposition areas. In practical engineering, for critical locations prone to deposition, such as elbows, diameter changes, and valves, the spacing between measurement sections can be appropriately reduced to three times the pipeline inner diameter.

[0023] Four acoustic emission sensors are arranged at each measurement section, installed at the top, bottom, left, and right sides of the pipe section. The four sensors are evenly distributed at 90-degree angles along the circumference; this symmetrical arrangement effectively distinguishes the spatial distribution of the slurry within the pipe section. The sensors at the top and bottom are used to detect vertical flow differences, which is particularly important in scenarios where particles settle under gravity. The sensors at the left and right sides are used to detect horizontal flow differences, identifying flow deviations caused by pipe inclination or centrifugal force at bends. The acoustic emission sensors are fixed to the outer surface of the pipe wall using magnetic adsorption or welding, and a coupling agent is coated between the sensor and the pipe wall to ensure effective sound wave transmission.

[0024] The acoustic emission sensor operates within a frequency range of 100 kHz to 1 MHz. The lower limit of 100 kHz avoids low-frequency mechanical vibration interference commonly found in piping systems, while the upper limit of 1 MHz covers the high-frequency elastic wave components generated by particle collisions. For solid particles with diameters ranging from 0.1 mm to 5 mm, the dominant frequency of the acoustic emission signal generated by their collision with the pipe wall is typically concentrated between 150 kHz and 600 kHz; the aforementioned frequency range can completely capture these characteristic frequency components. The sampling frequency is set to 2 MHz. According to the Nyquist sampling theorem, this sampling frequency is twice the highest operating frequency, which avoids signal aliasing and preserves complete frequency information. In some applications with higher requirements for high-frequency components, the sampling frequency can be increased to 5 MHz.

[0025] After the acoustic emission signal is acquired, time-frequency analysis is performed first to extract the frequency band energy vector. The continuously acquired acoustic emission signal is divided into analysis windows with a time length of 50 milliseconds. The selection of the analysis window length needs to balance the contradiction between time resolution and frequency resolution: too short a window will reduce frequency resolution and make it difficult to distinguish the energy differences between adjacent frequency bands; too long a window will reduce time resolution and fail to respond promptly to rapid changes in the flow regime. A window length of 50 milliseconds corresponds to a basic frequency resolution of 20 Hz, which can meet the needs of slurry flow regime monitoring, while ensuring that the system can output 20 analysis results per second, providing good real-time performance. In application scenarios where the flow regime changes are relatively slow, the analysis window length can be extended to 100 milliseconds to obtain higher frequency resolution.

[0026] Short-time Fourier Transform (SFT) was performed on the acoustic emission signal within each analysis window to obtain the time-frequency spectrum. The SFT calculation process is as follows: the signal within the analysis window is multiplied by a Hanning window function to reduce spectral leakage, and then a Fast Fourier Transform (FFT) is performed on the windowed signal. For a signal with a sampling frequency of 2 MHz and a window length of 50 milliseconds, each analysis window contains 100,000 sampling points, and the corresponding spectrum is obtained after the FFT. Considering that the operating frequency range of the acoustic emission sensor is 100 kHz to 1 MHz, only the spectral data within this frequency range is retained for subsequent analysis.

[0027] In the time-spectrum diagram, the frequency axis is divided into eight equal-width bands. The operating frequency range is 100 kHz to 1 MHz, with a total bandwidth of 900 kHz. After being divided into eight bands, each band has a bandwidth of 112.5 kHz. Specifically, the first band covers 100 kHz to 212.5 kHz, the second band covers 212.5 kHz to 325 kHz, and so on, with the eighth band covering 787.5 kHz to 900 kHz. Dividing the frequency axis into multiple bands instead of using the complete spectrum is mainly for two reasons: First, the acoustic emission signals generated by the collision of particles of different sizes have different dominant frequency characteristics. Collisions of large particles are dominated by low-frequency components, while collisions of small particles are dominated by high-frequency components. Band division can preserve this particle size-related frequency characteristic. Second, discretizing the continuous spectrum into finite-dimensional feature vectors can significantly reduce the computational complexity of subsequent state recognition algorithms. The number of frequency bands was optimized. Too few frequency bands would result in the loss of frequency details, while too many frequency bands would introduce redundant information and increase the computational burden.

[0028] The energy values ​​within each frequency band are statistically analyzed, forming an 8-dimensional frequency band energy vector. The frequency band energy is calculated by summing the squares of the amplitudes at all frequency points within that band. Let the... The frequency range of each frequency band is to ,in Indicates the first The lower limit frequency of each frequency band Indicates the first The upper limit frequency of the first frequency band, then the second frequency band... Energy value of each frequency band The energy vector is obtained by summing the squares of the spectral amplitudes at all discrete frequency points within the frequency band. The resulting 8-dimensional frequency band energy vector is denoted as... ,in to These correspond to the energy values ​​of frequency bands 1 through 8, respectively.

[0029] Simultaneously with time-frequency analysis, envelope detection is performed on the acoustic emission signals within each analysis window to statistically count acoustic emission events. The purpose of envelope detection is to extract the amplitude envelope of the acoustic emission signal, thereby identifying single particle collision events. The specific processing procedure is as follows: first, a Hilbert transform is performed on the original acoustic emission signal to obtain an analytic signal; then, the magnitude of the analytic signal is taken as the envelope signal. The Hilbert transform converts a real signal into a complex analytic signal, where the imaginary and real parts are orthogonal; the magnitude of the complex signal is the instantaneous amplitude envelope.

[0030] The amplitude threshold is set to three times the root mean square (RMS) value of the background noise. The RMS value of the background noise is obtained by collecting acoustic emission signals and calculating their RMS value under static conditions where the pipeline is not conveying slurry. Using three times the RMS value as the threshold statistically filters out 99.7% of Gaussian noise components, ensuring that the identified acoustic emission events are indeed caused by particle collisions rather than background noise. When the envelope signal rises from below the amplitude threshold to above the amplitude threshold, it is recorded as the start of an acoustic emission event; when the envelope signal falls from above the amplitude threshold to below the amplitude threshold, it is recorded as the end of that acoustic emission event. The acoustic emission event count value within a single analysis window is counted, i.e., the total number of acoustic emission events detected within that window. In industrial sites with complex noise environments, the amplitude threshold can be adjusted to four or five times the RMS value of the background noise to improve the reliability of detection.

[0031] For four acoustic emission sensors at the same measurement cross section, the above time-frequency analysis and envelope detection processing were performed respectively to obtain their respective 8-dimensional frequency band energy vectors and acoustic emission event counts. The frequency band energy vector of the acoustic emission sensor located at the top position is denoted as the top frequency band energy vector. The frequency band energy vector of the acoustic emission sensor located at the bottom is denoted as the bottom frequency band energy vector. The frequency band energy vector of the acoustic emission sensor located on the left is denoted as the left-side frequency band energy vector. The frequency band energy vector of the acoustic emission sensor located on the right is denoted as the right-side frequency band energy vector. Each frequency band energy vector is an 8-dimensional vector.

[0032] The difference between the top and bottom frequency band energy vectors is calculated to obtain the vertical energy difference vector. The difference operation is a subtraction of components one by one, that is... The result remains an 8-dimensional vector. The vertical energy difference vector reflects the degree of asymmetry in the vertical distribution of particle collision energy within the pipe cross-section. Under normal flow conditions, the slurry is uniformly suspended, and the particle collision energy at the top and bottom is similar, with each component of the vertical energy difference vector approaching zero. When slurry begins to deposit, particle movement in the bottom region is hindered, and collision energy decreases, causing a positive shift in the components of the vertical energy difference vector. This difference characteristic is significant for early identification of deposition because it directly reflects the change in particle spatial distribution under the influence of gravity.

[0033] Calculate the difference between the energy vector of the left frequency band and the energy vector of the right frequency band to obtain the horizontal energy difference vector. The calculation method is the same as that for the vertical energy difference vector, i.e. The horizontal energy difference vector is used to detect horizontal flow asymmetry within a pipe cross-section. Under normal operating conditions in a straight pipe section, the components of the horizontal energy difference vector are typically close to zero. However, downstream of pipe bends, due to centrifugal force, the slurry may deflect outwards from the bend, resulting in a significantly non-zero value for the horizontal energy difference vector. By monitoring changes in the horizontal energy difference vector, flow deviation and localized deposition caused by bend effects or pipe tilt can be identified.

[0034] The top, bottom, left, and right frequency band energy vectors, as well as the vertical and horizontal energy difference vectors, are sequentially concatenated to form a 48-dimensional cross-sectional energy feature vector. The concatenation order is fixed to ensure that subsequent state recognition algorithms can correctly interpret the physical meaning of each feature component. The 48-dimensional cross-sectional energy feature vector can be represented as follows: Each sub-vector is 8-dimensional, totaling 48 dimensions. This feature construction method not only preserves the original energy information of each sensor, but also explicitly extracts the asymmetric features of spatial distribution through interpolation, providing rich input information for subsequent state determination.

[0035] The acoustic emission event counts from the four acoustic emission sensors are combined to form a 4-dimensional event count vector. Let the acoustic emission event count of the acoustic emission sensor located at the top be [value missing]. The acoustic emission event count value of the acoustic emission sensor located at the bottom is... The acoustic emission event count value of the acoustic emission sensor located on the left is... The acoustic emission event count value of the acoustic emission sensor located on the right is... Then the 4-dimensional event counting vector is Event counts characterize the activity level of particle collisions from another dimension, complementing the frequency band energy vector. The frequency band energy vector reflects the intensity distribution of collisions, while the event counts reflect the frequency distribution of collisions. Combining the two provides a more comprehensive characterization of the pipeline flow regime.

[0036] Calculate the difference between the maximum and minimum count values ​​of the four acoustic emission events, and record this difference as the event count range. The event count range is obtained through... The calculation yielded, where This indicates the operation of finding the maximum value. This indicates the minimum value operation. The event count range is a scalar indicator that measures the spatial uniformity of acoustic emission events within a cross-section. Under normal flow conditions, the event count values ​​detected by the four sensors are similar, resulting in a small event count range. When local deposition occurs, the event count values ​​of the sensors corresponding to the deposition area are significantly lower than those of the other sensors, increasing the event count range. Compared to the directional energy difference vector, the event count range is a global uniformity indicator, not distinguishing specific non-uniform directions, but it is simpler to calculate and more sensitive to deposition phenomena.

[0037] Calculate the sum of the counts of the four acoustic emission events, and record this sum as the total number of cross-sectional events. The total number of cross-sectional events is determined by... The total number of cross-sectional events is calculated to reflect the overall particle collision activity at the measurement cross-section. Under stable slurry concentration and flow velocity conditions, the total number of cross-sectional events should remain within a relatively stable range. A continuous downward trend in the total number of cross-sectional events may indicate that the deposition area is expanding or the deposition layer thickness is increasing, leading to a reduction in the effective flow cross-section and a decrease in particle collision opportunities. The total number of cross-sectional events also serves as a benchmark for calculating the relative value of the event count range, making the range values ​​comparable under different operating conditions.

[0038] The 48-dimensional cross-sectional energy feature vector, the 4-dimensional event count vector, the event count range, and the total number of cross-sectional events are concatenated to form a 54-dimensional pipeline flow regime observation vector. This 54-dimensional pipeline flow regime observation vector can be represented as follows: ,in This is a 48-dimensional cross-sectional energy eigenvector. It is a 4-dimensional event counting vector. For the event count range, This represents the total number of events at the cross-section. The 54-dimensional pipeline flow regime observation vector is a complete mathematical representation of the flow regime characteristics of a single measurement cross-section within a single analysis window, encompassing information from three levels: frequency domain characteristics, spatial distribution characteristics, and statistical characteristics.

[0039] In practical applications, when multiple measurement sections are arranged on the pipeline, each measurement section independently generates its own 54-dimensional pipeline flow state observation vector. For distributed monitoring systems, the pipeline flow state observation vectors of each measurement section can be aggregated to a central processing unit for unified analysis, or the state identification can be performed independently in the local processing unit of each measurement section. For operating conditions with significant temperature changes, the propagation characteristics of acoustic emission signals may be affected by temperature. In this case, temperature compensation can be performed on the signal before feature extraction, or temperature measurements can be added as auxiliary feature components to the pipeline flow state observation vector.

[0040] The blockage process in a slurry pipeline does not occur instantaneously, but rather proceeds through a series of gradual intermediate states. From normal transport to complete blockage, the flow pattern inside the pipeline sequentially progresses through stages such as particle settling, gradual thickening of the deposition layer, and continuous reduction of the effective flow area. This gradual evolutionary characteristic makes the blockage state exhibit a significant temporal dependence, meaning that the current state is closely related to the state at the previous moment. Hidden Markov Models (HMMs) are naturally well-suited for describing such temporal processes with state transition characteristics. Their core idea is to model the internal state of the system, which cannot be directly observed, as a hidden state, and infer the evolution sequence of the hidden state through observable external signals. In this embodiment, the actual deposition state inside the pipeline cannot be directly measured, but it can be indirectly sensed through signals collected by an acoustic emission sensor array—a typical application scenario for HMMs.

[0041] When constructing a hidden Markov model for pipeline blockage, the first step is to define the hidden state space. Based on the physical characteristics of the slurry pipeline blockage process and engineering experience, the hidden state space is divided into four hidden states: normal flow state, light deposition state, severe deposition state, and pre-blockage state. These four hidden states are arranged in ascending order of blockage risk, corresponding to different stages of the pipeline from safe operation to impending blockage.

[0042] Normal flow state characterizes the ideal operating condition of uniform suspension and transport of slurry in a pipeline. Under this state, solid particles reach dynamic equilibrium under the combined action of fluid carrying force and gravity, and the particles are approximately uniformly distributed within the pipeline cross-section, with no obvious deposition. From the perspective of acoustic emission signals, the characteristics of normal flow state are: the degree of particle collision activity is similar in all directions of the pipeline cross-section, and the signal energy and event count values ​​detected by the four sensors at the top, bottom, left, and right are in a statistically balanced state.

[0043] The light deposition stage represents an early phase of blockage evolution. When the flow rate decreases, the slurry concentration increases, or the particle size enlarges, some particles begin to lose their suspension and settle towards the bottom of the pipeline under gravity. At this stage, a thin layer of deposit begins to form at the bottom of the pipeline, but its thickness is small and has not yet significantly affected the pipeline's transport capacity. Acoustic emission signals exhibit spatial heterogeneity in this state: the collision frequency and intensity decrease in the bottom region due to slower particle movement, while the top and side regions maintain higher collision activity. This vertical asymmetry is a typical characteristic of the light deposition stage.

[0044] Severe deposition indicates that the deposition layer has developed to a considerable extent. The thickness of the deposition layer at the bottom of the pipeline continues to increase, the effective flow cross-section decreases significantly, and the pipeline's transport resistance increases. In this state, not only is the acoustic emission signal in the bottom region of the pipeline significantly attenuated, but the deposition area may also extend to the lower half of the pipeline sidewall. The signals detected by the acoustic emission sensors exhibit a distinct silent zone characteristic, meaning that the detection signals of one or more sensors remain at a low level for an extended period. Occasional deposition layer detachment events will produce sudden, strong signals, but the overall collision frequency remains low.

[0045] The precursory phase of blockage is a dangerous stage in a pipeline where complete blockage is imminent. At this point, the deposit thickness is nearly proportional to the pipeline's inner diameter, the flow cross-section decreases sharply, and abnormal pressure distribution occurs within the pipeline. The slurry is forced to flow at high speed through narrow channels, making the flow highly unstable. Acoustic emission signals in this state exhibit violent fluctuations: the total number of cross-sectional events may change abruptly, and the energy in each frequency band may show a rapid, monotonic change trend. These signal characteristics reflect the rapid deterioration of the flow state within the pipeline, indicating that blockage is imminent.

[0046] For each hidden state, it is necessary to establish observation feature judgment conditions corresponding to the 54-dimensional pipeline flow state observation vector. These judgment conditions map the continuous multidimensional observation vector into discrete state judgment results, which is the basis for the hidden Markov model to perform state inference.

[0047] refer to Figure 1 This figure visually reflects the clustering characteristics and discrimination boundaries of the 54-dimensional pipeline flow observation vector in the simplified two-dimensional feature space. For ease of visualization and principle explanation, the diagram uses two of the most representative feature dimensions to construct a coordinate system: the horizontal axis represents the total number of cross-sectional events S, and the vertical axis represents the proportion of the vertical energy difference vector in the total energy of the corresponding frequency band. Each scatter point in the figure represents a flow observation sample extracted within an analysis window; different types of scatter points correspond to the four different hidden states defined in the Hidden Markov Model. In the lower right region of the coordinate system, data points representing the normal flow state are densely distributed. In this state, the high-concentration slurry flows uniformly in suspension within the pipeline, and the collisions between solid particles and the pipe wall remain essentially balanced in all directions. Therefore, the total number of cross-sectional events S remains at a high level, usually around 1000 times, indicating that particle collisions are active; at the same time, the proportion of vertical energy difference is extremely low, with the vast majority of sample points located in the range of 5% to 10%, far below the 15% judgment threshold. This verifies that the energy received by the top sensor and the bottom sensor is basically the same, and no particle concentration stratification caused by gravity has formed inside the tube.

[0048] As the flow pattern evolves, the data points migrate to the upper left of the coordinate system, entering a region of light deposition. Data points in this region show a slight decrease in the total number of cross-sectional events (S), with an average of approximately 900 events, reflecting reduced particle activity due to sedimentation. A more significant feature is the increase in the vertical axis value, with data points mainly distributed within the 15% to 40% vertical energy difference range. The key threshold of 15% is clearly marked by a dashed line in the figure; it is the boundary for distinguishing between normal flow and light deposition. When data points cross this line, it means that the energy detected by the bottom sensor is significantly lower than that of the top sensor, indicating the presence of a thin layer of deposition at the bottom of the pipe. Continuing to the upper left, data points representing severe deposition form a distinct cluster. The total number of cross-sectional events (S) in this region decreases sharply to around 400 events, indicating that the thick deposition layer at the bottom of the pipe has severely suppressed particle collision behavior, forming a so-called silent zone. Simultaneously, the vertical energy difference percentage further increases and generally exceeds 40%. The second dashed line in the figure represents a threshold of 40%, which effectively separates light deposition from severe deposition and higher-risk conditions. In this region, although slurry still flows through the top flow channels, the bottom is occupied by sediment, resulting in an extremely asymmetrical energy distribution in the vertical direction.

[0049] In the leftmost and uppermost regions of the coordinate system, data points representing the precursory state of blockage are scattered. These data points exhibit the most discrete distribution characteristics, reflecting the extreme instability of the flow regime in this state. The total number of cross-sectional events, S, decreases to an extremely low level, typically less than 200, and fluctuates dramatically between adjacent windows. The proportion of vertical energy difference reaches its peak, exceeding 80% in some samples. This extreme data distribution pattern corresponds to the critical critical condition before complete blockage of the pipeline, at which point the effective flow cross-section is extremely small and the flow field is turbulent. Through this mapping in a two-dimensional feature space, this embodiment clearly reveals the evolutionary path from normal flow to the precursory blockage, verifying the effectiveness and separability of the observation feature judgment conditions in distinguishing different latent states.

[0050] The criteria for determining the observation characteristics under normal flow conditions focus on examining the uniformity of acoustic emission signal distribution within the pipe cross-section. Specific criteria include three aspects: event count range. Less than the total number of cross-sectional events 20%; Vertical energy difference vector The absolute values ​​of each component are less than 15% of the total energy of the corresponding frequency band; the horizontal energy difference vector The absolute values ​​of each component are all less than 15% of the total energy of the corresponding frequency band. For the first... Each frequency band, the total energy of which is defined as the sum of the energy values ​​of that frequency band across the four sensors, i.e. ,in Indicates the top sensor at the Energy value of each frequency band Indicates the bottom sensor at the first Energy value of each frequency band Indicates the left sensor at the first Energy value of each frequency band Indicates the right sensor at the first The energy values ​​for each frequency band. The selection of the two threshold values, 20% and 15%, is based on engineering experience with slurry transport systems. Under normal operating conditions, due to turbulent fluctuations and measurement noise, the detected values ​​of each sensor cannot be completely equal, and allowing a certain range of fluctuations is reasonable. When all three conditions are met simultaneously, the current observation is determined to conform to the characteristics of a normal flow state.

[0051] The observational criteria for determining a lightly deposited state aim to capture early signs of deposition. The criteria use an OR logic to connect two sub-conditions: the event count range. Greater than or equal to the total number of cross-sectional events 20% and less than the total number of cross-sectional events 50%; or the vertical energy difference vector The observation is considered to have at least two components whose absolute values ​​are greater than or equal to 15% and less than 40% of the total energy of the corresponding frequency band. The first sub-condition detects that the spatial non-uniformity of the distribution is beginning to intensify but has not yet reached a severe level, from the perspective of event count. The second sub-condition detects vertical asymmetry from the perspective of frequency band energy, requiring at least two frequency bands to show a moderate degree of energy difference. This is because anomalies in a single frequency band may originate from noise or interference, while anomalies in multiple frequency bands simultaneously are more indicative of a change in the flow regime. The 40% upper limit threshold is used to distinguish between mild deposition and more severe states. When either of the two sub-conditions is met, the current observation is determined to conform to the characteristics of a mild deposition state.

[0052] The criteria for determining severe depositional conditions focus on the changes in the acoustic characteristics and spectral structure of the depositional area. Specifically, the criteria are: at least one of the four acoustic emission sensors has an acoustic emission event count value less than the total number of cross-sectional events. The first condition detects the presence of sensor locations with significantly reduced collision activity. A 10% threshold means the sensor detects less than one-tenth of the total number of collision events across the cross-section, indicating the corresponding area is covered by a deposition layer. The second condition uses changes in the spectral structure for further verification: the high-frequency band includes bands 5 to 8, and the low-frequency band includes bands 1 to 4. When severe deposition occurs in a region, collisions of fine particles almost cease, while occasional collisions of large particles or deposition layer detachment mainly generate low-frequency signals, thus the high-frequency energy decreases significantly relative to the low-frequency energy. Both conditions must be met simultaneously, and cross-validation using frequency domain features improves the reliability of the determination.

[0053] The observational characteristics for identifying precursory blockage states focus on the abrupt changes in the detected signal. The criteria also employ an OR logic to connect two sub-conditions: the total number of cross-sectional events. The first sub-condition captures abrupt changes in the total number of cross-sectional events if: the change in the total number of cross-sectional events exceeds 100% of the total number of cross-sectional events in the previous analysis window; or more than half of the components in the 48-dimensional cross-sectional energy eigenvector show a monotonically decreasing trend in the total number of cross-sectional events in the previous three consecutive analysis windows, with a cumulative decrease exceeding 60% of the initial value. Let the total number of cross-sectional events in the five consecutive analysis windows be... , , , , The condition is satisfied if the absolute value of the difference between any two adjacent values ​​among these five values ​​exceeds 100% of the preceding value. This abrupt change typically stems from severe instability in the flow regime prior to blockage. The second sub-condition captures the continuous decay trend of the energy characteristic. Let the values ​​of a certain component in three consecutive analysis windows be... , , If satisfied and If the value of this component decreases monotonically and the cumulative decrease exceeds 60%, then this component exhibits a monotonically decreasing trend. When more than 24 components in the 48-dimensional cross-sectional energy eigenvector simultaneously satisfy this trend condition, it indicates that the pipeline flow regime is deteriorating across the board. Satisfying either of the two sub-conditions determines that the current observation conforms to the characteristics of a pre-blockage state.

[0054] After establishing the observation feature determination criteria, for the 54-dimensional pipeline flow state observation vector obtained in the current analysis window, it is sequentially checked whether it satisfies the observation feature determination criteria for the above four hidden states. The test results are recorded in the form of a vector that satisfies the current observation conditions. The vector that satisfies the current observation conditions is a 4-dimensional vector, denoted as... ,in Corresponding to normal flow conditions, Corresponding to a lightly deposited state, Corresponding to severe sedimentary conditions, This corresponds to the precursory state of a blockage. When the observation characteristic judgment condition of a certain hidden state is met, the element at the corresponding position takes the value of 1; when it is not met, the element at the corresponding position takes the value of 0. Since the judgment conditions of each hidden state may overlap, multiple positions in the vector that satisfy the current observation condition may simultaneously take the value of 1, or all positions may take the value of 0.

[0055] Setting the state transition permission matrix is ​​a crucial step in Hidden Markov Modeling (HMM). Unlike traditional HMMs that use probabilistic state transition matrices, this embodiment employs a deterministic state transition permission matrix. The values ​​of 0 and 1 explicitly define which state transitions are physically permissible and which are prohibited. This design stems from the physical laws governing slurry pipeline blockage: deposition is a gradual process, and states do not evolve abruptly. For example, a pipeline cannot directly jump from a normal flow state to a pre-blockage state; it must pass through stages of mild and severe deposition. The state transition permission matrix explicitly encodes this physical constraint into the model, avoiding unreasonable state jumps caused by observational noise.

[0056] State transition permission matrix It is a 4x4 matrix, with matrix elements Indicates from the first The hidden state transitions to the first Whether the recessive state is allowed, among which and The values ​​range from 1 to 4, corresponding to normal flow, light deposition, severe deposition, and pre-blockage conditions, respectively. When The time indicates that it is allowed from the first The hidden state transitions to the first Hidden state, when The time indicates that the transfer is prohibited.

[0057] A transition from normal flow to both normal flow and slight deposition is permissible. This means that when a pipe is in a normal flow state, it will either continue to flow normally or begin to show signs of slight deposition. A direct transition from a normal flow state to a severe deposition state or a state in the early stages of blockage is prohibited, as deposition requires time to develop.

[0058] A state of slight deposition allows for transition to normal flow, a state of slight deposition, and a state of severe deposition. Slight deposition can be reversed to normal flow with increased flow velocity or other interventions; it can also be maintained; however, if conditions continue to deteriorate, it may develop into a state of severe deposition. Direct transition from a state of slight deposition to a state of impending blockage is prohibited.

[0059] Severe deposition can progress to mild deposition, severe deposition, and pre-blockage conditions. Severe deposition can be alleviated to mild deposition with slurry removal intervention; it can maintain the current state; if it continues to deteriorate, it will enter the pre-blockage state. Direct restoration from a severe deposition state to normal flow is prohibited, as the removal of the deposition layer is also a gradual process.

[0060] The pre-blocking state allows for a transition to both severe deposition and pre-blocking states. A pre-blocking state can be alleviated to severe deposition with emergency drainage intervention; alternatively, it can remain in the pre-blocking state until further action is taken. A direct transition from a pre-blocking state to normal flow or light deposition is prohibited, as such a multi-stage recovery is physically impossible to achieve instantaneously.

[0061] The matrix elements corresponding to the allowed transitions mentioned above are all set to 1, while the remaining matrix elements are all set to 0. The complete form of the state transition permission matrix is: The rows of the matrix correspond to the normal flow state, light deposition state, severe deposition state, and pre-blockage state from top to bottom, and the columns of the matrix correspond to the same order from left to right.

[0062] The process of performing hidden state sequence inference begins with setting the initial hidden state. When the system starts up or the pipeline begins transporting slurry, the initial hidden state is set to a normal flow state. This initialization assumption is based on common engineering practices: the pipeline should have been cleaned and inspected before startup, and the initial state should be a normal, deposit-free state. In certain special scenarios, such as restarting the pipeline after a shutdown, it may be necessary to set different initial hidden states based on the state before shutdown or the results of manual inspection.

[0063] For each new analysis window, the hidden state sequence inference is performed according to the following steps. First, the 54-dimensional pipeline flow state observation vector for the current analysis window is obtained, and the vector satisfying the current observation conditions is generated according to the aforementioned method. Then, based on the hidden state from the previous time step, query the state transition permission matrix. Obtain the set of hidden states that are allowed to be reached at the current time. Let the hidden state at the previous time be the th... If there are several hidden states, then the set of hidden states that can be reached at the current time is all states that satisfy the condition. The A set consisting of hidden states.

[0064] In the set of allowed hidden states, the hidden state whose corresponding position in the vector satisfies the current observation condition is 1 is selected as a candidate. When exactly one hidden state in the set of allowed hidden states has a corresponding position in the vector satisfying the current observation condition being 1, that hidden state is the hidden state identification result at the current time.

[0065] When multiple latent states meet the criteria, the latent state with the highest state number is selected as the latent state identification result at the current moment. The state numbers are arranged as follows: normal flow state 1, light deposition state 2, severe deposition state 3, and pre-blockage state 4. The strategy of selecting the latent state with the highest number reflects the safety-first design philosophy: when the observation data simultaneously meets the judgment criteria for multiple states, the state with the higher risk is prioritized to trigger slurry drainage intervention measures in a timely manner, erring on the side of over-warning rather than overlooking risks.

[0066] When the current observation conditions for all hidden states in the allowed set of hidden states satisfy the condition that the vector value is 0, the hidden state from the previous time step remains unchanged. This situation may occur when the flow characteristics are in the fuzzy region of the state boundary. In this case, maintaining the original state is a conservative but robust strategy to avoid frequent state jumps due to temporary observational anomalies.

[0067] The following example illustrates the process of hidden state sequence inference. Assuming the hidden state at the previous time step was a lightly deposited state, querying the state transition permission matrix yields the set of allowed hidden states at the current time step: normal flow, lightly deposited, and heavily deposited. The current observation conditions generated from the observation data in the current analysis window satisfy the following vector: This indicates that the current observation simultaneously meets the criteria for both a lightly sedimentary state and a severely sedimentary state. Within the set of allowed hidden states, both the lightly sedimentary state and the severely sedimentary state are located at position 1, meaning there are multiple hidden states that meet the criteria. According to the rule of selecting the hidden state with the largest state number, the severely sedimentary state is numbered 3, which is greater than the lightly sedimentary state numbered 2. Therefore, the hidden state identification result at the current moment is a severely sedimentary state.

[0068] Calculating the probability of pipeline blockage risk requires comprehensive consideration of the latent state identification results over a recent period, rather than relying solely on the state at a single moment. This design is based on two considerations: firstly, the state identification at a single moment may be affected by observation noise or transient interference, potentially leading to misjudgment; secondly, the risk probability should reflect the overall trend and persistence of the pipeline state, as transient abnormal states and persistent abnormal states have different risk implications.

[0069] refer to Figure 2The horizontal axis of the graph represents time in seconds, ranging from 0 to 200 seconds. The vertical axis represents the latent state number, numbered 0, 1, 2, and 3 from bottom to top, corresponding to four latent states: normal flow, light deposition, severe deposition, and pre-blockage state. A solid blue line plots the change in latent state identification results over time, with the vertical axis representing the latent state number at the current moment. To more intuitively display the distribution areas of different latent states, different background colors are used to fill the different states: green for normal flow, yellow for light deposition, orange for severe deposition, and red for pre-blockage.

[0070] From the perspective of temporal evolution, during the period from 0 to 80 seconds, the latent state identification result remains at 0, indicating that the pipeline is in a normal flow state. At this time, the slurry is uniformly suspended and transported in the pipeline, and the signal energy and event count values ​​detected by each acoustic emission sensor are in a statistically balanced state, satisfying the observation characteristic judgment conditions of the normal flow state. At 80 seconds, the latent state identification result jumps from 0 to 1, indicating that the pipeline enters a light deposition state. A blue vertical dashed line is drawn at this moment in the figure, with the text "Activate high-frequency small pulse," indicating that the high-frequency small pulse slurry discharge mode is immediately activated to intervene after detecting the light deposition state. During the period from 80 to 120 seconds, the latent state identification result fluctuates around 1, indicating that the pipeline continues to be in a light deposition state. At 120 seconds, the latent state identification result jumps from 1 to 2, indicating that the pipeline enters a severe deposition state. A purple vertical dashed line is drawn at this moment in the figure, with the text "Activate low-frequency large pulse," indicating that the low-frequency large pulse slurry discharge mode is activated to provide a stronger slurry discharge effect after detecting the severe deposition state.

[0071] From 120 to 160 seconds, the latent state identification result remained at 2, indicating that the deposition situation continued to worsen. At 160 seconds, the latent state identification result jumped from 2 to 3, indicating that the pipeline entered a pre-blockage state. A dark red vertical dashed line is drawn at this moment in the figure, labeled "Frequency Sweep Pulse Activation," indicating that the frequency sweep pulse slurry drainage mode was immediately activated after detecting the pre-blockage state. The resonant frequency of the deposition layer was found through frequency sweeping and continuously excited at the resonant frequency to maximize the slurry drainage effect. From 160 to 180 seconds, the latent state identification result remained at 3, indicating that the pipeline was in a dangerous stage of impending blockage. After 180 seconds, with the continued action of the frequency sweep pulse slurry drainage mode, the deposition layer gradually disintegrated and detached, and the latent state identification result began to recover, first dropping from 3 to 2, indicating that the pipeline recovered from the pre-blockage state to a severe deposition state. Around 220 seconds, the latent state identification result continued to drop from 2 to 1, indicating that the pipeline further recovered from a severe deposition state to a mild deposition state. After 260 seconds, the latent state identification result finally returned to 0, indicating that the pipeline had returned to normal flow and the slurry discharge control had achieved the expected results. The entire time-series curve shows the complete process of the pipeline gradually deteriorating from a normal state to the pre-blockage stage, and then gradually recovering to a normal state through graded pulse slurry discharge intervention.

[0072] A risk level lookup table is established to map four latent states to numerical risk level values. Normal flow corresponds to a risk level value of 0, light deposition to 1, severe deposition to 2, and pre-blockage state to 3. The risk level values ​​reflect the proximity of each state to blockage; higher values ​​indicate higher risk. The sequence of latent state identification results for the most recent 10 consecutive analysis windows is obtained. The total time span for these 10 analysis windows is 500 milliseconds, a timescale sufficient to filter out transient interference without delaying risk warnings due to excessively long time spans. In applications requiring more sensitive responses, the number of analysis windows can be reduced to 5; in scenarios with significant noise interference, the number can be increased to 20.

[0073] The risk level lookup table is consulted to convert the 10 hidden state identification results into 10 corresponding risk level values. Let these 10 risk level values ​​be... ,in Corresponding to the earliest analysis window, The corresponding analysis window.

[0074] Pipeline blockage risk probability The calculation is as follows: Add the maximum value among the 10 risk level values ​​to the arithmetic mean of the 10 risk level values, then divide by 6. ;in This represents the maximum value among the 10 risk level values. This represents the arithmetic mean of 10 risk level values. The denominator is 6 because the maximum possible value of the maximum value is 3, the maximum possible value of the arithmetic mean is also 3, and the maximum sum of the two is 6. Dividing by 6 normalizes the probability of pipeline blockage risk to the range of 0 to 1.

[0075] This calculation formula is designed to consider both peak risk and average risk. The maximum value term ensures that even if a high-risk state occurs only at a single moment, it will be reflected in the risk probability, preventing severe risks from being masked by the averaging effect. The arithmetic mean term reflects the persistence of the risk state within a time window; situations where a high-risk state is maintained for an extended period will receive a higher risk probability assessment. Through the combination of the two, the risk probability can remain sensitive to both sudden high-risk events and persistent medium-risk states.

[0076] refer to Figure 3 The horizontal axis of this graph also represents time, in seconds, ranging from 0 to 200 seconds. Figure 2 Maintain consistency. The vertical axis represents the probability of pipe blockage. The value ranges from 0 to 1.0. A higher risk probability value indicates a higher risk of pipe blockage. The graph uses a thick red solid line to depict the evolution of the risk probability over time, with a red semi-transparent fill below the curve to enhance visual appeal. The risk probability is calculated based on the hidden state identification results of the most recent 10 consecutive analysis windows; the specific calculation formula is as follows: ,in arrive These represent the risk level values ​​corresponding to the most recent 10 analysis windows: 0 for normal flow, 1 for light deposition, 2 for severe deposition, and 3 for pre-blockage conditions. This calculation formula, through a combination of a maximum value term and an arithmetic mean term, can simultaneously remain sensitive to both sudden high-risk events and persistent moderate-risk conditions. Two horizontal dashed lines are plotted in the figure as risk threshold reference lines: the green dashed line corresponds to a risk probability of 0.3 and is marked as the safe threshold, while the orange dashed line corresponds to a risk probability of 0.5 and is marked as the warning threshold.

[0077] The background area is divided into three zones based on risk level: a low-risk zone with a light green background (0-0.3), a medium-risk zone with a light yellow background (0.3-0.5), and a high-risk zone with a light red background (0.5-1.0). From a temporal perspective, during the period from 0 to 80 seconds, because the pipeline is in a normal flow state, the latent state identification results for all 10 analysis windows are 0, corresponding to 0 for all 10 risk level values. According to the calculation formula, the risk probability is 0, and the curve is close to the horizontal axis during this period. After the pipeline enters a state of slight deposition at 80 seconds, the latent state identification results for the 10 analysis windows gradually change from all 0s to containing more and more 1s, and the risk probability begins to rise. Between 80 and 120 seconds, the risk probability gradually rises from near 0 to approximately 0.33, and the curve crosses the green safety threshold line, entering the medium-risk zone. After the pipeline enters a state of severe deposition at 120 seconds, the risk probability continues to rise.

[0078] Between 120 and 160 seconds, the risk probability continued to rise from approximately 0.33 to between 0.5 and 0.6, with the curve crossing the orange warning threshold and entering the high-risk zone. After the pipeline entered a pre-blockage state at 160 seconds, the maximum value rapidly jumped to 3, as the risk level corresponding to this state was 3, and the risk probability quickly increased to above 0.6. After 180 seconds, with the action of the frequency sweep pulse slurry discharge mode, the latent state identification results began to recover, and the risk probability correspondingly began to decrease. The curve first gradually decreased from the high-risk zone, then fell back to the medium-risk zone between 200 and 220 seconds, and continued to decrease to the low-risk zone between 240 and 260 seconds, finally reducing the risk probability to below 0.3 after 260 seconds and remaining stable. The entire risk probability evolution curve fully reflects the dynamic change process of pipeline blockage risk from low to high and then from high to low, providing a quantitative basis for the start and stop timing of pulse slurry discharge control.

[0079] In actual operation, the probability of pipeline blockage risk is updated with each new analysis window. When the pipeline is in a normal flow state, all 10 risk level values ​​are 0, and the probability of pipeline blockage risk is 0. When the pipeline transitions from a normal flow state to a slightly deposited state and remains there, over time, the 10 risk level values ​​gradually change from all 0 to all 1, and the probability of pipeline blockage risk gradually increases from 0 to approximately 0.33. When the pipeline enters a pre-blockage state, even if only one analysis window is in this state, the maximum value immediately jumps to 3, and the probability of pipeline blockage risk increases rapidly, thereby triggering subsequent emergency slurry drainage control.

[0080] For distributed monitoring systems with multiple measurement sections, each measurement section independently calculates its own pipeline blockage risk probability. The overall risk probability of the entire pipeline can be taken as the maximum value of the risk probabilities of each measurement section, so as to respond promptly when a high risk occurs at any section. In some scenarios that require more refined control, the risk probabilities of each measurement section can also be weighted and aggregated, and measurement sections located at high-risk locations such as elbows and diameter changes can be assigned higher weights.

[0081] In alternative implementations, the risk level values ​​can be adjusted to suit different application needs. For example, for scenarios where particular attention is paid to severe deposition and precursory blockage states, the risk level value for severe deposition can be adjusted to 3, and the risk level value for precursory blockage states can be adjusted to 5, thereby widening the gap between high-risk and low-risk states. Accordingly, the denominator also needs to be adjusted to the new maximum possible value. The formula for calculating the risk probability can also take other forms, such as using only the maximum value or only the weighted average, but it is necessary to ensure that the calculation results can effectively distinguish different risk levels and match the trigger thresholds of subsequent control strategies.

[0082] The core idea of ​​pulsed slurry discharge control is to generate periodic flow velocity fluctuations in the pipeline, utilizing the shear and inertial forces generated by the pulsating flow field to act on the deposit layer, causing the deposited particles to resuspend or peel off from the pipe wall. Different degrees of deposition require pulse excitation with different characteristics: slight deposition only requires small-amplitude high-frequency disturbances to prevent further deterioration; thick deposit layers that have already formed require large-amplitude low-frequency impacts to disrupt their structure; and for critical situations where blockage is imminent, it is necessary to find the mechanical weak points of the deposit layer and apply targeted excitation. This embodiment automatically selects the matching pulsed slurry discharge control mode based on the latent state identification results, achieving graded response and precise intervention.

[0083] Configuring a pulse slurry discharge actuator is the hardware foundation for implementing pulse control. The pulse slurry discharge actuator comprises three core components: an accumulator, a pulse control valve, and a pulse generator controller. The accumulator is installed on a bypass branch of the main pipeline. Its function is to pre-store a certain amount of high-pressure fluid, which is rapidly released the moment the pulse control valve opens, thereby generating a pressure pulse in the pipeline. The accumulator pre-charge pressure is set to 1.5 times the normal operating pressure of the pipeline. This ratio ensures that the accumulator releases a sufficient pressure increment without damaging the pipeline system due to excessive pressure. For slurry delivery pipelines with a normal operating pressure of 2 MPa, the accumulator pre-charge pressure is correspondingly set to 3 MPa. The accumulator volume is determined based on the pipeline inner diameter and the required pulse intensity. For pipelines with an inner diameter of 200 mm, the accumulator volume is typically selected from 10 to 20 liters.

[0084] The pulse control valve is installed on the connecting pipeline between the accumulator and the main pipeline, and is an electromagnetic quick-opening valve. The opening response time of the electromagnetic quick-opening valve is less than 30 milliseconds; this rapid response characteristic is crucial for generating a steep pressure pulse leading edge. An excessively long response time will result in a slow pressure rise and a weakened pulse effect. The diameter of the pulse control valve needs to be selected to balance flow capacity and response speed, typically one-quarter to one-third of the main pipeline's inner diameter. The pulse generator controller receives the implicit state recognition result from the state recognition circuit and generates the opening and closing timing signals of the pulse control valve according to a preset control strategy. The pulse generator controller can be implemented using a programmable logic controller or a dedicated timing signal generation circuit.

[0085] When the latent state identification result indicates normal flow, the pulse control valve remains closed, and the pipeline maintains normal continuous conveying conditions. At this time, no slurry discharge intervention is required, the slurry is stably conveyed at the designed flow rate, and the particles remain in a uniform suspension state.

[0086] When the latent state identification result indicates a light deposition state, a high-frequency small-pulse slurry discharge mode is activated. In the light deposition stage, the sediment layer is thin and loosely structured, with relatively weak adhesion between particles and to the pipe wall. At this stage, high-frequency but small-amplitude pulse excitation is used to generate continuous flow velocity disturbance in the deposition area, preventing further particle settling and causing particles that have already begun to settle to re-enter a suspended state. Although the periodic shear stress generated by the high-frequency pulse has a small single amplitude, its cumulative effect is significant, effectively suppressing further deposition.

[0087] The pulse frequency of the high-frequency small-pulse discharge mode is 15 Hz, and the duty cycle is 30%. A frequency of 15 Hz means that 15 complete pulse cycles are generated per second, and the duration of a single cycle is... Approximately 67 milliseconds. Duty cycle. Defined as the ratio of the pulse control valve's opening time to the duration of a single cycle, a 30% duty cycle corresponds to the pulse control valve being open for 20 milliseconds and then closed for 47 milliseconds in each cycle. Opening time pass Calculated, i.e. milliseconds; shutdown time Subtract the start time from the cycle duration, i.e. millisecond.

[0088] The 15 Hz pulse frequency was chosen to take into account the response characteristics of the slurry flow. If the frequency is too low, the interval between adjacent pulses is too long, giving particles in the deposition area sufficient time to settle again; if the frequency is too high, the flow generated by the previous pulse has not yet fully developed before the next pulse arrives, causing interference between the pulses. 15 Hz is within the effective response band of the slurry pipeline system, producing clear velocity pulsations. A 30% duty cycle results in a relatively short opening time for the pulse control valve, limiting the amount of fluid released, and generating velocity pulsations with an amplitude of approximately 10% to 20% of the steady-state velocity. This amplitude is sufficient to disturb loosely deposited particles without causing excessive pressure shocks to the pipeline system.

[0089] When the latent state identification result indicates a severe deposition state, the low-frequency large-pulse slurry discharge mode is activated. In the severe deposition stage, the deposit layer is already quite thick, and the particles have formed a strong bond structure through prolonged contact, establishing a firm adhesion to the pipe wall. Disrupting this solidified deposit layer requires a sufficiently large instantaneous shear force; small-amplitude high-frequency disturbances are insufficient. The low-frequency large-pulse mode extends the duration of each pulse by reducing the pulse frequency, allowing the accumulator to release more fluid, thereby generating stronger flow velocity pulsations and greater shear stress in the pipeline.

[0090] The low-frequency, large-pulse discharge mode has a pulse frequency of 2 Hz and a duty cycle of 50%. A 2 Hz pulse frequency corresponds to a single cycle duration of 500 milliseconds, and a 50% duty cycle means that the pulse control valve is open for 250 milliseconds and then closed for 250 milliseconds within each cycle. Compared to the high-frequency, small-pulse mode, the opening time is extended from 20 milliseconds to 250 milliseconds, giving the accumulator ample time to release a larger volume of fluid, resulting in flow velocity pulsation amplitudes that can reach 40% to 60% of the steady-state flow velocity.

[0091] The strong shearing force generated by low-frequency, high-pulse pulses can directly strip away the deposited layer on the pipe wall. When the pulse control valve opens, the high-pressure fluid in the accumulator rapidly rushes into the main pipeline, creating a strong downstream scouring effect in the deposition area; when the pulse control valve closes, the pipeline pressure drops, the flow rate decreases, and the surface of the deposited layer undergoes a stress unloading. This periodic loading-unloading process is similar to a fatigue failure mechanism, gradually disintegrating the internal structure of the deposited layer. The 2 Hz frequency ensures sufficient recovery time between each pulse, allowing the accumulator to refill with high-pressure fluid and prepare for the next pulse.

[0092] When the latent state identification result indicates a pre-blockage state, the frequency sweep pulse slurry removal mode is activated. A pre-blockage state means that a severely obstructive deposit structure has formed inside the pipeline, which may not be effectively removed by conventional fixed-frequency pulses. The deposit layer, as a particle aggregate, possesses specific mechanical properties and a natural frequency. When the external excitation frequency approaches the natural frequency of the deposit layer, resonance occurs. The vibration amplitude of the deposit layer in the resonant state is much greater than in the non-resonant state, making it more prone to structural damage and overall detachment. The design concept of the frequency sweep pulse slurry removal mode is based on this resonance mechanism: by scanning within a certain frequency range, the frequency that can excite the deposit layer to resonate is found, and then continuous excitation at that frequency maximizes the slurry removal effect.

[0093] The sweep frequency start frequency of the sweep pulse discharge mode 1 Hz, sweep termination frequency 20 Hz, frequency step value The frequency step size is 0.5 Hz. The frequency sweep range covers the entire spectrum from low-frequency large pulses to high-frequency small pulses, ensuring that the resonant frequencies of different types of deposits can be captured. The 0.5 Hz frequency step provides sufficient frequency resolution, with a total of 39 frequency points from 1 Hz to 20 Hz to be tested.

[0094] The frequency sweep process begins at a sweep start frequency of 1 Hz. The pulse control valve is driven at the current frequency for 10 complete cycles, during which the average amplitude of the acoustic emission signal acquired by the acoustic emission sensor array is recorded. Ten cycles are chosen to allow the pulse effect to fully establish and reach a steady-state response, while avoiding excessively long test times. The average amplitude of the acoustic emission signal reflects the response intensity of the deposition layer under the current frequency excitation: when the excitation frequency approaches the resonant frequency of the deposition layer, the vibration of the deposition layer intensifies, particle collisions and shedding events increase, and the amplitude of the acoustic emission signal increases accordingly.

[0095] After completing the test at the current frequency, increase the current frequency by one frequency step, from 1 Hz to 1.5 Hz, and then repeat the above test process. Continue in this manner until the current frequency reaches the sweep termination frequency of 20 Hz. The entire sweep process requires testing 39 frequency points, with each frequency point tested for 10 cycles. For example, at 1 Hz, a single cycle is 1000 milliseconds, and 10 cycles are 10 seconds; at 20 Hz, a single cycle is 50 milliseconds, and 10 cycles are 0.5 seconds. The total duration of the entire sweep process is approximately 3 to 4 minutes.

[0096] After completing the tests at all frequency points from the start frequency to the end frequency of the frequency sweep, the average amplitude of the acoustic emission signal at each frequency point was compared, and the frequency point with the largest average amplitude of the acoustic emission signal was determined as the resonant frequency of the deposition layer. The maximum average amplitude of the acoustic emission signal means that the response of the deposited layer is most intense under excitation at that frequency, which is a typical characteristic of the resonant frequency.

[0097] The resonant frequency of the deposition layer is used as the operating frequency of the pulse control valve, with a duty cycle set to 40%. It operates continuously until the probability of pipeline blockage decreases to below 0.3. The 40% duty cycle is a compromise between the 30% duty cycle of the high-frequency small pulse mode and the 50% duty cycle of the low-frequency large pulse mode, suitable for resonant slurry discharge scenarios. Under continuous excitation at the resonant frequency, the deposition layer gradually disintegrates and detaches under the cumulative vibration. The acoustic emission sensor array can detect signal changes during the deposition layer removal process, and the latent state identification results are updated accordingly.

[0098] During the pulsed slurry discharge control process, the system continuously acquires acoustic emission signals and extracts a 54-dimensional pipeline flow state observation vector according to the method in step one, and continuously updates the latent state identification results and pipeline blockage risk probability according to the method in step two. This closed-loop control structure enables real-time monitoring and evaluation of the slurry discharge effect. When the pipeline blockage risk probability decreases to below 0.3 and the latent state identification result returns to the normal flow state and remains at this level for 20 analysis windows, it indicates that the slurry discharge intervention has achieved the expected effect and the pipeline flow state has returned to normal. At this point, the current pulsed slurry discharge control mode is exited, and the normal continuous conveying operation is resumed. The 20 analysis windows correspond to a duration of 1 second, and this continuity requirement avoids premature exit due to instantaneous fluctuations.

[0099] In optional implementations, the parameters of each pulse slurry discharge mode can be adjusted according to the characteristics of the specific pipeline system. For slurries with high viscosity, the frequency of the high-frequency small pulse mode can be reduced to 10 Hz, and the duty cycle can be increased to 40% to generate a stronger disturbance effect. For systems with long pipeline lengths, the frequency of the low-frequency large pulse mode can be further reduced to 1 Hz to ensure that the pulse wave can propagate to the distant deposition area. The frequency sweep range can also be optimized based on historical operating data. If it is known that the resonant frequency of the deposition layer of a certain type of slurry is usually concentrated in a specific range, the frequency sweep range can be narrowed to accelerate the search speed of the resonant frequency.

[0100] While specific embodiments of the present invention have been described above, those skilled in the art should understand that these specific embodiments are merely illustrative. Those skilled in the art can omit, substitute, and modify the details of the above methods and systems in various ways without departing from the principles and essence of the present invention. For example, combining the above method steps to perform substantially the same function and achieve substantially the same result according to substantially the same method falls within the scope of the present invention. Therefore, the scope of the present invention is defined only by the appended claims.

Claims

1. A method for analyzing the risk of blockage in high-concentration slurry pipelines and for automatic slurry discharge control, characterized in that, The method includes: Step 1: Install an acoustic emission sensor array on the outer wall of the high-concentration slurry delivery pipeline along its axial direction, with multiple acoustic emission sensors arranged at each measurement section; perform time-frequency analysis on the acoustic emission signals to extract the frequency band energy vector, perform envelope detection processing on the acoustic emission signals, and count the acoustic emission events; calculate the directional energy difference vector based on the frequency band energy vectors of each acoustic emission sensor at the same measurement section, and form a section energy feature vector; calculate the event count range and the total number of events at the section based on the acoustic emission event count values; concatenate the section energy feature vector, event count vector, event count range, and total number of events at the section to form the pipeline flow state observation vector; Step 2: Construct a hidden Markov model for pipeline blockage, defining four hidden states: normal flow state, light deposition state, severe deposition state, and pre-blockage state. Establish observation feature judgment conditions for each hidden state; perform hidden state sequence inference based on pipeline flow observation vector and state transition permission matrix to obtain hidden state identification results, and calculate pipeline blockage risk probability based on hidden state identification results. Step 3: Select the corresponding pulse slurry discharge control mode based on the latent state identification result. When the latent state identification result is a light deposition state, start the high-frequency small pulse slurry discharge mode. When the latent state identification result is a severe deposition state, start the low-frequency large pulse slurry discharge mode. When the latent state identification result is a pre-blockage state, start the frequency sweep pulse slurry discharge mode to determine the resonance frequency of the deposition layer and continue to work at the resonance frequency of the deposition layer.

2. The method according to claim 1, characterized in that, In step one, four acoustic emission sensors are arranged at each measurement section. The four acoustic emission sensors are located at the top, bottom, left and right positions of the pipe section, respectively. The axial spacing between adjacent measurement sections is five times the inner diameter of the pipe. The operating frequency range of the acoustic emission sensors is 100kHz to 1MHz, and the sampling frequency is 2MHz.

3. The method according to claim 1, characterized in that, In step one, the specific process of time-frequency analysis is as follows: the continuously acquired acoustic emission signal is divided into analysis windows with a time length of 50ms. A short-time Fourier transform is performed on the acoustic emission signal in each analysis window to obtain a time-frequency spectrum. In the time-frequency spectrum, the frequency axis is divided into 8 equal-width frequency bands. The energy value in each frequency band is counted to form an 8-dimensional frequency band energy vector. The specific process of envelope detection is as follows: the amplitude threshold is set to 3 times the root mean square value of the background noise. When the envelope signal exceeds the amplitude threshold, it is recorded as an acoustic emission event. The count value of acoustic emission events in a single analysis window is counted.

4. The method according to claim 2, characterized in that, In step one, the directional energy difference vector includes a vertical energy difference vector and a horizontal energy difference vector. The vertical energy difference vector is the difference between the frequency band energy vector of the acoustic emission sensor located at the top and the frequency band energy vector of the acoustic emission sensor located at the bottom. The horizontal energy difference vector is the difference between the frequency band energy vector of the acoustic emission sensor located on the left and the frequency band energy vector of the acoustic emission sensor located on the right. The frequency band energy vectors of the acoustic emission sensors located at the top, bottom, left, and right, the vertical energy difference vector, and the horizontal energy difference vector are sequentially concatenated to form a 48-dimensional cross-sectional energy feature vector. The 48-dimensional cross-sectional energy feature vector, the 4-dimensional event count vector, the event count range, and the total number of cross-sectional events are concatenated to form a 54-dimensional pipeline flow state observation vector.

5. The method according to claim 4, characterized in that, In step two, the observation characteristic determination conditions include: For normal flow conditions, the observation characteristic determination conditions are: the event count range is less than 20% of the total number of events in the cross-section, and the absolute value of each component in the vertical energy difference vector is less than 15% of the total energy of the corresponding frequency band, and the absolute value of each component in the horizontal energy difference vector is less than 15% of the total energy of the corresponding frequency band; For slightly deposited conditions, the observation characteristic determination conditions are: the event count range is greater than or equal to 20% and less than 50% of the total number of events in the cross-section, or at least two components in the vertical energy difference vector have an absolute value greater than or equal to 15% and less than 40% of the total energy of the corresponding frequency band. The observation characteristics for severe deposition are: at least one of the four acoustic emission sensors has an acoustic emission event count value less than 10% of the total number of cross-sectional events, and the sum of the energy of the four high-frequency bands in the frequency band energy vector corresponding to that sensor is less than 30% of the sum of the energy of the four low-frequency bands. The observation characteristics for blockage precursors are: the change in the total number of cross-sectional events between five consecutive adjacent analysis windows exceeds 100% of the total number of cross-sectional events in the previous analysis window, or more than half of the components in the 48-dimensional cross-sectional energy feature vector show a monotonically decreasing trend in three consecutive adjacent analysis windows and the cumulative decrease exceeds 60% of the initial value.

6. The method according to claim 1, characterized in that, In step two, the state transition permission matrix is ​​a 4x4 matrix with matrix elements having values ​​of 0 or 1. The normal flow state allows transitions to the normal flow state and the slightly deposited state; the slightly deposited state allows transitions to the normal flow state, the slightly deposited state, and the severely deposited state; the severely deposited state allows transitions to the slightly deposited state, the severely deposited state, and the pre-blockage state; and the pre-blockage state allows transitions to the severely deposited state and the pre-blockage state. The matrix elements corresponding to the above-mentioned permitted transitions have a value of 1, and the other matrix elements have a value of 0.

7. The method according to claim 6, characterized in that, In step two, the specific process of inferring the hidden state sequence is as follows: based on the pipeline flow state observation vector, check whether the observation feature judgment conditions of the four hidden states are met in turn, generate the current observation condition satisfaction vector, the current observation condition satisfaction vector is a 4-dimensional vector, when the observation feature judgment condition of a certain hidden state is met, the corresponding position is set to 1, and when it is not met, the corresponding position is set to 0. The initial hidden state is set as the normal flow state. For each new analysis window, the state transition permission matrix is ​​queried based on the hidden state of the previous time step to obtain the set of hidden states that can be reached at the current time step. In the set of hidden states that can be reached, the hidden state whose corresponding position in the current observation condition satisfaction vector is 1 is selected. When there are multiple hidden states that satisfy the conditions, the hidden state with the largest state number is selected as the hidden state identification result. When the current observation condition satisfaction vector corresponding to all hidden states in the set of hidden states that can be reached is 0, the hidden state of the previous time step is kept unchanged.

8. The method according to claim 1, characterized in that, In step two, the calculation process for the probability of pipeline blockage risk is as follows: establish a risk level lookup table, with normal flow state corresponding to risk level value 0, slight deposition state corresponding to risk level value 1, severe deposition state corresponding to risk level value 2, and blockage precursor state corresponding to risk level value 3. Obtain the hidden state identification result sequence of the most recent 10 consecutive analysis windows, query the risk level lookup table to obtain the corresponding 10 risk level values, add the maximum value among the 10 risk level values ​​to the arithmetic mean of the 10 risk level values, and divide by 6 to obtain the pipeline blockage risk probability.

9. The method according to claim 1, characterized in that, In step three, the pulse frequency of the high-frequency small-pulse slurry discharge mode is 15Hz and the duty cycle is 30%; the pulse frequency of the low-frequency large-pulse slurry discharge mode is 2Hz and the duty cycle is 50%; the sweep frequency pulse slurry discharge mode has a sweep frequency start frequency of 1Hz, a sweep frequency end frequency of 20Hz, and a frequency step value of 0.5Hz. Pulses are generated at each frequency and the operation lasts for 10 complete cycles. During this period, the average amplitude of the acoustic emission signal is recorded, and the frequency point with the largest average amplitude of the acoustic emission signal is determined as the resonant frequency of the sedimentation layer. Pulses are continuously generated at the resonant frequency of the sedimentation layer and the duty cycle is set to 40%.

10. The method according to claim 1, characterized in that, In step three, a pulse slurry discharge actuator is configured. The pulse slurry discharge actuator includes an accumulator, a pulse control valve, and a pulse generator controller. The pre-charge pressure of the accumulator is set to 1.5 times the normal operating pressure of the pipeline. The pulse control valve is an electromagnetic quick-opening valve with an opening response time of less than 30ms. During the execution of pulse slurry discharge control, the hidden state identification results and the pipeline blockage risk probability are continuously updated. When the pipeline blockage risk probability decreases to below 0.3 and the hidden state identification results return to the normal flow state and are maintained for 20 analysis windows, the current pulse slurry discharge control mode is exited and the normal continuous conveying condition is restored.

Images