Erosion corrosion early warning method, device and equipment for tail gas pipeline of reduction furnace

Abstract

The invention discloses an erosion corrosion early warning method, device and equipment for a tail gas pipeline of a reduction furnace. The method comprises the following steps: firstly, acquiring multi-source real-time data of a target monitoring point in the tail gas pipeline of the reduction furnace; based on the multi-source real-time data, obtaining a future silicon powder mass concentration prediction sequence and a future airflow velocity prediction sequence; inputting the data into a pre-constructed dynamic wear rate model and an effective impact energy accumulation model, and respectively calculating to obtain a first future pipe wall residual thickness sequence and a second future pipe wall residual thickness sequence; comparing the predicted pipe wall residual thickness at each future moment in the future pipe wall residual thickness sequence with a preset pipe wall residual thickness safety threshold, and determining a first predicted residual life and a second predicted residual life; and finally, based on the first predicted residual life and the second predicted residual life, generating early warning information of the pipeline eroded and corroded by the silicon powder. According to the invention, high-precision and foresight early warning of the erosion corrosion risk of the tail gas pipeline of the reduction furnace can be realized.

Description

Technical Field

[0001] This invention relates to the field of polysilicon production technology, and in particular to a method, apparatus and equipment for early warning of erosion corrosion in the tail gas pipeline of a reduction furnace. Background Technology

[0002] In the polysilicon production process, the reduction furnace, as the core reaction device, is highly susceptible to side reactions in its high-temperature environment, which can easily induce precursors such as trichlorosilane to generate a large amount of micron-sized atomized silicon powder. These high-hardness silicon powder particles are carried at high speed through the downstream pipeline system in the high-temperature exhaust gas. Especially in areas where the flow path changes abruptly, such as bends and diameter changes, the drastic change in airflow direction causes the silicon powder to exert a continuous and intense physical impact on the pipe wall, leading to significant erosion and corrosion damage.

[0003] Such damage is characterized by its high degree of concealment and rapid development. If intervention measures are not taken in time, it can easily lead to local perforation of pipelines, leakage of media, or even pipe rupture accidents, posing a serious threat to production safety and the continuous and stable operation of the equipment. Therefore, real-time monitoring and risk warning of the scouring and corrosion status of the reduction furnace tail gas pipeline is a key link in ensuring the inherent safety of polysilicon production lines.

[0004] However, the pipeline corrosion monitoring and early warning technologies currently used in industry have significant limitations in application and are difficult to meet the requirements of high-reliability operation and maintenance, mainly in the following three aspects:

[0005] (1) The process parameters and structural state data are disconnected, lacking an effective correlation mechanism. Existing atomized silicon powder monitoring systems are mainly used to optimize the chemical reaction efficiency in reduction furnaces. The process parameters such as silicon powder concentration and airflow velocity output by these systems do not establish a physical or data-driven quantitative relationship with the evolution of pipe wall thickness. Furthermore, pipe wall thickness monitoring generally relies on manual ultrasonic sampling after periodic shutdowns, which cannot achieve continuous online sensing. Both types of data have gaps in both time and space dimensions, making it difficult to support the dynamic characterization of the corrosion process.

[0006] (2) Lack of dynamic prediction models based on multi-source inputs, resulting in severe delays in early warning. Although some production lines have deployed eddy current or ultrasonic thickness measurement devices in key areas, the existing systems can only provide a "snapshot" of the wall thickness at the current moment and have failed to construct a corrosion rate evolution model that integrates multi-dimensional operating parameters such as silicon powder mass concentration, airflow velocity, and particle size. Due to the lack of the ability to predict future wall thickness trends, the system cannot predict future failure risks based on the current operating status and often only triggers an alarm when the wall thickness approaches the safety threshold, resulting in the loss of the best window for predictive maintenance.

[0007] (3) Harsh operating environment restricts the reliability and accuracy of monitoring methods. The tail gas pipeline of the reduction furnace is in a high-temperature (usually exceeding 500℃) and high-concentration gas-solid two-phase flow environment for a long time. Conventional online thickness measurement sensors (such as ordinary ultrasonic probes) are easily affected by thermal attenuation, dust adhesion and signal scattering under such conditions, resulting in a significant reduction in signal-to-noise ratio and making it difficult to guarantee the stability and repeatability of the measurement. Low-quality monitoring data further limits the accuracy of corrosion assessment and prediction models.

[0008] In summary, existing technologies are essentially still in the passive response stage. Even if some real-time data is acquired, it is still impossible to achieve high-precision and forward-looking early warning of the risk of erosion and corrosion in the tail gas pipeline of the reduction furnace. Summary of the Invention

[0009] The technical problem to be solved by this invention is to address the aforementioned shortcomings of the prior art by proposing a method, device, and equipment for early warning of erosion corrosion in reduction furnace tail gas pipelines. This method can achieve high-precision and forward-looking early warning of the risk of erosion corrosion in reduction furnace tail gas pipelines.

[0010] In a first aspect, the present invention provides a method for early warning of erosion corrosion in a reduction furnace tail gas pipeline, the method comprising the following steps:

[0011] Acquire multi-source real-time data from target monitoring points in the tail gas pipeline of the reduction furnace; the multi-source real-time data includes historical silicon powder mass concentration time series data, historical airflow velocity time series data, current silicon powder mass concentration data, current airflow velocity data, and median particle size of silicon powder;

[0012] Based on historical silicon powder mass concentration time series data and historical airflow velocity time series data, and combined with current silicon powder mass concentration and current airflow velocity data, the changes in silicon powder mass concentration and airflow velocity within a set prediction period in the future are predicted, resulting in a future silicon powder mass concentration prediction sequence and a future airflow velocity prediction sequence.

[0013] The future silicon powder mass concentration prediction sequence and the future airflow velocity prediction sequence are input into a pre-built dynamic wear rate model to calculate the first future pipe wall remaining thickness sequence; and the median particle size of silicon powder, the future silicon powder mass concentration prediction sequence and the future airflow velocity prediction sequence are input into a pre-built effective impact energy accumulation model to calculate the second future pipe wall remaining thickness sequence.

[0014] The predicted remaining pipe wall thickness at each future time in the first future pipe wall remaining thickness sequence and the second future pipe wall remaining thickness sequence are compared with a preset pipe wall remaining thickness safety threshold to determine the first predicted remaining lifetime corresponding to the first future pipe wall remaining thickness sequence and the second predicted remaining lifetime corresponding to the second future pipe wall remaining thickness sequence.

[0015] Based on the first and second predicted remaining lifetimes, early warning information on silicon powder erosion and corrosion of the pipeline is generated and output.

[0016] Furthermore, based on historical silicon powder mass concentration time-series data and historical airflow velocity time-series data, and combined with current silicon powder mass concentration and current airflow velocity data, the changes in silicon powder mass concentration and airflow velocity within a set prediction period in the future are predicted, resulting in a future silicon powder mass concentration prediction sequence and a future airflow velocity prediction sequence, specifically including:

[0017] The current silicon powder mass concentration data and the current airflow velocity data are appended to the end of the historical silicon powder mass concentration time series data and the historical airflow velocity time series data, respectively, to form an expanded silicon powder mass concentration time series dataset and an expanded airflow velocity time series dataset.

[0018] The changes in silicon powder mass concentration and airflow velocity within a set prediction period in the future are predicted using either mean prediction or long short-term memory network prediction methods.

[0019] The mean prediction method is as follows: calculate the arithmetic mean of the extended silicon powder mass concentration time series dataset as the predicted value of future silicon powder mass concentration, calculate the arithmetic mean of the extended airflow velocity time series dataset as the predicted value of future airflow velocity, and extend the predicted values ​​of future silicon powder mass concentration and future airflow velocity to sequences with the same length as the set prediction period in the future, so as to obtain the predicted sequence of future silicon powder mass concentration and the predicted sequence of future airflow velocity.

[0020] The prediction method of the Long Short-Term Memory Network is as follows: the extended silicon powder mass concentration time series dataset and the extended airflow velocity time series dataset are input into the pre-trained Long Short-Term Memory Network model, and the Long Short-Term Memory Network model outputs the future silicon powder mass concentration prediction sequence and the future airflow velocity prediction sequence.

[0021] Furthermore, based on the first and second predicted remaining lifetimes, early warning information for silicon powder erosion corrosion of the pipeline is generated, specifically including:

[0022] Determine the difference between the first predicted remaining lifetime and the second predicted remaining lifetime;

[0023] If the difference between the first predicted remaining lifetime and the second predicted remaining lifetime is less than a preset tolerance threshold, a consistency warning message is generated. The consistency warning message includes a unified predicted remaining lifetime in days and a maintenance time window recommendation. The warning message includes the consistency warning message.

[0024] If the difference between the first predicted remaining lifetime and the second predicted remaining lifetime is greater than or equal to a preset tolerance threshold, a difference warning message is generated. The difference warning message includes the predicted remaining lifetime of the two models, prediction uncertainty prompts, and operation and maintenance inspection suggestions; the warning message includes the difference warning message.

[0025] Furthermore, the method also includes:

[0026] Obtain the currently measured remaining pipe wall thickness data;

[0027] The difference between the measured remaining pipe wall thickness data and the preset safe threshold for remaining pipe wall thickness is calculated to obtain the safe margin value for remaining pipe wall thickness.

[0028] Based on the safety margin value of the remaining pipe wall thickness, a health index reflecting the structural condition of the pipeline is generated.

[0029] Furthermore, after generating health indicators reflecting the structural condition of the pipeline, the method also includes:

[0030] Acquire current silicon powder mass concentration data, current airflow velocity data, current measured pipe wall remaining thickness data, health index, first predicted remaining lifespan, second predicted remaining lifespan, early warning information, first future pipe wall remaining thickness sequence, second future pipe wall remaining thickness sequence, real-time calculated cumulative pipe wall remaining thickness reduction, real-time calculated cumulative effective impact energy, and real-time pipe wall remaining thickness value calculated based on the current measured pipe wall remaining thickness data and the preset pipe wall remaining thickness safety threshold;

[0031] The display interface includes: current silicon powder mass concentration data and its corresponding real-time trend curve, current airflow velocity data and its corresponding real-time trend curve, current measured remaining pipe wall thickness data and its corresponding historical trend curve, real-time remaining pipe wall thickness value, cumulative remaining pipe wall thickness reduction, cumulative effective impact energy, prediction curves corresponding to the first future remaining pipe wall thickness sequence, prediction curves corresponding to the second future remaining pipe wall thickness sequence, health indicators, first predicted remaining lifespan, second predicted remaining lifespan, and warning information.

[0032] Furthermore, the method also includes:

[0033] Obtain the historical pipe wall remaining thickness data sequence; perform a difference operation on the historical pipe wall remaining thickness data sequence and divide it by the sampling time interval to obtain the historical pipe wall remaining thickness instantaneous wear rate sequence.

[0034] The historical pipe wall remaining thickness instantaneous wear rate sequence, historical silicon powder mass concentration time series data, and historical airflow velocity time series data were subjected to natural logarithmic transformation;

[0035] Using the instantaneous wear rate sequence of the remaining pipe wall thickness after natural logarithmic transformation as the dependent variable, and the time series data of the historical airflow velocity and the time series data of the historical silicon powder mass concentration after natural logarithmic transformation as independent variables, a multiple linear regression analysis was performed to determine the velocity exponent parameter of the dynamic wear rate model, as well as the composite parameter of the comprehensive wear coefficient and the impact angle correction coefficient.

[0036] The first model is constructed based on the velocity exponent parameter and the composite parameter.

[0037] The first model was determined to be the dynamic wear rate model.

[0038] Furthermore, the first model is determined to be a dynamic wear rate model, including:

[0039] Obtain validation set data, which includes silicon powder mass concentration data and airflow velocity data within a preset validation time range;

[0040] The silicon powder mass concentration data and airflow velocity data in the validation set data are input into the first model to calculate the first pipe wall remaining thickness prediction sequence. Based on the pipe wall remaining thickness data in the validation set data, the first coefficient of determination and the first mean absolute error between the first pipe wall remaining thickness prediction sequence and the actual data are calculated.

[0041] If the first coefficient of determination is greater than or equal to the preset first coefficient threshold, and the first mean absolute error is less than or equal to the preset first error threshold, then the first model is determined to have a significant relationship, and the first model is determined to be a dynamic wear rate model.

[0042] Furthermore, determining the first model as a dynamic wear rate model also includes:

[0043] If the first coefficient of determination is less than the first coefficient threshold, or the first mean absolute error is greater than the first error threshold, then the first model is determined to have no significant relationship, and the first model is reconstructed.

[0044] Furthermore, the method also includes:

[0045] Based on historical silicon powder mass concentration time series data, historical airflow velocity time series data, and silicon powder median particle size data, a historical cumulative effective impact energy sequence was obtained.

[0046] Based on the historical time series data of remaining pipe wall thickness and the initial remaining pipe wall thickness value, the difference between the initial remaining pipe wall thickness value and the measured remaining pipe wall thickness value at each time point is calculated to obtain the historical cumulative remaining pipe wall thickness reduction sequence.

[0047] The historical cumulative remaining pipe wall thickness reduction sequence and the historical cumulative effective impact energy sequence were respectively subjected to natural logarithmic transformation;

[0048] Using the historical cumulative wall thickness reduction sequence after natural logarithmic transformation as the dependent variable and the historical cumulative effective impact energy sequence after natural logarithmic transformation as the independent variable, a univariate linear regression analysis was performed to determine the exponential and coefficient parameters of the effective impact energy accumulation model.

[0049] The second model is constructed based on the exponential parameter and the coefficient parameter;

[0050] The second model was determined to be the effective impact energy accumulation model.

[0051] Furthermore, the second model was determined to be an effective impact energy accumulation model, including:

[0052] Obtain validation set data, which includes silicon powder mass concentration data, airflow velocity data, median particle size of silicon powder, and remaining pipe wall thickness data within a preset validation time range;

[0053] The silicon powder mass concentration data, airflow velocity data, and median particle size data of silicon powder in the validation set data are input into the effective impact energy accumulation model to calculate the second pipe wall remaining thickness prediction sequence. Based on the pipe wall remaining thickness data in the validation set data, the second coefficient of determination and the second mean absolute error between the second pipe wall remaining thickness prediction sequence and the actual data are calculated.

[0054] If the second coefficient of determination is greater than or equal to the preset second coefficient threshold, and the second mean absolute error is less than or equal to the preset second error threshold, then the effective impact energy accumulation model is determined to have a significant relationship, and the second model is determined to be the effective impact energy accumulation model.

[0055] Furthermore, determining the second model as an effective impact energy accumulation model also includes:

[0056] If the second coefficient of determination is less than the second coefficient threshold, or the second mean absolute error is greater than the second error threshold, then the effective impact energy accumulation model is determined to have no significant relationship, and the second model is reconstructed.

[0057] Secondly, the present invention provides an early warning device for erosion and corrosion of a reduction furnace tail gas pipeline, the device comprising:

[0058] The acquisition unit is used to acquire multi-source real-time data of target monitoring points in the tail gas pipeline of the reduction furnace; wherein, the multi-source real-time data includes historical silicon powder mass concentration time series data, historical airflow velocity time series data, current silicon powder mass concentration data, current airflow velocity data, and median particle size of silicon powder;

[0059] The prediction unit, connected to the acquisition unit, is used to predict the changes in silicon powder mass concentration and airflow velocity within a set prediction period in the future, based on historical silicon powder mass concentration time-series data and historical airflow velocity time-series data, and combined with current silicon powder mass concentration data and current airflow velocity data, to obtain a future silicon powder mass concentration prediction sequence and a future airflow velocity prediction sequence.

[0060] The calculation unit, connected to the prediction unit and the acquisition unit respectively, is used to input the future silicon powder mass concentration prediction sequence and the future airflow velocity prediction sequence into the pre-built dynamic wear rate model to calculate the first future pipe wall remaining thickness sequence; and to input the silicon powder median particle size data, the future silicon powder mass concentration prediction sequence and the future airflow velocity prediction sequence into the pre-built effective impact energy accumulation model to calculate the second future pipe wall remaining thickness sequence.

[0061] The early warning generation unit, connected to the calculation unit, is used to compare the predicted remaining pipe wall thickness at each future time in the first future pipe wall remaining thickness sequence and the second future pipe wall remaining thickness sequence with a preset pipe wall remaining thickness safety threshold to determine the first predicted remaining lifespan corresponding to the first future pipe wall remaining thickness sequence and the second predicted remaining lifespan corresponding to the second future pipe wall remaining thickness sequence; based on the first predicted remaining lifespan and the second predicted remaining lifespan, it generates early warning information for silicon powder erosion corrosion of the pipe and outputs the early warning information.

[0062] Thirdly, the present invention provides an electronic device comprising a memory and a processor, wherein the memory stores a computer program, and when the processor runs the computer program stored in the memory, the processor executes the scouring and corrosion early warning method for the tail gas pipeline of the reduction furnace as described in the first aspect.

[0063] This invention, by fusing multi-source real-time data and constructing and applying a complementary prediction system of dual models (dynamic wear rate model and effective impact energy accumulation model), can achieve high-precision and forward-looking early warning of erosion corrosion risks in reduction furnace tail gas pipelines. This effectively solves the problems of data fragmentation, delayed early warning, and poor adaptability to harsh operating conditions in existing technologies. Specific beneficial effects are as follows:

[0064] 1. The dual-model synergy and complementarity enhance the comprehensiveness and reliability of early warning.

[0065] By simultaneously applying the dynamic wear rate model and the effective impact energy accumulation model, pipeline wear can be predicted from different time scales and physical perspectives. The dynamic wear rate model, based on current and predicted operating parameters, can quickly calculate the future trend of remaining pipe wall thickness, enabling real-time tracking of short-term wear conditions. The effective impact energy accumulation model, combined with silica particle size information, assesses the total long-term wear from the perspective of accumulated energy, resulting in more stable predictions and greater representativeness of long-term trends. The synergistic effect of these two models allows the early warning system to both promptly reflect fluctuations in operating conditions and grasp the macroscopic development trend of wear, leading to more comprehensive and reliable early warning conclusions.

[0066] 2. By integrating operating condition prediction and wear model, proactive early warning is achieved.

[0067] By first performing time-series predictions of silicon powder mass concentration and airflow velocity, and then inputting the prediction results into a wear model, this invention enables dynamic prediction of the future remaining thickness of the pipeline. This allows the system to anticipate whether the actual remaining thickness of the pipe wall will reach or exceed the safety threshold before it does, thus issuing early warnings before potential risks occur. This changes the passive mode of traditional technology, which can only issue alarms when the wall thickness is close to its limit, and buys valuable time for taking predictive maintenance measures.

[0068] 3. Provides multi-dimensional prediction results to support refined risk management.

[0069] This invention outputs two independent sequences of future pipe wall remaining thickness and calculates two predicted remaining lifetimes (first predicted remaining lifetime and second predicted remaining lifetime). These two predictions can reveal the pipeline's risk status from different dimensions. Decision-makers can comprehensively evaluate the consistency of these two results; if they are similar, confidence in the predictions is enhanced; if discrepancies exist, they can serve as a warning of risk uncertainty, triggering further verification of data quality or operational stability. Furthermore, based on comparisons of different thresholds, different warning levels can be defined, providing a refined decision-making basis for implementing tiered responses and developing differentiated maintenance strategies.

[0070] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of the present invention, nor is it intended to limit the scope of the invention. Other features of the invention will become readily apparent from the following description. Attached Figure Description

[0071] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used together with the embodiments of the invention to explain the invention and do not constitute a limitation thereof. The above and other features and advantages will become more apparent to those skilled in the art from the detailed description of exemplary embodiments with reference to the accompanying drawings, in which:

[0072] Figure 1 A schematic diagram of the scouring and corrosion early warning method for the tail gas pipeline of the reduction furnace provided in an embodiment of the present invention;

[0073] Figure 2 This is a flowchart of the scouring and corrosion early warning system for the tail gas pipeline of the reduction furnace provided in an embodiment of the present invention.

[0074] Figure 3 This is a framework diagram for erosion and corrosion early warning of the tail gas pipeline of the reduction furnace provided in an embodiment of the present invention;

[0075] Figure 4 A schematic diagram of the scouring and corrosion early warning device for the tail gas pipeline of the reduction furnace provided in an embodiment of the present invention;

[0076] Figure 5 This is an architectural diagram of an electronic device provided in an embodiment of the present invention.

[0077] Reference numerals: 10, acquisition unit; 20, prediction unit; 30, calculation unit; 40, early warning generation unit; 100, processor; 200, memory. Detailed Implementation

[0078] It is understood that the specific embodiments and accompanying drawings described herein are merely for explaining the invention and are not intended to limit the invention.

[0079] It is understood that, without conflict, the various embodiments and features in the embodiments of the present invention can be combined with each other.

[0080] It is understood that, for ease of description, only the parts related to the present invention are shown in the accompanying drawings, while the parts unrelated to the present invention are not shown in the drawings.

[0081] It is understood that each unit or module involved in the embodiments of the present invention may correspond to only one entity structure, or may be composed of multiple entity structures, or multiple units or modules may be integrated into one entity structure.

[0082] It is understood that, without conflict, the functions and steps marked in the flowcharts and block diagrams of this invention may occur in a different order than that marked in the accompanying drawings.

[0083] It is understood that the flowcharts and block diagrams of this invention illustrate the possible architecture, functions, and operations of systems, apparatuses, devices, and methods according to various embodiments of this invention. Each block in the flowchart or block diagram may represent a unit, module, program segment, or code, containing executable instructions for implementing the specified function. Furthermore, each block or combination of blocks in the block diagram and flowchart can be implemented using a hardware-based system to achieve the specified function, or using a combination of hardware and computer instructions.

[0084] It is understood that the units and modules involved in the embodiments of the present invention can be implemented by software or by hardware. For example, the units and modules can be located in a processor.

[0085] Example 1:

[0086] The erosion corrosion early warning method for reduction furnace tail gas pipelines provided in this embodiment is mainly applied to the safety monitoring and predictive maintenance scenarios of reduction furnace tail gas pipeline systems in the industrial production of silicon-based materials such as polysilicon, organosilicon, and semiconductor materials. In this scenario, micron-sized high-hardness silicon powder generated during the high-temperature synthesis reaction in the reduction furnace continuously impacts downstream pipelines (especially at bends and diameter changes) with the high-speed tail gas, causing erosion corrosion of the pipe walls. This method integrates multi-source data such as silicon powder concentration, airflow velocity, and wall thickness in real time, and uses a dual-model approach (dynamic wear rate model and effective impact energy accumulation model) to collaboratively predict corrosion trends, enabling a shift from passive shutdown maintenance to proactive predictive maintenance. Specifically, this method can provide early warning of the remaining lifespan of the pipeline several days to weeks in advance, guiding production units to arrange maintenance or replacement within the planned shutdown window. This avoids unplanned production stoppages, material leaks, safety accidents, and environmental risks caused by sudden thinning or perforation of the pipeline, significantly improving production continuity and intrinsic safety. Meanwhile, the multi-dimensional indicators output by this method, such as real-time corrosion rate, cumulative damage, and health status, also provide data support for process optimization (such as adjusting the intake air ratio and flow rate) to extend pipeline life.

[0087] like Figure 1 As shown, the scouring and corrosion early warning method for the tail gas pipeline of the reduction furnace provided in this embodiment specifically includes steps S1 to S5.

[0088] Step S1: Obtain multi-source real-time data of the target monitoring point in the tail gas pipeline of the reduction furnace; wherein, the multi-source real-time data includes historical silicon powder mass concentration time series data, historical airflow velocity time series data, current silicon powder mass concentration data, current airflow velocity data, and median particle size of silicon powder.

[0089] In practice, acquiring multi-source real-time data is a systematic process that integrates sensor deployment, data acquisition and preprocessing, and historical data accumulation. The specific implementation methods are as follows:

[0090] (1) First, a sensor network is deployed at the physical level. An atomization monitor is installed on the downstream pipe of each reduction furnace outlet to continuously and in real-time measure and output the mass concentration of silicon powder in the exhaust gas, typically in grams per cubic meter (g / m³). Simultaneously, a high-temperature resistant ultrasonic thickness gauge array is installed at critical locations with the highest risk of erosion corrosion (e.g., bends and diameter changes in the exhaust main pipe) to non-invasively measure the remaining thickness of the pipe wall in real-time, typically in millimeters (mm). Furthermore, existing or separately installed flow meters on the pipeline are used to obtain the gas flow rate, in meters per second (m / s). For the characteristic size of silicon powder particles, real-time measurement is performed using an online laser particle size analyzer, or model estimation is performed based on specific production process operating parameters (such as furnace temperature, pressure, and raw material ratio) to obtain the representative median particle size of silicon powder, in micrometers (μm). The median particle size of silicon powder (standard symbol D50) is a key physical property parameter characterizing the average size of a particle group. Specifically, it refers to the mass median particle size, meaning the particle size at which the cumulative mass distribution in a silicon powder sample reaches 50% of the total particle size; that is, 50% of the particles by mass have a particle size smaller than this value, and the other 50% have a particle size larger than this value. As the most representative average size indicator, D50 is crucial in this invention—the erosion corrosion energy of silicon powder on the pipe wall is directly related to the particle kinetic energy, which is significantly affected by its mass. Therefore, introducing D50 as a key input parameter into the effective impact energy accumulation model can more accurately quantify the silicon powder erosion intensity and is one of the core elements for improving the model's prediction accuracy.

[0091] (2) Secondly, real-time processing and quality control are performed at the data stream level. The raw signals (such as current, voltage, and waveform) collected by the aforementioned sensors are initially converted into a digital time-series data stream. To ensure the accuracy of subsequent analysis, the raw data must be preprocessed. The steps include: applying digital filters (such as moving average or low-pass filtering) to smooth random noise; and identifying and removing outliers (such as invalid data caused by sensor momentary failure or electromagnetic interference) through statistical methods (such as based on standard deviation or interquartile range). This step aims to improve the signal-to-noise ratio and reliability of the data, providing a high-quality data source for model input.

[0092] (3) Finally, the historical sequence required for analysis is constructed in the time dimension. The historical silicon powder mass concentration time series data and the historical airflow velocity time series data are continuously accumulated on the time axis by the above-mentioned preprocessed data stream. In order to meet the training and initialization requirements of the prediction model, a sufficiently long historical period of data needs to be obtained. Specifically, the number of samples of historical time series data should reach a statistically effective scale, such as no less than 100 consecutive time series sample points (corresponding to hourly data of the past 100 hours). This lower limit can be adjusted according to the fluctuation frequency of the actual working conditions and the user's specific requirements for prediction stability. The current silicon powder mass concentration data and the current airflow velocity data refer to the instantaneous values ​​obtained at the latest sampling time and after preprocessing. The median particle size of silicon powder is usually used as a relatively stable physical property parameter input and can be updated according to the periodic measurement results of the particle size analyzer or the process status.

[0093] Step S2: Based on historical silicon powder mass concentration time series data and historical airflow velocity time series data, and combined with current silicon powder mass concentration and current airflow velocity data, predict the changes in silicon powder mass concentration and airflow velocity within a set prediction period in the future, and obtain the future silicon powder mass concentration prediction sequence and the future airflow velocity prediction sequence.

[0094] As a specific implementation method, the process of predicting the future silicon powder mass concentration and airflow velocity is carried out as follows: First, the current silicon powder mass concentration data and the current airflow velocity data are appended to the end of their corresponding historical time series data, thereby forming an expanded silicon powder mass concentration time series dataset and an expanded airflow velocity time series dataset, thus constructing a complete input sequence. Then, one of two methods, mean prediction or Long Short-Term Memory (LSTM) network prediction, is used to predict the parameter changes within a set future time period. If the mean prediction method is used, the steps are as follows: calculate the arithmetic mean of the two expanded datasets, using this as a single, constant predicted value for the future silicon powder mass concentration and the future airflow velocity; then, each of these predicted values ​​is copied and expanded to generate a sequence of constant value with the same length as the set future prediction time period, thus obtaining the future silicon powder mass concentration prediction sequence and the future airflow velocity prediction sequence, respectively. If a long short-term memory network (LSTM) prediction method is used, the steps are as follows: input the expanded silicon powder mass concentration time series dataset and the expanded airflow velocity time series dataset into a pre-trained LSTM network model; based on the learned temporal evolution law, the model directly outputs the future silicon powder mass concentration prediction sequence and the future airflow velocity prediction sequence corresponding to the future set time period, with dynamic numerical changes.

[0095] Step S3: Input the future silicon powder mass concentration prediction sequence and the future airflow velocity prediction sequence into the pre-built dynamic wear rate model to calculate the first future pipe wall remaining thickness sequence; and input the median particle size of silicon powder, the future silicon powder mass concentration prediction sequence and the future airflow velocity prediction sequence into the pre-built effective impact energy accumulation model to calculate the second future pipe wall remaining thickness sequence.

[0096] As a specific implementation method, the method further includes the following steps: First, obtain the historical pipe wall remaining thickness data sequence; then, perform a difference operation on the sequence and divide the difference result by the corresponding sampling time interval to obtain the instantaneous wear rate sequence of the historical pipe wall remaining thickness; subsequently, perform natural logarithmic transformation on the instantaneous wear rate sequence, the historical silicon powder mass concentration time series data, and the historical airflow velocity time series data respectively to linearize the power law relationship; based on this, take the instantaneous wear rate sequence after natural logarithmic transformation as the dependent variable, and the airflow velocity time series data and silicon powder mass concentration time series data after natural logarithmic transformation as independent variables, perform multiple linear regression analysis to determine two key parameters in the dynamic wear rate model: the exponential parameter of airflow velocity, and the composite parameter of the comprehensive wear coefficient and the impact angle correction coefficient; finally, based on the identified velocity exponential parameter and the composite parameter, construct and establish a first model, and use the first model as the dynamic wear rate model for describing the pipe wear process. By constructing the first model, the originally complex and difficult-to-model wear process is transformed into a linear regression problem that can be analyzed by mathematical methods. This provides a calculable and reproducible model basis for predicting the wear state of pipe walls, and significantly improves the objectivity and operability of wear analysis.

[0097] In a more specific implementation, the process of determining the first model as a dynamic wear rate model includes the following steps: First, obtaining validation set data, which includes silicon powder mass concentration data and airflow velocity data collected within a preset validation time range; next, inputting the silicon powder mass concentration data and airflow velocity data from the validation set data into the first model to calculate a first pipe wall remaining thickness prediction sequence; then, based on the actually measured pipe wall remaining thickness data in the validation set data, calculating a first coefficient of determination and a first mean absolute error between the first pipe wall remaining thickness prediction sequence and the actual data; finally, if the first coefficient of determination is greater than or equal to a preset first coefficient threshold, and the first mean absolute error is less than or equal to a preset first error threshold, then it is determined that there is a significant correlation between the first model and the actual data, and the first model is determined as the finally usable dynamic wear rate model. By objectively measuring the model's prediction accuracy and generalization ability through quantitative indicators such as the coefficient of determination and mean absolute error, the risk of the model performing well only on training data but failing in practical applications is effectively avoided, ensuring that the finally determined dynamic wear rate model has practical application value and reliability.

[0098] In a more specific implementation, the process of determining the first model as a dynamic wear rate model further includes the following steps: if the calculated first coefficient of determination is less than a preset first coefficient threshold, or the first mean absolute error is greater than a preset first error threshold, then it is determined that there is no statistically significant relationship between the first model and the actual data. In this case, the model building process needs to be re-executed to generate a new first model. By setting a feedback and reconstruction mechanism when model validation fails, it is ensured that only models that pass rigorous statistical tests can be ultimately adopted. This closed-loop design significantly enhances the robustness and adaptability of the modeling process, preventing model failure due to data anomalies, parameter mismatches, or changes in operating conditions, thereby continuously optimizing the applicability and accuracy of the model and ensuring the long-term stability and effectiveness of the wear prediction system.

[0099] As a specific implementation method, the method further includes: calculating and generating a historical cumulative effective impact energy sequence based on historical silicon powder mass concentration time-series data, historical airflow velocity time-series data, and median particle size data of silicon powder. Simultaneously, based on historical pipe wall remaining thickness time-series data and initial pipe wall remaining thickness values, a historical cumulative pipe wall remaining thickness reduction sequence is obtained by calculating the difference between the measured pipe wall remaining thickness value and the initial value at each time point. Next, the two sequences are subjected to natural logarithmic transformation. Using the historical cumulative pipe wall remaining thickness reduction sequence after natural logarithmic transformation as the dependent variable and the historical cumulative effective impact energy sequence after natural logarithmic transformation as the independent variable, a univariate linear regression analysis is performed to determine the exponential and coefficient parameters of the effective impact energy accumulation model. Based on the determined exponential and coefficient parameters, a second model is constructed, and this second model is ultimately determined to be the effective impact energy accumulation model. This implementation modeled the pipeline wear process as a physical evolution under the effect of energy accumulation. By establishing a quantitative relationship between accumulated impact energy and accumulated wear, it revealed the constitutive correlation between long-term wear and operating load, improved the physical interpretability and engineering applicability of wear prediction, and provided more robust theoretical and model support for pipeline life assessment and preventive maintenance.

[0100] In a more specific implementation, the process of determining the second model as an effective impact energy accumulation model includes: based on a validation set containing data on silicon powder mass concentration, airflow velocity, median silicon powder particle size, and remaining pipe wall thickness during the validation period, the first three types of data are input into the model to calculate a second predicted sequence for the remaining pipe wall thickness. Then, by comparing the predicted sequence with the actual data, a second coefficient of determination and a second mean absolute error are calculated. If the second coefficient of determination is not lower than a preset second coefficient threshold and the second mean absolute error is not higher than a preset second error threshold, the model is determined to be statistically significant, thus establishing the second model as a verified and reliable effective impact energy accumulation model. This step, through rigorous independent verification and quantitative evaluation, ensures that the model is not only theoretically self-consistent but also possesses predictive accuracy and engineering applicability under actual working conditions. It provides a statistically verifiable and reliable basis for the long-term evolution analysis of pipeline wear, significantly improving the credibility and deployment value of the wear prediction model.

[0101] As a more specific implementation method, determining the second model as an effective impact energy accumulation model further includes: if the calculated second coefficient of determination is lower than a preset second coefficient threshold, or the second mean absolute error exceeds a preset second error threshold, then it is determined that the model does not have a statistically significant relationship with the actual data. In this case, the construction steps need to be re-executed to generate a new second model. By setting strict verification standards and a feedback reconstruction mechanism, it is ensured that only models with sufficient prediction accuracy and statistical significance will be ultimately adopted, thereby improving the rigor of the modeling process and the reliability of the model output. At the same time, it also enables the overall modeling system to have the ability to self-correct and continuously optimize, effectively ensuring the long-term applicability and stability of the model under real complex working conditions.

[0102] For example, the expression for the first model is as shown in formula (1):

[0103] (1)

[0104] In formula (1):

[0105] Instantaneous wear rate of wall thickness (unit: mm / s);

[0106] : Silica powder mass concentration (unit: kg / m³) ; Indicates silicon powder concentration;

[0107] Airflow velocity (unit: m / s);

[0108] The speed index is a key parameter to be determined.

[0109] Impact angle correction factor;

[0110] Comprehensive wear coefficient (unit: mm³·S) n-2 / kg).

[0111] Optionally, , and The specific value is determined through multiple linear regression.

[0112] For example, the expression for the cumulative wall thickness reduction ΔW is shown in formula (2):

[0113] (2)

[0114] In formula (2),

[0115] Cumulative wall thickness reduction (unit: mm);

[0116] It represents the cumulative effective impact energy, which is a dimensionless integral quantity that characterizes the sum of historical impact energies;

[0117] The regression coefficients represent the model values ​​used to adjust the influence of cumulative effective impact energy on the cumulative wall thickness reduction (ΔW).

[0118] β represents the model regression coefficient, used to adjust the degree of influence of cumulative effective impact energy on cumulative wall thickness reduction (ΔW) (exponential term).

[0119] For example, the expression for the second model is as shown in formula (3):

[0120] (3)

[0121] In formula (3):

[0122] Indicates the first Silicon powder mass concentration at each time step;

[0123] Indicates the first Airflow velocity at each time step;

[0124] Indicates the first Median particle size of silicon powder at each time step;

[0125] This indicates the time interval, that is, the duration of each time step;

[0126] T represents the total number of time steps, and the upper limit of the summation.

[0127] The validation and prediction phase of the second model requires validation of both the first model (which can be a dynamic wear rate model) and the second model (which can be an effective impact energy accumulation model) based on relevant formulas (the formulas can be those in the example) and a new dataset containing silicon powder concentration, airflow velocity in the pipe, and median particle size of silicon powder. The prediction accuracy of the model is quantified by calculating the coefficient of determination (R²) and the mean absolute error (MAE). If the R² and MAE results show no significant relationship between the model and the data, the most recent dataset should be selected for re-prediction; if a significant relationship exists, the model can be directly used to calculate the remaining pipe wall thickness and accumulated effective impact energy in real time.

[0128] In practice, to ensure the reliability of both the first and second models, in addition to the aforementioned verification methods, other verification methods can be selectively adopted based on specific scenario needs. These include: statistical hypothesis testing (including F-test and t-test), cross-validation (e.g., k-fold cross-validation), hold-out verification, business rule compliance verification, expert experience review, and model selection criteria such as AIC (Akaike Information Criterion) and BIC (Bayesian Information Criterion). These methods can comprehensively evaluate model performance from different dimensions: the F-test helps determine the significance of the overall regression relationship of the model; the t-test can analyze the statistical significance of the influence of each input variable; k-fold cross-validation can more robustly estimate model performance and alleviate overfitting problems under limited data conditions; AIC and BIC, by balancing model complexity and goodness of fit, help select the prediction model with the optimal structure. Furthermore, in industrial application scenarios, the experience judgment of engineering experts can be introduced, and the model output can be consistently verified in conjunction with on-site operation and maintenance rules, thereby ensuring that the model is not only mathematically reliable but also has sufficient safety and credibility in actual engineering operation.

[0129] Step S4: Compare the predicted remaining pipe wall thickness at each future time in the first future pipe wall remaining thickness sequence and the second future pipe wall remaining thickness sequence with the preset pipe wall remaining thickness safety threshold to determine the first predicted remaining lifetime corresponding to the first future pipe wall remaining thickness sequence and the second predicted remaining lifetime corresponding to the second future pipe wall remaining thickness sequence.

[0130] Step S5: Based on the first and second predicted remaining lifetimes, generate early warning information for silicon powder erosion corrosion of the pipeline and output the early warning information.

[0131] As a specific implementation method, the process of generating early warning information based on the first and second predicted remaining lifetimes is implemented as follows: First, the first and second predicted remaining lifetimes are compared for consistency. If the difference in days between the two is less than a preset tolerance threshold, a consistency early warning is generated. This information includes a unified predicted remaining lifetime in days after integrating the results of the two models, and provides specific maintenance time window suggestions accordingly. If the difference in days between the two is greater than or equal to the preset tolerance threshold, a difference early warning is generated. This information will simultaneously list the predicted remaining lifetimes of the two models, clearly indicate the uncertainty in the prediction, and provide further operation and maintenance inspection suggestions. Finally, the generated consistency or difference early warning information is output as the final early warning information for silicon powder erosion corrosion of pipelines.

[0132] As a specific implementation method, the method also includes a step of determining pipeline health. The specific implementation process is as follows: First, obtain the currently measured remaining pipe wall thickness data. Next, calculate the difference between the measured remaining pipe wall thickness data and a preset safe threshold for remaining pipe wall thickness, thereby obtaining a safe margin value for remaining pipe wall thickness. Finally, based on this safe margin value, through preset mapping or grading rules, generate a health index that can intuitively reflect the current structural integrity and safety reserve of the pipeline.

[0133] As a specific implementation method, after determining the pipeline health status, the method also includes data integration and visualization steps. The specific implementation process is as follows: First, the current silicon powder mass concentration data, current airflow velocity data, current measured remaining pipe wall thickness data, health indicators, first predicted remaining lifespan, second predicted remaining lifespan, early warning information, first future remaining pipe wall thickness sequence, second future remaining pipe wall thickness sequence, real-time calculated cumulative remaining pipe wall thickness reduction, real-time calculated cumulative effective impact energy, and real-time remaining pipe wall thickness value calculated based on the current measured remaining pipe wall thickness data and a preset safety threshold are all sent to the front-end display module. Subsequently, the front-end display module generates and displays a comprehensive visualization interface based on all the received data. The interface displays the following: the current silicon powder mass concentration data and its corresponding real-time trend curve; the current airflow velocity data and its corresponding real-time trend curve; the current measured remaining pipe wall thickness data and its corresponding historical trend curve; the real-time remaining pipe wall thickness value; the cumulative remaining pipe wall thickness reduction value; the cumulative effective impact energy value; the prediction curve corresponding to the first future remaining pipe wall thickness sequence; the prediction curve corresponding to the second future remaining pipe wall thickness sequence; health indicators displayed in numerical or grade form; the first predicted remaining lifespan; the second predicted remaining lifespan; and warning information clearly displayed in text or graphic form.

[0134] To clearly demonstrate the practical application effect of the method described in this embodiment, the following section combines the operating data of the tail gas pipeline of a polysilicon production plant's reduction furnace with... Figure 2 The overall process shown provides a step-by-step explanation of the entire process of model construction, verification, and early warning.

[0135] 1. Data Acquisition and Preprocessing:

[0136] In this embodiment, valid monitoring data was collected over a continuous 10-day operating cycle (approximately 864,000 seconds in total), with a sampling interval of 1 second. To build and validate the model, the data from the first 7 days was used as the training set for identifying and fitting model parameters; the data from the remaining 3 days served as an independent test set to evaluate the model's prediction accuracy and generalization ability. As shown in Table 1, an example dataset containing 120 time points (5-minute intervals) was extracted from the original data, fully demonstrating the data structure used for model building and validation.

[0137] 2. Model Construction and Parameter Identification:

[0138] The collected data points were equally distributed for model training (the first 60 points) and model validation (the last 60 points). Wear rate (dW / dt): calculated by dividing the difference in wall thickness values ​​between adjacent time points by the time interval (300 seconds), i.e. .

[0139] Cumulative energy (E_eff): According to the formula The calculation starts from time point 1 and is accumulated.

[0140] Cumulative thinning amount (ΔW): obtained by subtracting the current measured wall thickness from the initial wall thickness (10.0 mm), i.e. .

[0141] Table 1: Example of complete data used for model parameter identification and validation (120 time points)

[0142] Time point (minutes) Flow velocity v (m / s) Silicon powder concentration C (g / m³) Median particle size D50 (μm) Measured wall thickness W (mm) <![CDATA[Calculate the wear rate dW / dt (10⁻ 6 mm / s)]]> <![CDATA[Accumulated energy E_eff (×10 6 )]]> Cumulative thinning amount ΔW (mm) 1 24.8 78.5 6.1 9.99712 10.43 2.5 0.00288 2 25.1 82.3 5.9 9.99401 10.67 5.05 0.00599 3 25.4 85.7 6.2 9.99005 11.98 8.48 0.00995 4 25.6 88.9 6 9.98626 11.65 12 0.01374 5 24.9 91.5 6.3 9.98245 11.83 15.36 0.01755 6 24.7 94.2 5.8 9.97872 11.1 18.69 0.02128 7 24.5 96.8 6.1 9.9749 11.87 22.04 0.0251 8 24.3 99.4 5.9 9.97113 11.63 25.39 0.02887 9 24.1 102 6.2 9.96729 12.22 28.76 0.03271 10 23.9 104.6 6 9.96347 11.95 32.12 0.03653 11 24.2 101.2 5.8 9.95978 11.28 35.4 0.04022 12 24.5 97.8 6 9.95612 11.52 38.76 0.04388 13 24.8 94.4 6.1 9.95249 11.77 42.2 0.04751 14 25.1 91 6.3 9.94889 12.02 45.72 0.05111 15 25.4 87.6 6 9.94532 11.68 49.16 0.05468 16 25.7 84.2 5.9 9.94178 11.83 52.68 0.05822 17 26 80.8 6.2 9.93827 12.48 56.34 0.06173 18 26.3 77.4 5.9 9.93479 11.83 59.89 0.06521 19 26.6 74 5.7 9.93134 11.18 63.33 0.06866 20 26.9 70.6 5.5 9.92792 10.53 66.66 0.07208 21 27.2 67.2 5.3 9.92453 9.88 69.88 0.07547 22 26.9 69.1 5.4 9.92117 10.23 73.18 0.07883 23 26.6 71 5.5 9.91784 10.58 76.56 0.08216 24 26.3 72.9 5.6 9.91454 10.93 80.02 0.08546 25 26 74.8 5.7 9.91127 11.28 83.56 0.08873 26 25.7 76.7 5.8 9.90803 11.63 87.18 0.09197 27 25.4 78.6 5.9 9.90482 11.98 90.88 0.09518 28 25.1 80.5 6 9.90164 12.33 94.66 0.09836 29 24.8 82.4 6.1 9.89849 12.68 98.52 0.10151 30 24.5 84.3 6.2 9.89537 13.03 102.46 0.10463 31 24.8 81.5 6 9.89238 12.37 106.32 0.10762 32 25.1 78.7 5.9 9.88942 12.12 110.22 0.11058 33 25.4 75.9 5.8 9.88649 11.87 114.14 0.11351 34 25.7 73.1 5.7 9.88359 11.62 118.08 0.11641 35 26 70.3 5.6 9.88072 11.37 122.04 0.11928 36 26.3 67.5 5.5 9.87788 11.12 126.02 0.12212 37 26.6 64.7 5.4 9.87507 10.87 130.02 0.12493 38 26.9 61.9 5.3 9.87229 10.62 134.04 0.12771 39 27.2 59.1 5.2 9.86954 10.37 138.08 0.13046 40 27.5 56.3 5.1 9.86682 10.12 142.14 0.13318 41 27.2 57.9 5.2 9.86413 10.37 146.2 0.13587 42 26.9 59.5 5.3 9.86147 10.62 150.3 0.13853 43 26.6 61.1 5.4 9.85884 10.87 154.44 0.14116 44 26.3 62.7 5.5 9.85624 11.12 158.62 0.14376 45 26 64.3 5.6 9.85367 11.37 162.84 0.14633 46 25.7 65.9 5.7 9.85113 11.62 167.1 0.14887 47 25.4 67.5 5.8 9.84862 11.87 171.4 0.15138 48 25.1 69.1 5.9 9.84614 12.12 175.74 0.15386 49 24.8 70.7 6 9.84369 12.37 180.12 0.15631 50 24.5 72.3 6.1 9.84127 12.62 184.54 0.15873 51 24.8 70.2 6 9.83898 12.37 188.92 0.16102 52 25.1 68.1 5.9 9.83672 12.12 193.28 0.16328 53 25.4 66 5.8 9.83449 11.87 197.62 0.16551 54 25.7 63.9 5.7 9.83229 11.62 201.94 0.16771 55 26 61.8 5.6 9.83012 11.37 206.24 0.16988 56 26.3 59.7 5.5 9.82798 11.12 210.52 0.17202 57 26.6 57.6 5.4 9.82587 10.87 214.78 0.17413 58 26.9 55.5 5.3 9.82379 10.62 219.02 0.17621 59 27.2 53.4 5.2 9.82174 10.37 223.24 0.17826 60 27.5 51.3 5.1 9.81972 10.12 227.44 0.18028 61 27.2 52.6 5.2 9.81773 10.37 231.64 0.18227 62 26.9 53.9 5.3 9.81577 10.62 235.86 0.18423 63 26.6 55.2 5.4 9.81384 10.87 240.1 0.18616 64 26.3 56.5 5.5 9.81194 11.12 244.36 0.18806 65 26 57.8 5.6 9.81007 11.37 248.64 0.18993 66 25.7 59.1 5.7 9.80823 11.62 252.94 0.19177 67 25.4 60.4 5.8 9.80642 11.87 257.26 0.19358 68 25.1 61.7 5.9 9.80464 12.12 261.6 0.19536 69 24.8 63 6 9.80289 12.37 265.96 0.19711 70 24.5 64.3 6.1 9.80117 12.62 270.34 0.19883 71 24.8 62.5 6 9.79948 12.37 274.68 0.20052 72 25.1 60.7 5.9 9.79782 12.12 279 0.20218 73 25.4 58.9 5.8 9.79619 11.87 283.3 0.20381 74 25.7 57.1 5.7 9.79459 11.62 287.58 0.20541 75 26 55.3 5.6 9.79302 11.37 291.84 0.20698 76 26.3 53.5 5.5 9.79148 11.12 296.08 0.20852 77 26.6 51.7 5.4 9.78997 10.87 300.3 0.21003 78 26.9 49.9 5.3 9.78849 10.62 304.5 0.21151 79 27.2 48.1 5.2 9.78704 10.37 308.68 0.21296 80 27.5 46.3 5.1 9.78562 10.12 312.84 0.21438 81 27.2 47.4 5.2 9.78423 10.37 317 0.21577 82 26.9 48.5 5.3 9.78287 10.62 321.18 0.21713 83 26.6 49.6 5.4 9.78154 10.87 325.38 0.21846 84 26.3 50.7 5.5 9.78024 11.12 329.6 0.21976 85 26 51.8 5.6 9.77897 11.37 333.84 0.22103 86 25.7 52.9 5.7 9.77773 11.62 338.1 0.22227 87 25.4 54 5.8 9.77652 11.87 342.38 0.22348 88 25.1 55.1 5.9 9.77534 12.12 346.68 0.22466 89 24.8 56.2 6 9.77419 12.37 351 0.22581 90 24.5 57.3 6.1 9.77307 12.62 355.34 0.22693 91 24.8 55.8 6 9.77198 12.37 359.66 0.22802 92 25.1 54.3 5.9 9.77092 12.12 363.96 0.22908 93 25.4 52.8 5.8 9.76989 11.87 368.24 0.23011 94 25.7 51.3 5.7 9.76889 11.62 372.5 0.23111 95 26 49.8 5.6 9.76792 11.37 376.74 0.23208 96 26.3 48.3 5.5 9.76698 11.12 380.96 0.23302 97 26.6 46.8 5.4 9.76607 10.87 385.16 0.23393 98 26.9 45.3 5.3 9.76519 10.62 389.34 0.23481 99 27.2 43.8 5.2 9.76434 10.37 393.5 0.23566 100 27.5 42.3 5.1 9.76352 10.12 397.64 0.23648 101 27.2 43.2 5.2 9.76273 10.37 401.78 0.23727 102 26.9 44.1 5.3 9.76197 10.62 405.94 0.23803 103 26.6 45 5.4 9.76124 10.87 410.12 0.23876 104 26.3 45.9 5.5 9.76054 11.12 414.32 0.23946 105 26 46.8 5.6 9.75987 11.37 418.54 0.24013 106 25.7 47.7 5.7 9.75923 11.62 422.78 0.24077 107 25.4 48.6 5.8 9.75862 11.87 427.04 0.24138 108 25.1 49.5 5.9 9.75804 12.12 431.32 0.24196 109 24.8 50.4 6 9.75749 12.37 435.62 0.24251 110 24.5 51.3 6.1 9.75697 12.62 439.94 0.24303 111 24.8 50 6 9.75648 12.37 444.26 0.24352 112 25.1 48.7 5.9 9.75602 12.12 448.56 0.24398 113 25.4 47.4 5.8 9.75559 11.87 452.84 0.24441 114 25.7 46.1 5.7 9.75519 11.62 457.1 0.24481 115 26 44.8 5.6 9.75482 11.37 461.34 0.24518 116 26.3 43.5 5.5 9.75448 11.12 465.56 0.24552 117 26.6 42.2 5.4 9.75417 10.87 469.76 0.24583 118 26.9 40.9 5.3 9.75389 10.62 473.94 0.24611 119 27.2 39.6 5.2 9.75364 10.37 478.1 0.24636 120 27.5 38.3 5.1 9.75342 10.12 482.24 0.24658

[0143] The parameters of the first model are identified using a multiple linear regression equation, the formula of which is as follows (expression 1a below):

[0144]

[0145] The results obtained by linear regression on the training set are shown in Table 2.

[0146] Table 2: Results of Multiple Linear Regression for the First Model

[0147]

[0148] Therefore, we get:

[0149] Speed ​​Index Composite parameters Based on the obtained parameters, construct the first model, whose expression is (as shown in expression 1b below):

[0150]

[0151] Next, the parameters of the second model are solved:

[0152] Perform the univariate linear regression described in formula (3) using the training set data, i.e., for and Perform fitting:

[0153] The results are shown in Table 3.

[0154] Table 3: Results of the univariate linear regression of the second model

[0155]

[0156] Thus, we obtain: the index ,coefficient Therefore, the expression for the second model can be obtained as (expression 2c below):

[0157] .

[0158] 3. Model Validation and Prediction:

[0159] Use the test set data to validate the first and second models.

[0160] The validation results of the first model will be on the test set. and Substitute the values ​​into the model to calculate the predicted wear rate, compare it with the reference value, and obtain the results in Table 4.

[0161] Table 4: Validation results of the first model on the test set

[0162] Statistics of the entire test set numerical values Coefficient of determination R² 0.943 Mean Absolute Error (MAE) <![CDATA[1.85×10⁻ 6 mm / s]]> Root Mean Square Error (RMSE) <![CDATA[2.41×10⁻ 6 mm / s]]>

[0163] The cumulative calculation for each time point of the test set Substitute into model (2c) to predict cumulative thinning amount The results were compared with the reference values, and Table 5 shows the results.

[0164] Table 5: Validation results of the second model on the test set

[0165] Statistics of the entire test set numerical values Coefficient of determination R² 0.998 Mean Absolute Error (MAE) 0.0047mm Root Mean Square Error (RMSE) 0.0061mm

[0166] Verification results: The first model (R²=0.943) can accurately track the dynamic changes in wear rate, and the second model (R²=0.998) can predict the cumulative wear amount with extremely high accuracy. Both models have been validated and can be used for practical early warning.

[0167] Optionally, for predicting the future silicon powder concentration C and silicon powder flow rate v, the average of historical data is generally chosen. If the changing trend of both is considered, LSTM can be used for prediction.

[0168] In this embodiment, a prediction example is given:

[0169] Mean method:

[0170] From the data in Table 1, we can calculate:

[0171] Average flow velocity v avg ≈25.8m / s

[0172] Average silicon powder concentration C avg ≈63.5g / m 3 Converted to silicon powder mass concentration =0.0635kg / m 3 The results of the iterative calculations are shown in Table 6.

[0173] Table 6: Model Iteration Results

[0174] Predicted number of days Model 1: Remaining wall thickness (mm) <![CDATA[Second Model: Cumulative Energy (10 6 )]]> Second model: Remaining wall thickness (mm) Daily wear and tear (mm) Cumulative wear (mm) 0 (Current) 9.75342 482.24 9.75342 - 0.24658 5 9.15503 573.74 9.15067 0.1197 0.84497 10 8.55664 665.24 8.54764 0.1197 1.44336 15 7.95825 756.74 7.94433 0.1197 2.04175 20 7.35986 848.24 7.34075 0.1197 2.64014 21 7.24016 866.54 7.22076 0.1197 2.75984 22 7.12046 884.84 7.10074 0.1197 2.87954 23 7.00076 903.14 6.98069 0.1197 2.99924 24 6.88106 921.44 6.86060 0.1197 3.11894

[0175] It can be observed that the wall thickness of the first model was less than 7 mm on day 24, and the wall thickness of the second model was less than 7 mm on day 23.

[0176] As shown in this example, the predicted remaining lifespan is approximately 23 days. The system will generate a report stating: "Based on the current operating trend, the exhaust manifold elbow is expected to reach the maintenance threshold in 23 days (±2 days). It is recommended to schedule inspection or replacement during the planned downtime window (e.g., days 20-25)." Furthermore, if the two model predictions are inconsistent, the system will trigger an intermediate alert indicating "Prediction uncertainty." This alerts maintenance personnel:

[0177] Check the quality of sensor data.

[0178] Pay attention to whether there are drastic changes in process conditions.

[0179] More frequent short-term monitoring or on-site verification may be required.

[0180] A method for predicting silicon powder concentration C and silicon powder flow rate v based on LSTM time series:

[0181] Using the last 60 sets of data in Table 1, we used LSTM to identify the changing trends of silicon powder concentration C and silicon powder flow rate v to make predictions. The results are shown in Table 7.

[0182] Table 7: Silicon powder concentration, silicon powder flow rate and model iteration results

[0183] Days C(g / m³) v(m / s) Model 1: Daily Wear Amount (mm) Model 1: Remaining wall thickness (mm) <![CDATA[Second model: daily energy increment (10 6 )]]> <![CDATA[Second Model: Cumulative Energy (10 6 )]]> Second model: Remaining wall thickness (mm) 0 - - - 9.75342 - 482.24 9.75342 1 63.8 25.8 0.1193 9.63412 18.25 500.49 9.63463 2 64.2 25.9 0.1205 9.51362 18.55 519.04 9.51413 3 64.5 26.0 0.1217 9.39192 18.85 537.89 9.39243 4 64.9 26.1 0.1232 9.26872 19.20 557.09 9.26923 5 65.2 26.2 0.1245 9.14422 19.50 576.59 9.14473 6 65.5 26.2 0.1248 9.01942 19.60 596.19 9.01993 7 65.8 26.3 0.1260 8.89342 19.90 616.09 8.89393 8 66.1 26.4 0.1273 8.76612 20.25 636.34 8.76663 9 66.4 26.4 0.1275 8.63862 20.35 656.69 8.63913 10 66.6 26.5 0.1285 8.51012 20.60 677.29 8.51063 11 66.8 26.5 0.1288 8.38132 20.70 697.99 8.38183 12 67.0 26.5 0.1292 8.25212 20.85 718.84 8.25263 13 67.1 26.5 0.1294 8.12272 20.90 739.74 8.12323 14 67.2 26.5 0.1296 7.99312 20.95 760.69 7.99363 15 67.3 26.5 0.1298 7.86332 21.00 781.69 7.86383 16 67.4 26.5 0.1300 7.73332 21.05 802.74 7.73383 17 67.5 26.5 0.1302 7.60312 21.10 823.84 7.60363 18 67.5 26.5 0.1302 7.47292 21.10 844.94 7.47343 19 67.5 26.5 0.1302 7.34272 21.10 866.04 7.34323 20 67.5 26.5 0.1302 7.21252 21.10 887.14 7.21303 21 67.5 26.5 0.1302 7.08232 21.10 908.24 7.08283 22 67.5 26.5 0.1302 6.95212 21.10 929.34 6.95263 23 67.5 26.5 0.1302 6.82192 21.10 950.44 6.82243 24 67.5 26.5 0.1302 6.69172 21.10 971.54 6.69223 25 67.5 26.5 0.1302 6.56152 21.10 992.64 6.56203 26 67.5 26.5 0.1302 6.43132 21.10 1013.74 6.43183 27 67.5 26.5 0.1302 6.30112 21.10 1034.84 6.30163 28 67.5 26.5 0.1302 6.17092 21.10 1055.94 6.17143 29 67.5 26.5 0.1302 6.04072 21.10 1077.04 6.04123 30 67.5 26.5 0.1302 5.91052 21.10 1098.14 5.91103

[0184] 4. Early warning generation.

[0185] The prediction results show that the first future remaining pipe wall thickness sequence calculated by the first model is predicted to fall below the safety threshold of 7 mm for the first time on day 22. Simultaneously, the second future remaining pipe wall thickness sequence calculated by the second model is also predicted to fall below this safety threshold for the first time on day 22. The prediction results from the two independent models are highly consistent, both indicating that the pipeline's remaining safe life is approximately 22 days. This example demonstrates that, based on the collaborative prediction of the two models, the system can draw clear and mutually corroborating risk warning conclusions.

[0186] Figure 3 The system demonstrates the complete logic from data acquisition to front-end display. Real-time data collected by the atomization monitor, ultrasonic thickness gauge, and flow meter are stored in the "Data Storage" module. The "Training Module" uses historical data to build and validate the aforementioned models. The "Calculation Module" receives the validated models and, combined with real-time / predicted operating data, calculates indicators such as remaining wall thickness and cumulative damage. The "Early Warning Module" generates early warning information based on the predicted remaining lifespan of the two models (e.g., "Maintenance threshold is expected to be reached in 23 days; inspection is recommended between days 20-25"). The "Health Module" calculates the current health status of the pipeline. All results are presented to the user via the "Front-end Display" module. If a model fails, the "Emergency Status Module" switches to directly displaying the sensor-measured wall thickness to ensure continuous monitoring.

[0187] This embodiment provides an early warning method for erosion corrosion in reduction furnace tail gas pipelines. By integrating multi-source real-time data acquisition, a dynamic wear rate model, and an effective impact energy accumulation model, it achieves accurate prediction of the pipeline's remaining wall thickness and future wear trends. The method first deploys a sensor network to acquire key parameters including silicon powder mass concentration, airflow velocity, and median particle size, and performs data preprocessing to ensure the quality of the input data to the model. Next, it uses historical data to train a Long Short-Term Memory (LSTM) network or employs a mean prediction method to predict parameter changes over future periods. Based on these predictions, two independent future pipeline wall thickness sequences are calculated using the dynamic wear rate model and the effective impact energy accumulation model, respectively. Finally, these two sequences are compared with a safety threshold to determine the pipeline's remaining lifespan and generate early warning information. This method not only accurately tracks the changing trend of the wear rate but also predicts the cumulative wear amount with high precision, thus providing a scientific basis for pipeline maintenance and significantly improving the safety and efficiency of industrial production. The synergistic use of the two models enhances the reliability and accuracy of the prediction results, helps to promptly identify potential risks, guides reasonable maintenance planning, and thereby reduces unplanned downtime and repair costs.

[0188] Example 2:

[0189] This embodiment provides an early warning device for erosion corrosion in a reduction furnace tail gas pipeline. By integrating data acquisition, operating condition prediction, dual-model calculation, and collaborative early warning functions, it achieves intelligent and proactive judgment of erosion corrosion risks. The following is a detailed description of the device's composition and functions.

[0190] like Figure 4 As shown, the device specifically includes the following four units: an acquisition unit 10, a prediction unit 20, a calculation unit 30, and an early warning generation unit 40. The specific functions of each unit and their interconnections are as follows:

[0191] 1. Obtain Unit 10:

[0192] The acquisition unit 10 is used to acquire multi-source real-time data from the target monitoring points. This data forms the basis for all subsequent analysis and prediction, including:

[0193] Historical silicon powder mass concentration time series data (unit: kg / m³).

[0194] Historical airflow velocity time series data (unit: m / s);

[0195] Current silicon powder mass concentration (unit: kg / m³);

[0196] Current airflow velocity (unit: m / s);

[0197] Median particle size of silicon powder (unit: μm).

[0198] 2. Prediction Unit 20:

[0199] The prediction unit 20 is connected to the acquisition unit 10 to receive multi-source real-time data. The core task of this unit is to predict the trends of future operating parameters using historical and current data. Specifically, it includes:

[0200] By using historical silicon powder mass concentration time series data and historical airflow velocity time series data, combined with the current real-time silicon powder mass concentration and airflow velocity, the changing trends of silicon powder mass concentration and airflow velocity within a specified future prediction period are predicted.

[0201] It outputs future silicon powder mass concentration prediction sequences and future airflow velocity prediction sequences, enabling a "look-ahead" view of key operating parameters and providing necessary predictive information for subsequent model calculations.

[0202] 3. Calculation Unit 30:

[0203] The computing unit 30 is connected to both the prediction unit 20 and the acquisition unit 10, and is used to receive future prediction sequences and current particle size data to perform parallel computation of the dual-path physical model, specifically including:

[0204] First future pipe wall remaining thickness sequence: Inputting the predicted future silicon powder mass concentration sequence and the predicted future airflow velocity sequence into a pre-constructed dynamic wear rate model, a time series representing the instantaneous wear evolution of the pipe wall remaining thickness is obtained. Second future pipe wall remaining thickness sequence: Inputting the median silicon powder particle size, the predicted future silicon powder mass concentration sequence, and the predicted future airflow velocity sequence into a pre-constructed effective impact energy accumulation model, a pipe wall remaining thickness sequence reflecting cumulative damage changes is calculated.

[0205] These two sequences predict changes in pipe wall thickness over a certain period of time from different physical perspectives, providing a basis for multi-dimensional risk assessment.

[0206] 4. Early Warning Generation Unit 40:

[0207] The early warning generation unit 40 is connected to the computing unit 30 to receive its output future wall thickness prediction sequence. Its main function is to achieve intelligent diagnosis and risk early warning, specifically including:

[0208] For each future moment in the "First Future Remaining Pipe Wall Thickness Sequence" and the "Second Future Remaining Pipe Wall Thickness Sequence," the predicted remaining pipe wall thickness data is compared with a set safety threshold. The time point at which the thickness might first fall below the safety threshold in both prediction models is calculated, resulting in the "First Predicted Remaining Life" and the "Second Predicted Remaining Life." Based on the difference between these two predicted remaining lifespans (e.g., comparing their consistency or adopting a conservative value strategy), a "Silica Powder Erosion Corrosion Pipeline Early Warning Information" is generated, including risk level, warning prompts, and maintenance recommendations.

[0209] In summary, the scouring and corrosion early warning device for the tail gas pipeline of the reduction furnace provided in this embodiment, through the coordinated work of the above four units, constitutes a complete closed-loop system from data perception, operating condition prediction, model calculation to intelligent decision-making; it realizes automatic perception, accurate prediction and early warning of the risk of scouring and corrosion of the tail gas pipeline of the reduction furnace, with high reliability and foresight, providing a scientific basis for maintenance management and ensuring the safe and stable operation of the production line.

[0210] The apparatus in this embodiment is capable of performing the method in Embodiment 1.

[0211] Example 3:

[0212] like Figure 5 As shown, this embodiment provides an electronic device, which includes a memory 200 and a processor 100. The memory 200 stores a computer program. When the processor 100 runs the computer program stored in the memory 200, the processor 100 executes the scouring and corrosion early warning method for the tail gas pipeline of the reduction furnace according to Embodiment 1.

[0213] It is understood that the above embodiments are merely exemplary implementations used to illustrate the principles of the present invention, and the present invention is not limited thereto. For those skilled in the art, various modifications and improvements can be made without departing from the spirit and essence of the present invention, and these modifications and improvements are also considered to be within the scope of protection of the present invention.

Claims

1. A method for early warning of erosion corrosion in a reduction furnace tail gas pipeline, characterized in that, The method includes the following steps: Acquire multi-source real-time data of target monitoring points in the tail gas pipeline of the reduction furnace; wherein, the multi-source real-time data includes historical silicon powder mass concentration time series data, historical airflow velocity time series data, current silicon powder mass concentration data, current airflow velocity data, and median particle size of silicon powder; Based on the historical silicon powder mass concentration time series data and the historical airflow velocity time series data, and combined with the current silicon powder mass concentration and the current airflow velocity data, the changes in silicon powder mass concentration and airflow velocity within a set prediction period in the future are predicted to obtain the future silicon powder mass concentration prediction sequence and the future airflow velocity prediction sequence. The predicted sequence of future silicon powder mass concentration and the predicted sequence of future airflow velocity are input into a pre-built dynamic wear rate model to calculate the first future pipe wall remaining thickness sequence; and the predicted sequence of future silicon powder median particle size, the predicted sequence of future silicon powder mass concentration and the predicted sequence of future airflow velocity are input into a pre-built effective impact energy accumulation model to calculate the second future pipe wall remaining thickness sequence. The predicted remaining pipe wall thickness at each future time in the first future pipe wall remaining thickness sequence and the second future pipe wall remaining thickness sequence are compared with a preset pipe wall remaining thickness safety threshold to determine the first predicted remaining lifetime corresponding to the first future pipe wall remaining thickness sequence and the second predicted remaining lifetime corresponding to the second future pipe wall remaining thickness sequence. Based on the first predicted remaining lifetime and the second predicted remaining lifetime, an early warning message for silicon powder erosion corrosion of the pipeline is generated and output.

2. The method for early warning of erosion and corrosion of the tail gas pipeline of the reduction furnace according to claim 1, characterized in that, The process involves using historical silicon powder mass concentration time-series data and historical airflow velocity time-series data, combined with current silicon powder mass concentration and current airflow velocity data, to predict changes in silicon powder mass concentration and airflow velocity within a predetermined prediction period in the future, resulting in a future silicon powder mass concentration prediction sequence and a future airflow velocity prediction sequence. Specifically, this includes: The current silicon powder mass concentration data and the current airflow velocity data are appended to the end of the historical silicon powder mass concentration time series data and the historical airflow velocity time series data, respectively, to form an expanded silicon powder mass concentration time series dataset and an expanded airflow velocity time series dataset. The changes in silicon powder mass concentration and airflow velocity within a set prediction period are predicted using either mean prediction or long short-term memory network prediction. The mean prediction method is as follows: calculate the arithmetic mean of the extended silicon powder mass concentration time series dataset as the predicted value of future silicon powder mass concentration, calculate the arithmetic mean of the extended airflow velocity time series dataset as the predicted value of future airflow velocity, and extend the predicted value of future silicon powder mass concentration and the predicted value of future airflow velocity into sequences with the same length as the set future prediction period, to obtain the predicted sequence of future silicon powder mass concentration and the predicted sequence of future airflow velocity. The Long Short-Term Memory (LSTM) network prediction method is as follows: the extended silicon powder mass concentration time series dataset and the extended airflow velocity time series dataset are input into a pre-trained LTM network model, and the LTM network model outputs the future silicon powder mass concentration prediction sequence and the future airflow velocity prediction sequence.

3. The method for early warning of erosion and corrosion of the tail gas pipeline of the reduction furnace according to claim 1, characterized in that, The generation of early warning information for silicon powder erosion corrosion of pipelines based on the first predicted remaining lifetime and the second predicted remaining lifetime specifically includes: Determine the difference between the first predicted remaining lifetime and the second predicted remaining lifetime; If the difference is less than a preset tolerance threshold, a consistency warning message is generated. The consistency warning message includes a unified prediction of the remaining lifespan in days and a maintenance time window recommendation. The warning message includes the consistency warning message. If the difference is greater than or equal to a preset tolerance threshold, a difference warning message is generated. The difference warning message includes the predicted remaining lifetime of the two models, prediction uncertainty prompts, and operation and maintenance inspection suggestions; the warning message includes the difference warning message.

4. The method for early warning of erosion and corrosion of the tail gas pipeline of the reduction furnace according to claim 1, characterized in that, The method further includes: Obtain the currently measured remaining pipe wall thickness data; The difference between the measured remaining pipe wall thickness data and the preset safe threshold for remaining pipe wall thickness is calculated to obtain the safe margin value for remaining pipe wall thickness. Based on the remaining thickness safety margin of the pipe wall, a health index reflecting the structural condition of the pipeline is generated.

5. The method for early warning of erosion and corrosion of the tail gas pipeline of the reduction furnace according to claim 4, characterized in that, After generating the health index reflecting the structural condition of the pipeline, the method further includes: The system acquires the current silicon powder mass concentration data, the current airflow velocity data, the current measured remaining pipe wall thickness data, the health index, the first predicted remaining lifespan, the second predicted remaining lifespan, the early warning information, the first future remaining pipe wall thickness sequence, the second future remaining pipe wall thickness sequence, the cumulative remaining pipe wall thickness reduction calculated in real time, the cumulative effective impact energy calculated in real time, and the real-time remaining pipe wall thickness value calculated based on the current measured remaining pipe wall thickness data and the preset remaining pipe wall thickness safety threshold. The display interface includes the following: the current silicon powder mass concentration data and its corresponding real-time trend curve; the current airflow velocity data and its corresponding real-time trend curve; the current measured remaining pipe wall thickness data and its corresponding historical trend curve; the real-time remaining pipe wall thickness value; the cumulative remaining pipe wall thickness reduction; the cumulative effective impact energy; the prediction curve corresponding to the first future remaining pipe wall thickness sequence; the prediction curve corresponding to the second future remaining pipe wall thickness sequence; the health index; the first predicted remaining lifespan; the second predicted remaining lifespan; and the warning information.

6. The method for early warning of erosion corrosion in the tail gas pipeline of the reduction furnace according to any one of claims 1 to 5, characterized in that, The method further includes: Obtain a historical pipe wall remaining thickness data sequence; perform a difference operation on the historical pipe wall remaining thickness data sequence and divide it by the sampling time interval to obtain a historical pipe wall remaining thickness instantaneous wear rate sequence; The historical pipe wall remaining thickness instantaneous wear rate sequence, the historical silicon powder mass concentration time series data, and the historical airflow velocity time series data are subjected to natural logarithmic transformation; Using the instantaneous wear rate sequence of the remaining pipe wall thickness after natural logarithmic transformation as the dependent variable, and the time series data of the historical airflow velocity and the time series data of the historical silicon powder mass concentration after natural logarithmic transformation as independent variables, a multiple linear regression analysis was performed to determine the velocity exponent parameter of the dynamic wear rate model, as well as the composite parameter of the comprehensive wear coefficient and the impact angle correction coefficient. Based on the velocity exponent parameter and the composite parameter, a first model is constructed. The first model is determined to be the dynamic wear rate model.

7. The method for early warning of erosion and corrosion of the tail gas pipeline of the reduction furnace according to claim 6, characterized in that, Determining the first model as the dynamic wear rate model includes: Obtain validation set data, which includes silicon powder mass concentration data and airflow velocity data within a preset validation time range; The silicon powder mass concentration data and airflow velocity data in the validation set data are input into the first model to calculate the first pipe wall remaining thickness prediction sequence. Based on the pipe wall remaining thickness data in the validation set data, the first coefficient of determination and the first mean absolute error between the first pipe wall remaining thickness prediction sequence and the actual data are calculated. If the first determining coefficient is greater than or equal to a preset first coefficient threshold, and the first average absolute error is less than or equal to a preset first error threshold, then the first model is determined to have a significant relationship, and the first model is determined to be the dynamic wear rate model.

8. The method for early warning of erosion and corrosion of the tail gas pipeline of the reduction furnace according to claim 7, characterized in that, The step of determining the first model as the dynamic wear rate model further includes: If the first coefficient of determination is less than the first coefficient threshold, or the first mean absolute error is greater than the first error threshold, then the first model is determined to have no significant relationship, and the first model is reconstructed.

9. The method for early warning of erosion corrosion in the tail gas pipeline of the reduction furnace according to any one of claims 1 to 5, characterized in that, The method further includes: Based on the historical time series data of remaining pipe wall thickness and the initial remaining pipe wall thickness value, the difference between the initial remaining pipe wall thickness value and the measured remaining pipe wall thickness value at each time point is calculated to obtain the historical cumulative remaining pipe wall thickness reduction sequence. The historical cumulative remaining pipe wall thickness reduction sequence and the historical cumulative effective impact energy sequence are respectively subjected to natural logarithmic transformation; Using the historical cumulative wall thickness reduction sequence after natural logarithmic transformation as the dependent variable and the historical cumulative effective impact energy sequence after natural logarithmic transformation as the independent variable, a univariate linear regression analysis was performed to determine the exponential and coefficient parameters of the effective impact energy accumulation model. Based on the exponential parameters and the coefficient parameters, a second model is constructed. The second model is determined to be the effective impact energy accumulation model.

10. The method for early warning of erosion corrosion in the tail gas pipeline of the reduction furnace according to claim 9, characterized in that, The determination that the second model is the effective impact energy accumulation model includes: Obtain validation set data, which includes silicon powder mass concentration data, airflow velocity data, median particle size of silicon powder, and remaining pipe wall thickness data within a preset validation time range; The silicon powder mass concentration data, airflow velocity data, and median particle size data of silicon powder in the validation set data are input into the effective impact energy accumulation model to calculate the second pipe wall remaining thickness prediction sequence. Based on the pipe wall remaining thickness data in the validation set data, the second coefficient of determination and the second mean absolute error between the second pipe wall remaining thickness prediction sequence and the actual data are calculated. If the second determining coefficient is greater than or equal to the preset second coefficient threshold, and the second average absolute error is less than or equal to the preset second error threshold, then the effective impact energy accumulation model is determined to have a significant relationship, and the second model is determined to be the effective impact energy accumulation model.

11. The method for early warning of erosion corrosion in the tail gas pipeline of the reduction furnace according to claim 10, characterized in that, The determination of the second model as the effective impact energy accumulation model further includes: If the second coefficient of determination is less than the second coefficient threshold, or the second mean absolute error is greater than the second error threshold, then the effective impact energy accumulation model is determined to have no significant relationship, and the second model is reconstructed.

12. A scouring and corrosion early warning device for a reduction furnace tail gas pipeline, characterized in that, The device includes: The acquisition unit is used to acquire multi-source real-time data of target monitoring points in the tail gas pipeline of the reduction furnace; wherein, the multi-source real-time data includes historical silicon powder mass concentration time series data, historical airflow velocity time series data, current silicon powder mass concentration data, current airflow velocity data, and median particle size of silicon powder. The prediction unit, connected to the acquisition unit, is used to predict the changes in silicon powder mass concentration and airflow velocity within a set prediction period in the future, based on the historical silicon powder mass concentration time series data and the historical airflow velocity time series data, and in combination with the current silicon powder mass concentration data and the current airflow velocity data, so as to obtain the future silicon powder mass concentration prediction sequence and the future airflow velocity prediction sequence. The calculation unit, connected to the prediction unit and the acquisition unit respectively, is used to input the future silicon powder mass concentration prediction sequence and the future airflow velocity prediction sequence into a pre-built dynamic wear rate model to calculate a first future pipe wall remaining thickness sequence; and to input the median particle size data of silicon powder, the future silicon powder mass concentration prediction sequence and the future airflow velocity prediction sequence into a pre-built effective impact energy accumulation model to calculate a second future pipe wall remaining thickness sequence. An early warning generation unit, connected to the calculation unit, is used to compare the predicted remaining pipe wall thickness at each future time in the first future pipe wall remaining thickness sequence and the second future pipe wall remaining thickness sequence with a preset pipe wall remaining thickness safety threshold, to determine the first predicted remaining lifetime corresponding to the first future pipe wall remaining thickness sequence and the second predicted remaining lifetime corresponding to the second future pipe wall remaining thickness sequence; based on the first predicted remaining lifetime and the second predicted remaining lifetime, to generate early warning information for silicon powder erosion corrosion of the pipe, and to output the early warning information.

13. An electronic device comprising a memory and a processor, characterized in that, The memory stores a computer program. When the processor runs the computer program stored in the memory, the processor executes the scouring and corrosion early warning method for the tail gas pipeline of the reduction furnace according to any one of claims 1 to 11.

Images