Equipment corrosion evaluation and life prediction method and application

Abstract

The invention relates to the technical field of equipment monitoring, in particular to an equipment corrosion evaluation and life prediction method and application, and the method comprises the following steps: deploying a sensor network in an easily-corroded area of coal chemical equipment, and collecting multi-dimensional data; carrying out abnormal value elimination, data compression, time synchronization and space-time alignment preprocessing on the collected multi-source data; image features are extracted through a convolutional neural network, processed data are analyzed through an LSTM-attention model, and a fuzzy comprehensive evaluation matrix is established to evaluate the corrosion level; a physical model based on the Faraday electrolysis law and a data driving model based on the Transform network are constructed, and the residual life is predicted through Bayesian network fusion output and Monte Carlo simulation. Through fusion of multi-source data and an intelligent algorithm, accurate evaluation of the corrosion state of the equipment and accurate prediction of the residual life are realized, and safe and efficient operation of the coal chemical equipment is guaranteed.

Description

Technical Field

[0001] The present invention relates to the technical field of equipment monitoring, and in particular to an equipment corrosion assessment and life prediction method and application. Background Art

[0002] In the coal chemical industry, equipment operates for long periods of time in extreme environments characterized by high temperatures (200-1500°C), high pressures (5-30 MPa), and rich corrosive media such as CO2, H2S, and Cl-. Taking a water-coal slurry gasifier as an example, its inner wall is subjected to long-term erosion at temperatures exceeding 1400°C. At the same time, the electrolyte solution formed by acidic gases such as H2S and HCN and condensed water, in conjunction with the mechanical wear of coal ash, results in a compound damage of "high-temperature corrosion + electrochemical corrosion + erosion." In ammonia synthesis units, the risk of hydrogen corrosion faced by the tower wall is particularly prominent. Under high temperature and high pressure, H2 molecules penetrate the steel lattice, reacting with carbon to form methane, leading to decarburization. At the same time, pitting corrosion caused by Cl- rust accelerates local failure.

[0003] The corrosion problem in pipeline systems is even more complex. For example, in a low-temperature methanol wash unit, at -40°C, acidic gases such as H2S and COS dissolve in the methanol solution, forming a corrosive medium. Fluid scour exacerbates this at pipe elbows and reducers, leading to turbulent corrosion. During downtime for maintenance, residual media comes into contact with air, generating sulfuric acid, which triggers under-scale corrosion. According to industry statistics, unplanned shutdowns due to corrosion in coal chemical plants account for an average of 28% of all plant shutdowns annually, with direct economic losses from a single shutdown reaching tens of millions of yuan.

[0004] At present, traditional assessment methods have significant limitations: single-point corrosion rate monitoring based on linear polarization resistance can only capture the uniform corrosion state and cannot reflect the coupling effect of local corrosion and various corrosion forms; the empirical life prediction model that relies on historical data fitting is difficult to adapt to the operating condition fluctuations caused by the frequent load changes of coal chemical equipment. For example, a coal-to-olefin project used traditional methods to predict the pipeline life of 8 years, but perforation and leakage occurred after 5 years of actual operation, exposing the failure risk of traditional assessment systems under complex working conditions. Therefore, there is an urgent need to build an assessment system that can dynamically quantify the interaction of multiple corrosion mechanisms to achieve accurate protection of coal chemical equipment throughout its life cycle. Summary of the Invention

[0005] The purpose of this invention is to provide a method and application for equipment corrosion assessment and life prediction. By integrating multi-source data with intelligent algorithms, accurate assessment of the equipment corrosion status and accurate prediction of the remaining life can be achieved, thereby ensuring the safe and efficient operation of coal chemical equipment.

[0006] To achieve the above objectives, the present invention provides a method for equipment corrosion assessment and life prediction, comprising the following steps:

[0007] S1. Multi-source data collection

[0008] Deploy sensor networks in corrosion-prone areas of coal chemical equipment to collect multi-dimensional data on equipment surface corrosion current density, environmental parameters, wall thickness, stress, and surface temperature images;

[0009] S2. Data Preprocessing

[0010] The multi-source data collected by S1 is transmitted to the edge computing node for outlier removal, data compression, time synchronization and space-time alignment preprocessing;

[0011] S3. Corrosion feature extraction and evaluation

[0012] Convolutional neural networks are used to extract image features, and the LSTM-attention model is used to analyze the data processed by S2, and a fuzzy comprehensive evaluation matrix is ​​established to assess the corrosion level.

[0013] S4. Remaining life prediction

[0014] Construct a physical model based on Faraday's electrolysis law and a data-driven model based on the Transformer network, fused the output through a Bayesian network, and combined it with Monte Carlo simulation to predict the remaining life;

[0015] S5. Dynamic calibration and method support

[0016] Model calibration is triggered based on offline detection data, corrosion level changes, and remaining life thresholds, and maintenance methods are formulated according to warning levels.

[0017] Preferably, in S1, the sensor network includes a linear polarization resistance sensor, a sampling and analysis sensor, a pH sensor, a temperature and humidity sensor, an ultrasonic thickness gauge, a strain gauge, and an infrared thermal imager.

[0018] Preferably, in S2, data outliers are eliminated based on the 3σ principle, sliding window mean filtering is used to compress high-frequency corrosion current density data, time synchronization is achieved through the NTP protocol, and a timestamp-based Kalman filter algorithm is used for spatiotemporal alignment and integration.

[0019] Preferably, in S3, a convolutional neural network is used to process the temperature image collected by the infrared thermal imager to extract the image features of the rust layer thickness and crack distribution; the corrosion rate, environmental parameters, equipment wall thickness data, strain data, and equipment surface temperature image data are analyzed through the LSTM-attention model to identify the coupling relationship between key influencing factors and the corrosion rate.

[0020] Preferably, in S3, the evaluation indicators of the fuzzy comprehensive evaluation matrix include corrosion rate, corrosion depth, environmental erosion index, and stress corrosion risk, and the weights of each indicator are 0.4, 0.3, 0.2, and 0.1, respectively. The corrosion level is determined by Z-score normalization, construction of the membership matrix, and calculation of the comprehensive evaluation vector. The corrosion level includes five levels: intact, mild corrosion, moderate corrosion, severe corrosion, and failure risk.

[0021] More preferably, in S3, the corrosion rate is a synergistic corrosion rate that establishes a synergistic effect of CO2 corrosion, H2S corrosion, and erosion corrosion;

[0022] CO2 corrosion is:

[0023]

[0024] Where, is the CO2 corrosion rate, mm / a; k1 is the CO2 corrosion rate constant, mm / (a·MPa·K); is the CO2 partial pressure, MPa; E a is the activation energy, kJ / mol, R is the ideal gas constant, T is the temperature in °C; ηpH = 10 0.1(4-pH) , is the pH value influence coefficient, and corrosion is significantly accelerated when pH ≤ 4.

[0025] H2S corrosion is:

[0026]

[0027] Where, is the H2S corrosion rate, mm / a; k2 is the H2S corrosion rate constant, mm / (a·√ppm); cH2S is the H2S concentration, ppm; c Cl - is the chloride ion concentration, ppm.

[0028] Erosion corrosion is:

[0029] υ crosion =k3·u 2 ·α·ρ s

[0030] Where, crosion is the erosion corrosion rate, mm / a; k3 is the erosion corrosion coefficient, mm / (a·(m / s) 2 kg / m 3 ); u is the fluid velocity, m / s; α is the solid particle volume fraction, %; ρ s is the particle density, kg / m 3 .

[0031] The synergistic corrosion rate is:

[0032]

[0033] Where, 协同 is the synergistic corrosion rate, mm / a; k4 is the synergistic effect coefficient m 3 / (a 2 ·MPa·ppm), which reflects the interaction acceleration effect between CO2 and H2S, and takes a value of 0.01 to 0.05.

[0034] More preferably, in S3, the image features are quantified by a convolutional neural network, and the fractal dimension D and crack density δ of the rust layer in the corrosion area are output;

[0035] The fractal dimension D of the rust layer is calculated using the box dimension algorithm to calculate the surface roughness of the rust layer;

[0036] The crack density δ is:

[0037]

[0038] Where, L crack is the total length of the crack, mm; A ROI is the area of ​​the region of interest, mm 2 .

[0039] More preferably, in S3, a method including image features and corrosion rate υ is constructed. 协同 , environmental parameters The 10-dimensional eigenvector of stress and strain (σ, ε):

[0040]

[0041] More preferably, in S3, the model construction of the comprehensive evaluation vector includes:

[0042] (1) Standardization of indicators

[0043]

[0044] When i=1, x1 is the corrosion rate; when i=2, x2 is the corrosion depth; when i=3, x3 is the environmental corrosion index; when i=4, x4 is the stress corrosion risk;

[0045] (2) Membership function

[0046] The trapezoidal distribution function is used to calculate the membership of each level μ j (x' i )(j=1~5, corresponding to 5 corrosion levels):

[0047]

[0048] Where a j , b j , cj is the threshold parameter of each level;

[0049] (3) Comprehensive evaluation vector

[0050]

[0051] Where ω1, ω2, ω3, and ω4 are the weight coefficients of each evaluation index respectively.

[0052] Preferably, in S4, the theoretical corrosion depth is calculated by combining the physical model with the material properties of the equipment, and the Transformer network is used to input historical corrosion depth, environmental parameters, operating conditions and maintenance record data to establish a data-driven model to predict the corrosion depth in the next 1-10 years.

[0053] More preferably, in S4, the physical model is to construct a theoretical corrosion rate basic equation based on Faraday's electrolysis law:

[0054]

[0055] Where, 理论 is the theoretical corrosion rate, mm / a; K is a constant, K = 3.27 × 10 -3 ;i corr is the corrosion current density, μA / cm 2 ; M is the molar mass of the metal, g / mol; n is the oxidation state number; ρ is the metal density, g / cm 3 ;

[0056] More preferably, in S4, the calculation formula for the theoretical corrosion depth is:

[0057] d 理论 =υ 理论 t+d0

[0058] Where t is time, years; d0 is the initial corrosion depth.

[0059] More preferably, in S4, the calculation formula for predicting the corrosion depth is:

[0060]

[0061] Preferably, in S4, the outputs of the physical model and the data-driven model are dynamically fused through a Bayesian network, and then the Monte Carlo simulation is used to sample uncertain parameters and run it 100,000 times to generate a probability distribution of corrosion depth. The remaining life is defined as the time quantile when the corrosion depth reaches the critical value, and the 5% quantile, 50% quantile, and 95% quantile are output.

[0062] More preferably, the output of the dynamic fusion of the physical model and the data-driven model through the Bayesian network is:

[0063] Where, is the final corrosion assessment result obtained after fusion;

[0064] d 理论 is the theoretical corrosion depth calculated based on the physical model;

[0065] d 预测 The corrosion depth predicted by the data-driven model;

[0066] is a dynamic weight function, and the parameters k, t0 are determined by Bayesian optimization, with an initial value of ω = 0.5.

[0067] Preferably, in S5, when offline detection data is uploaded, or the corrosion level rises for three consecutive days and the predicted remaining life is lower than the warning threshold, model calibration is triggered, and the model calibration is performed by a model parameter fine-tuning algorithm based on stochastic gradient descent.

[0068] Preferably, in S5, the warning levels include yellow warning, orange warning and red warning. The maintenance method is to increase the sensor sampling frequency and start weekly infrared inspections during yellow warnings, expand the ultrasonic thickness measurement detection range during orange warnings, and conduct a comprehensive analysis of the corrosive components of the medium, and formulate a replacement plan and arrange shutdown maintenance during red warnings.

[0069] The present invention also provides an application of the above-mentioned equipment corrosion assessment and life prediction method, which is applied to the corrosion assessment and life prediction of gasifiers, synthesis towers, separation towers, and process pipelines in coal chemical equipment. The coal chemical equipment is in a high-temperature, high-pressure, corrosive medium environment containing CO2 and H2S.

[0070] Beneficial effects of the present invention:

[0071] (1) The present invention adopts the above-mentioned equipment corrosion assessment and life prediction method and application, and comprehensively obtains equipment corrosion-related information through multi-source data fusion and collection, breaking through the limitations of traditional single parameter monitoring and being able to accurately reflect the equipment corrosion status under the synergistic effect of multiple corrosion mechanisms;

[0072] (2) The present invention adopts the above-mentioned equipment corrosion assessment and life prediction method and application, adopts a hybrid prediction framework that combines physical models with data-driven models, combines Bayesian networks and Monte Carlo simulation, takes into account the physical and chemical nature of corrosion, and utilizes the laws of historical data, significantly improving the accuracy and reliability of remaining life prediction;

[0073] (3) The present invention adopts the above-mentioned equipment corrosion assessment and life prediction method and application, dynamic calibration mechanism and early warning maintenance strategy, which can timely adjust the prediction model and maintenance plan according to the actual operation of the equipment, effectively reduce the risk of equipment failure caused by corrosion, and improve the safety and economy of coal chemical equipment.

[0074] The technical solution of the present invention is further described in detail below through examples. DETAILED DESCRIPTION

[0075] The present invention will be further described below with reference to the following embodiments. Unless otherwise defined, technical or scientific terms used herein shall have the same meanings as those commonly understood by persons of ordinary skill in the art to which the present invention pertains. The above-mentioned features or features described in the specific examples of the present invention may be combined in any manner. These specific embodiments are intended only to illustrate the present invention and are not intended to limit the scope of the present invention.

[0076] Example

[0077] The present invention provides a method for equipment corrosion assessment and life prediction, comprising the following steps:

[0078] S1. Multi-source data collection

[0079] A sensor network is deployed in corrosion-prone areas of coal chemical equipment to collect multi-dimensional data on equipment surface corrosion current density, environmental parameters, wall thickness, stress, and surface temperature images.

[0080] The sensor network includes a linear polarization resistance sensor, a chloride ion concentration sensor, a pH sensor, a temperature and humidity sensor, an ultrasonic thickness gauge, a strain gauge, and an infrared thermal imager. The linear polarization resistance sensor collects real-time corrosion current density data on the equipment surface for corrosion rate calculation. Environmental sensors, such as the chloride ion concentration sensor, pH sensor, temperature and humidity sensor, obtain environmental parameters such as the chloride concentration, pH value, temperature, and humidity of the equipment's environment. An ultrasonic thickness gauge regularly measures the equipment's wall thickness, and a strain gauge monitors strain data in areas of stress concentration. The infrared thermal imager regularly collects temperature images of the equipment surface to identify areas of abnormal temperature due to corrosion.

[0081] S2. Data Preprocessing

[0082] The multi-source data collected by S1 is transmitted to the edge computing node for outlier removal, data compression, time synchronization and space-time alignment preprocessing.

[0083] Outliers in corrosion current density, wall thickness, and other data were removed based on the 3σ principle. High-frequency corrosion current density data was compressed using a sliding window mean filter (5-minute window size). Sensor data was synchronized using the NTP protocol to ensure a timestamp error of less than 10ms. A spatiotemporal alignment algorithm (such as a timestamp-based Kalman filter) was used to integrate heterogeneous data and construct the equipment corrosion state vector.

[0084] S3. Corrosion feature extraction and evaluation

[0085] The temperature images collected by the infrared thermal imager are processed using a convolutional neural network to extract the image features of rust layer thickness and crack distribution.

[0086] The image features are quantified through convolutional neural networks, and the fractal dimension D and crack density δ of the rust layer in the corrosion area are output;

[0087] The fractal dimension D of the rust layer is calculated using the box dimension algorithm to calculate the surface roughness of the rust layer;

[0088] The crack density δ is:

[0089]

[0090] Where, L crack is the total length of the crack, mm; A ROI is the area of ​​the region of interest, mm 2 .

[0091] More preferably, in S3, a method including image features and corrosion rate υ is constructed. 协同 、Environmental parameters( cH2S, c Cl- , T, pH), stress and strain (σ, ε) 10-dimensional feature vector:

[0092] The corrosion rate, environmental parameters, equipment wall thickness data, strain data, and equipment surface temperature image data after S2 processing are analyzed through the LSTM-attention model to identify the coupling relationship between key influencing factors and corrosion rate.

[0093] The corrosion rate is the synergistic corrosion rate that establishes the synergistic effect of CO2 corrosion, H2S corrosion, and erosion corrosion;

[0094] CO2 corrosion is:

[0095]

[0096] Where, is the CO2 corrosion rate, mm / a; k1 is the CO2 corrosion rate constant, mm / (a·MPa·K); is the CO2 partial pressure, MPa; Ea is the activation energy, kJ / mol, R is the ideal gas constant, T is the temperature in °C; ηpH = 10 0.1(4-pH) , is the pH value influence coefficient, and corrosion is significantly accelerated when pH ≤ 4.

[0097] H2S corrosion is:

[0098]

[0099] Where, is the H2S corrosion rate, mm / a; k2 is the H2S corrosion rate constant, mm / (a·√ppm); cH2S is the H2S concentration, ppm; c Cl- is the chloride ion concentration, ppm.

[0100] Erosion corrosion is:

[0101] υ crosion =k3·u 2 ·α·ρ s

[0102] Where, corision is the erosion corrosion rate, mm / a; k3 is the erosion corrosion coefficient, mm / (a·(m / s) 2 kg / m 3 ); u is the fluid velocity, m / s; α is the solid particle volume fraction, %; ρ s is the particle density, kg / m 3 .

[0103] The synergistic corrosion rate is:

[0104]

[0105] Where, 协同 is the synergistic corrosion rate, i.e., the corrosion rate, mm / a; k4 is the synergistic effect coefficient m 3 / (a 2 ·MPa·ppm), which reflects the interaction acceleration effect between CO2 and H2S, and takes a value of 0.01 to 0.05.

[0106] A fuzzy comprehensive evaluation matrix was established to evaluate the corrosion level. The evaluation indicators included corrosion rate, corrosion depth, environmental corrosion index, and stress corrosion risk. The weights of each indicator were 0.4, 0.3, 0.2, and 0.1, respectively. The corrosion level was determined by Z-score normalization, construction of a membership matrix, and calculation of a comprehensive evaluation vector. The corrosion level includes five levels: intact, mild corrosion, moderate corrosion, severe corrosion, and failure risk. The specific steps are as follows:

[0107] (1) Index standardization: The corrosion rate, corrosion depth, environmental erosion index, and stress corrosion risk are normalized by Z-score. The formula is:

[0108]

[0109] When i=1, x1 is the corrosion rate; when i=2, x2 is the corrosion depth; when i=3, x3 is the environmental corrosion index; when i=4, x4 is the stress corrosion risk;

[0110] (2) Membership function

[0111] The trapezoidal distribution function is used to calculate the membership of each indicator to the five corrosion levels μ j (x' i )(j=1~5, corresponding to 5 corrosion levels):

[0112]

[0113] Where a j , b j , c j is the threshold parameter of each level;

[0114] For example, the membership function of the corrosion rate is:

[0115]

[0116] (3) Comprehensive evaluation vector

[0117]

[0118] Where ω1, ω2, ω3, and ω4 are the weight coefficients of each evaluation index, respectively [W = 0.4, 0.3, 0.2, and 0.1].

[0119] S4. Remaining life prediction

[0120] A physical model based on Faraday's electrolysis law and a data-driven model based on the Transformer network were constructed. The remaining life was predicted by fusion output of the Bayesian network and combined with Monte Carlo simulation.

[0121] The theoretical corrosion depth is calculated by combining the physical model with the material properties of the equipment. The physical model is based on the Faraday electrolysis law to construct the basic equation for the theoretical corrosion rate:

[0122]

[0123] Where, 理论 is the theoretical corrosion rate, mm / a; K is a constant, K = 3.27 × 10 -3 ;i corris the corrosion current density, μA / cm 2 ; M is the molar mass of the metal, g / mol; n is the oxidation state number; ρ is the metal density, g / cm 3 .

[0124] Carbon steel (M = 55.85 g / mol, n = 2, ρ = 7.87 g / cm 3 ) as an example, if the measured corrosion current density i corr =15μA / cm 2 ), then the theoretical corrosion rate:

[0125]

[0126] The calculation formula for the theoretical corrosion depth is:

[0127] d 理论 =0.018·t+d0

[0128] Where t is time, years; d0 is the initial corrosion depth.

[0129] Using the Transformer network, historical corrosion depth, environmental parameters, operating conditions, and maintenance record data are input to establish a data-driven model to predict the corrosion depth in the next 1-10 years. The calculation formula for predicting corrosion depth is:

[0130]

[0131] For example, if the corrosion depth sequence of a pipeline in the first five years is input as [0.1, 0.3, 0.6, 1.0, 1.5] mm, the model predicts that the corrosion depth in the sixth year will be 1.9 mm.

[0132] Dynamically fuse the outputs of the physical model and the data-driven model through a Bayesian network:

[0133]

[0134] Where, is the final corrosion assessment result obtained after fusion;

[0135] d 理论 is the theoretical corrosion depth calculated based on the physical model;

[0136] d 预测 The corrosion depth predicted by the data-driven model;

[0137] is a dynamic weight function, and the parameters k, t0 are determined by Bayesian optimization, with an initial value of ω = 0.5.

[0138] When the current weight is 0.6,

[0139] Monte Carlo simulation was then used to sample uncertain parameters such as the corrosion rate fluctuation range (±20%) and Cl- concentration change (±15%), and it was run 100,000 times to generate a probability distribution of corrosion depth. The remaining life was defined as the time quantile when the corrosion depth reached the critical value, and the 5% quantile (P5), 50% quantile (P50), and 95% quantile (P95) were output.

[0140] S5. Dynamic calibration and method support

[0141] Model calibration is triggered based on offline inspection data, corrosion level changes, and remaining life thresholds: θ is the model parameter, α=0.01 is the learning rate, and L is the mean square error loss function.

[0142] The warning levels and maintenance strategies are as follows:

[0143] Model calibration is triggered when offline inspection data is uploaded, or when the corrosion level increases for three consecutive days and the predicted remaining life falls below the warning threshold. Model calibration is a model parameter fine-tuning algorithm based on stochastic gradient descent. Global calibration is automatically performed once a month, and local fine-tuning is performed weekly. Warning levels include yellow warning: remaining life ≤ 3 years, orange warning: remaining life ≤ 2 years, and red warning: remaining life ≤ 1 year. The maintenance method is to increase the sensor sampling frequency to 1 time per minute during yellow warnings and initiate weekly infrared inspections. During orange warnings, the ultrasonic thickness measurement detection range is expanded, and the number of detection points is increased by 50% to cover all suspected corrosion areas. A comprehensive analysis of the corrosive components of the medium (such as CO2, H2S, and Cl- concentration fluctuations) is performed, and process parameters are adjusted to suppress corrosion. During red warnings, a replacement plan is formulated and downtime for maintenance is arranged.

[0144] Application Examples

[0145] The present invention also provides an application of the equipment corrosion assessment and life prediction method of the above embodiment. In 2016, it was applied to the corrosion assessment of the top pipeline of the carbon dioxide separation tower of a coal chemical gasification unit. The pipeline transported syngas containing CO2 / H2S, and the operating parameters were as follows:

[0146] cH2S=600ppm,c Cl- =180ppm, T=200℃, conveying speed u=6m / s;

[0147] The material is 304L, the initial wall thickness is 8mm, and the critical corrosion depth is 2.4mm (30% of the wall thickness).

[0148] The specific steps include:

[0149] S1. Deploy sensor networks to collect multi-source data

[0150] Three linear polarization resistance sensors, two sampling and analysis sensors, and two temperature and humidity sensors are deployed in corrosion-prone areas such as pipeline welds and corners; two strain gauges are installed at the welds where stress is concentrated in the pipeline; the wall thickness of the pipeline is measured at six points using an ultrasonic thickness gauge every month; and a comprehensive temperature image of the pipeline is collected using an infrared thermal imager every week.

[0151] S2. Data Processing

[0152] The multi-source data collected by S1 is transmitted to the edge computing node via the LoRaWAN wireless protocol. The collected corrosion current density data is processed by outlier removal and sliding window mean filtering. The environmental parameters and strain data are time synchronized and spatially aligned to form a pipeline corrosion state vector and store it in the InfluxDB time series database.

[0153] S3. Corrosion feature extraction and evaluation

[0154] The images captured by the infrared thermal imager were input into a convolutional neural network, which identified a corroded area on the inner wall of the pipeline with a rust layer thickness of approximately 0.3 mm. The time series data was analyzed using an LSTM-attention model, which determined that CO2 corrosion and H2S corrosion were the key factors affecting the current corrosion rate.

[0155]

[0156] υ crosion =k3·6 2 0.05 2500 = 0.045 mm / a

[0157] The calculated synergistic corrosion rate is υ 协同 =0.101+0.15+0.03×0.101×0..15=0.30mm / a.

[0158] According to the fuzzy comprehensive evaluation matrix, the comprehensive evaluation vector is calculated to determine the current corrosion level of the chemical pipeline.

[0159] Through fuzzy comprehensive evaluation, B=[0.1, 0.2, 0.3, 0.4, 0.0] was obtained, which was judged to be level 4 (severe corrosion).

[0160] S4. Remaining life prediction

[0161] The theoretical corrosion depth was calculated using a physical model, and combined with parameters such as the current corrosion current density, the theoretical corrosion rate was obtained to be 0.075 mm / a.

[0162] The Transformer network of the data-driven model predicts that the corrosion depth will increase by 0.12mm in the next year based on historical data from the past five years. The Bayesian network fuses the results of the physical model and the data model to obtain the fused corrosion depth prediction value. After running the Monte Carlo simulation 100,000 times, the fused model prediction is output. In 2024 (the eighth year of actual use), the corrosion depth was detected to be 2.3 mm, with an error of <5%, verifying the effectiveness of the method.

[0163] S5. Dynamic calibration and method support

[0164] Six months after the sensors were deployed, real-time data was collected through the sensor network, pre-processed by edge computing, and then fed into an intelligent model, enabling dynamic corrosion level assessment and remaining life prediction. After one year of operation, the system successfully issued warnings for three moderate corrosion risks, avoiding unplanned downtime and reducing equipment maintenance costs by 25%, validating the engineering practicality of the method.

[0165] Therefore, the present invention adopts the above-mentioned equipment corrosion assessment and life prediction method and application, which can effectively realize the corrosion assessment and life prediction of coal chemical equipment under the synergistic action of multiple corrosion mechanisms, and has good application effect and promotion value.

[0166] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit the same. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that they can still modify or replace the technical solutions of the present invention with equivalents, and these modifications or equivalent replacements cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for equipment corrosion assessment and life prediction, characterized in that: The following steps are involved: S1. Multi-source data collection Deploy sensor networks in corrosion-prone areas of coal chemical equipment to collect multi-dimensional data on equipment surface corrosion current density, environmental parameters, wall thickness, stress, and surface temperature images; S2. Data Preprocessing The multi-source data collected by S1 is transmitted to the edge computing node for outlier removal, data compression, time synchronization and space-time alignment preprocessing; S3. Corrosion feature extraction and evaluation Convolutional neural networks are used to extract image features, and the LSTM-attention model is used to analyze the data processed by S2, and a fuzzy comprehensive evaluation matrix is ​​established to assess the corrosion level. S4. Remaining life prediction Construct a physical model based on Faraday's electrolysis law and a data-driven model based on the Transformer network, fused the output through a Bayesian network, and combined it with Monte Carlo simulation to predict the remaining life; S5. Dynamic calibration and method support Model calibration is triggered based on offline detection data, corrosion level changes, and remaining life thresholds, and maintenance methods are formulated according to warning levels.

2. The equipment corrosion assessment and life prediction method according to claim 1, characterized in that: In S1, the sensor network includes a linear polarization resistance sensor, a sampling and analysis sensor, a pH sensor, a temperature and humidity sensor, an ultrasonic thickness gauge, a strain gauge, and an infrared thermal imager.

3. The method for equipment corrosion assessment and life prediction according to claim 1, characterized in that: In S2, data outliers are eliminated based on the 3σ principle, and the high-frequency corrosion current density data are compressed using a sliding window mean filter. Time synchronization is achieved through the NTP protocol, and the timestamp-based Kalman filter algorithm is used for spatiotemporal alignment and integration.

4. The equipment corrosion assessment and life prediction method according to claim 1, characterized in that: In S3, a convolutional neural network is used to process the temperature images collected by the infrared thermal imager to extract the image features of the rust layer thickness and crack distribution. The corrosion rate, environmental parameters, equipment wall thickness data, strain data, and equipment surface temperature image data are analyzed through the LSTM-attention model to identify the coupling relationship between key influencing factors and the corrosion rate.

5. The equipment corrosion assessment and life prediction method according to claim 1, characterized in that: In S3, the evaluation indicators of the fuzzy comprehensive evaluation matrix include corrosion rate, corrosion depth, environmental erosion index, and stress corrosion risk. The weights of each indicator are 0.4, 0.3, 0.2, and 0.1, respectively. The corrosion level is determined by Z-score normalization, construction of the membership matrix, and calculation of the comprehensive evaluation vector. The corrosion level includes five levels: intact, mild corrosion, moderate corrosion, severe corrosion, and failure risk.

6. The equipment corrosion assessment and life prediction method according to claim 1, characterized in that: In S4, the theoretical corrosion depth is calculated by combining the physical model with the equipment material properties. The corrosion depth in the next 1-10 years is predicted by inputting historical corrosion depth, environmental parameters, operating conditions and maintenance record data into the data-driven model.

7. The equipment corrosion assessment and life prediction method according to claim 1, characterized in that: In S4, the outputs of the physical model and the data-driven model are dynamically fused through the Bayesian network, and Monte Carlo simulation is used to sample uncertain parameters. The outputs of the physical model and the data-driven model are dynamically fused through the Bayesian network 100,000 times. The remaining life is defined as the time quantile when the corrosion depth reaches the critical value, and the 5% quantile, 50% quantile, and 95% quantile are output.

8. The equipment corrosion assessment and life prediction method according to claim 1, characterized in that: In S5, when offline detection data is uploaded, or the corrosion level increases for three consecutive days and the predicted remaining life is lower than the warning threshold, model calibration is triggered. Model calibration is performed using a model parameter fine-tuning algorithm based on stochastic gradient descent.

9. The equipment corrosion assessment and life prediction method according to claim 1, characterized in that: In S5, the warning levels include yellow warning, orange warning and red warning. The maintenance method is to increase the sensor sampling frequency and start weekly infrared inspections during yellow warnings; expand the ultrasonic thickness measurement detection range during orange warnings and conduct a comprehensive analysis of the corrosive components of the medium; formulate a replacement plan and arrange shutdown maintenance during red warnings.

10. An application of the equipment corrosion assessment and life prediction method according to any one of claims 1 to 9, characterized in that: It is used for corrosion assessment and life prediction of gasifiers, synthesis towers, separation towers and process pipelines in coal chemical equipment. Coal chemical equipment is in a high temperature, high pressure and corrosive medium environment containing CO2 and H2S.

Images