System for online monitoring of corrosion state of sulfuric acid plant

Abstract

The application discloses an online monitoring system for corrosion state of a sulfuric acid device, and relates to the technical field of corrosion monitoring of industrial equipment.The system comprises an external reference calibration module for providing corrosion baseline data and initial calibration of sensors; a key window observation module integrating a corrosion-resistant transparent window, an internal reference sensor array and an industrial camera system, which is used for collecting local corrosion data and visual features; a distributed multi-parameter sensor network module distributed throughout the device; and a time-series corrosion evolution model based on a hierarchical self-attention mechanism, which can perform deep spatio-temporal correlation analysis and fusion processing on multi-source heterogeneous data. Through multi-level collaborative calibration, deep fusion of visual and sensing information and intelligent prediction algorithms, the application significantly improves the accuracy and early warning capability of corrosion monitoring, predicts the remaining service life of the device, optimizes maintenance decisions, and thus effectively guarantees the safe, long-period and economic operation of the sulfuric acid device.

Description

Technical Field

[0001] This invention relates to the field of industrial equipment corrosion monitoring technology, specifically an online monitoring system for the corrosion status of sulfuric acid plants. Background Technology

[0002] Sulfuric acid, as an important basic chemical raw material, faces severe corrosion problems during its production, storage, and use. Its equipment (such as reactors, heat exchangers, pipelines, and storage tanks) are exposed to highly corrosive media for extended periods. Corrosion not only leads to material loss and performance degradation, shortening equipment lifespan, but can also cause safety accidents such as equipment leaks and failures, resulting in huge economic losses and environmental risks. Therefore, real-time and accurate online monitoring and trend prediction of the corrosion status of sulfuric acid plants are of paramount importance for ensuring the safe and stable operation of the plants, optimizing maintenance strategies, and extending equipment lifespan.

[0003] Currently, there are numerous studies and applications of monitoring technologies for equipment corrosion. For example, Chinese invention patent CN110849796B discloses a corrosion monitoring system belonging to the field of mechanical technology. This corrosion monitoring system is used to monitor the corrosion of the metal under test. It includes a power supply component, a processing component, a Hall sensor component, and a resistance probe component. The Hall sensor component and the resistance probe component are connected in parallel at both ends of the power supply component and are both located around the metal under test. The power supply component, the Hall sensor component, and the resistance probe component are all connected to the processing component. This patented solution solves the problems of long time required and low efficiency in determining corrosion rate in the traditional on-site plate method by combining the resistance probe method and the Hall sensor method. It also attempts to explore the relationship between corrosion rate and magnetic induction intensity.

[0004] For example, Chinese invention patent CN113640199B discloses a corrosion monitoring method for chemical equipment, which includes obtaining the corrosion rate C using n different methods, determining the confidence level and validity corresponding to each corrosion rate, calculating the posterior probability by combining the prior probability and conditional test probability of the corrosion damage state, and finally determining the thinning probability POFthin. Based on this, the thinning damage factor is calculated to classify and warn of corrosion. This method focuses on assessing the thinning risk of equipment by integrating multiple corrosion rate measurement results and using a probabilistic statistical model.

[0005] The existing technologies mentioned above, by combining multiple sensing methods (such as CN110849796B using resistance probes and Hall sensors) or fusing multiple corrosion rate measurement results and applying probabilistic models (such as CN113640199B assessing thinning risk), have improved the efficiency of corrosion monitoring or the quantitative level of risk assessment to a certain extent. However, for application scenarios such as sulfuric acid plants with complex operating conditions such as high temperature, high pressure, strong corrosiveness, multiphase flow, and variable process parameters, there are still certain limitations, specifically reflected in:

[0006] Lack of highly reliable in-situ multimodal data anchors and dynamic calibration mechanisms: Existing technologies mostly rely on indirect, single, or limited combinations of sensor data, making it difficult to confirm the specific morphology, type, and early subtle changes of corrosion in real time and intuitively. At the same time, in harsh sulfuric acid environments, the stability and accuracy of long-term sensor operation are easily affected. Existing technologies generally lack effective mechanisms for dynamic in-situ calibration combined with direct observation, making it difficult to ensure the continuous reliability of long-term monitoring data.

[0007] Insufficient accuracy in predicting corrosion evolution trends and remaining lifespan under complex operating conditions: The corrosion process in sulfuric acid plants is often the result of nonlinear and time-varying coupling effects of multiple factors (such as temperature, concentration, flow rate, impurities, stress, etc.). Existing technologies often use simplified models or rely on experience when predicting future corrosion states and remaining equipment lifespan, making it difficult to fully explore and utilize the complex spatiotemporal dependencies and deep correlations between multi-source heterogeneous monitoring data (such as electrochemical data, visual feature data, and process parameter data), resulting in a need to improve prediction accuracy and reliability.

[0008] There is a lack of early and accurate identification and quantitative assessment methods for localized corrosion such as pitting and crevice corrosion: In sulfuric acid plants, localized corrosion forms such as pitting, crevice corrosion, and stress corrosion cracking are often more insidious and sudden than uniform corrosion, and are also more harmful. Existing technologies are still insufficient in their ability to identify, locate, quantitatively assess, and predict the risks of such localized corrosion in the early online stage, making it difficult to meet the needs of refined and forward-looking management of high-risk corrosion forms.

[0009] There is a lack of quantitative analysis and feedback guidance on corrosion behavior under the synergistic influence of multiple process parameters: Although existing technologies also monitor process parameters, they are still incomplete in revealing and quantifying how multiple key process parameters (such as sulfuric acid concentration, operating temperature, and medium flow rate) synergistically affect corrosion rate and mode, and in how to use this quantitative understanding to guide process optimization and proactively control corrosion in a closed-loop approach.

[0010] Therefore, there is an urgent need to develop a new system and method that can overcome the above limitations and achieve more comprehensive, accurate and intelligent online monitoring and prediction of the corrosion status of sulfuric acid plants. Summary of the Invention

[0011] The purpose of this invention is to overcome the shortcomings of the prior art and propose an online monitoring system for the corrosion status of sulfuric acid plants to solve the aforementioned problems.

[0012] The objective of this invention is achieved through the following technical solution: an online monitoring system for the corrosion status of a sulfuric acid plant, comprising: an external reference calibration module, including a standard container for containing a sulfuric acid solution of known concentration and an external reference sensor installed inside;

[0013] The critical viewing window observation module is installed at a critical location in the sulfuric acid plant. The critical viewing window observation module includes a transparent viewing window that is resistant to sulfuric acid corrosion, an internal reference sensor array set adjacent to its inner surface, and an industrial camera system that acquires internal images through the transparent viewing window.

[0014] A distributed multi-parameter sensor network module, comprising multiple distributed sensors located at different locations on the device;

[0015] The data acquisition and transmission module is used to synchronously acquire and timestamp multi-source data from multiple distributed sensors, including external reference calibration module, internal reference sensor array, visual features processed by industrial camera system, and distributed sensor network, at preset uniform time intervals.

[0016] The central data processing and intelligent analysis platform is configured with a time-series erosion evolution model based on a hierarchical self-attention mechanism. The central data processing and intelligent analysis platform is configured as follows:

[0017] Multi-source data is time-coded and feature-processed to form a data packet sequence;

[0018] Based on the first-layer multi-head self-attention mechanism, the data packet sequence is analyzed to generate a context-aware state representation vector.

[0019] A second-layer multi-head self-attention mechanism analysis is performed on the context-aware state representation vectors of multiple consecutive time intervals to generate a global erosion state representation vector.

[0020] The global corrosion state characterization vector monitoring device is used to monitor the current corrosion state and predict future corrosion trends or remaining service life.

[0021] The central data processing and intelligent analysis platform is further configured to: perform initial absolute calibration of each sensor in the system using data from the external reference calibration module, and combine the dynamic correlation analysis results between the readings of the internal reference sensor array of the key window observation module and the visual feature data collected and processed by the industrial camera system to perform dynamic in-situ relative calibration or drift correction of the distributed sensors in the distributed multi-parameter sensor network module, thereby forming a multi-level collaborative calibration mechanism.

[0022] The transparent window of the key viewing module is made of special quartz glass, sapphire, or modified fluoroplastic. The industrial camera system includes an independently controllable lighting unit and an image analysis artificial intelligence module configured to extract structured visual features, including the area percentage of the corroded area, color histogram, defect count, or turbidity quantification of the sulfuric acid medium.

[0023] The internal reference sensor array includes at least three internal reference sensors, and the central data processing and intelligent analysis platform is configured to use the average readings of the at least three internal reference sensors as a reference for the local corrosion status of the viewing window observation anchor point area.

[0024] The external benchmark calibration module is further equipped with temperature control and flow rate simulation devices to generate baseline data on the corrosion rate of the material under known sulfuric acid concentration, temperature and flow rate conditions. The baseline data is used to parameterize a time-series corrosion evolution model based on a hierarchical self-attention mechanism.

[0025] The central data processing and intelligent analysis platform performs feature processing on multi-source data, including: extracting time-domain and frequency-domain features from electrochemical noise data, and extracting color, texture, and defect coordinates and size features from visual data from industrial camera systems, as well as features identified by the target detection model.

[0026] The central data processing and intelligent analysis platform is further configured with multiple prediction heads connected to the output of the time-series corrosion evolution model based on a hierarchical self-attention mechanism. The prediction heads are used to output local corrosion rate, remaining wall thickness, and risk level for different failure modes.

[0027] The central data processing and intelligent analysis platform is further configured to automatically generate actionable maintenance suggestions based on the monitored current corrosion status, predicted future corrosion trends, and remaining service life, or to trigger multi-level alarms when a preset risk threshold is reached.

[0028] The time-series corrosion evolution model based on the hierarchical self-attention mechanism of the central data processing and intelligent analysis platform is constructed and optimized by using historical datasets including multi-source data and labeled data such as offline equipment detection results, equipment maintenance records, or manually labeled corrosion events corresponding to the historical datasets for supervised learning or self-supervised learning pre-training.

[0029] The distributed multi-parameter sensor network module includes electrochemical sensors, temperature sensors, pressure sensors, and flow rate sensors. The central data processing and intelligent analysis platform is configured to use data from the temperature, pressure, and flow rate sensors, combined with baseline data from an external benchmark calibration module, to quantify the synergistic effect of sulfuric acid concentration, temperature, and flow rate on the corrosion rate.

[0030] The beneficial effects of this invention are:

[0031] 1. The system integrates an external benchmark calibration module, a key window observation module, a distributed multi-parameter sensor network module, a data acquisition and transmission module, and a central data processing and intelligent analysis platform, realizing comprehensive and multi-dimensional monitoring of the corrosion status of sulfuric acid plants. Compared with traditional single-sensor monitoring methods, this invention can integrate electrochemical, visual, and process parameter data to provide more comprehensive corrosion information.

[0032] 2. The key window observation module significantly improves the accuracy and reliability of local corrosion observation through a corrosion-resistant transparent window, an enhanced industrial camera system, and an internal reference sensor array. The external reference calibration module generates high-quality baseline data through environmental control, providing dynamic calibration basis for distributed sensors and enhancing the data reliability of the entire sensor network.

[0033] 3. The central data processing platform employs algorithms and deep feature extraction technology, combined with a time-series corrosion evolution model based on a hierarchical self-attention mechanism, to achieve accurate prediction of corrosion rate, remaining wall thickness, and risks of various failure modes. The prediction head outputs specific risk levels for different failure modes, improving the guidance and practicality of the prediction results.

[0034] 4. Synergistic impact analysis function: quantifies the independent and interactive effects of process parameters such as temperature, pressure, and flow rate on corrosion, providing operators with a scientific basis to optimize the combination of process parameters, thereby slowing down corrosion and extending equipment life without significantly affecting production efficiency.

[0035] 5. This invention provides maintenance personnel with targeted inspection and prevention suggestions through data insights and multi-dimensional risk assessment, optimizes resource allocation, reduces unplanned downtime, and supports proactive adjustment of process parameters based on synergistic impact analysis results, thereby enhancing the initiative and effectiveness of corrosion control.

[0036] 6. Through deep feature extraction and baseline data analysis, the system not only monitors corrosion symptoms but also reveals microscopic corrosion signs and driving factors, which helps to optimize anti-corrosion strategies from a mechanistic perspective and provides theoretical support for long-term equipment management. Attached Figure Description

[0037] Figure 1 This is a system architecture diagram of the present invention;

[0038] Figure 2 This is a timing diagram for the present invention. Detailed Implementation

[0039] The technical solution of the present invention will be clearly and completely described below with reference to the embodiments. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0040] It should be noted that the directional concepts of "left", "right", "up", "down", "front", "back", "inner", and "outer" in the following scheme are all relative directions, and will not be listed one by one here.

[0041] Example 1:

[0042] like Figure 1 and Figure 2 As shown in the figure, this embodiment discloses a specific implementation plan of a corrosion monitoring and prediction system for sulfuric acid plants. The system aims to achieve high-precision and reliable corrosion status assessment and trend prediction through a multi-level calibration mechanism and local corrosion monitoring of key parts. The system is suitable for high-risk areas of sulfuric acid plants, such as the bottom of the reactor and pipe bends. The system performs initial sensor calibration through an external reference calibration module, achieves dynamic calibration by combining visual-sensor data from the key window observation module, and uses an intelligent analysis platform to monitor corrosion status and predict future trends.

[0043] This system includes the following main modules: external benchmark calibration module, key window observation module, distributed multi-parameter sensor network module, data acquisition and transmission module, and central data processing and intelligent analysis platform.

[0044] An external reference calibration module is used to provide corrosion rate reference data to ensure the initial measurement accuracy of the sensor. This module includes a standard container lined with corrosion-resistant materials (such as PTFE or stainless steel) and filled with a known concentration sulfuric acid solution (e.g., 95-98 wt%) calibrated by titration with a concentration error of ±0.01%. The container is equipped with a temperature control device (accuracy ±0.1℃, range 20-120℃) and a flow rate simulation device (flow rate 0.1-1 m / s) to simulate typical operating conditions of a sulfuric acid plant.

[0045] An external reference sensor, including a linear polarization resistance (LPR) probe and an electrochemical noise (ECN) electrode, is installed inside the container. The sensor is made of Hastelloy C-276 and has a measurement range of 0.01–10 mm / year with an accuracy of ±5%. These sensors are used to generate calibration parameters and corrosion baseline data.

[0046] The critical viewing window module is installed in critical parts of the sulfuric acid plant (such as the bottom of the reactor or pipe bends) for high-precision localized corrosion monitoring. This module includes the following components:

[0047] The transparent window is made of special quartz glass with a diameter of 50-100 mm and a thickness of 10-15 mm. It can withstand temperatures up to 600℃ and pressures up to 10 MPa. The window is fixed by Hastelloy flanges and fluororubber gaskets to ensure sealing under high temperature and high pressure conditions.

[0048] The internal reference sensor array contains three miniature sensors (LPR probe, ECN electrode, and temperature sensor) made of Hastelloy C-276. They are installed near the inner surface of the transparent window and have the same measurement range as the external reference sensor. The central data processing and intelligent analysis platform takes the average value of the three sensor readings as the local corrosion state reference.

[0049] The industrial camera system is equipped with a high-definition industrial camera (1920x1080 resolution, 30 fps frame rate, 60 dB dynamic range). The housing is made of corrosion-resistant materials. The system has a built-in adjustable white LED illumination unit (wavelength 450-650nm). The camera is connected to an image analysis module based on convolutional neural networks (CNN). The U-Net model is used to segment the corrosion area, the YOLOv5 model is used to detect pitting corrosion, and the HSV spatial statistical color histogram is used to extract structured visual features, including the percentage of corrosion area, color histogram, pitting corrosion count, and sulfuric acid medium turbidity quantification.

[0050] The distributed multi-parameter sensor network module contains multiple distributed sensors, including LPR probes, ECN electrodes, temperature sensors, and pressure sensors. These sensors are installed at locations such as the inlet of the sulfuric acid plant, heat exchangers, and straight sections of pipelines. The sensors are made of corrosion-resistant alloys, and their measurement range is consistent with that of the internal reference sensors. They are used to monitor global corrosion and process parameters.

[0051] The data acquisition and transmission module adopts an embedded data acquisition system (sampling rate 100 Hz, 16-bit ADC) and transmits data via industrial Ethernet. The system synchronously acquires multi-source data according to monitoring requirements (such as electrochemical data every minute, visual data every 5 minutes, and process parameters every 10 seconds) and adds a timestamp (accuracy ±1 ms) to each data point.

[0052] The central data processing and intelligent analysis platform is responsible for data processing, calibration, analysis, and prediction. It utilizes high-performance servers (configured with multi-core CPUs, 64 GB RAM, and GPU acceleration). The software stores data based on a time-series database, and the core analysis algorithms are implemented using Python and PyTorch. It employs a hierarchical self-attention mechanism-based time-series erosion evolution model, the structure of which includes:

[0053] The first layer of self-attention mechanism consists of 8 attention heads, which take a 256-dimensional feature vector as input and output a 512-dimensional context-aware state representation vector (defined as a high-dimensional vector that integrates multi-source features).

[0054] The second layer of self-attention mechanism processes 144 time points within 24 hours (one every 10 minutes) and outputs a 512-dimensional global erosion state representation vector (defined as a comprehensive vector reflecting spatiotemporal evolution).

[0055] Work process

[0056] The system's operation process includes multi-level calibration, local corrosion monitoring, data processing, result output, and model optimization.

[0057] Before system startup, all sensors are calibrated in an external reference calibration module:

[0058] The sensor was placed in a standard container and exposed to a known sulfuric acid environment (concentration 98 wt%, temperature 60°C, flow rate 0.5 m / s).

[0059] An external reference sensor measures the corrosion rate (e.g., 0.1 mm / year) as a baseline value.

[0060] The data acquisition module collects sensor readings, and the central platform fits the calibration curve through linear regression (R²>0.95) and calculates the calibration coefficient. For example, if a sensor reading is 0.12 mm / year, the calibration coefficient is 0.1 / 0.12 = 0.833.

[0061] The calibration parameters are stored in a database and used for subsequent data processing.

[0062] During system operation, the key window observation module provides calibration reference:

[0063] The industrial camera system acquires images every 5 minutes, and the image analysis module extracts visual features (such as 5% corrosion area).

[0064] An internal reference sensor array measures the corrosion current density and takes the average value (e.g., 10 μA / cm²) as a local reference.

[0065] The central platform uses a multilayer perceptron (MLP, 3 layers, with 128 neurons in the hidden layer) to map visual features with sensor readings and generate reference values.

[0066] Compare the distributed sensor reading (e.g., 12 μA / cm²) with the reference value. If the deviation is >10%, calculate the correction factor (e.g., 10 / 12 = 0.833) and adjust the reading.

[0067] Calibration is performed every 4 hours, and a correction log is recorded.

[0068] The key view observation module is responsible for monitoring localized corrosion.

[0069] The transparent window provides an internal visual channel for the device. The camera captures images, and the image analysis module extracts features such as corrosion area and pitting number. The internal reference sensor array measures parameters and calculates average values.

[0070] The central platform cross-validates visual features and sensor data to confirm the corrosion status.

[0071] The central platform performs the following steps:

[0072] Data cleaning was performed using the 3σ criterion to remove electrochemical noise outliers.

[0073] Feature extraction, calculation of standard deviation (σ=√(Σ(x_i-μ)^2 / N)) and power spectral density of electrochemical noise (calculated by FFT); extraction of corrosion area (U-Net segmentation) and pitting count (YOLOv5 detection) from visual data.

[0074] Time encoding adds a timestamp to each data point to form a data packet sequence.

[0075] First-level self-attention mechanism:

[0076] Input a 256-dimensional feature vector (including electrochemical, visual, and process parameter features), 8 attention heads, and output a 512-dimensional context-aware state representation vector.

[0077] Second-level self-attention mechanism:

[0078] Input 144 context-aware state representation vectors within 24 hours, analyze temporal dependencies, and output a 512-dimensional global corrosion state representation vector.

[0079] The global vector is processed by a feedforward neural network (2 layers, 256 neurons) to output the current corrosion rate, wall thickness, and corrosion trend for the next 30 days.

[0080] Model training and optimization

[0081] The time-series corrosion evolution model is constructed through the following steps:

[0082] Dataset: Collected 1 year of operational data (approximately 1 million records), including electrochemical noise (μA / cm²), visual features (area %), daily wall thickness measurements (mm), and annotation of 500 pitting events.

[0083] Supervised learning: using the mean squared error loss function, Adam optimizer (learning rate 0.001, batch size 32), trained for 100 epochs.

[0084] Self-supervised pre-training: Using a time series prediction task, masking 50% of the data points, and pre-training for 50 epochs.

[0085] Incremental learning: The model is updated monthly with new data, and an experience replay mechanism is adopted.

[0086] Maintenance suggestions and alarms

[0087] The central platform generates maintenance recommendations based on the corrosion status and prediction results, such as "It is recommended to check the pipe bends within 7 days and increase the corrosion inhibitor by 10%".

[0088] Triggering multi-level alarms:

[0089] Warning (yellow): Corrosion rate > 0.2 mm / year; General alarm (orange): Wall thickness < safety margin.

[0090] Emergency alarm (red): If the number of pitting defects is greater than 10 or the remaining lifespan is less than 30 days, the alarm will be notified via interface, SMS, and audible and visual alarm.

[0091] The innovative algorithm formula for dynamically adjusting the remaining service life of equipment is as follows:

[0092]

[0093] Symbol explanation:

[0094] RUL adj : Remaining useful life (years) after dynamic risk adjustment.

[0095] RUL pred : Remaining useful life (in years) directly predicted by the time-series corrosion evolution model based on the global corrosion state characterization vector.

[0096] Additional lifespan loss (in years) caused by actual operating conditions deviating from ideal or design conditions (such as over-temperature or over-concentration) from the last assessment time t0 to the current time t.

[0097] R S The comprehensive corrosion risk index (range 0-1) is output by the model and takes into account factors such as the current corrosion rate, local corrosion tendency, and historical damage accumulation.

[0098] M conf Model prediction confidence (range 0-1) reflects the model's confidence level in its prediction results, calculated based on output variance and consistency with historical data.

[0099] w m Model confidence affects the weighting factor, with a typical value of 0.1-0.5.

[0100] R S,crit : The preset critical corrosion risk index, with a typical value of 0.8.

[0101] ε: A small constant to prevent the denominator from being zero, typically 0.01.

[0102] tanh(·): Hyperbolic tangent function, used to smooth and constrain the risk adjustment factor within the 0-2 interval.

[0103] Formula Explanation:

[0104] This formula dynamically adjusts the remaining lifespan of the equipment through two parts:

[0105] Lifetime Loss Correction: From the initial predicted value RUL pred The additional lifespan loss caused by changes in operating conditions (such as temperature and concentration) is subtracted from the total lifespan.

[0106] Risk and confidence level adjustment: Adjusted using the tanh function based on the comprehensive corrosion risk index R. S and model confidence M conf The ratio of R to dynamically adjust lifetime prediction, when R S Approaching the critical value R S,crit Or M conf When the value is low, the adjustment factor decreases, resulting in more conservative predictions. For example, if RUL... pred = 5 years, operating loss 0.2 years, R S = 0.7, M conf = 0.9, then RUL adj With a timeframe of approximately 4.5 years, this formula improves the robustness of predictions and is particularly suitable for high-risk operating conditions.

[0107] To improve the reliability and conservatism of the prediction, the system uses a global corrosion state characterization vector to calculate the remaining service life with dynamic risk adjustment. This value is calculated using the formula mentioned above, taking into account operating condition deviations and model confidence. When the corrosion risk increases or the prediction confidence decreases, the adjusted service life prediction will become more conservative, thereby improving the robustness of the system.

[0108] Application Cases

[0109] At a certain pipe bend, the system has been running for one month, and the following data has been recorded:

[0110] Initial calibration: LPR probe reading 0.12 mm / year, 0.1 mm / year after calibration; Dynamic calibration: Visual characteristics show corrosion area of ​​5%, sensor reading 12 μA / cm², 9.8 μA / cm² after calibration; Monitoring results: Current corrosion rate 0.15 mm / year, wall thickness 8.5 mm; Predicted results: Corrosion area to increase to 6% within 30 days, wall thickness to decrease by 0.05 mm. Maintenance recommendations: Inspect elbows and adjust process parameters.

[0111] The key window observation module combines visual features and sensor data to significantly improve monitoring accuracy. Compared with the 15-20% error of traditional electrochemical monitoring, the error of this system is usually controlled within 5-10%. For example, after calibration, the corrosion current density error is reduced from 15% to 5%.

[0112] The multi-level calibration mechanism ensures sensor data consistency through initial calibration and dynamic correction, with drift error typically controlled within 5-10%, superior to the 10-15% error of traditional single calibration methods. The system can predict corrosion trends 15-30 days in advance, reducing unplanned downtime by 20-40%, effectively preventing equipment failure caused by accelerated corrosion. Based on accurate monitoring and prediction, maintenance recommendations optimize maintenance plans, typically reducing maintenance costs by 10-20%, improving equipment operating efficiency, timely detection of potential localized corrosion hazards, significantly reducing the risk of leakage caused by corrosion, and enhancing the safety of equipment operation.

[0113] The specific effects described above vary depending on the operating conditions (such as sulfuric acid concentration and equipment material). Those skilled in the art can adjust the parameters according to actual needs to achieve similar effects.

[0114] This embodiment achieves high-precision monitoring and trend prediction of localized corrosion in sulfuric acid plants through external benchmark calibration, key window observation, multi-level calibration mechanism and intelligent analysis, providing reliable protection for safe equipment operation and economic benefits.

[0115] Example 2:

[0116] like Figure 1 and Figure 2 As shown in the figure, this embodiment describes in detail the preferred configuration and specific implementation of the key window observation module and the external reference calibration module in the sulfuric acid plant corrosion monitoring and prediction system. These modules improve the accuracy of local corrosion monitoring, data reliability and the prediction capability of the core AI model. This embodiment is applicable to key parts of sulfuric acid plants (such as the bottom of the reactor and pipe bends) in high temperature, high pressure or complex chemical environments.

[0117] The key viewing window observation module is installed in the key parts of the device for high-precision local corrosion monitoring. In this embodiment, its core components have been optimized.

[0118] For corrosion-resistant transparent window materials, suitable for high-temperature, high-pressure, and highly corrosive environments (temperature 150-200°C, pressure 0.5-1.0MPa, sulfuric acid concentration ≥98%), sapphire (single crystal α-Al2O3) is used, which has excellent mechanical strength (Mohs hardness 9), optical transmittance (ultraviolet to mid-infrared) and chemical resistance.

[0119] For medium temperature and pressure conditions (temperature <120°C, pressure <0.5 MPa), special quartz glass (such as Corning 7980) is used, which has excellent thermal shock resistance and high chemical purity.

[0120] For chemical environments (such as those containing fluorine impurities), modified fluoroplastics (such as PFA or FEP) are used, which are highly corrosion resistant and suitable for complex shape requirements.

[0121] The viewing window is fixed by a corrosion-resistant alloy (such as Hastelloy C-276) flange and sealed with expanded polytetrafluoroethylene (ePTFE) or flexible graphite composite gaskets to ensure reliability under high temperature and pressure. The size of the viewing window is adjusted according to the working conditions (such as diameter 50-100 mm, thickness 10-20 mm).

[0122] The lighting unit of the industrial camera system is equipped with an independently controllable white LED lighting unit (wavelength 450-650nm), which adjusts the brightness according to the turbidity of the medium or surface reflection through pulse width modulation (PWM) to reduce glare.

[0123] The image analysis AI module is integrated into the central data processing and intelligent analysis platform. It receives 1920x1080 resolution RGB images and uses the following algorithms to extract structured visual features:

[0124] The percentage of corroded area is segmented using the U-Net model, outputting the proportion of the corroded area to the total area (0-100%). The color histogram is based on the HSV space to statistically analyze the color distribution of corrosion products or media, reflecting the corrosion type or turbidity changes. The pitting count is detected using the YOLOv5 model, outputting the number and distribution density. The turbidity quantification of sulfuric acid media is analyzed through gray-level co-occurrence matrix analysis of image texture to semi-quantitatively assess turbidity (range 0-1).

[0125] Using a labeled image dataset (approximately 10,000 images containing different degrees of corrosion and working conditions), U-Net and YOLOv5 models were trained through supervised learning, achieving an accuracy of 95%.

[0126] The internal reference sensor array is equipped with three identical linear polarized resistance (LPR) probes, made of the same material as the main body of the equipment (such as 316L stainless steel or Hastelloy), and installed in a triangular layout on the inner wall of the transparent window. The measurement range is 0.01-10 mm / year, with an accuracy of ±5%.

[0127] The central data processing and intelligent analysis platform simultaneously collects readings from three probes (corrosion current density, μA / cm²). If a reading deviates from other readings by more than 20%, it is considered abnormal and discarded. The arithmetic mean of the remaining readings is taken as the benchmark for local corrosion status. For example, if the readings from the three probes are 10, 10.5, and 12 μA / cm², after discarding 12, the average value is 10.25 μA / cm².

[0128] The external reference calibration module provides corrosion baseline data and sensor calibration references.

[0129] Environmental control devices

[0130] Temperature control employs a high-precision thermostat (such as a circulating oil bath), with a control range of 20-180°C and an accuracy of ±0.2°C. This is achieved through a PID controller and a PT100 platinum resistance sensor. Flow rate simulation is configured with a corrosion-resistant magnetic stirrer or a micro pump, with a flow rate range of 0-1 m / s and an accuracy of ±0.01 m / s. It is monitored using a Coriolis mass flow meter.

[0131] Corrosion Rate Baseline Data Generation and Application

[0132] Standard material specimens (with the same material as the equipment, such as Hastelloy C-276) and external reference sensors (LPR probe, ECN electrode) are placed in a known sulfuric acid environment (concentration 95-98%, temperature 60-120°C, flow rate 0.1-0.5 m / s), and electrochemical parameters (such as corrosion current density and polarization resistance) are continuously recorded. After the experiment, the corrosion rate is verified by the weight loss method.

[0133] The relationship between corrosion rate and environmental parameters is fitted using the least squares method, with the formula k = A·exp(-Ea / RT)·C^n, where k is the corrosion rate (mm / year), A is the pre-exponential factor, Ea is the activation energy (e.g., 50 kJ / mol), R is the gas constant, T is the absolute temperature (K), C is the sulfuric acid concentration (wt%), and n is the concentration exponent (e.g., 1.2).

[0134] Baseline data provides prior knowledge for time-series corrosion evolution models, initializes parameters such as Ea and n, verifies the agreement between model predictions and experimental data with an error of <5%, and guides the selection of time-domain characteristics of electrochemical noise (such as standard deviation) as model input.

[0135] Work process

[0136] The transparent window provides an internal visual channel for the equipment. The industrial camera acquires images every 5 minutes, and the image analysis module extracts visual features (such as corrosion area of ​​5% and pitting of 3 per cm²). The internal reference sensor array measures the corrosion current density and takes the average value (such as 10.25 μA / cm²). The central data processing and intelligent analysis platform maps the visual features and sensor readings through a multilayer perceptron (MLP, 3 layers, 128 hidden neurons) to generate calibration reference values.

[0137] The external benchmark calibration module works by simulating a sulfuric acid environment (e.g., 98% concentration, 60°C, 0.5 m / s) in a standard container, recording the corrosion rate (e.g., 0.1 mm / year) using an external reference sensor, generating calibration parameters, and fitting the baseline data to model parameters (e.g., Ea = 50 kJ / mol) for training the central platform model.

[0138] Optimal window materials and advanced image analysis enhance the accuracy of visual features, redundant sensor design ensures the stability of anchor point data, and the external benchmark calibration module enhances the model's understanding of environmental factors through baseline data. The key window module provides highly reliable local data, and the external benchmark module provides basic corrosion laws. The combination of these two enables the time-series corrosion evolution model to more accurately predict corrosion behavior under complex working conditions.

[0139]

[0140] Symbol explanation:

[0141] I VC Overall visual corrosion index (0-100)

[0142] F vis,i (t): The i-th visual feature value (e.g., corrosion area %)

[0143] F vis,i,baseline Characteristic health baseline value (e.g., initial area 0%)

[0144] F vis,i,scale Normalized scale factor (e.g., area scale 100%)

[0145] Feature weights (0.1-0.5)

[0146] p i Sensitivity index (1-2)

[0147] Count defect Number of pitting pits (per cm²)

[0148] S defect Average pitting size (mm)

[0149] Defect weight (0.2)

[0150] ε: Small constant (0.01)

[0151] This formula quantifies the degree of corrosion by weighted nonlinear combination of visual features and taking into account the number and severity of pitting. For example, if the corrosion area increases from 0% to 5%, with 3 pits / cm² and a size of 0.1 mm, then I VC≈ 10. This index can be used as a model input or cross-validated with sensor readings to improve monitoring accuracy.

[0152] Application Cases

[0153] At a pipe bend (sulfuric acid concentration 98%, temperature 100°C), the system operated for one month:

[0154] Visual characteristics: Under the preferred sapphire window, the U-Net segmentation corrosion area is 5%, with an error of 2% (compared to 5% for traditional quartz glass). The readings of the three LPR probes are 10, 10.5, and 12 μA / cm², with an average of 10.25 μA / cm². The stability (variance) is improved by 15%. Baseline data: The fitted corrosion rate is k = 0.1·exp(-50 / RT), and the model prediction error is <5%. Calibration effect: The dynamic calibration frequency is reduced from once a month to once a quarter.

[0155] Optimized window materials and advanced image analysis reduce visual feature extraction error from 5-10% in traditional methods to 2-5%. Redundant design of three sensors and average value processing improve data stability, and reading variance is typically reduced by 10-20%, which is better than the 15-25% variance of a single sensor. Baseline data optimizes the time-series corrosion evolution model, which typically improves prediction accuracy by 10-20% and achieves an accuracy of 85-95%, which is better than the 70-80% of traditional statistical models.

[0156] Optimal configuration ensures stable operation of the system in high temperature, high pressure or impurity environments, improving operational stability by 15-25%.

[0157] The specific effects described above vary depending on the operating conditions (such as sulfuric acid concentration and equipment material). Those skilled in the art can adjust the configuration according to actual needs to achieve similar effects.

[0158] This embodiment significantly improves the local observation accuracy, data reliability, and prediction capability of the sulfuric acid plant corrosion monitoring and prediction system through the optimized configuration of the key window observation module and the external reference calibration module.

[0159] Example 3:

[0160] like Figure 1 and Figure 2 As shown, this embodiment is based on the system architecture of embodiment 1 and the preferred configuration of the key observation and calibration modules of embodiment 2. It elaborates in detail the algorithms, refined data analysis and application extension functions of the central data processing and intelligent analysis platform (hereinafter referred to as the "central platform"). These functions enable in-depth insight into corrosion status, multi-failure mode risk assessment and quantification of the synergistic impact of process parameters, thereby improving the accuracy and decision support capabilities of the corrosion monitoring and prediction system for sulfuric acid plants. This embodiment is applicable to key parts of high-temperature and high-pressure sulfuric acid plants (such as the bottom of the reactor and pipe bends).

[0161] The central platform performs deep feature extraction on multi-source data (electrochemical noise, visual data, and process parameters) to generate a data packet sequence that is input into the time-series corrosion evolution model.

[0162] When the distributed multi-parameter sensing network module or internal reference sensor array contains an electrochemical noise sensor, the central platform performs the following processing on the electrochemical noise time series (corrosion potential noise V, current noise I):

[0163] Time-domain characteristics: noise resistance (R) n ):R n = σ V / σ I , range 10³-10 5 Ω·cm², which is related to the reciprocal of the corrosion rate.

[0164] Standard deviation (σ) V , σ I : σ = √(Σ(x) i -μ)² / N) reflects the fluctuation range, with typical values ​​of 0.1-10 μV or nA.

[0165] Kurtosis: Kurt = Σ(x i -μ) 4 / (Nσ 4 -3, range -2 to 10, high value indicates pitting outbreak.

[0166] Localization index (LI): LI = σ I / I rms I_rms = √(ΣI_i² / N), ranging from 0 to 1, LI>0.5 indicates localized corrosion.

[0167] Frequency domain characteristics:

[0168] Power spectral density (PSD): Calculated by Fast Fourier Transform (FFT), analyzing low-frequency amplitude (10⁻³-10⁻¹Hz, reflecting corrosion rate) and roll-off slope (-20 to -40 dB / decade, reflecting film characteristics).

[0169] The features are calculated using Python signal processing libraries (such as SciPy) to form a 256-dimensional ECN feature vector.

[0170] Based on the image analysis module of Example 2, the central platform further enhances visual feature extraction:

[0171] Color features: Calculate the HSV color histogram (256 bins) and color moments (mean, variance, skewness), and train a ResNet-18 classifier to identify corrosion products (e.g., Fe2O3 reddish-brown, probability 0-1).

[0172] Texture features: Contrast (0-1), correlation (0-1), and energy (0-1) are extracted using the gray-level co-occurrence matrix (GLCM), and surface roughness is described by a Gabor filter (4 directions, 3 scales).

[0173] Defect characteristics: Using the YOLOv7 model, the input is a 1920x1080 RGB image, and the output is the bounding box, number (pieces / cm²), and average size (mm) of pitting pits. The model is trained with 10,000 labeled images (including pitting and cracks) and has an accuracy of 95%.

[0174] The features form a 256-dimensional visual feature vector, which is then input into the temporal erosion evolution model.

[0175] The temporal corrosion evolution model of the central platform is based on a hierarchical self-attention mechanism. The second layer outputs a 512-dimensional global corrosion state representation vector (G, which integrates spatiotemporal features) and connects multiple prediction heads.

[0176] A local corrosion rate prediction head predicts the local corrosion rate (mm / year) for the next 24 hours, 7 days, and 30 days. It uses a 2-layer fully connected neural network (FFN, 256 neurons), with input G and output corrosion rate at each monitoring point. Training uses the mean squared error loss function and the Adam optimizer (learning rate 0.001) for 100 epochs.

[0177] The remaining wall thickness prediction head, combined with the initial wall thickness and historical corrosion amount, predicts the remaining wall thickness (mm) for the next 1 month, 6 months, and 1 year. FFN predicts the wall thickness reduction rate (mm / month), and the remaining wall thickness is calculated by integration.

[0178] The failure mode risk prediction head assesses the risk level (low, medium, high) of pitting corrosion and uniform thinning. FFN outputs classification probabilities (Softmax activation). The training data contains 500 pitting corrosion events, with an accuracy of 90%.

[0179] The central platform utilizes process parameter sensors (temperature, pressure, flow rate) from the distributed sensor network module and baseline data from the external benchmark calibration module to quantify the synergistic effects of multiple factors.

[0180] Sensor configuration

[0181] Temperature sensor: Armored PT100 RTD, accuracy ±0.1°C, range 20-200°C; Pressure sensor: Corrosion-resistant pressure transmitter, accuracy ±0.5%, range 0-2 MPa; Flow rate sensor: Electromagnetic flow meter, accuracy ±0.5%, range 0-2 m / s.

[0182] Corrosion rate (R_real), temperature (T), pressure (P), and flow rate (V) are aligned by timestamps (accuracy ±1 ms). A multivariate nonlinear regression model is used, with the formula R_real = β0 + β1T + β2P + β3V + β4TP + ​​β5TV + β6PV, where β_i is the regression coefficient. The model is fitted using the least squares method. An external benchmark module provides the influence of a single factor (e.g., R_base_T = 0.1·exp(-50 / RT)). The difference between the linear superposition of R_real and R_base is compared to quantify the synergistic effect, generating contribution weights (e.g., "temperature 40%, flow rate 30%)" and optimization suggestions (e.g., "reduce flow rate by 10% to slow down corrosion").

[0183] Work process

[0184] Data acquisition: Collect ECN, vision, and process parameter data every minute.

[0185] Feature extraction: Generate a 256-dimensional ECN and visual feature vector, combined with a 64-dimensional process parameter vector.

[0186] Model processing: The first layer of self-attention (8 heads, 512-dimensional output) fuses multi-source features, and the second layer analyzes 24-hour time series (144 time points) to generate a 512-dimensional vector G.

[0187] Predictive output: The prediction head outputs corrosion rate, wall thickness, and risk level.

[0188] Collaborative analysis: Regression models quantify the impact of process parameters and generate optimization suggestions.

[0189]

[0190] Symbol explanation:

[0191] Γ sync (T,P,V): Synergistic effect factor of process parameters (temperature T, pressure P, flow rate V) (unitless).

[0192] CR actual (T,P,V): Actual corrosion rate (mm / year), measured by distributed sensors.

[0193] CR base The baseline corrosion rate (mm / year) is provided by an external baseline calibration module.

[0194] ΔP i : The deviation of the i-th process parameter (e.g., ΔT, unit °C; ΔP, unit MPa; ΔV, unit m / s).

[0195] α i : The linear influence coefficient of the i-th process parameter, with a typical value of 0.01-0.1.

[0196] β ij : The interaction coefficient between the i-th and j-th process parameters, with a typical value of 0.005-0.05.

[0197] f int (P i , P j ): Interaction function, describing the nonlinear interaction between parameters.

[0198] ε: A small constant to prevent the denominator from being zero, typically 0.001.

[0199] Formula Explanation:

[0200] This formula compares the actual corrosion rate CR... actual Compared with the corrosion rate prediction based on baseline conditions and parameter deviations, the synergistic effect of process parameters such as temperature T, pressure P, and flow rate V is quantified. The prediction model includes a linear term (α). i · ΔP i ) and interaction items (β) ij · ΔP i · ΔP j · f int (P i , P j ), Synergistic effect factor Γ sync Reflecting the strength of actual synergy:

[0201] Γ sync >1: There is a synergistic effect between the parameters to promote corrosion.

[0202] Γ sync <1: There is a synergistic effect between the parameters to inhibit corrosion.

[0203] Γ sync ≈ 1: No significant synergistic effect.

[0204] For example, if CR actual = 0.2 mm / year, the predicted value is 0.15 mm / year, then Γ syncThe value ≈ 1.33 indicates a synergistic effect of 33%, suggesting that process parameters need to be adjusted (such as reducing the flow rate by 10%) to mitigate corrosion. This formula supports active corrosion control and improves the accuracy of equipment life prediction.

[0205] Application Cases

[0206] At a pipe bend (sulfuric acid concentration 98%, temperature 100°C, flow velocity 0.5 m / s):

[0207] ECN characteristics: R_n=10 4 Ω·cm², LI=0.6 (localized corrosion), visual characteristics: corrosion area 5%, 3 pits / cm², YOLOv7 accuracy 95%, prediction results: 30-day corrosion rate 0.2 mm / year, wall thickness 8.45 mm, medium risk of pitting corrosion, synergistic effects: temperature contributes 40%, flow rate 30%, it is recommended to reduce the flow rate by 10%.

[0208] Enriching features enhances model resolution, improving accuracy by 10%. Multidimensional risk assessment optimizes maintenance priorities, provides a basis for process optimization, and extends equipment lifespan.

[0209] Deep feature extraction reveals signs of microscopic corrosion, improving feature resolution by 10-20%, which is 5-10% better than traditional methods. The prediction head improves the accuracy of pitting corrosion risk prediction to 85-95%, which is better than 70-80% of traditional statistical models. Collaborative impact analysis quantifies the contribution weights of temperature, flow rate, etc. (e.g., temperature 40%), provides optimization suggestions, and reduces maintenance costs by 10-15%. Multi-failure mode prediction optimizes resource allocation and reduces unplanned downtime by 20-30%. The algorithm adapts to complex working conditions and improves stability by 15-25%.

[0210] The above effects have been verified through simulation experiments. The specific performance varies depending on the working conditions. Those skilled in the art can adjust the parameters according to actual needs to achieve similar effects.

[0211] This embodiment significantly improves the accuracy and decision support capability of the sulfuric acid plant corrosion monitoring and prediction system through the algorithm, refined analysis, and quantification of the synergistic effects of process parameters on the central data processing platform. Those skilled in the art can implement this technical solution based on the above description.

[0212] The above description is merely a preferred embodiment of the present invention. It should be understood that the present invention is not limited to the forms disclosed herein and should not be construed as excluding other embodiments. It can be used in various other combinations, modifications, and environments, and can be modified within the scope of the concept described herein through the above teachings or related technologies or knowledge. Modifications and variations made by those skilled in the art that do not depart from the spirit and scope of the present invention should be within the protection scope of the appended claims.

Claims

1. An online monitoring system for the corrosion status of a sulfuric acid plant, characterized in that, include: The external reference calibration module includes a standard container for holding a sulfuric acid solution of known concentration and an internally mounted external reference sensor. The critical viewing window observation module is installed at a critical location in the sulfuric acid plant. The critical viewing window observation module includes a transparent viewing window that is resistant to sulfuric acid corrosion, an internal reference sensor array disposed adjacent to its inner surface, and an industrial camera system that acquires internal images through the transparent viewing window. A distributed multi-parameter sensor network module, comprising multiple distributed sensors located at different locations on the device; The data acquisition and transmission module is used to synchronously acquire and timestamp multi-source data from the external reference calibration module, the internal reference sensor array, the industrial camera system, and the multiple distributed sensors at preset uniform time intervals. The central data processing and intelligent analysis platform is configured with a time-series erosion evolution model based on a hierarchical self-attention mechanism. The central data processing and intelligent analysis platform is configured as follows: The multi-source data is time-coded and feature-processed to form a data packet sequence; The data packet sequence is analyzed based on the first-layer multi-head self-attention mechanism to generate a context-aware state representation vector. A second-layer multi-head self-attention mechanism analysis is performed on the context-aware state representation vectors of multiple consecutive time intervals to generate a global erosion state representation vector. The global corrosion state characterization vector is used to monitor the current corrosion state of the equipment and predict future corrosion trends or remaining service life. The central data processing and intelligent analysis platform is further configured to: perform initial absolute calibration on each sensor in the system using the data from the external reference calibration module, and combine the dynamic correlation analysis results between the readings of the internal reference sensor array of the key window observation module and the visual feature data collected and processed by the industrial camera system to perform dynamic in-situ relative calibration or drift correction on the distributed sensors in the distributed multi-parameter sensor network module, thereby forming a multi-level collaborative calibration mechanism. The industrial camera system includes an image analysis artificial intelligence module configured to extract structured visual features, including the area percentage of the corroded area, color histogram, defect count, or quantitative value of sulfuric acid medium turbidity.

2. The online monitoring system for the corrosion status of a sulfuric acid plant according to claim 1, characterized in that: The transparent window of the key viewing module is made of special quartz glass, sapphire, or modified fluoroplastic, and the industrial camera system includes an independently controllable lighting unit.

3. The online monitoring system for the corrosion status of a sulfuric acid plant according to claim 1, characterized in that: The internal reference sensor array includes at least three internal reference sensors, and the central data processing and intelligent analysis platform is configured to use the average reading of the at least three internal reference sensors as a local corrosion state reference for the viewing window observation anchor area.

4. The online monitoring system for the corrosion status of a sulfuric acid plant according to claim 1, characterized in that: The external benchmark calibration module is further configured with a temperature control and flow rate simulation device, which is used to generate baseline data of the corrosion rate of the material under known sulfuric acid concentration, temperature and flow rate conditions. The baseline data is used to parameterize the time-series corrosion evolution model based on the hierarchical self-attention mechanism.

5. The online monitoring system for the corrosion status of a sulfuric acid plant according to claim 1, characterized in that: The central data processing and intelligent analysis platform performs feature processing on the multi-source data, including: extracting time-domain and frequency-domain features from electrochemical noise data, and extracting color, texture, and defect coordinates and size features identified by the target detection model from the visual data from the industrial camera system.

6. The online monitoring system for the corrosion status of a sulfuric acid plant according to claim 1, characterized in that: The central data processing and intelligent analysis platform is further configured with multiple prediction heads connected to the output of the time-series corrosion evolution model based on the hierarchical self-attention mechanism. The prediction heads are used to output local corrosion rate, remaining wall thickness, and risk level for different failure modes.

7. The online monitoring system for the corrosion status of a sulfuric acid plant according to claim 1, characterized in that: The central data processing and intelligent analysis platform is further configured to automatically generate actionable maintenance suggestions based on the monitored current corrosion status, predicted future corrosion trends, and remaining service life, or to trigger multi-level alarms when a preset risk threshold is reached.

8. The online monitoring system for the corrosion status of a sulfuric acid plant according to claim 1, characterized in that: The time-series corrosion evolution model based on the hierarchical self-attention mechanism of the central data processing and intelligent analysis platform is constructed and optimized by using a historical dataset including the multi-source data and labeled data of offline equipment detection results, equipment maintenance records or manually labeled corrosion events corresponding to the historical dataset for supervised learning or self-supervised learning pre-training.

9. The online monitoring system for the corrosion status of a sulfuric acid plant according to claim 1, characterized in that: The distributed multi-parameter sensing network module includes an electrochemical sensor, a temperature sensor, a pressure sensor, and a flow rate sensor. The central data processing and intelligent analysis platform is configured to use the data from the temperature sensor, pressure sensor, and flow rate sensor, combined with the baseline data from the external reference calibration module, to quantify the synergistic effect of sulfuric acid concentration, temperature, and flow rate on the corrosion rate.

Images