Reaction kettle safety monitoring system for dangerous chemical workshop production
By combining shape memory alloy reinforced sealing rings with the Attention-LSTM model, real-time monitoring and early warning were achieved, solving the problem of reactor sealing failure, enabling early prediction and timely intervention, and avoiding safety accidents and property losses.
Patent Information
- Application Number
- CN202511084067.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-08-04
- Publication Date
- 2025-11-18
AI Technical Summary
Existing static seals on reactors are prone to aging under high temperature and strong corrosion conditions, leading to seal failure. Existing monitoring methods have low accuracy and are difficult to detect sealing abnormalities in the early stages, resulting in media leakage and safety accidents.
A safety monitoring system combining shape memory alloy reinforced sealing rings and an Attention-LSTM model is used. The shape memory alloy expands at high temperatures to enhance the sealing force, the monitoring module collects data in real time, and the Attention-LSTM model makes predictions, providing graded early warnings and timely intervention.
Effectively prevents seal failure, detects seal degradation in a timely manner, avoids media leakage and safety accidents, and reduces property damage.
Smart Images
Figure BDA0005531872350000061
Abstract
Description
TECHNICAL FIELD
[0001] The present application relates to the field of reaction kettle safety monitoring, in particular to a reaction kettle safety monitoring system for dangerous chemical workshop production. BACKGROUND
[0002] As the core equipment of chemical production, the static sealing performance of the reaction kettle is directly related to the safety of production, especially in high-temperature and strong-corrosion conditions such as nitration, oxidation, fluorination, diazotization and chlorination reactions, the failure of static sealing may lead to medium leakage and cause serious accidents such as explosion and poisoning. Moreover, in the continuous production process, once the medium leakage occurs, it will cause greater property loss.
[0003] The traditional static sealing of the reaction kettle relies on the elastic deformation of the sealing ring to achieve sealing, but the high-temperature environment easily leads to aging and elastic decline of the sealing ring, insufficient sealing surface pressure, and further causes sealing failure. The existing sealing monitoring method has low monitoring accuracy and cannot monitor the early abnormal state of sealing; more often, it is monitored after the sealing leakage has occurred, which leads to passive treatment of the leakage problem. Therefore, the present application provides a reaction kettle safety monitoring system for dangerous chemical workshop production. SUMMARY
[0004] The purpose of the present application is to provide a reaction kettle safety monitoring system for dangerous chemical workshop production to solve the problem of single detection scheme in the background art.
[0005] To achieve the above purpose, the present application provides the following technical scheme:
[0006] A reaction kettle safety monitoring system for dangerous chemical workshop production, comprising a memory alloy enhanced sealing ring, a monitoring module, a data processing module, a hierarchical early warning module and a data storage module;
[0007] The memory alloy enhanced sealing ring is arranged at the flange connection of the reaction kettle, and a memory alloy is embedded inside. When the temperature rises to a preset temperature, an expansion force is generated to push the sealing ring substrate to extrude the flange sealing surface, thereby enhancing the sealing surface pressure and enhancing the sealing performance;
[0008] The monitoring module comprises a pressure sensor, a temperature sensor and a displacement sensor for real-time monitoring of the sealing ring data;
[0009] The data processing module adopts an Attention-LSTM model to calculate the data of the temperature sensor, the pressure sensor and the displacement sensor and outputs the calculated prediction results; the prediction results are divided into normal, slight abnormality and serious abnormality;
[0010] The hierarchical early warning module sets safety, primary warning and secondary warning corresponding to the normal, slight abnormality and serious abnormality of the data processing module;
[0011] The data storage module adopts a cloud storage server to store various monitoring data, neural network output results and alarm records.
[0012] Preferably, the preset expansion temperature of the memory alloy matches the lower limit of the normal working temperature of the reaction kettle; the sealing ring base material is made of high and low temperature resistant and corrosion resistant material, and the memory alloy wires are uniformly distributed in the inner side or the middle layer.
[0013] Preferably, the temperature sensor is embedded in the circumference of the sealing ring to collect temperature data at the sealing ring in real time; the pressure sensor is embedded in the outer edge of the sealing surface in contact with the flange to monitor the sealing surface pressure in real time; the displacement sensor is arranged on the flange to collect the relative opening distance of the flange in real time.
[0014] Preferably, when the output result is normal, no early warning is triggered; when the output result is slightly abnormal, a first-level early warning is automatically triggered, and alarm information is automatically sent to the monitoring center and the mobile terminal of the staff in real time; when the output result is seriously abnormal, a second-level early warning is automatically triggered, and an audible and visual alarm is started based on the first-level early warning to remind the workshop personnel to evacuate, and a shutdown instruction is sent to the reaction kettle control system in real time.
[0015] Preferably, when the number of cyclic expansions of the memory alloy wire exceeds the critical value of the set expansion number, the data storage module generates a sealing ring replacement reminder.
[0016] Preferably, the Attention-LSTM model includes an LSTM layer, an Attention layer and a fully connected layer, and the prediction result is output by calculating the monitoring data.
[0017] Preferably, the LSTM layer captures long-term dependencies by processing time series data of temperature, pressure and displacement.
[0018] Preferably, the Attention layer weights the hidden state sequence output by the LSTM to highlight the information of the key time step and outputs a context vector.
[0019] Preferably, the context vector output by the Attention layer is input into the fully connected layer to output a prediction result; the prediction result is one of normal, slightly abnormal and seriously abnormal.
[0020] Preferably, the data storage module supports online updating of the training data of the Attention-LSTM model to optimize the performance of the model.
[0021] Compared with the prior art, the beneficial effects of the present invention are as follows: by utilizing the phase change expansion characteristics of the shape memory alloy-reinforced sealing ring, the sealing surface pressure is automatically increased when the reactor temperature rises to the lower operating limit, effectively compensating for the elastic decay problem of the sealing ring under high temperature conditions, and preventing sealing failure from the physical structure. While the sealing decay problem is detected in time through the Attention-LSTM model, the shape memory alloy can still ensure the sealing effect, providing sufficient time for maintenance and replacement of the sealing ring, thereby achieving timely detection and intervention to avoid safety accidents and associated property losses caused by media leakage.
[0022] The Attention-LSTM fusion algorithm is adopted to capture the long-term temporal dependence of temperature / pressure / displacement through the LSTM layer and combine it with the Attention layer to dynamically focus on the time points of abnormal signs, so as to achieve early prediction of seal decay. Detailed Implementation
[0023] Example 1:
[0024] This invention provides a technical solution: a safety monitoring system for reaction vessels used in hazardous chemical workshops, comprising a shape memory alloy reinforced sealing ring, a monitoring module, a data processing module, a graded early warning module, and a data storage module.
[0025] A shape memory alloy reinforced sealing ring is installed at the flange connection of the reactor, with a shape memory alloy, such as a nickel-titanium alloy wire, embedded inside. The preset expansion temperature of the nickel-titanium alloy wire matches the lower limit of the reactor's normal operating temperature. When the temperature rises to the preset temperature, the nickel-titanium alloy wire undergoes a phase change, generating expansion force that pushes the sealing ring substrate against the flange sealing surface, increasing the sealing surface pressure and enhancing the sealing performance. When the temperature falls below the preset temperature, the alloy wire contracts, and the substrate relies on its own elasticity to maintain a basic seal. The sealing ring substrate can be made of fluororubber or silicone rubber, which has high and low temperature resistance and corrosion resistance. The shape memory alloy wire is evenly distributed on the inner or middle layer and can be bonded to the substrate through a surface microporous injection process, avoiding direct contact with the medium inside the reactor.
[0026] The monitoring module includes a pressure sensor, a temperature sensor, and a displacement sensor.
[0027] The temperature sensor is embedded around the sealing ring to collect temperature data at the sealing ring in real time with an accuracy of ±1℃; the temperature sensor is a miniature temperature sensor that is resistant to high temperature and corrosion.
[0028] The pressure sensor is embedded in the outer edge of the sealing surface where the sealing ring contacts the flange, with an accuracy of ±0.05MPa, and monitors the pressure on the sealing surface in real time; the pressure sensor adopts a miniature pressure sensor that is resistant to high temperature and corrosion.
[0029] The displacement sensor is installed on the flange to collect the relative opening distance of the flange in real time, with an accuracy of less than 0.005 mm.
[0030] The data processing module uses the Attention-LSTM model (also known as the Attention-LSTM neural network model) to calculate the data from the temperature sensor, pressure sensor, and displacement sensor and output the calculated prediction results; the prediction results are divided into three types: normal, slightly abnormal, and severely abnormal.
[0031] The tiered early warning module is configured with three levels: safety, Level 1 warning, and Level 2 warning, corresponding to normal, minor anomaly, and severe anomaly, respectively. When the output result is normal, no warning is triggered. When the output result is a minor anomaly, a Level 1 warning is automatically triggered, and alarm information (including abnormal parameters, fault type, and reactor number) is automatically sent remotely to the monitoring center and staff mobile terminals. When the output result is a severe anomaly, a Level 2 warning is automatically triggered, simultaneously activating the audible and visual alarms to alert workshop personnel to evacuate and remotely sending a shutdown command to the reactor control system. A sealing pressure ≤10% is considered normal; a sealing pressure drop >10% is considered a minor anomaly; and a sealing pressure drop >30% or opening displacement >0.1mm is considered a severe anomaly.
[0032] The data storage module utilizes a cloud storage server to store various monitoring data (including the number of cyclic expansion cycles of the shape memory alloy wire, maintenance records, running time, temperature, sealing surface pressure, flange opening, etc.), neural network output results, and alarm records. It supports online updates of the neural network model's training data. Specifically, through a feedback loop between real-time monitoring data and early warning results, new samples are automatically labeled and the model is incrementally trained to optimize model performance. When the number of cyclic expansion cycles of the shape memory alloy wire exceeds a pre-set critical value, a sealing ring replacement reminder is generated. Specifically, a temperature sensor records the number of times the preset expansion temperature is reached, and this count is accumulated in the data storage module.
[0033] The Attention-LSTM model is used to improve the accuracy of early prediction of sealing anomalies. First, the LSTM layer processes time-series data of temperature, pressure, and displacement (specifically using Z-score normalization) to output the hidden state at each time step, capturing long-short-term dependencies. Then, the Attention layer, through weight allocation (e.g., assigning higher weights to key time points such as sudden temperature increases and pressure drops), focuses on anomaly features, reduces the error in predicting failure probability, and identifies early sealing anomalies in advance.
[0034] The Attention layer weights the hidden state sequence output by the LSTM, highlighting information from key time steps and outputting a context vector.
[0035] Finally, the context vector c output by the Attention layer is input into a fully connected layer, followed by a Softmax activation function, which outputs the probability distributions for three categories: normal, slight anomaly, and severe anomaly. Slight anomaly corresponds to early degradation of sealing performance (such as slow pressure drop or uneven temperature distribution), while severe anomaly corresponds to impending or already occurring leakage (such as sudden pressure drop or excessive opening displacement).
[0036] All parameters in this system model are configurable variables. In actual deployment, they need to be dynamically set according to the specific operating conditions of the reactor, historical data, and safety standards.
[0037] Example 2
[0038] Based on Example 1, the calculation formulas for the LSTM layer, Attention layer, and fully connected layer are as follows:
[0039] The formula for calculating LSTM layers is: h t =tanh(C t )⊙(W O x t +U O h t-1 +b o ),
[0040] C t =σ(W f x t +U f h t-1 +b f )⊙C t-1 +σ(W i x t +U i h t-1 +b i )⊙tanh(W c x t +U c h t-1 +b c ).
[0041] Where x t The input vector at time step t (with dimension d) in ), d in h represents the input feature dimension. t Let d be the LSTM hidden state at time step t (with dimension d). h ), d h C represents the hidden state dimension of the LSTM. t Let d be the LSTM cell state at time step t (dimension d) h );h t-1 This is the hidden state from the previous time step; W f Wi W c W o U is the weight matrix of the input features. f U i U c U o Let b be the weight matrix of the hidden state. f b i b c b o σ is the bias term; σ is the sigmoid activation function (output range [0,1]), tanh is the hyperbolic tangent activation function, and ⊙ is element-wise multiplication.
[0042] The formula for calculating the Attention layer is:
[0043]
[0044] Where h t Let e be the LSTM hidden state of the LSTM layer at time step t (t = 1, 2, ..., T, where T is the sequence length). t Let α be the attention score for the t-th hidden state. t The attention weights for the t-th hidden state (satisfying) ), W a Let υ be the attention weight matrix, b be the attention score vector, and b be the attention weight matrix. a is the bias term, and c is the final output context vector (which integrates key sequence information).
[0045] The formula for calculating a fully connected layer is:
[0046] y = Softmax(z) = p(normal / minor abnormality / serious abnormality), z = W y ·c+b y Among them, W y Let b be the weight matrix. y is the bias term, and p is the output result.
Claims
1. A safety monitoring system for reaction vessels used in hazardous chemical production workshops, characterized in that: It includes shape memory alloy reinforced sealing rings, monitoring modules, data processing modules, graded early warning modules, and data storage modules; The shape memory alloy reinforced sealing ring is installed at the flange connection of the reactor. The shape memory alloy is embedded inside. When the temperature rises to the preset temperature, it generates expansion force, which pushes the sealing ring substrate to the flange sealing surface, thereby increasing the sealing surface pressure and enhancing the sealing performance. Monitoring module: Includes pressure sensor, temperature sensor and displacement sensor, to monitor sealing ring data in real time; Data processing module: The Attention-LSTM model is used to calculate the prediction results from the temperature sensor, pressure sensor and displacement sensor data and output the calculated prediction results; the prediction results are divided into three types: normal, slightly abnormal and severely abnormal. The tiered early warning module is set to safety, Level 1 warning, and Level 2 warning, corresponding to normal, minor anomaly, and severe anomaly in the data processing module; The data storage module uses a cloud storage server to store various monitoring data, neural network output results, and alarm records.
2. The safety monitoring system for a reaction vessel used in a hazardous chemical workshop according to claim 1, characterized in that: The preset expansion temperature of the shape memory alloy matches the lower limit of the normal operating temperature of the reactor; the sealing ring substrate is made of a material that is resistant to high and low temperatures and corrosion, and the shape memory alloy wires are evenly distributed on the inner or middle layer.
3. The safety monitoring system for a reaction vessel used in a hazardous chemical workshop according to claim 1, characterized in that: A temperature sensor is embedded around the sealing ring to collect temperature data at the sealing ring in real time. The pressure sensor is embedded in the outer edge of the sealing surface where the sealing ring contacts the flange to monitor the pressure on the sealing surface in real time; the displacement sensor is set on the flange to collect the relative opening distance of the flange in real time.
4. The safety monitoring system for a reaction vessel used in a hazardous chemical workshop according to claim 1, characterized in that: When the output result is normal, no warning is triggered. When the output result is slightly abnormal, a level one warning is automatically triggered, and the alarm information is automatically sent remotely to the monitoring center and the staff's mobile terminal. When the output result is seriously abnormal, a level two warning is automatically triggered. On the basis of the level one warning, the audible and visual alarm is activated at the same time to remind workshop personnel to evacuate and to remotely send a shutdown command to the reactor control system.
5. The safety monitoring system for a reaction vessel used in a hazardous chemical workshop according to claim 1, characterized in that: When the number of expansion cycles of the shape memory alloy wire exceeds the set critical value, the data storage module generates a reminder to replace the sealing ring.
6. The safety monitoring system for a reaction vessel used in a hazardous chemical workshop according to claim 1, characterized in that: The Attention-LSTM model consists of an LSTM layer, an Attention layer, and a fully connected layer. It calculates and outputs prediction results by analyzing the monitored data.
7. A safety monitoring system for a reaction vessel used in a hazardous chemical workshop according to claim 6, characterized in that: LSTM layers capture long-term dependencies by processing time-series data of temperature, pressure, and displacement.
8. A safety monitoring system for a reaction vessel used in a hazardous chemical workshop according to claim 7, characterized in that: The Attention layer weights the hidden state sequence output by the LSTM, highlighting information from key time steps and outputting a context vector.
9. A safety monitoring system for a reaction vessel used in a hazardous chemical workshop according to claim 8, characterized in that: The context vector output from the Attention layer is input into the fully connected layer, and the prediction result is output; the prediction result is one of three types: normal, slightly abnormal, and severely abnormal.
10. A safety monitoring system for a reaction vessel used in a hazardous chemical workshop according to claim 1, characterized in that: The data storage module supports online updates of the Attention-LSTM model's training data, optimizing model performance.