A vulcanization process for rubber tracks

Through phased boosting and dynamic temperature adjustment combined with LSTM model and fuzzy control, the temperature inhomogeneity problem during rubber track vulcanization is solved, the automation and accuracy of the vulcanization process is improved, and product quality and production efficiency are ensured.

CN118769581BActive Publication Date: 2024-12-06LAI ZHOU SHI LAI SUO ZHI PIN YOU XIAN GONG SI
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411241268.4
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-09-05
Publication Date
2024-12-06
Estimated Expiration
2044-09-05

AI Technical Summary

Technical Problem

During the vulcanization process of rubber tracks, temperature unevenness in the boosting and heating stages affects the stability of the crosslinking reaction, resulting in uneven heating inside and outside the vulcanized tank, affecting product quality.

Method used

The staged boost and dynamic temperature adjustment strategy are adopted, combined with the LSTM model and fuzzy control, through the temperature penalty terms and intelligent monitoring of the time dimension, the uniformity and stability of the temperature field in the vulcanized tank are ensured, and the efficiency of the nitrogen cooling system is optimized.

Benefits of technology

Improves the automation and accuracy of the vulcanization process of rubber tracks, ensures consistency of product quality and production efficiency, and improves wear resistance, shock resistance and low noise performance.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118769581B_ABST
    Figure CN118769581B_ABST
Patent Text Reader

Abstract

The invention relates to the field of rubber technology, and specifically discloses a vulcanization process of a rubber track, which is used to solve the problem of how to use cold nitrogen to keep the temperature in a vulcanization tank constant while boosting the pressure in the vulcanization tank. The process includes batching, mixing and opening, calendering and molding, first vulcanization, joint processing, and second vulcanization. In the boosting process of the first and second vulcanization, a dynamic adjustment strategy of cold nitrogen input flow and temperature set to maintain the temperature stability in the tank is proposed; through the staged boosting and dynamic temperature adjustment strategy, combined with an LSTM model and fuzzy control, the uniformity and stability of the temperature field in the boosting stage in the vulcanization tank are achieved; through the temperature penalty item and intelligent monitoring in the time dimension, the production automation and accuracy are improved, the efficiency of the nitrogen cooling system is evaluated, and the high quality and production efficiency of the rubber track are ensured.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of rubber technology, and more specifically, to a vulcanization process of a rubber crawler. Background Art

[0002] Rubber tracks have excellent wear resistance, shock resistance and low noise performance. They are an important alternative to wheeled vehicles in agricultural machinery, engineering vehicles and military equipment. Rubber tracks are usually made of natural rubber or synthetic rubber and their derivatives. The vulcanization process is to mix rubber with a vulcanizing agent and react chemically at a preset temperature and pressure to form a cross-linked structure, thereby improving the hardness, elasticity and wear resistance of the rubber. The preparation process of jointless rubber tracks mainly includes batching, mixing, molding, first vulcanization, joint treatment, second vulcanization, cooling and inspection. The first vulcanization is mainly to preliminarily vulcanize and mold the rubber mixture to ensure that the rubber has basic physical and chemical properties, and the main purpose of the second vulcanization is to ensure that the rubber at the joint is completely vulcanized to form a jointless overall structure. The vulcanization tank is an indispensable production equipment for the two vulcanizations. The process flow includes preparation, loading, heating stage, boosting stage, vulcanization stage, heat preservation and pressure maintenance, cooling and pressure reduction, and discharging. In the rubber track vulcanization process In the process, the pressurization and heating stages can be carried out simultaneously or heated first and then pressurized according to the specific process requirements and equipment configuration. The method of heating first and then pressurizing is currently more commonly used. After the vulcanizer is heated to the set temperature, constant temperature pressurization is required. This is because the cross-linking reaction of the rubber material is most stable at a constant temperature. If the temperature in the tank changes during the pressurization process, it will lead to uneven reaction rate and affect the stability of the cross-linking reaction. At the same time, the rubber is heated unevenly inside and outside, and the cross-linking reaction is inconsistent, which will affect the uniformity of rubber track vulcanization and product quality. According to the thermodynamics and gas dynamics principles of the adiabatic compression process of ideal gas in the vulcanizer, the initial temperature in the tank is 160°C, and nitrogen is used to increase the pressure in the tank from 1 MPa to 15 MPa. The average free path between gas molecules in the tank will decrease, the molecular collision frequency will increase, and the temperature will rise to 742°C. Therefore, how to use cold nitrogen to keep the temperature in the tank constant while pressurizing the vulcanizer has become a problem that needs to be solved urgently. Summary of the invention

[0003] In order to overcome the above-mentioned defects of the prior art, the present invention provides a vulcanization process for a rubber track, which realizes the uniformity and stability of the temperature field in the boosting stage in the vulcanization tank through a staged boosting and dynamic temperature adjustment strategy, combined with an LSTM model and fuzzy control, and improves production automation and accuracy through temperature penalty terms and intelligent monitoring in the time dimension, evaluates the efficiency of the nitrogen cooling system, and ensures the high quality and production efficiency of the rubber track.

[0004] To achieve the above object, the present invention provides the following technical solutions:

[0005] A vulcanization process for a rubber track includes the following steps:

[0006] Step 1, batching, mixing and open mixing: adding rubber base material, vulcanizing agent, accelerator, reinforcing agent and softener in a mixing mill according to the formula ratio, controlling the mixing temperature at 70-90° C., mixing for 5-10 minutes, transferring the mixed rubber material to an open mixing mill, and adjusting the thickness and uniformity of the rubber material;

[0007] Step 2, calendering and molding: the mixture obtained in step 1 is pressed into a sample with a thickness of 2 to 5 mm by a calender, and cut into several sections according to the design size of the crawler;

[0008] Step 3, first vulcanization: Place the sample formed in step 2 evenly in the customized heating mold in the vulcanization tank, gradually and evenly increase the temperature in the tank to the set temperature, use the temperature sensor to monitor the temperature in the tank in real time, start the cold nitrogen precooling system to cool the nitrogen to the initial set temperature, gradually introduce cold nitrogen to gradually increase the pressure in the tank to the set pressure, monitor the temperature in the tank in real time, dynamically adjust the nitrogen flow and temperature to keep the set temperature in the tank, maintain the set temperature and pressure in the tank after the pressurization is completed, vulcanize for 45 minutes, maintain the set temperature and pressure for 30 minutes after the end, then gradually reduce the temperature in the tank at a rate of 3 to 5 ° C / minute, and gradually reduce the pressure in the tank at a rate of 0.5 to 1 MPa / minute. After the temperature and pressure are within the safe range, open the tank door to take out the rubber track, conduct quality inspection, and take qualified products;

[0009] Step 4, joint processing: cut the rubber track obtained in step 3 according to the needs, fill the unvulcanized rubber mixture at the joint, apply chloroprene adhesive and pre-vulcanize;

[0010] Step 5, second vulcanization: place the track treated in step 4 evenly in the vulcanization tank, and perform secondary vulcanization with the dynamic temperature adjustment strategy in the tank during the cold nitrogen pressurization process in step 3, then take out the vulcanized rubber track product, conduct quality inspection, and take out qualified products.

[0011] In step 3, cold nitrogen gas is gradually introduced to gradually increase the pressure in the tank to the set pressure, which is divided into an initial low-pressure boosting stage, a medium-pressure boosting stage and a high-pressure boosting stage. In the low-pressure boosting stage, the boost pressure difference accounts for 20% of the difference between the target pressure value and the current pressure value, and the duration is 15 minutes. In the medium-pressure boosting stage, the boost pressure difference accounts for 40% of the difference between the target pressure value and the current pressure value, and the duration is 20 minutes. In the high-pressure boosting stage, the boost pressure difference accounts for 40% of the difference between the target pressure value and the current pressure value, and the duration is 25 minutes.

[0012] As a further scheme of the present invention, in step 1, the specific operation process is: take 100 parts of styrene-butadiene rubber and add it to an internal mixer to start mixing, add 30-60 parts of carbon black, 5-10 parts of paraffin oil, 1-3 parts of rosin, and 1-2 parts of antioxidant 4020 successively, mix for 5-10 minutes, add 3-5 parts of zinc oxide and 1-2 parts of stearate successively, control the mixing temperature at 70-90°C, then add 2-3 parts of sulfur, 0.5-1.5 parts of N-cyclohexyl-2-benzothiazole sulfonamide, and 0.2-0.8 parts of tetramethylthiuram disulfide to the mixture, continue mixing until all ingredients are fully mixed, transfer the mixture to an open mill for further mixing and adjust the thickness and uniformity of the mixture.

[0013] As a further solution of the present invention, the specific implementation process of step 2 includes:

[0014] Step 21, initial setting: adjust the roll gap of the calender to 2-5 mm, start the intelligent vision system, and calibrate the camera and laser rangefinder;

[0015] Step 22, feeding and calendering: gradually feeding the mixed material into the feed port of the calender, starting calendering, controlling the roller temperature of the calender between 60 and 80° C., and adjusting the roller speed of the calender to be able to uniformly press the rubber material into a sample sheet of the required thickness;

[0016] Step 23, intelligent monitoring and feedback: the intelligent visual system takes real-time photos of the rubber and sample during the calendering process, analyzes the sample thickness through the image processing algorithm, and measures the thickness in real time with a laser rangefinder. The measured data is fed back to the control system of the calender to automatically adjust the roller gap and roller speed to make the sample thickness uniform;

[0017] Step 24, sample conveying and cooling: the calendered sample is conveyed to the cutting area by a conveyor belt. During the cooling process, the sample is rapidly cooled at 1.5-2°C / min in the initial cooling stage for 5-10 minutes, gradually cooled at 1°C-1.5°C / min in the intermediate cooling stage for 10-20 minutes, and slowly cooled at 0.5°C-1°C / min in the final cooling stage until the sample temperature reaches the ambient temperature;

[0018] Step 25, intelligent cutting control: Use a laser rangefinder and image processing technology to accurately and automatically locate the cutting position of the sample, dynamically adjust the parameters of the cutting tool based on the real-time feedback of the visual system, adjust the parameters of the cutting tool based on the design size of the crawler, position the sample in the cutting area, and use the cutting tool for cutting. The visual system monitors the cracks, bubbles and thickness of the sample in real time. When defects are found, it automatically issues an alarm and marks the defective sample for removal or rework.

[0019] As a further solution of the present invention, the dynamic temperature adjustment strategy in the tank during the cold nitrogen pressurization process in step 3 and step 5 is:

[0020] Step 31, data collection and preprocessing: real-time collection of temperature and pressure data of each monitoring point within the observation time window of each pressurization stage in the vulcanizer, and data standardization and serialization;

[0021] Step 32, LSTM model training: input historical data, train the LSTM model, and add the temperature uniformity and stability penalty items in the autoclave in the time dimension;

[0022] Step 33, model prediction: input the temperature and pressure data in the autoclave at each pressurization stage in the current time window, and use the trained LSTM model to predict the input flow and input temperature combination of cold nitrogen within the preset threshold range, as well as the ratio of actual input cooling capacity to theoretical input cooling capacity;

[0023] Step 34, fuzzy control feedback adjustment: monitor the temperature and pressure data in the vulcanizer in real time, and according to the fuzzy control rules, traverse the cold nitrogen input flow rate and input temperature combination input in step 33 to dynamically adjust the nitrogen input flow rate and input temperature to ensure that the spatial uniformity and stability of the temperature field in the tank meet the preset requirements.

[0024] In step 33, according to the principle of gas thermodynamics, when the pressure is increased in the autoclave, the process is a variable non-adiabatic process, and the temperature rise is calculated using the following formula:

[0025] ;

[0026] Where: is the temperature rise caused by voltage increase, is the initial temperature in Kelvin, For the final pressure, is the initial pressure, is the adiabatic index of nitrogen, which is equal to 1.4.

[0027] Then calculate the heat exchange between the autoclave and the environment:

[0028] ;

[0029] Where: is the heat exchange between the autoclave and the environment, is the heat transfer coefficient, determined experimentally, is the heat transfer area, which is equal to the surface area of ​​the autoclave. , are the temperature inside the autoclave and the ambient temperature outside the autoclave, respectively. is the time interval.

[0030] The theoretical input cooling capacity is equal to the heat generated by the temperature rise caused by the pressure increase minus the heat exchange between the autoclave and the environment. Calculating the ratio of the actual input cooling capacity to the theoretical input cooling capacity can help evaluate the efficiency of the nitrogen cooling system under different operating conditions. By analyzing the ratio, changes in the heat transfer coefficient of the autoclave, environmental impacts, and degradation of equipment performance can be identified, so that appropriate compensation measures can be taken.

[0031] As a further scheme of the present invention, in step 32, the LSTM model includes an input layer, a first LSTM layer, a second LSTM layer, a fully connected layer 1, a fully connected layer 2, and an output layer, wherein the first LSTM layer and the second LSTM layer contain 64 units and 32 units respectively, the first LSTM layer and the second LSTM layer both contain an input gate, a forget gate, an output gate and candidate cell states, cell state updates and hidden state updates, and the fully connected layer 1 and the fully connected layer 2 contain 50 neurons and 25 neurons respectively.

[0032] The first layer of LSTM contains 64 units, which are used to extract the nonlinear relationship between the temperature and pressure parameters in the vulcanizer and the nitrogen input temperature. 64 units can provide sufficient capacity to capture the complex time series characteristics of the input data, especially the dynamic changes involving temperature and pressure. 64 units are a balance point that can capture complex time series characteristics without causing excessive computational costs. The second layer of LSTM contains 32 units, which are used to extract the nonlinear relationship between the temperature and pressure parameters in the vulcanizer and the nitrogen input flow rate. The 32 units of the second layer of LSTM are used to further extract high-level features based on capturing the output features of the first layer of LSTM. The number is halved to reduce the dimension and avoid overfitting. Reducing the number of units helps to improve the efficiency of model training and reasoning while maintaining sufficient expressiveness. The fully connected layer 1 contains 50 neurons for further processing the output of the second layer LSTM. The 50 neurons provide sufficient nonlinear conversion capabilities to map the features extracted by the LSTM to a higher-dimensional feature space. The 50 neurons can effectively process the data output by the LSTM layer while controlling the complexity and computational cost of the model. The fully connected layer 2 contains 25 neurons for further processing the output of the fully connected layer 1. The dimensionality is further reduced on the basis of the fully connected layer 1. The 25 neurons help to extract higher-level features while avoiding excessive compression of information. The 25 neurons provide a simplified structure to help further filter and extract key information, providing optimized input for the final output layer.

[0033] The input layer receives input data of the current time step, including the temperature and pressure parameters in the vulcanizer, the input temperature and flow rate of nitrogen, and the attribute parameters of the vulcanizer (tank volume, tank material thermal conductivity, specific heat capacity of gas in the tank, ambient temperature, nitrogen density, nitrogen dynamic viscosity, nitrogen basic flow rate, nitrogen flow rate preset threshold range, nitrogen temperature preset threshold range).

[0034] As a further solution of the present invention, in the first layer LSTM, the activation function of the input gate is:

[0035] ;

[0036] Where: is the output of the first layer LSTM input gate, , are the nitrogen input temperature change growth rate and lag time reaction parameters, obtained through training data, is the weight matrix of the input gate, is the hidden state of the previous time step, is the input of the current time step, is the bias vector of the input gate;

[0037] The temperature change during the vulcanization process has complex nonlinear characteristics. The traditional activation function cannot fully capture these complex changes. This function is a double exponential function that can effectively capture the slow growth and hysteresis effects, which is very important for modeling the nonlinear characteristics of the temperature change process.

[0038] , Respectively represent the growth rate and lag time of nitrogen input temperature change. These parameters can be obtained through training data, so that the model can better adapt to the actual temperature change pattern. This customized activation function can more accurately reflect the dynamic characteristics of temperature change. The shape of the function allows smooth processing of temperature changes, thereby avoiding sudden changes in temperature during the change process. Smoothing of temperature changes helps to improve the stability and accuracy of model prediction. Using this function as the activation function of the input gate can enhance the LSTM model's ability to extract specific features in time series data, especially when the input data has an obvious growth pattern and lag effect. In the vulcanization process, the accuracy of temperature control is crucial. By using this activation function, the LSTM model can better adapt to complex working conditions and improve the accuracy and reliability of temperature control. Since the activation function takes into account the growth rate and lag time of temperature change, the model can dynamically adjust the output of the input gate to better reflect the actual temperature changes under the current working conditions. Traditional activation functions will cause overfitting problems on complex data, while the double exponential characteristics of this function can better adapt to the data distribution of temperature changes, thereby reducing the risk of overfitting.

[0039] The activation function of the forget gate is:

[0040] ;

[0041] Where: is the output of the first layer LSTM forget gate, is the weight matrix of the forget gate, is the bias vector of the forget gate;

[0042] This activation function can capture the complex nonlinear characteristics of the input data. The temperature and pressure changes in the vulcanization process are highly nonlinear, and traditional activation functions cannot fully capture these complex relationships. The combined nonlinear function can better describe these changes. The hyperbolic cosine function is symmetrical and grows rapidly when the input is large. It can capture the multi-peak characteristics and local extreme values ​​in the data, which is very important when dealing with temperature and pressure changes. The hyperbolic tangent function has a compression characteristic when dealing with the numerical range, so that the output value is limited between -1 and 1, with a smooth transition, which is suitable for processing the output of the forget gate and ensuring the stability of the state update. The combined function can provide a smooth nonlinear transformation, which helps to smooth the output of the forget gate during the state update process, avoid mutations, and ensure the stability of the model. By combining nonlinear functions, the model can more effectively selectively forget unimportant information while retaining important historical information. It is very critical for the LSTM model to process long-term series data. In the vulcanization process, the temperature changes are complex and nonlinear. Using this activation function can better adapt to these changes and improve the prediction accuracy of the model. It can smoothly process the input data, avoid drastic fluctuations when the model updates the state, and improve the stability and reliability of the model.

[0043] The activation function of the output gate is:

[0044] ;

[0045] Where: is the activation function of the first layer LSTM output gate, is the weight matrix of the output gate, is the bias vector of the output gate;

[0046] By applying an exponential function to the input and using logarithms, this activation function can expand the dynamic range of values ​​so that the output can cover a wider range of values, which is particularly important when dealing with complex temperature and pressure changes in the vulcanization process. The function provides smooth and continuous output, which helps avoid sudden changes during the state update process and improves the stability of the model. When performing exponential operations, the values ​​can become very large. Using logarithmic functions can compress the range of values, avoid overflow problems, and improve the stability of numerical calculations. This function combines exponential and logarithmic operations to provide nonlinear mapping, which is suitable for processing complex time series data and capturing the nonlinear relationship between input and output. In the vulcanization process, the change in temperature is nonlinear. This activation function can better capture and reflect these complex changes and improve the prediction accuracy of the model. By using a smooth and continuous activation function, it can avoid drastic fluctuations during the state update process and ensure the stability of the model's output. Exponential operations can cause numerical overflow problems, while logarithmic operations can compress the range of values ​​to avoid such problems and ensure the stability of numerical calculations.

[0047] The activation function of the candidate cell state is:

[0048] ;

[0049] Where: is the output of the candidate cell state, is the weight matrix of the candidate cell state, is the bias matrix of the candidate cell state.

[0050] This function is very effective in dealing with cumulative errors. It can smoothly accumulate changes in input signals and avoid mutations in time series. This smooth accumulation feature is very important for changes in temperature and pressure. It can effectively deal with cumulative errors in input data and capture complex nonlinear relationships in input data. Through this activation function, LSTM can better reflect complex patterns in temperature and pressure changes. Its output range of [-1,1] can provide good numerical stability and avoid numerical instability caused by excessively large or small output values. It is essential for precise control of the vulcanization process. Temperature changes have a cumulative effect, and this function can well describe this cumulative effect. By smoothing and integrating the input signal, the model can capture the long-term trend and cumulative effect of temperature changes.

[0051] In the first layer of LSTM, the cell state update formula is: ,in, Update the value of the cell state at the current moment, Update the value of the cell state at the previous moment;

[0052] The hidden state update formula is: , Updates the value for the hidden state.

[0053] As a further solution of the present invention, in the second layer LSTM, the activation function of the input gate is:

[0054] ;

[0055] Where: is the output of the second layer LSTM input gate, , are the nitrogen flow resistance and flow rate relationship adjustment coefficients, obtained through training data, is the weight matrix of the input gate, is the hidden state of the previous time step, For the first layer LSTM at time step The hidden state output of is the bias vector of the input gate;

[0056] By accurately describing the relationship between resistance and flow rate during nitrogen flow, the model can optimize nitrogen input flow and temperature, and improve the temperature and pressure control accuracy during vulcanization. Complex nonlinear activation functions can better handle nonlinear relationships in actual operations and improve the prediction accuracy of the model. Combining the hidden state of the previous time step and the input of the current time step, the model can better capture the dynamic changes in time series data and improve the prediction ability. The relationship between flow resistance and flow rate can be obtained through experimental data and theoretical formulas. Parameters , It is obtained by fitting the training data and can accurately reflect this relationship. Introducing the square root function in the denominator can smooth the input data, avoid values ​​that are too large or too small, and provide stable output.

[0057] The activation function of the forget gate is:

[0058] ;

[0059] Where: is the output of the second layer LSTM forget gate, is the friction coefficient of cold nitrogen, determined experimentally, is the diameter of the cold nitrogen input pipe, is the acceleration due to gravity, is the weight matrix of the forget gate, is the bias vector of the forget gate;

[0060] The friction coefficient determined experimentally reflects the friction characteristics of nitrogen when it flows in the pipeline. This parameter has a direct impact on the resistance to flow. The input pipe diameter of cold nitrogen is an important physical parameter that directly affects the flow rate and flow resistance. In fluid mechanics, the acceleration of gravity is a basic constant that affects the flow behavior of the fluid. By introducing these physical parameters into the activation function, the physical characteristics of nitrogen flow can be more accurately reflected, enabling the model to better capture the complex relationships in actual operations. The square root function is used to smooth the input data to avoid output values ​​that are too large or too small, thereby providing a stable output, which is particularly important when dealing with the relationship between resistance and flow rate when nitrogen flows. Each parameter in the activation function has a clear physical meaning, which enhances the interpretability and transparency of the model, making the model not only a black box, but also able to explain the influence of each parameter in the flow process. By combining the hidden state of the previous time step and the input of the current time step, the model can better capture the dynamic changes in the time series data and improve the predictive ability of nitrogen flow control.

[0061] The activation function of the output gate is:

[0062] ;

[0063] Where: is the activation function of the second layer LSTM output gate, is the weight matrix of the output gate, is the bias vector of the output gate, is the density of nitrogen, which is a known parameter;

[0064] The density of nitrogen is a known physical parameter that reflects the quality characteristics of nitrogen at a specific temperature and pressure. By introducing nitrogen density, the actual physical characteristics of nitrogen flow can be more accurately reflected. The square root function is used to smooth the input data to avoid output values ​​that are too large or too small, thereby providing a stable output, which is particularly important when dealing with complex fluid mechanics problems. By introducing the ratio of square root and nitrogen density in the activation function, it can ensure that the output value fluctuates within a reasonable range, thereby avoiding numerical instability. Each parameter in the activation function has a clear physical meaning, which enhances the interpretability and transparency of the model, making the model not only a black box, but also able to explain the influence of each parameter in the flow process. It is a common form in fluid mechanics, used to describe the energy characteristics of fluids, and is convenient for capturing energy changes in nitrogen flow, so as to more accurately predict and control nitrogen flow rate and temperature. The activation function combines the hidden state of the previous time step with the hidden state of the first layer LSTM in the current time step, which can comprehensively consider the comprehensive influence of historical information and current input in time series data, and improve the prediction ability of the model. By accurately describing the energy characteristics and flow rate relationship of nitrogen flow, the model can optimize the nitrogen input flow rate and temperature, and improve the temperature and pressure control accuracy during the vulcanization process. The complex nonlinear activation function can better handle the nonlinear relationship in actual operation and improve the prediction accuracy of the model. Combining the hidden state of the previous time step and the input of the current time step, the model can better capture the dynamic changes in time series data and improve the prediction ability.

[0065] The activation function of the candidate cell state is:

[0066] ;

[0067] Where: is the output of the candidate cell state, is the weight matrix of the candidate cell state, is the bias matrix of the candidate cell state, is the basic flow rate of nitrogen, is the average of the maximum and minimum values ​​of the flow rate threshold range for nitrogen, is the dynamic viscosity of nitrogen, which is a known parameter.

[0068] Through this activation function, the physical characteristics of nitrogen flow can be reflected, the flow characteristics can be smoothed, stable output and control range can be provided, flow characteristics and energy characteristics can be captured, historical information and current input can be comprehensively considered, and the prediction accuracy and stability of temperature and pressure control in the vulcanization process can be improved. It can better adapt to the complex conditions in the vulcanization process and ensure the stability and efficiency of the process.

[0069] In the second layer of LSTM, the cell state update formula is: ,in, Update the value of the cell state at the current moment, Update the value of the cell state at the previous moment;

[0070] The hidden state update formula is: , Updates the value for the hidden state.

[0071] As a further solution of the present invention, in step 32, in the loss function of the LSTM model, the temperature uniformity penalty term in the time dimension is:

[0072] ;

[0073] Where: is the penalty term for temperature uniformity in the time dimension, is the index of the data point in the time window, is the total number of data points in the time window, In the time window The temperature of the data point is the average of all temperature measurement points in the autoclave. is the mean temperature of all data points in the time window;

[0074] The temperature stability penalty term in the time dimension is:

[0075] ;

[0076] Where: is the temperature stability penalty term in the time dimension, is the temperature value of the data point before the current data point in the time window.

[0077] The penalty for temperature uniformity in the time dimension is to ensure the uniform distribution of the temperature in the vulcanizer in the time dimension. By minimizing the difference between the temperature of each data point and the average temperature in the time window, the temperature fluctuation can be reduced and the temperature uniformity can be ensured. During the vulcanization process, temperature uniformity is crucial for the uniform vulcanization of rubber. Uneven temperature will lead to inconsistent product quality and affect the performance and durability of the rubber track. By penalizing temperature fluctuations, local overheating or overcooling can be avoided. Local overheating or overcooling will lead to different vulcanization degrees in different parts of the rubber track, affecting the overall performance. The penalty for temperature stability in the time dimension is to ensure the stability of the temperature in the vulcanizer in the time dimension. By minimizing the temperature change between adjacent time points, the temperature fluctuation can be reduced and the temperature stability can be ensured. The drastic temperature fluctuation will lead to stress concentration and thermal damage to the rubber material during the vulcanization process, thereby affecting the product quality. By penalizing the temperature change between adjacent time points, a smooth temperature transition can be achieved. Smooth temperature changes help reduce temperature stress, ensure that the rubber material is heated evenly during the vulcanization process, and avoid stress concentration and material deformation. By ensuring the uniformity and stability of temperature, the quality of rubber tracks during the vulcanization process can be improved and the performance and durability of the product can be ensured; by reducing temperature fluctuations and achieving smooth temperature changes, the energy consumption and time during the vulcanization process can be optimized and production efficiency can be improved; the control of temperature uniformity and stability helps to ensure the stability of the vulcanization process and reduce the uncertainty caused by temperature fluctuations, thereby improving the controllability and reliability of the production process.

[0078] As a further solution of the present invention, in step 32, in the loss function in the LSTM model, the loss function is:

[0079] ;

[0080] Where: is the loss function value, Index of the temperature monitoring point in the autoclave. In the time window Data point The actual temperature monitoring value at each location, In the time window Data point The predicted temperature value at each location, is the total number of temperature monitoring points, , are the weight coefficients of the temperature uniformity penalty term and the temperature stability penalty term in the time dimension, which are set empirically.

[0081] As a further solution of the present invention, in the fuzzy rules used in step 34, the input variables of the fuzzy control are temperature error, temperature change rate and temperature distribution uniformity, wherein the elements in the fuzzy set of temperature error are negative large (NB, temperature error <-5°C), negative small (NS1, -1> temperature error ≥-5°C), zero (ZE1, 1°C> temperature error ≥-1°C), positive small (PS1, 5°C> temperature error ≥1°C), and positive large (PB1 temperature error>5°C), and the elements in the fuzzy set of temperature change rate are fast decreasing (ND, temperature change rate <-2°C / min), slow decreasing (NS, -0.5°C / min), and slow decreasing (NS, -0.5°C / min). 〉temperature change rate ≥ -2℃ / min), unchanged (ZE2, 0.5℃ / min≥temperature change rate ≥ -0.5℃ / min), slow rise (PS2, 2℃ / min≥temperature change rate>0.5℃ / min), fast rise (PD, temperature change rate>2℃ / min), the elements in the fuzzy set of temperature distribution uniformity (the degree of temperature difference between the temperatures of each measuring point in the reaction vulcanization tank, which is the standard deviation of the temperature measuring point) are: uneven (NU, temperature distribution uniformity>2℃), relatively uniform (QY, 2℃≥temperature distribution uniformity>1℃), uniform (Y, temperature distribution uniformity≤1℃), the fuzzy control output variables are nitrogen flow adjustment and nitrogen temperature adjustment, and the fuzzy set elements of nitrogen flow adjustment include low flow (L1, nitrogen flow adjustment <-5L / min), medium flow (M1, -5 L / min≤nitrogen flow adjustment≤5 L / min), high flow rate (H1, nitrogen flow rate adjustment>5L / min), the fuzzy set elements of nitrogen temperature adjustment include low temperature (L2, nitrogen temperature adjustment<-5℃), medium temperature (M2, -5℃≤nitrogen temperature adjustment≤5℃), high temperature (H2, nitrogen temperature adjustment>5℃), evaluate the input data based on the fuzzy rules set by expert knowledge and experience, obtain the activation strength of each rule, use fuzzy logic to infer the activated rules, generate fuzzy output, calculate the membership of each output variable in the fuzzy set according to the activation strength, and use the centroid method to convert the fuzzy output into precise control output.

[0082] Compared with the prior art, the technical effect of the process proposed by the present invention is as follows: the present invention monitors and adjusts the nitrogen flow and temperature in the vulcanizer in real time through a staged boosting and dynamic temperature adjustment strategy, combined with an LSTM model and fuzzy control, to ensure the uniformity and stability of the temperature field of the vulcanizer during the staged boosting process, and improves the automation and accuracy of the production process by introducing temperature uniformity and stability penalty terms in the time dimension, as well as an intelligent monitoring and feedback mechanism, calculates the ratio of actual input cooling capacity to theoretical input cooling capacity, evaluates the efficiency of the nitrogen cooling system, and promptly identifies changes in the heat transfer coefficient of the vulcanizer, environmental impacts, and a decline in equipment performance, so as to facilitate the adoption of corresponding compensation measures, further ensuring the stability of the temperature in the vulcanizer during the boosting process, helping to improve the wear resistance, shock resistance, and low noise performance of the rubber track by optimizing the two vulcanization effects of the rubber track, and ensuring the consistency of product quality and production efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0083] Figure 1 The present invention is a vulcanization process flow chart of a rubber crawler;

[0084] Figure 2 The network topology diagram of the LSTM model mentioned in the present invention. DETAILED DESCRIPTION

[0085] The following will be combined with the accompanying drawings in the present invention to clearly and completely describe the technical solution in the present invention. Obviously, the described technical solution is only a part of the present invention but not all of it. Based on the technical solution in the present invention, all other technical solutions obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0086] like Figure 1 As shown, a vulcanization process of a rubber crawler proposed by the present invention includes the following process:

[0087] Step 1, batching, mixing and open mixing: adding rubber base material, vulcanizing agent, accelerator, reinforcing agent and softener in an internal mixer according to the formula ratio, controlling the mixing temperature at 70-90° C. for 5-10 minutes, transferring the mixed rubber to an open mixer, and adjusting the thickness and uniformity of the rubber;

[0088] Step 2, calendering and molding: the mixture obtained in step 1 is pressed into a sample with a thickness of 2 to 5 mm by a calender, and cut into several sections according to the design size of the crawler;

[0089] Step 3, first vulcanization: Place the sample formed in step 2 evenly in the customized heating mold in the vulcanization tank, gradually and evenly increase the temperature in the tank to the set temperature, use the temperature sensor to monitor the temperature in the tank in real time, start the cold nitrogen precooling system to cool the nitrogen to the initial set temperature, gradually introduce cold nitrogen to gradually increase the pressure in the tank to the set pressure, monitor the temperature in the tank in real time, dynamically adjust the nitrogen flow and temperature to keep the set temperature in the tank, maintain the set temperature and pressure in the tank after the pressurization is completed, vulcanize for 45 minutes, maintain the set temperature and pressure for 30 minutes after the end, then gradually reduce the temperature in the tank at a rate of 3 to 5 ° C / minute, and gradually reduce the pressure in the tank at a rate of 0.5 to 1 MPa / minute. After the temperature and pressure are within the safe range, open the tank door to take out the rubber track, conduct quality inspection, and take qualified products;

[0090] Step 4, joint processing: cut the rubber track obtained in step 3 according to the needs, fill the unvulcanized rubber mixture at the joint, apply chloroprene adhesive and pre-vulcanize;

[0091] Step 5, second vulcanization: place the track treated in step 4 evenly in the vulcanization tank, and perform secondary vulcanization with the dynamic temperature adjustment strategy in the tank during the cold nitrogen pressurization process in step 3, then take out the vulcanized rubber track product, conduct quality inspection, and take out qualified products.

[0092] The present invention uses a staged boosting and dynamic temperature adjustment strategy, combined with an LSTM model and fuzzy control, to monitor and adjust the nitrogen flow and temperature in a vulcanizer in real time, ensure the uniformity and stability of the temperature field of the vulcanizer during the staged boosting process, introduce temperature uniformity and stability penalty items in the time dimension, and an intelligent monitoring and feedback mechanism, thereby improving the automation and accuracy of the production process, calculating the ratio of actual input cooling capacity to theoretical input cooling capacity, evaluating the efficiency of the nitrogen cooling system, timely identifying changes in the heat transfer coefficient of the vulcanizer, environmental impacts, and degradation of equipment performance, so as to facilitate taking corresponding compensation measures, further ensuring the stability of the temperature in the vulcanizer during the boosting process, helping to improve the wear resistance, shock resistance, and low noise performance of the rubber track by optimizing the secondary vulcanization effect of the rubber track, and ensuring the consistency of product quality and production efficiency.

[0093] In step 3, cold nitrogen gas is gradually introduced to gradually increase the pressure in the tank to the set pressure. The boosting stage is divided into an initial low-pressure boosting stage, a medium-pressure boosting stage and a high-pressure boosting stage. In the low-pressure boosting stage, the boost pressure difference accounts for 20% of the difference between the target pressure value and the current pressure value, and the duration is 15 minutes. In the medium-pressure boosting stage, the boost pressure difference accounts for 40% of the difference between the target pressure value and the current pressure value, and the duration is 20 minutes. In the high-pressure boosting stage, the boost pressure difference accounts for 40% of the difference between the target pressure value and the current pressure value, and the duration is 25 minutes.

[0094] It should be noted that in step 1, the specific operation process is: add 100 parts of styrene-butadiene rubber into an internal mixer and start mixing, add 30-60 parts of carbon black, 5-10 parts of paraffin oil, 1-3 parts of rosin, and 1-2 parts of antioxidant 4020 successively, mix for 5-10 minutes, add 3-5 parts of zinc oxide and 1-2 parts of stearate successively, control the mixing temperature at 70-90°C, then add 2-3 parts of sulfur, 0.5-1.5 parts of N-cyclohexyl-2-benzothiazole sulfonamide, and 0.2-0.8 parts of tetramethylthiuram disulfide to the mixture, continue mixing until all ingredients are fully mixed, transfer the mixture to an open mill for further mixing and adjust the thickness and uniformity of the mixture.

[0095] It should be noted that the specific implementation process of step 2 includes:

[0096] Step 21, initial setting: adjust the roll gap of the calender to 2-5 mm, start the intelligent vision system, and calibrate the camera and laser rangefinder;

[0097] Step 22, feeding and calendering: gradually feeding the mixed material into the feed port of the calender, starting calendering, controlling the roller temperature of the calender between 60 and 80° C., and adjusting the roller speed of the calender to be able to uniformly press the rubber material into a sample sheet of the required thickness;

[0098] Step 23, intelligent monitoring and feedback: the intelligent visual system takes real-time photos of the rubber and sample during the calendering process, analyzes the sample thickness through the image processing algorithm, and measures the thickness in real time with a laser rangefinder. The measured data is fed back to the control system of the calender to automatically adjust the roller gap and roller speed to make the sample thickness uniform;

[0099] Step 24, sample conveying and cooling: the calendered sample is conveyed to the cutting area by a conveyor belt. During the process, the sample is rapidly cooled at 1.5-2°C / min in the initial cooling stage for 5-10 minutes, gradually cooled at 1-1.5°C / min in the intermediate cooling stage for 10-20 minutes, and slowly cooled at 0.5-1°C / min in the final cooling stage until the sample temperature reaches the ambient temperature;

[0100] Step 25, intelligent cutting control: Use a laser rangefinder and image processing technology to automatically and accurately locate the cutting position of the sample, dynamically adjust the parameters of the cutting tool based on the real-time feedback of the visual system, adjust the parameters of the cutting tool based on the design size of the crawler, position the sample in the cutting area, and use the cutting tool for cutting. The visual system monitors the cracks, bubbles and thickness of the sample in real time. When defects are found, it automatically issues an alarm and marks the defective sample for removal or rework.

[0101] It should be noted that the dynamic temperature adjustment strategy in the tank during the cold nitrogen pressurization process in step 3 and step 5 is:

[0102] Step 31, data collection and preprocessing: real-time collection of temperature and pressure data of each monitoring point within the observation time window of each pressurization stage in the vulcanizer, and data standardization and serialization;

[0103] Step 32, LSTM model training: input historical data, train the LSTM model, and add the temperature uniformity and stability penalty items in the autoclave in the time dimension;

[0104] Step 33, model prediction: input the temperature and pressure data in the autoclave at each pressurization stage in the current time window, and use the trained LSTM model to predict the input flow and input temperature combination of cold nitrogen within the preset threshold range, as well as the ratio of actual input cooling capacity to theoretical input cooling capacity;

[0105] Step 34, fuzzy control feedback adjustment: monitor the temperature and pressure data in the vulcanizer in real time, and according to the fuzzy control rules, traverse the cold nitrogen input flow rate and input temperature combination input in step 33 to dynamically adjust the nitrogen input flow rate and input temperature to ensure that the spatial uniformity and stability of the temperature field in the tank meet the preset requirements.

[0106] In step 33, according to the principle of gas thermodynamics, when the pressure is increased in the autoclave, the process is a variable non-adiabatic process, and the temperature rise is calculated using the following formula:

[0107] ;

[0108] Where: is the temperature rise caused by voltage increase, is the initial temperature in Kelvin, For the final pressure, is the initial pressure, is the adiabatic index of nitrogen, which is equal to 1.4.

[0109] Then calculate the heat exchange between the autoclave and the environment:

[0110] ;

[0111] Where: is the heat exchange between the autoclave and the environment, is the heat transfer coefficient, determined experimentally, is the heat transfer area, which is equal to the surface area of ​​the autoclave. , are the temperature inside the autoclave and the ambient temperature outside the autoclave, respectively. is the time interval.

[0112] The theoretical input cooling capacity is equal to the heat generated by the temperature rise caused by the pressure increase minus the heat exchange between the autoclave and the environment. Calculating the ratio of the actual input cooling capacity to the theoretical input cooling capacity can help evaluate the efficiency of the nitrogen cooling system under different operating conditions. By analyzing the ratio, changes in the heat transfer coefficient of the autoclave, environmental impacts, and degradation of equipment performance can be identified, so that appropriate compensation measures can be taken.

[0113] It should be noted that if Figure 2 As shown, in step 32, the LSTM model includes an input layer, a first LSTM layer, a second LSTM layer, a fully connected layer 1, a fully connected layer 2, and an output layer, wherein the first LSTM layer and the second LSTM layer contain 64 units and 32 units, respectively, the first LSTM layer and the second LSTM layer both contain an input gate, a forget gate, an output gate, and candidate cell states, cell state updates, and hidden state updates, and the fully connected layer 1 and the fully connected layer 2 contain 50 neurons and 25 neurons, respectively.

[0114] The first layer of LSTM contains 64 units, which are used to extract the nonlinear relationship between the temperature and pressure parameters in the vulcanizer and the nitrogen input temperature. 64 units can provide sufficient capacity to capture the complex time series characteristics of the input data, especially those involving dynamic changes in temperature and pressure. 64 units are a balance point that can capture complex time series characteristics without causing excessive computational costs. The second layer of LSTM contains 32 units, which are used to extract the nonlinear relationship between the temperature and pressure parameters in the vulcanizer and the nitrogen input flow rate. The 32 units of the second layer of LSTM are used to further extract high-level features based on capturing the output features of the first layer of LSTM. The number is halved to reduce the dimension and avoid overfitting. Reducing the number of units helps to improve the efficiency of model training and reasoning while maintaining sufficient expressiveness. The fully connected layer 1 contains 50 neurons for further processing the output of the second layer LSTM. The 50 neurons provide sufficient nonlinear conversion capabilities to map the features extracted by the LSTM to a higher-dimensional feature space. The 50 neurons can effectively process the data output by the LSTM layer while controlling the complexity and computational cost of the model. The fully connected layer 2 contains 25 neurons for further processing the output of the fully connected layer 1. The dimensionality is further reduced on the basis of the fully connected layer 1. The 25 neurons help to extract higher-level features while avoiding excessive compression of information. The 25 neurons provide a simplified structure to help further filter and extract key information, providing optimized input for the final output layer.

[0115] The input layer receives input data of the current time step, including the temperature and pressure parameters in the vulcanizer, the input temperature and flow rate of nitrogen, and the attribute parameters of the vulcanizer (tank volume, tank material thermal conductivity, specific heat capacity of gas in the tank, ambient temperature, nitrogen density, nitrogen dynamic viscosity, nitrogen basic flow rate, nitrogen flow rate preset threshold range, nitrogen temperature preset threshold range).

[0116] It should be noted that in the first layer of LSTM, the activation function of the input gate is:

[0117] ;

[0118] Where: is the output of the first layer LSTM input gate, , are the nitrogen input temperature change growth rate and lag time reaction parameters, obtained through training data, is the weight matrix of the input gate, is the hidden state of the previous time step, is the input of the current time step, is the bias vector of the input gate;

[0119] The temperature change during the vulcanization process has complex nonlinear characteristics. The traditional activation function cannot fully capture these complex changes. This function is a double exponential function that can effectively capture the slow growth and hysteresis effects, which is very important for modeling the nonlinear characteristics of the temperature change process.

[0120] , Respectively represent the growth rate and lag time of nitrogen input temperature change. These parameters can be obtained through training data, so that the model can better adapt to the actual temperature change pattern. This customized activation function can more accurately reflect the dynamic characteristics of temperature change. The shape of the function allows smooth processing of temperature changes, thereby avoiding sudden changes in temperature during the change process. Smoothing of temperature changes helps to improve the stability and accuracy of model prediction. Using this function as the activation function of the input gate can enhance the LSTM model's ability to extract specific features in time series data, especially when the input data has an obvious growth pattern and lag effect. In the vulcanization process, the accuracy of temperature control is crucial. By using this activation function, the LSTM model can better adapt to complex working conditions and improve the accuracy and reliability of temperature control. Since the activation function takes into account the growth rate and lag time of temperature change, the model can dynamically adjust the output of the input gate to better reflect the actual temperature changes under the current working conditions. Traditional activation functions will cause overfitting problems on complex data, while the double exponential characteristics of this function can better adapt to the data distribution of temperature changes, thereby reducing the risk of overfitting.

[0121] The activation function of the forget gate is:

[0122] ;

[0123] Where: is the output of the first layer LSTM forget gate, is the weight matrix of the forget gate, is the bias vector of the forget gate;

[0124] This activation function can capture the complex nonlinear characteristics of the input data. The temperature and pressure changes in the vulcanization process are highly nonlinear, and traditional activation functions cannot fully capture these complex relationships. The combined nonlinear function can better describe these changes. The hyperbolic cosine function is symmetrical and grows rapidly when the input is large. It can capture the multi-peak characteristics and local extreme values ​​in the data, which is very important when dealing with temperature and pressure changes. The hyperbolic tangent function has a compression characteristic when dealing with the numerical range, so that the output value is limited between -1 and 1, with a smooth transition, which is suitable for processing the output of the forget gate and ensuring the stability of the state update. The combined function can provide a smooth nonlinear transformation, which helps to smooth the output of the forget gate during the state update process, avoid mutations, and ensure the stability of the model. By combining nonlinear functions, the model can more effectively selectively forget unimportant information while retaining important historical information. It is very critical for the LSTM model to process long-term series data. In the vulcanization process, the temperature changes are complex and nonlinear. Using this activation function can better adapt to these changes and improve the prediction accuracy of the model. It can smoothly process the input data, avoid drastic fluctuations when the model updates the state, and improve the stability and reliability of the model.

[0125] The activation function of the output gate is:

[0126] ;

[0127] Where: is the activation function of the first layer LSTM output gate, is the weight matrix of the output gate, is the bias vector of the output gate;

[0128] By applying an exponential function to the input and then taking the logarithmic value with base 2, this activation function can expand the dynamic range of the value, so that the output can cover a wider range of values, which is particularly important when dealing with complex temperature and pressure changes in the vulcanization process. The function provides smooth and continuous output, which helps to avoid sudden changes in the state update process and improve the stability of the model. When performing exponential operations, the values ​​will become very large. Using a logarithmic function can compress the range of values, avoid overflow problems, and improve the stability of numerical calculations. This function combines exponential and logarithmic operations to provide nonlinear mapping, which is suitable for processing complex time series data and capturing the nonlinear relationship between input and output. In the vulcanization process, the change in temperature is nonlinear. This activation function can better capture and reflect these complex changes and improve the prediction accuracy of the model. By using a smooth and continuous activation function, it can avoid drastic fluctuations in the state update process and ensure the stability of the model output. Exponential operations can cause numerical overflow problems, while logarithmic operations can compress the range of values ​​to avoid such problems and ensure the stability of numerical calculations.

[0129] The activation function of the candidate cell state is:

[0130] ;

[0131] Where: is the output of the candidate cell state, is the weight matrix of the candidate cell state, is the bias matrix of the candidate cell state.

[0132] This function is very effective in dealing with cumulative errors. It can smoothly accumulate changes in input signals and avoid mutations in time series. This smooth accumulation feature is very important for changes in temperature and pressure. It can effectively deal with cumulative errors in input data and capture complex nonlinear relationships in input data. Through this activation function, LSTM can better reflect complex patterns in temperature and pressure changes. Its output range is between [-1,1], which can provide good numerical stability and avoid numerical instability caused by excessive or small output values. It is crucial for precise control of the vulcanization process. Temperature changes have a cumulative effect, and this function can well describe this cumulative effect. By smoothing and integrating the input signal, the model can capture the long-term trend and cumulative effect of temperature changes.

[0133] In the first layer of LSTM, the cell state update formula is: ,in, Update the value of the cell state at the current moment, Update the value of the cell state at the previous moment;

[0134] The hidden state update formula is: , Updates the value for the hidden state.

[0135] It should be noted that in the second layer of LSTM, the activation function of the input gate is:

[0136] ;

[0137] Where: is the output of the second layer LSTM input gate, , are the nitrogen flow resistance and flow rate relationship adjustment coefficients, obtained through training data, is the weight matrix of the input gate, is the hidden state of the previous time step, For the first layer LSTM at time step The hidden state output of is the bias vector of the input gate;

[0138] By accurately describing the relationship between resistance and flow rate during nitrogen flow, the model can optimize nitrogen input flow and temperature, and improve the temperature and pressure control accuracy during vulcanization. Complex nonlinear activation functions can better handle nonlinear relationships in actual operations and improve the prediction accuracy of the model. Combining the hidden state of the previous time step and the input of the current time step, the model can better capture the dynamic changes in time series data and improve the prediction ability. The relationship between flow resistance and flow rate can be obtained through experimental data and theoretical formulas. Parameters , It is obtained by fitting the training data and can accurately reflect this relationship. Introducing the square root function in the denominator can smooth the input data, avoid values ​​that are too large or too small, and provide stable output.

[0139] The activation function of the forget gate is:

[0140] ;

[0141] Where: is the output of the second layer LSTM forget gate, is the friction coefficient of cold nitrogen, determined experimentally, is the diameter of the cold nitrogen input pipe, is the acceleration due to gravity, is the weight matrix of the forget gate, is the bias vector of the forget gate;

[0142] The friction coefficient determined experimentally reflects the friction characteristics of nitrogen when it flows in the pipeline. This parameter has a direct impact on the resistance to flow. The input pipe diameter of cold nitrogen is an important physical parameter that directly affects the flow rate and flow resistance. In fluid mechanics, the acceleration of gravity is a basic constant that affects the flow behavior of the fluid. By introducing these physical parameters into the activation function, the physical characteristics of nitrogen flow can be more accurately reflected, enabling the model to better capture the complex relationships in actual operations. The square root function is used to smooth the input data to avoid output values ​​that are too large or too small, thereby providing a stable output, which is particularly important when dealing with the relationship between resistance and flow rate when nitrogen flows. Each parameter in the activation function has a clear physical meaning, which enhances the interpretability and transparency of the model, making the model not only a black box, but also able to explain the influence of each parameter in the flow process. By combining the hidden state of the previous time step and the input of the current time step, the model can better capture the dynamic changes in the time series data and improve the predictive ability of nitrogen flow control.

[0143] The activation function of the output gate is:

[0144] ; Where: is the activation function of the second layer LSTM output gate, is the weight matrix of the output gate, is the bias vector of the output gate, is the density of nitrogen, which is a known parameter;

[0145] The density of nitrogen is a known physical parameter that reflects the quality characteristics of nitrogen at a specific temperature and pressure. By introducing nitrogen density, the actual physical characteristics of nitrogen flow can be more accurately reflected. The square root function is used to smooth the input data to avoid output values ​​that are too large or too small, thereby providing a stable output, which is particularly important when dealing with complex fluid mechanics problems. By introducing the ratio of square root and nitrogen density in the activation function, it can ensure that the output value fluctuates within a reasonable range, thereby avoiding numerical instability. Each parameter in the activation function has a clear physical meaning, which enhances the interpretability and transparency of the model, making the model not only a black box, but also able to explain the influence of each parameter in the flow process. It is a common form in fluid mechanics, used to describe the energy characteristics of fluids, and is convenient for capturing energy changes in nitrogen flow, so as to more accurately predict and control nitrogen flow rate and temperature. The activation function combines the hidden state of the previous time step with the hidden state of the first layer LSTM in the current time step, which can comprehensively consider the comprehensive influence of historical information and current input in time series data, and improve the prediction ability of the model. By accurately describing the energy characteristics and flow rate relationship of nitrogen flow, the model can optimize the nitrogen input flow rate and temperature, and improve the temperature and pressure control accuracy during the vulcanization process. The complex nonlinear activation function can better handle the nonlinear relationship in actual operation and improve the prediction accuracy of the model. Combining the hidden state of the previous time step and the input of the current time step, the model can better capture the dynamic changes in time series data and improve the prediction ability.

[0146] The activation function of the candidate cell state is:

[0147] ;

[0148] Where: is the output of the candidate cell state, is the weight matrix of the candidate cell state, is the bias matrix of the candidate cell state, is the basic flow rate of nitrogen, is the average of the maximum and minimum values ​​of the flow rate threshold range for nitrogen, is the dynamic viscosity of nitrogen, which is a known parameter.

[0149] Through this activation function, the physical characteristics of nitrogen flow can be reflected, the flow characteristics can be smoothed, stable output and control range can be provided, flow characteristics and energy characteristics can be captured, historical information and current input can be comprehensively considered, and the prediction accuracy and stability of temperature and pressure control in the vulcanization process can be improved. It can better adapt to the complex conditions in the vulcanization process and ensure the stability and efficiency of the process.

[0150] In the second layer of LSTM, the cell state update formula is: ,in, Update the value of the cell state at the current moment, Update the value of the cell state at the previous moment;

[0151] The hidden state update formula is: , Updates the value for the hidden state.

[0152] It should be noted that in step 32, in the loss function of the LSTM model, the temperature uniformity penalty term in the time dimension is:

[0153] ;

[0154] Where: is the penalty term for temperature uniformity in the time dimension, is the index of the data point in the time window, is the total number of data points in the time window, In the time window The temperature of the data point is the average of all temperature measurement points in the autoclave. is the mean temperature of all data points in the time window;

[0155] The temperature stability penalty term in the time dimension is:

[0156] ;

[0157] Where: is the temperature stability penalty term in the time dimension, is the temperature value of the data point before the current data point in the time window.

[0158] The penalty for temperature uniformity in the time dimension is to ensure the uniform distribution of the temperature in the vulcanizer in the time dimension. By minimizing the difference between the temperature of each data point and the average temperature in the time window, the temperature fluctuation can be reduced and the temperature uniformity can be ensured. During the vulcanization process, temperature uniformity is crucial for the uniform vulcanization of rubber. Uneven temperature will lead to inconsistent product quality and affect the performance and durability of the rubber track. By penalizing temperature fluctuations, local overheating or overcooling can be avoided. Local overheating or overcooling will lead to different vulcanization degrees in different parts of the rubber track, affecting the overall performance. The penalty for temperature stability in the time dimension is to ensure the stability of the temperature in the vulcanizer in the time dimension. By minimizing the temperature change between adjacent time points, the temperature fluctuation can be reduced and the temperature stability can be ensured. The drastic temperature fluctuation will lead to stress concentration and thermal damage to the rubber material during the vulcanization process, thereby affecting the product quality. By penalizing the temperature change between adjacent time points, a smooth temperature transition can be achieved. Smooth temperature changes help reduce temperature stress, ensure that the rubber material is heated evenly during the vulcanization process, and avoid stress concentration and material deformation. By ensuring the uniformity and stability of temperature, the quality of rubber tracks during the vulcanization process can be improved and the performance and durability of the product can be ensured; by reducing temperature fluctuations and achieving smooth temperature changes, the energy consumption and time during the vulcanization process can be optimized and production efficiency can be improved; the control of temperature uniformity and stability helps to ensure the stability of the vulcanization process and reduce the uncertainty caused by temperature fluctuations, thereby improving the controllability and reliability of the production process.

[0159] It should be noted that in step 32, the loss function in the LSTM model is:

[0160] ;

[0161] Where: is the loss function value, Index of the temperature monitoring point in the autoclave. In the time window Data point The actual temperature monitoring value at each location, In the time window Data point The predicted temperature value at each location, is the total number of temperature monitoring points, , are the weight coefficients of the temperature uniformity penalty term and the temperature stability penalty term in the time dimension, which are set empirically.

[0162] It should be noted that in the fuzzy rules used in step 34, the input variables of fuzzy control are temperature error, temperature change rate and temperature distribution uniformity, among which the elements in the fuzzy set of temperature error are negative large (NB, temperature error <-5°C), negative small (NS1, -1> temperature error ≥-5°C), zero (ZE1, 1°C> temperature error ≥-1°C), positive small (PS1, 5°C> temperature error ≥1°C), and positive large (PB1 temperature error >5°C), and the elements in the fuzzy set of temperature change rate are fast decreasing (ND, temperature change rate <-2°C / min), slow decreasing (NS, -0.5°C / min 〉temperature change rate ≥ -2℃ / min), unchanged (ZE2, 0.5℃ / min≥temperature change rate ≥ -0.5℃ / min), slow rise (PS2, 2℃ / min≥temperature change rate>0.5℃ / min), fast rise (PD, temperature change rate>2℃ / min), the elements in the fuzzy set of temperature distribution uniformity (the degree of temperature difference between the temperatures of each measuring point in the reaction vulcanization tank, which is the standard deviation of the temperature measuring point) are: uneven (NU, temperature distribution uniformity>2℃), relatively uniform (QY, 2℃≥temperature distribution uniformity>1℃), uniform (Y, temperature distribution uniformity≤1℃), the fuzzy control output variables are nitrogen flow adjustment and nitrogen temperature adjustment, and the fuzzy set elements of nitrogen flow adjustment include low flow (L1, nitrogen flow adjustment <-5L / min), medium flow (M1, -5 L / min≤nitrogen flow adjustment≤5 L / min), high flow rate (H1, nitrogen flow rate adjustment>5L / min), the fuzzy set elements of nitrogen temperature adjustment include low temperature (L2, nitrogen temperature adjustment<-5℃), medium temperature (M2, -5℃≤nitrogen temperature adjustment≤5℃), high temperature (H2, nitrogen temperature adjustment>5℃), evaluate the input data based on the fuzzy rules set by expert knowledge and experience, obtain the activation strength of each rule, use fuzzy logic to infer the activated rules, generate fuzzy output, calculate the membership of each output variable in the fuzzy set according to the activation strength, and use the centroid method to convert the fuzzy output into precise control output.

[0163] The following are some examples of fuzzy rules, which can be adjusted and expanded according to specific needs and expert experience in actual applications:

[0164] Fuzzy rule 1: If the temperature error is negative and large (NB), the temperature change rate is decreasing rapidly (ND), and the temperature distribution uniformity is uneven (NU), then the nitrogen flow rate is adjusted to a high flow rate (H1) and the nitrogen temperature is adjusted to a low temperature (L2);

[0165] Fuzzy rule 2: If the temperature error is small negative (NS1), the temperature change rate is slow to decrease (NS), and the temperature distribution uniformity is relatively uniform (QY), then the nitrogen flow rate is adjusted to medium flow (M1) and the nitrogen temperature is adjusted to medium temperature (M2);

[0166] Fuzzy rule 3: if the temperature error is zero (ZE1), the temperature change rate is constant (ZE2), and the temperature distribution uniformity is uniform (Y), then the nitrogen flow rate is adjusted to medium flow (M1) and the nitrogen temperature is adjusted to medium temperature (M2);

[0167] Fuzzy rule 4: if the temperature error is positive and small (PS1), the temperature change rate is slow to rise (PS2), and the temperature distribution uniformity is relatively uniform (QY), then the nitrogen flow rate is adjusted to a low flow rate (L1) and the nitrogen temperature is adjusted to a high temperature (H2);

[0168] Fuzzy rule 5: If the temperature error is positive and large (PB1), the temperature change rate is rising fast (PD), and the temperature distribution uniformity is uneven (NU), then the nitrogen flow rate is adjusted to a low flow rate (L1) and the nitrogen temperature is adjusted to a high temperature (H2).

[0169] The fuzzy control strategy is specifically described in combination with fuzzy rule 1:

[0170] When the current nitrogen temperature error is -6℃, it belongs to the negative large (NB) in the temperature error fuzzy set, with a membership of 1. When the temperature change rate is -3℃ / min, it belongs to the fast decline (ND), with a membership of 1. When the temperature distribution uniformity is 2.5℃, it belongs to the uneven (NU), with a membership of 1. The activation strength of fuzzy rule 1 is based on the "AND" operation rule in fuzzy logic. The activation strength of fuzzy rule 1 is the minimum value of the membership of the three input conditions. The membership of the activated rule is transferred to the output variable, and the output is that the nitrogen flow rate is adjusted to a high flow rate ( H1), the membership is 1, the nitrogen temperature is adjusted to low temperature (L2), the membership is 1, the centroid method is used to convert the fuzzy output into precise output, the nitrogen flow adjustment range is [-10, 10] L / min, the temperature adjustment range is [-10, 10], and their respective specific distributions are {-10, -5, 0, 5, 10}, {-10, -5, 0, 5, 10}, for high flow (H1) and low temperature (L2) The membership is 1, and the values ​​represented are 10 and -10 respectively, then the precise value of nitrogen flow adjustment is , the precise value of nitrogen temperature adjustment is: .

[0171] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

[0172] Finally: The above description is only the technical solution of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A vulcanization process for a rubber track, characterized in that: The process includes: Step 1, batching, mixing and open mixing: adding rubber base material, vulcanizing agent, accelerator, reinforcing agent and softener in an internal mixer according to the formula ratio, controlling the mixing temperature at 70-90° C. for 5-10 minutes, transferring the mixed rubber to an open mixer, and adjusting the thickness and uniformity of the rubber; Step 2, calendering and molding: the mixture obtained in step 1 is pressed into a sample with a thickness of 2 to 5 mm by a calender, and cut into several sections according to the design size of the crawler; Step 3, first vulcanization: Place the sample formed in step 2 evenly in the customized heating mold in the vulcanization tank, gradually and evenly increase the temperature in the tank to the set temperature, use the temperature sensor to monitor the temperature in the tank in real time, start the cold nitrogen precooling system to cool the nitrogen to the initial set temperature, gradually introduce cold nitrogen to gradually increase the pressure in the tank to the set pressure, monitor the temperature in the tank in real time, dynamically adjust the nitrogen flow and temperature to keep the set temperature in the tank, maintain the set temperature and pressure in the tank after the pressurization is completed, vulcanize for 45 minutes, maintain the set temperature and pressure for 30 minutes after the end, then gradually reduce the temperature in the tank at a rate of 3 to 5°C / minute, and gradually reduce the pressure in the tank at a rate of 0.5 to 1 MPa / minute. After the temperature and pressure are within the safe range, open the tank door to take out the rubber track, conduct quality inspection, and take out qualified products; Step 4, joint processing: cut the rubber track obtained in step 3 according to the needs, fill the unvulcanized rubber mixture at the joint, apply chloroprene adhesive and pre-vulcanize; Step 5, second vulcanization: evenly place the track treated in step 4 in a vulcanization tank, perform secondary vulcanization with the same cold nitrogen pressurization process in step 3 using the dynamic temperature adjustment strategy in the tank, then take out the vulcanized rubber track product, conduct quality inspection, and take out qualified products; The dynamic temperature adjustment strategy in the tank during the cold nitrogen pressurization process in step 3 and step 5 is: Step 31, data collection and preprocessing: real-time collection of temperature and pressure data of each monitoring point within the observation time window of each pressurization stage in the vulcanizer, and data standardization and serialization; Step 32, LSTM model training: input historical data, train the LSTM model, and add the temperature uniformity and stability penalty items in the autoclave in the time dimension; Step 33, model prediction: input the temperature and pressure data in the autoclave at each pressurization stage in the current time window, and use the trained LSTM model to predict the input flow and input temperature combination of cold nitrogen within the preset threshold range, as well as the ratio of actual input cooling capacity to theoretical input cooling capacity; Step 34, fuzzy control feedback adjustment: real-time monitoring of the temperature and pressure data in the vulcanization tank, according to the fuzzy control rules, traversing the cold nitrogen input flow rate and input temperature combination input in step 33 to dynamically adjust the nitrogen input flow rate and input temperature, to ensure that the spatial uniformity and stability of the temperature field in the tank meet the preset requirements; In step 32, the LSTM model includes an input layer, a first LSTM layer, a second LSTM layer, a fully connected layer 1, a fully connected layer 2, and an output layer, wherein the first LSTM layer and the second LSTM layer include 64 units and 32 units, respectively, the first LSTM layer and the second LSTM layer both include an input gate, a forget gate, an output gate, and candidate cell states, cell state updates, and hidden state updates, and the fully connected layer 1 and the fully connected layer 2 include 50 neurons and 25 neurons, respectively; In the first layer of LSTM, the activation function of the input gate is: Where: is the output of the first layer LSTM input gate, α1 and β1 are the nitrogen input temperature change growth rate and lag time reaction parameters respectively, which are obtained through training data. is the weight matrix of the input gate, is the hidden state of the previous time step, x t is the input of the current time step, is the bias vector of the input gate; The activation function of the forget gate is: Where: is the output of the first layer LSTM forget gate, is the weight matrix of the forget gate, is the bias vector of the forget gate; The activation function of the output gate is: Where: is the activation function of the first layer LSTM output gate, is the weight matrix of the output gate, is the bias vector of the output gate; The activation function of the candidate cell state is: Where: is the output of the candidate cell state, is the weight matrix of the candidate cell state, is the bias matrix of the candidate cell state; In the second layer of LSTM, the activation function of the input gate is: Where: is the output of the second-layer LSTM input gate, α2 and β2 are the adjustment coefficients of nitrogen flow resistance and flow rate, respectively, obtained through training data. is the weight matrix of the input gate, is the hidden state of the previous time step, is the hidden state output of the first layer LSTM at time step t, is the bias vector of the input gate; The activation function of the forget gate is: Where: is the output of the second-layer LSTM forget gate, γ is the friction coefficient of cold nitrogen, determined by experiments, D is the input diameter of cold nitrogen, g is the acceleration of gravity, is the weight matrix of the forget gate, is the bias vector of the forget gate; The activation function of the output gate is: Where: is the activation function of the second layer LSTM output gate, is the weight matrix of the output gate, is the bias vector of the output gate, ρ is the density of nitrogen, which is a known parameter; The activation function of the candidate cell state is: Where: is the output of the candidate cell state, is the weight matrix of the candidate cell state, is the bias matrix of the candidate cell state, v is the basic flow rate of nitrogen, is the mean of the maximum and minimum values ​​of the nitrogen flow rate threshold range, and μ is the dynamic viscosity of nitrogen, which is a known parameter; In step 32, in the loss function of the LSTM model, the penalty term for temperature uniformity in the time dimension is: Where: P uniform,time is the temperature uniformity penalty term in the time dimension, l is the index of the data point in the time window, n is the total number of data points in the time window, T l is the temperature of the lth data point in the time window, is the mean of all temperature measurement points in the autoclave, is the mean temperature of all data points in the time window; The temperature stability penalty term in the time dimension is: Where: P stability,time is the temperature stability penalty term in the time dimension, T l-1 is the temperature value of the data point before the current data point in the time window.

2. A vulcanization process for a rubber crawler according to claim 1, characterized in that: In step 1, the specific operation process is: add 100 parts of styrene-butadiene rubber into an internal mixer and start mixing, add 30-60 parts of carbon black, 5-10 parts of paraffin oil, 1-3 parts of rosin, and 1-2 parts of antioxidant 4020 successively, mix for 5-10 minutes, add 3-5 parts of zinc oxide and 1-2 parts of stearate successively, control the mixing temperature at 70-90° C., then add 2-3 parts of sulfur, 0.5-1.5 parts of N-cyclohexyl-2-benzothiazole sulfonamide, and 0.2-0.8 parts of tetramethylthiuram disulfide to the mixture, continue mixing until all ingredients are fully mixed, transfer the mixture to an open mill for further mixing and adjust the thickness and uniformity of the mixture.

3. The vulcanization process of a rubber crawler according to claim 1, characterized in that: The specific implementation process of step 2 includes: Step 21, initial setting: adjust the roll gap of the calender to 2-5 mm, start the intelligent vision system, and calibrate the camera and laser rangefinder; Step 22, feeding and calendering: gradually feeding the mixed material into the feed port of the calender, starting calendering, controlling the roller temperature of the calender between 60 and 80° C., and adjusting the roller speed of the calender to be able to uniformly press the rubber material into a sample sheet of the required thickness; Step 23, intelligent monitoring and feedback: the intelligent visual system takes real-time photos of the rubber and sample during the calendering process, analyzes the sample thickness through the image processing algorithm, and measures the thickness in real time with a laser rangefinder. The measured data is fed back to the control system of the calender to automatically adjust the roller gap and roller speed to make the sample thickness uniform; Step 24, sample conveying and cooling: the calendered sample is conveyed to the cutting area by a conveyor belt. During the process, the sample is rapidly cooled at 1.5-2°C / min in the initial cooling stage for 5-10 minutes, gradually cooled at 1-1.5°C / min in the intermediate cooling stage for 10-20 minutes, and slowly cooled at 0.5-1°C / min in the final cooling stage until the sample temperature reaches the ambient temperature; Step 25, intelligent cutting control: Use a laser rangefinder and image processing technology to accurately and automatically locate the cutting position of the sample, dynamically adjust the parameters of the cutting tool based on the real-time feedback of the visual system, adjust the parameters of the cutting tool based on the design size of the crawler, position the sample in the cutting area, and use the cutting tool for cutting. The visual system monitors the cracks, bubbles and thickness of the sample in real time. When defects are found, it automatically issues an alarm and marks the defective sample for removal or rework.

4. The vulcanization process of a rubber crawler according to claim 1, characterized in that: In step 32, in the loss function in the LSTM model, the loss function is: Where: is the loss function value, k is the index of the temperature monitoring point in the autoclave, T l,k is the actual temperature monitoring value of the kth position of the lth data point in the time window, is the temperature prediction value of the kth position of the lth data point in the time window, m is the total number of temperature monitoring points, λ1 and λ2 are the weight coefficients of the temperature uniformity penalty term and the temperature stability penalty term in the time dimension, respectively, which are set empirically.

5. The vulcanization process of a rubber crawler according to claim 1, characterized in that: In the fuzzy rules used in step 34, the input variables of fuzzy control are temperature error, temperature change rate and temperature distribution uniformity, and the output variables of fuzzy control are nitrogen flow adjustment and nitrogen temperature adjustment. The fuzzy rules set based on expert knowledge and experience evaluate the input data, obtain the activation strength of each rule, use fuzzy logic to infer the activated rules, generate fuzzy output, and convert the fuzzy output into precise control output.