Polyurethane production reaction kettle abnormal condition automatic identification method and system

Abstract

The application discloses a polyurethane production reaction kettle abnormal condition automatic identification method and system, and belongs to the technical field of industrial process monitoring and fault diagnosis. The method comprises the following steps: collecting temperature sensor data, motor operation parameters and process setting parameters of the reaction kettle in real time; inputting the motor parameters and the process parameters into a pre-constructed thermal-mechanism model and a motor-viscosity regression model respectively, to obtain a first predicted temperature and a first predicted viscosity; updating the thermal-mechanism model based on the first predicted viscosity to obtain a second predicted temperature, and realizing model output verification by comparing the two predicted temperatures; after the verification, calculating the residual error between the first predicted temperature and the real-time temperature, and the first and second derivatives of the residual error with respect to time, and performing abnormality discrimination according to a multi-threshold rule. The application realizes early, accurate and reliable identification of abnormal conditions, and improves production safety and product qualification rate.

Description

Technical Field

[0001] This invention relates to the field of industrial process monitoring and fault diagnosis technology, specifically to an automatic identification method and system for abnormal operating conditions of polyurethane production reactors. Background Technology

[0002] Polyurethane production is a complex chemical process, and the operating status of its core reaction equipment, the reactor, directly determines the quality of the final product, production efficiency, and factory safety. Abnormal operating conditions within the reactor, such as thermal runaway, localized overheating, uneven stirring, and abnormal material viscosity, if not identified and addressed in a timely manner, can lead to product scrapping, production line shutdowns, and serious safety accidents.

[0003] However, the single-sensor threshold alarm method commonly used in existing technologies suffers from problems such as delayed warnings and high false alarm rates, making it impossible to achieve early and accurate intervention. On the other hand, although pure data-driven prediction models have been explored to some extent, they rely heavily on historical data, lack consideration of the reaction mechanism, and have poor generalization ability and interpretability when facing new operating conditions, making it difficult to meet the complex and ever-changing actual production needs. Summary of the Invention

[0004] This invention addresses the technical problems in existing technologies, such as delayed early warning and high false alarm rate due to reliance on single sensor threshold alarms, and weak generalization ability and poor interpretability of pure data-driven models due to lack of mechanistic constraints. It provides an automatic identification method and system for abnormal operating conditions of polyurethane production reactors.

[0005] The technical solution of the present invention to solve the above-mentioned technical problems is as follows:

[0006] In a first aspect, the present invention provides an automatic identification method for abnormal operating conditions of polyurethane production reactors, including:

[0007] Real-time acquisition of temperature sensor data, motor operating parameters, and process setting parameters of the target reactor;

[0008] The motor operating parameters and the process setting parameters are respectively input into the pre-constructed thermo-mechanical model and motor-viscosity regression model to predict the operating parameters and obtain the first predicted temperature and the first predicted viscosity.

[0009] The first predicted viscosity is verified and analyzed against the thermo-mechanical model to obtain the second predicted temperature, and the first predicted temperature is compared with the second predicted temperature to verify the predicted output.

[0010] If the predicted output verification passes, then anomaly identification is performed based on the first predicted temperature and the temperature sensing data.

[0011] Secondly, the present invention provides an automatic identification system for abnormal operating conditions of polyurethane production reactors, comprising:

[0012] The data acquisition module is used to collect temperature sensor data, motor operating parameters, and process setting parameters of the target reactor in real time.

[0013] The model prediction module is used to input the motor operating parameters and the process setting parameters into the pre-built thermo-mechanical model and motor-viscosity regression model respectively to predict the operating parameters, so as to obtain the first predicted temperature and the first predicted viscosity.

[0014] The verification analysis module is used to perform verification analysis based on the first predicted viscosity and the thermo-mechanical model, obtain the second predicted temperature, and compare the first predicted temperature with the second predicted temperature to perform prediction output verification.

[0015] An anomaly identification module is used to identify anomalies in operating conditions based on the first predicted temperature and the temperature sensing data when the prediction output verification passes.

[0016] The beneficial effects of this invention are:

[0017] Compared to existing technologies, this invention firstly integrates a thermo-mechanical model with a motor-viscosity regression model to achieve coordinated prediction of multiple key parameters, including temperature and viscosity, within the reactor, enhancing the comprehensive perception and accurate judgment of operating conditions. Secondly, it introduces an inter-model verification mechanism, feeding back the first predicted viscosity into the thermo-mechanical model to calculate the second predicted temperature. By comparing the two temperature predictions, the consistency of the model output results is self-verified, significantly improving the model's reliability and anti-interference performance. Thirdly, it combines the physical interpretability of the mechanistic model with the adaptability of the data-driven model to actual operating conditions, reducing reliance on large amounts of historical data and addressing the problem of insufficient prediction accuracy of traditional mechanistic models under complex and variable production conditions. Finally, it relies on residual analysis between the verified predicted temperature and real-time temperature sensing data to identify abnormal operating conditions, enabling the sensitive identification of early abnormal signals and realizing a shift from delayed alarms to early warnings, effectively improving the safety level of the production process and the product qualification rate. Attached Figure Description

[0018] Figure 1 A flowchart illustrating the automatic identification method for abnormal operating conditions of polyurethane production reactors provided by the present invention.

[0019] Figure 2 This is a schematic diagram of the automatic identification system for abnormal operating conditions of polyurethane production reactors provided by the present invention.

[0020] In the attached diagram, the components represented by each number are as follows:

[0021] Data acquisition module 11, model prediction module 12, verification and analysis module 13, anomaly identification module 14. Detailed Implementation

[0022] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0023] In the description of this invention, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of the stated features. In the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0024] In the description of this invention, the term "for example" is used to mean "used as an example, illustration, or description." Any embodiment described as "for example" in this invention is not necessarily to be construed as being more preferred or advantageous than other embodiments. The following description is provided to enable any person skilled in the art to make and use the invention. Details are set forth in the following description for purposes of explanation. It should be understood that those skilled in the art will recognize that the invention can be made without using these specific details. In other instances, well-known structures and processes will not be described in detail to avoid obscuring the description of the invention with unnecessary detail. Therefore, the invention is not intended to be limited to the embodiments shown, but is consistent with the broadest scope of the principles and features disclosed herein.

[0025] Example 1, as Figure 1 As shown, this embodiment of the invention provides an automatic identification method for abnormal operating conditions of polyurethane production reactors, including:

[0026] S10: Real-time acquisition of temperature sensor data, motor operating parameters, and process setting parameters of the target reactor;

[0027] Specifically, real-time acquisition of temperature sensor data, motor operating parameters, and process setting parameters of the target reactor, including:

[0028] Acquire the time-series temperature data of the main material in the target reactor and output it as the temperature sensing data;

[0029] The current-time curve, voltage-time curve, and speed-time curve of the stirring motor of the target reactor are collected synchronously, and the output is the motor operating parameters;

[0030] Based on the production management terminal, the real-time production parameters of the target reactor are obtained and output as the process setting parameters. The process setting parameters include at least the raw material formula parameters, reactor setting parameters, standard viscosity parameters, and intrinsic equipment parameters.

[0031] The target reactor refers to the reactor currently being inspected that is producing polyurethane. First, the time-series temperature data of the main materials in the target reactor is acquired. The main materials refer to the core raw material mixture undergoing the chemical reaction in the polyurethane reactor, including polyols, isocyanates, and other additives. The time-series temperature data is a collection of continuously recorded temperature measurements in chronological order. Since the temperature of the main materials directly reflects the intensity and state of the chemical reaction, any abnormal operating conditions, such as thermal runaway, uneven mixing, or reaction stagnation, will be directly reflected in temperature changes, serving as the core basis for operating condition identification. Therefore, it is necessary to install temperature sensors inside the reactor to continuously measure and record the temperature changes of the main materials inside, using this as temperature sensing data.

[0032] Secondly, the current-time curve, voltage-time curve, and speed-time curve of the stirring motor in the target reactor are simultaneously collected, and the output is the motor operating parameters. Specifically, the stirring motor is the motor that drives the stirring blades inside the reactor to ensure uniform mixing and good heat transfer of the materials; the current directly reflects the motor's load, and an abnormally high current often indicates excessive material viscosity or mechanical failures such as shaft seizure; the voltage reflects the stability of the motor's power supply; and the speed reflects the actual operating speed of the motor. The stirring motor is a core component of the reactor, and its operating status indirectly reflects the physical characteristics of the materials inside the reactor. The higher the material viscosity, the greater the stirring resistance, and the heavier the motor load, which is manifested in changes in parameters such as current and power. By collecting the current-time curve, voltage-time curve, and speed-time curve, the motor's operating status can be analyzed, and the motor operating parameters can be output.

[0033] Furthermore, based on the production management terminal, real-time production parameters of the target reactor are acquired and output as process setting parameters. The production management terminal refers to the computer system controlling the production process, which stores the production formula and process requirements for each batch of products. Process setting parameters refer to the set of preset parameters related to the current reactor acquired from the production management terminal, including at least: raw material formula parameters, referring to the proportions, quantities, and order of various components such as polyols, isocyanates, catalysts, and additives, which are the basis for reaction heat calculations and mechanistic models; reactor setting parameters, such as target temperature, stirring speed settings, pressure control values, and reaction time; standard viscosity parameters, referring to the theoretical viscosity reference value of the reactant system over time under the current formula and ideal process conditions; and intrinsic equipment parameters, referring to the inherent properties of the reactor itself, such as reactor volume, heat transfer area, impeller type, motor power and efficiency, etc., which are constants necessary for calculating heat transfer and stirring power in the mechanistic model. By collecting relevant preset parameters, initial conditions can be provided for the theoretical basis and model calculations of the current production batch.

[0034] S20: Input the motor operating parameters and the process setting parameters into the pre-built thermo-mechanical model and motor-viscosity regression model respectively to predict the operating parameters and obtain the first predicted temperature and the first predicted viscosity;

[0035] Specifically, the motor operating parameters and the process setting parameters are respectively input into a pre-constructed thermo-mechanical model and a motor-viscosity regression model to predict operating parameters and obtain a first predicted temperature and a first predicted viscosity, including:

[0036] Input the motor operating parameters into the motor-viscosity regression model to obtain the first predicted viscosity;

[0037] The thermodynamic-mechanism model is initialized based on the standard viscosity parameters and the intrinsic parameters of the equipment. The raw material formulation parameters and the reactor setting parameters are used as inputs to activate the initialized thermodynamic-mechanism model for predictive analysis and obtain the first predicted temperature.

[0038] The steps for constructing the thermodynamic-mechanism model include:

[0039] Establish a digital model of the physical structure of the target reactor;

[0040] Based on the prior property model and the physical structure digital model, an initial mechanism model is constructed, wherein the prior property model includes a polyurethane reaction kinetics model and a reactor heat transfer characteristic model.

[0041] Obtain standard reaction data of the target reactor under standard operating conditions, and use the standard reaction data to perform parameter correction and verification of the initial mechanism model to obtain the thermodynamic-mechanism model.

[0042] First, a digital physical model of the target reactor is established. This digital physical model is a virtual model created in a computer to accurately describe the geometry, physical properties, and spatial relationships of the reactor and its auxiliary equipment. It includes the reactor's geometric dimensions, such as internal diameter, height, and volume; heat transfer area, such as the jacket's heat exchange area and the surface area of ​​the coils; the stirring system, such as the type, diameter, and installation location of the stirring paddles; and material properties, such as the thermal conductivity and specific heat capacity of the reactor material; and the inner wall thickness. This digital physical model serves as the physical container and computational foundation for all subsequent mechanistic simulations.

[0043] Secondly, based on the prior property model and the digital model of the physical structure, an initial mechanism model is constructed. The prior property model includes a polyurethane reaction kinetic model and a reactor heat transfer characteristic model. The prior property model refers to a mathematical model established based on known and universal natural science laws, such as physical and chemical laws, used to describe the inherent properties of matter and the intrinsic laws of processes. Specifically, the polyurethane reaction kinetic model is a mathematical model describing the chemical reaction rate, used to define how the concentration and temperature of reactants, i.e., polyols and isocyanates, affect the reaction rate and heat release. The reactor heat transfer characteristic model is a mathematical model based on thermodynamic laws, used to describe how heat is transferred and dissipated. Combining the polyurethane reaction kinetic model and the reactor heat transfer characteristic model with the digital model of the physical structure means embedding the laws of chemical reaction and heat transfer into the digital twin from the previous step, enabling it to dynamically simulate the reaction process.

[0044] Finally, standard reaction data of the target reactor under standard operating conditions is obtained, and the initial mechanistic model is calibrated and verified using this data to obtain a thermodynamic-mechanistic model. Standard reaction data under standard operating conditions refers to high-quality historical data collected under strictly controlled and ideal production conditions, including: material temperature change curves over time, jacket temperature, motor power, etc. The standard reaction data is then input into the initial mechanistic model to predict the temperature curve, which is then compared with the actual temperature curve. Since the initial model is theoretical, there will inevitably be discrepancies between the two. A parameter estimation algorithm automatically adjusts some uncertain parameters in the model, such as the pre-exponential factor and activation energy in the reaction kinetic equation, or the overall heat transfer coefficient in the heat transfer model, so that the model's predicted curve approximates the actual temperature curve infinitely. The model obtained through these steps, after calibration and verification, capable of accurately simulating the actual behavior of the target reactor, is the thermodynamic-mechanistic model.

[0045] Meanwhile, the construction steps of the motor-viscosity regression model include:

[0046] Acquire multimodal motor sample data and corresponding sample material viscosity data, wherein the multimodal motor sample data includes at least the associated stored sample current, sample voltage and sample speed;

[0047] Based on the process setting parameters, the multimodal motor sample data and the sample material viscosity data are segmented and the learning rate weights are configured differently.

[0048] Using the sample current, sample voltage, and sample rotation speed as inputs, and the sample material viscosity data as supervision, the motor-viscosity regression model based on regression analysis is constructed and trained.

[0049] First, acquire multimodal motor sample data and corresponding sample material viscosity data. The multimodal motor sample data includes at least the associated stored sample current, sample voltage, and sample speed. Multimodal refers to measurements from multiple dimensions of the same device, collectively forming a complete picture of the motor's operating status. The sample material viscosity data refers to the measured viscosity values ​​of the material inside the reactor, acquired through direct measurement at the same time point as the motor sample data.

[0050] Secondly, the multimodal motor sample data and sample material viscosity data are segmented based on process setting parameters, and learning rate weights are configured differently. Specifically, the entire dataset is divided into several subsets according to the process setting parameters, including the current process sample segment and other sample segments. Based on the process setting parameters, the multimodal motor sample data and sample material viscosity data are traversed. When the process setting parameters of a historical sample completely match or are very close to the current process setting parameters, the sample is assigned to the current process sample segment. Samples that do not meet the above conditions are assigned to other sample segments. Furthermore, a high weight, such as 1.0, is assigned to the current process sample segment. During training, the model will focus on fitting this part of the data to ensure prediction accuracy under the current operating conditions. Low weights, such as 0.5, are assigned to other sample segments. This data is used to constrain the model and prevent it from going to extremes, but will not have an excessive impact on the model's dominant laws.

[0051] Furthermore, using sample current, sample voltage, and sample rotational speed as inputs, and sample material viscosity data as supervision, a motor-viscosity regression model based on regression analysis is constructed and trained. Specifically, the motor-viscosity regression model is constructed and trained based on historical data to predict the viscosity of materials in the reactor in real time according to motor operating parameters. This includes the following steps: First, the collected multimodal historical sample data, including sample current, sample voltage, sample rotational speed, and corresponding sample material viscosity, are preprocessed to remove invalid data points caused by equipment start-up and shutdown or sensor malfunctions. At the same time, the dimensional differences between various electrical parameters of the motor are eliminated to improve the stability of model training.

[0052] Secondly, for example, a neural network regression model is chosen. A neural network is a computational model that mimics the structure and function of a biological nervous system. It consists of a large number of artificial neurons connected together to form a network, capable of learning patterns from data and making predictions or classifications. Its basic structure includes an input layer, hidden layers, and an output layer. The input layer receives raw data; the hidden layers, located between the input and output layers, process information through the connection weights between neurons; and the output layer outputs the final result. The core principle of a neural network is that input data flows in from the input layer, is weighted and calculated by neurons in the hidden layers, and finally the prediction result is obtained from the output layer. Simultaneously, by comparing the error between the prediction result and the actual result, the connection weights of neurons in each layer are adjusted in reverse from the output layer to gradually reduce the error.

[0053] Specifically, the input layer receives preprocessed multi-dimensional motor operation feature vectors, including sample current, sample voltage, and sample speed. The hidden layer employs a two-layer or higher hidden structure and uses the ReLU activation function to fully learn the complex mapping relationship between motor parameters and material viscosity through nonlinear combination. The output layer uses a linear activation function to output a continuous scalar value, i.e., the predicted material viscosity value. The preprocessed complete dataset is randomly divided into a training set and a validation set according to a preset ratio, such as 7:3. The training set is used for updating model weights, and the validation set is used to monitor generalization performance. Simultaneously, the training set is segmented according to process setting parameters, and differentiated learning rate weights are configured for different data segments to enhance the model's ability to learn key operating conditions.

[0054] During training, weighted training set samples are input into the network: the input layer receives standardized motor parameters, which are propagated forward layer by layer through the hidden layers and nonlinearly transformed, resulting in the viscosity prediction value at the output layer. The mean square error (MSE) between the predicted and actual viscosity is calculated as the loss function. Using the backpropagation algorithm, the gradient of the loss with respect to the weights is calculated layer by layer. An adaptive moment estimation (Adam) optimizer is employed to dynamically adjust the network connection weights based on the gradient and the configured differential weights, gradually reducing the prediction error. Training is iterated continuously until the validation set error stabilizes and no longer decreases, and the prediction accuracy is greater than 90%. At this point, training terminates, ultimately yielding a high-precision motor-viscosity regression model.

[0055] In summary, both the thermodynamic-mechanical model and the motor-viscosity regression model have been constructed and trained. Next, the motor operating parameters are input into the motor-viscosity regression model to obtain the first predicted viscosity. Specifically, the real-time collected synchronous time-series signals, including the current-time curve, voltage-time curve, and speed-time curve of the stirring motor, are preprocessed and feature extracted according to the same specifications as in the training phase to form a standardized feature vector. This feature vector is then input into the trained neural network model. After receiving the data in the input layer, undergoing multiple nonlinear transformations in the hidden layer, and linear calculations in the output layer, a continuous viscosity prediction value, i.e., the first predicted viscosity, is directly output.

[0056] Furthermore, a thermodynamic-mechanism model is initialized based on standard viscosity parameters and intrinsic equipment parameters. Raw material formulation parameters and reactor setting parameters are used as inputs to activate the initialized thermodynamic-mechanism model for predictive analysis, obtaining the first predicted temperature. Specifically, firstly, the intrinsic equipment parameters and standard viscosity parameters are used as initial inputs to the material rheological properties, combined with a priori property model to initialize the parameters of the thermodynamic-mechanism model. Secondly, the raw material formulation parameters and reactor setting parameters are used as dynamic inputs to drive the model, calculating the temperature prediction curve of the reaction system over time, and extracting the current predicted temperature value as the first predicted temperature.

[0057] S30: Based on the first predicted viscosity and the thermo-mechanical model, a verification analysis is performed to obtain the second predicted temperature, and the first predicted temperature and the second predicted temperature are compared to perform a prediction output verification.

[0058] Specifically, based on the first predicted viscosity and the thermodynamic-mechanical model, a verification analysis is performed to obtain the second predicted temperature, and the prediction output is verified by comparing the first predicted temperature with the second predicted temperature, including:

[0059] The thermodynamic-mechanism model is updated by replacing the standard viscosity parameter with the first predicted viscosity.

[0060] Based on the updated thermo-mechanical model, iterative predictive analysis is performed to obtain the second predicted temperature;

[0061] Calculate the verification temperature residual between the second predicted temperature and the first predicted temperature, and determine whether the verification temperature residual meets the preset temperature residual threshold.

[0062] If the temperature residual of the verification meets the temperature residual threshold, the prediction output verification passes.

[0063] First, the thermodynamic-mechanistic model is updated by replacing the standard viscosity parameter with the first predicted viscosity. Since the standard viscosity parameter used during the initialization of the thermodynamic-mechanistic model is a theoretical reference value, an assumption under ideal operating conditions and a perfect formulation, but in actual production, factors such as batch differences in raw materials, environmental fluctuations, and slight fouling of equipment can cause the actual viscosity of the material to deviate from the standard, it is necessary to replace the standard viscosity parameter with the first predicted viscosity to correct the parameters of the thermodynamic-mechanistic model, making its current operating basis more consistent with actual conditions.

[0064] Secondly, iterative predictive analysis is performed based on the updated thermo-mechanical model to obtain the second predicted temperature. Then, the verification temperature residual between the second predicted temperature and the first predicted temperature is calculated, and it is determined whether the verification temperature residual meets the preset temperature residual threshold. Specifically, the first predicted temperature is the temperature value calculated by the thermo-mechanical model based on the theoretical assumption of standard viscosity parameters, while the first predicted viscosity is the viscosity estimate directly inferred by the data-driven model from real-time motor operating data. After updating the viscosity parameters in the thermo-mechanical model using the first predicted viscosity, the model has been corrected from an ideal assumption state to a state closer to the actual material properties. At this point, the second predicted temperature, re-output through iterative calculation, represents the theoretical temperature that should occur under the current inferred viscosity.

[0065] Furthermore, the temperature residual is verified, which is the difference between the second predicted temperature and the first predicted temperature. This residual reflects the degree of influence of the inconsistency between the actual inferred viscosity and the standard viscosity assumption on the temperature prediction. The temperature residual is the absolute difference between the second and first predicted temperatures, used to characterize the degree of deviation of the temperature prediction result. The temperature residual threshold is the boundary value of the temperature residual, set according to the specific production process, equipment characteristics, and quality control requirements, such as... .

[0066] If the residual exceeds the threshold, it indicates a significant deviation between the inferred viscosity and the standard assumption. This deviation has a significant impact on temperature prediction, suggesting that the operating conditions may be abnormal or the model assumptions may deviate from reality, requiring the triggering of subsequent verification failure processing procedures. If the residual is less than the preset threshold, it indicates that even if the viscosity inferred by the data-driven model is used, the temperature prediction result of the mechanism model has not changed significantly, thus verifying the consistency of the two model outputs and the stability of the current model operation. In this case, the prediction output verification is successful.

[0067] S40: If the prediction output verification passes, then anomaly identification is performed based on the first predicted temperature and the temperature sensing data.

[0068] Specifically, if the predicted output verification passes, then based on the first predicted temperature and the temperature sensing data, anomaly identification is performed, including:

[0069] Calculate the real-time temperature residual between the first predicted temperature and the real-time acquired temperature sensing data;

[0070] Calculate the first and second derivatives of the real-time temperature residual as a function of time;

[0071] Anomalies are determined according to anomaly triggering rules. If any one of the anomaly triggering rules is triggered, an abnormal operating condition is output. The anomaly triggering rules include:

[0072] The real-time temperature residual exceeds a preset first threshold; the first derivative exceeds a preset second threshold; and the second derivative exceeds a preset third threshold.

[0073] First, the real-time temperature residual between the first predicted temperature and the real-time acquired temperature sensor data is calculated. The real-time temperature residual is the absolute value of the difference between the first predicted temperature value and the actual measured value at the same time point. This residual quantitatively characterizes the instantaneous deviation between the temperature predicted by the thermodynamic-mechanistic model and the actual measured temperature, and is used to evaluate the model's prediction accuracy and operating status in real time. Second, the first and second derivatives of the real-time temperature residual as a function of time are calculated:

[0074]

[0075] Wherein, the first derivative The formula represents the rate of change of the real-time temperature residual, where The residual at the current moment, The residual is the value from the previous time step, and Δt is the sampling time interval; the second derivative is... The acceleration represents the rate of change of the real-time temperature residual, where... The first derivative at the current moment. It is the first derivative of the previous time step.

[0076] Furthermore, anomaly detection is performed based on anomaly triggering rules. Specifically, these rules include: real-time temperature residual exceeding a preset first threshold; first derivative exceeding a preset second threshold; and second derivative exceeding a preset third threshold. These thresholds are determined by professional technicians based on the historical data, process characteristics, and safety requirements of the specific reactor.

[0077] Among these, the most direct temperature anomaly detection is when the real-time temperature residual exceeds a preset first threshold. A large and persistent deviation between the model's predicted and actual measured values ​​indicates a departure from normal operation, corresponding to a serious fault or severe process disturbance. The first derivative captures the slope of the residual change; when it exceeds a preset second threshold, it signifies a rapidly widening deviation and an accelerated, uncontrolled reaction. The second derivative exceeding a preset third threshold is the most sensitive and proactive indicator, monitoring the rate of change. When it exceeds this threshold, it indicates a drastic reversal in the residual's trend, serving as the initial signal of a sudden and severe anomaly, providing early warning before the absolute deviation and rate of change reach dangerous levels. If any of the anomaly trigger rules are triggered, an abnormal operating condition is output.

[0078] In summary, the embodiments of this application have at least the following technical effects:

[0079] Compared to existing technologies, this invention constructs a prediction architecture that integrates mechanistic and data models. It employs a thermodynamic-mechanistic model and an electrodynamic-viscosity regression model for parallel computation and information complementarity, enabling high-precision collaborative inference of core parameters such as temperature and viscosity within the reactor. This effectively overcomes the limitations of single sensing methods or independent models, providing more comprehensive and reliable parameter data for identifying abnormal operating conditions. Secondly, this invention proposes a cross-validation mechanism between models. Using the first predicted viscosity obtained from the electrodynamic-viscosity regression model, the prediction results of the thermodynamic-mechanistic model are reverse-verified. By iteratively generating a second predicted temperature and calculating its residual with the first predicted temperature, online evaluation of the consistency and reliability of the model output is achieved, significantly reducing the risk of false alarms and false negatives.

[0080] Furthermore, this invention organically integrates the advantages of a highly interpretable mechanistic model and a highly adaptable data-driven model. The thermodynamic-mechanistic model is built upon physicochemical mechanisms, and its predictions have clear physical meaning and high reliability. The electrodynamic-viscosity regression model is trained using actual operating data and can characterize complex coupling relationships that are difficult to describe by mechanistic equations. The two work together to reduce dependence on the quantity and quality of historical data while also improving adaptability to different formulations and dynamic operating conditions.

[0081] In summary, this invention achieves earlier, more accurate, and more reliable anomaly diagnosis at the technical level, and has good practicality and stability at the engineering application level, providing reliable technical support for intelligent control and safe operation of polyurethane production processes.

[0082] Example 2, as Figure 2 As shown, based on the same inventive concept as the automatic identification method for abnormal operating conditions of polyurethane production reactors provided in Embodiment 1, this embodiment of the invention also provides an automatic identification system for abnormal operating conditions of polyurethane production reactors, including:

[0083] Data acquisition module 11 is used to acquire temperature sensor data, motor operating parameters and process setting parameters of the target reactor in real time;

[0084] Model prediction module 12 is used to input the motor operating parameters and the process setting parameters into the pre-built thermo-mechanical model and motor-viscosity regression model respectively to predict the operating parameters, so as to obtain the first predicted temperature and the first predicted viscosity.

[0085] The verification analysis module 13 is used to perform verification analysis based on the first predicted viscosity and the thermo-mechanical model, obtain the second predicted temperature, and compare the first predicted temperature with the second predicted temperature to perform prediction output verification.

[0086] The anomaly identification module 14 is used to identify anomalies in operating conditions based on the first predicted temperature and the temperature sensing data when the prediction output verification passes.

[0087] Specifically, the data acquisition module 11 is used for:

[0088] Real-time acquisition of temperature sensor data, motor operating parameters, and process setting parameters of the target reactor, including:

[0089] Acquire the time-series temperature data of the main material in the target reactor and output it as the temperature sensing data;

[0090] The current-time curve, voltage-time curve, and speed-time curve of the stirring motor of the target reactor are collected synchronously, and the output is the motor operating parameters;

[0091] Based on the production management terminal, the real-time production parameters of the target reactor are obtained and output as the process setting parameters. The process setting parameters include at least the raw material formula parameters, reactor setting parameters, standard viscosity parameters, and intrinsic equipment parameters.

[0092] Specifically, the model prediction module 12 is used for:

[0093] The motor operating parameters and the process setting parameters are respectively input into a pre-constructed thermo-mechanical model and a motor-viscosity regression model to predict operating parameters, and the first predicted temperature and the first predicted viscosity are obtained, including:

[0094] Input the motor operating parameters into the motor-viscosity regression model to obtain the first predicted viscosity;

[0095] The thermodynamic-mechanism model is initialized based on the standard viscosity parameters and the intrinsic parameters of the equipment. The raw material formulation parameters and the reactor setting parameters are used as inputs to activate the initialized thermodynamic-mechanism model for predictive analysis and obtain the first predicted temperature.

[0096] Specifically, the steps for constructing the thermodynamic-mechanism model include:

[0097] Establish a digital model of the physical structure of the target reactor;

[0098] Based on the prior property model and the physical structure digital model, an initial mechanism model is constructed, wherein the prior property model includes a polyurethane reaction kinetics model and a reactor heat transfer characteristic model.

[0099] Obtain standard reaction data of the target reactor under standard operating conditions, and use the standard reaction data to perform parameter correction and verification of the initial mechanism model to obtain the thermodynamic-mechanism model.

[0100] Specifically, the steps for constructing the motor-viscosity regression model include:

[0101] Acquire multimodal motor sample data and corresponding sample material viscosity data, wherein the multimodal motor sample data includes at least the associated stored sample current, sample voltage and sample speed;

[0102] Based on the process setting parameters, the multimodal motor sample data and the sample material viscosity data are segmented and the learning rate weights are configured differently.

[0103] Using the sample current, sample voltage, and sample rotation speed as inputs, and the sample material viscosity data as supervision, the motor-viscosity regression model based on regression analysis is constructed and trained.

[0104] Specifically, the verification analysis module 13 is used for:

[0105] Based on the first predicted viscosity and the thermodynamic-mechanical model, a verification analysis is performed to obtain the second predicted temperature. The prediction output is then verified by comparing the first predicted temperature with the second predicted temperature, including:

[0106] The thermodynamic-mechanism model is updated by replacing the standard viscosity parameter with the first predicted viscosity.

[0107] Based on the updated thermo-mechanical model, iterative predictive analysis is performed to obtain the second predicted temperature;

[0108] Calculate the verification temperature residual between the second predicted temperature and the first predicted temperature, and determine whether the verification temperature residual meets the preset temperature residual threshold.

[0109] If the temperature residual of the verification meets the temperature residual threshold, the prediction output verification passes.

[0110] The anomaly detection module 14 is specifically used for:

[0111] If the predicted output verification passes, then based on the first predicted temperature and the temperature sensing data, anomaly identification is performed, including:

[0112] Calculate the real-time temperature residual between the first predicted temperature and the real-time acquired temperature sensing data;

[0113] Calculate the first and second derivatives of the real-time temperature residual as a function of time;

[0114] Anomalies are determined according to anomaly triggering rules. If any one of the anomaly triggering rules is triggered, an abnormal operating condition is output. The anomaly triggering rules include:

[0115] The real-time temperature residual exceeds a preset first threshold; the first derivative exceeds a preset second threshold; and the second derivative exceeds a preset third threshold.

[0116] It should be noted that the order of the embodiments described above is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. Furthermore, the above description focuses on specific embodiments of this specification. Additionally, the processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired results. In some implementations, multitasking and parallel processing are possible or may be advantageous.

[0117] The above description is only a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

[0118] This specification and accompanying drawings are merely illustrative examples of this application and are intended to cover any and all modifications, variations, combinations, or equivalents within the scope of this application. Clearly, those skilled in the art can make various alterations and modifications to this application without departing from its scope. Therefore, if such modifications and modifications fall within the scope of this application and its equivalents, this application intends to include such modifications and modifications.

Claims

1. A method for automatically identifying abnormal working conditions of a polyurethane production reactor, characterized in that, The method comprises the steps of: real-time acquisition of temperature sensor data, motor operating parameters and process setting parameters of a target reaction kettle; inputting the motor operating parameters and the process setting parameters into a pre-constructed thermal-mechanism model and a motor-viscosity regression model respectively for working condition parameter prediction to obtain a first predicted temperature and a first predicted viscosity; carrying out verification analysis according to the first predicted viscosity and the thermal-mechanism model to obtain a second predicted temperature, and comparing the first predicted temperature and the second predicted temperature for prediction output verification; if the prediction output verification passes, carrying out working condition abnormality identification based on the first predicted temperature and the temperature sensor data; wherein the verification analysis according to the first predicted viscosity and the thermal-mechanism model to obtain a second predicted temperature, and the comparison of the first predicted temperature and the second predicted temperature for prediction output verification comprises: replacing the standard viscosity parameter with the first predicted viscosity to update the thermal-mechanism model; carrying out iterative prediction analysis based on the updated thermal-mechanism model to obtain the second predicted temperature; calculating a verification temperature residual of the second predicted temperature and the first predicted temperature, and determining whether the verification temperature residual meets a preset temperature residual threshold; if the verification temperature residual meets the temperature residual threshold, the prediction output verification passes; wherein the working condition abnormality identification based on the first predicted temperature and the temperature sensor data comprises: calculating a real-time temperature residual between the first predicted temperature and the real-time acquired temperature sensor data; calculating a first-order derivative and a second-order derivative of the real-time temperature residual with respect to time; carrying out abnormality identification according to abnormality triggering rules, and if any of the abnormality triggering rules is triggered, outputting working condition abnormality, wherein the abnormality triggering rules comprise: the real-time temperature residual exceeds a preset first threshold; the first-order derivative exceeds a preset second threshold; and the second-order derivative exceeds a preset third threshold.

2. The polyurethane production reactor abnormal condition automatic identification method of claim 1, wherein, real-time acquisition of temperature sensor data, motor operating parameters and process setting parameters of a target reaction kettle, comprising: acquiring time sequence temperature data of main materials of the target reaction kettle, and outputting the temperature sensor data; synchronously acquiring current-time curve, voltage-time curve and speed-time curve of the stirring motor of the target reaction kettle, and outputting the motor operating parameters; based on a production management terminal, acquiring real-time production parameters of the target reaction kettle, and outputting the process setting parameters, wherein the process setting parameters at least include raw material formula parameters, reaction kettle setting parameters, standard viscosity parameters and equipment intrinsic parameters.

3. The polyurethane production reactor abnormal condition automatic identification method of claim 2, wherein, inputting the motor operating parameters and the process setting parameters into a pre-constructed thermal-mechanism model and a motor-viscosity regression model respectively for working condition parameter prediction to obtain a first predicted temperature and a first predicted viscosity, comprising: inputting the motor operating parameters into the motor-viscosity regression model to obtain the first predicted viscosity; Initialize the thermal-mechanism model based on the standard viscosity parameter and the equipment intrinsic parameter, and input the raw material formula parameter and the reaction kettle setting parameter as input, activate the initialized thermal-mechanism model to perform prediction analysis, and obtain the first predicted temperature.

4. The polyurethane production reactor abnormal condition automatic identification method of claim 1, wherein, The construction step of the thermal-mechanism model comprises: establishing a digital model of the physical structure of the target reaction kettle; Based on the prior physical property model and the digital model of the physical structure, an initial mechanism model is constructed, wherein the prior physical property model comprises a polyurethane reaction kinetics model and a reaction kettle heat transfer characteristic model; Obtain the standard reaction data of the target reaction kettle under standard working conditions, and correct and verify the initial mechanism model through the standard reaction data to obtain the thermal-mechanism model.

5. The polyurethane production reactor abnormal condition automatic identification method of claim 1, wherein, The construction step of the motor-viscosity regression model comprises: Obtain multi-modal motor sample data and corresponding sample material viscosity data, wherein the multi-modal motor sample data at least includes sample current, sample voltage and sample speed associated with stored sample data; Based on the process setting parameter, the multi-modal motor sample data and the sample material viscosity data are segmented and divided, and the learning rate weight is differentiated configured; Taking the sample current, sample voltage and sample speed as input, and taking the sample material viscosity data as supervision, the motor-viscosity regression model based on regression analysis is constructed and trained.

6. The polyurethane production reactor abnormal condition automatic identification system, characterized in that, For performing the method of any one of claims 1-5, comprising: A data acquisition module for real-time acquisition of temperature sensing data, motor operating parameters and process setting parameters of the target reaction kettle; A model prediction module for inputting the motor operating parameters and the process setting parameters into the pre-constructed thermal-mechanism model and motor-viscosity regression model respectively to perform working condition parameter prediction to obtain a first predicted temperature and a first predicted viscosity; A verification analysis module for verifying analysis according to the first predicted viscosity and the thermal-mechanism model to obtain a second predicted temperature, and comparing the first predicted temperature and the second predicted temperature to perform prediction output verification; An abnormality identification module for identifying working condition abnormalities based on the first predicted temperature and the temperature sensing data when the prediction output verification is passed.

Images