Intelligent fault diagnosis system for high-speed continuous rolling cold pilger mill based on Internet of Things

Abstract

The invention belongs to the technical field of industrial equipment fault diagnosis, and discloses an intelligent fault diagnosis system for an Internet of Things high-speed continuous rolling cold pilger mill, and the system comprises a real-time sensing module which is used for collecting the operation state data of the cold pilger mill in real time in the working process of the cold pilger mill; the cold pilger mill operation state data comprises cold pilger mill state data, operation condition data, electrical parameter data and working environment data; the self-adaptive feature extraction module is used for performing feature extraction on the cold pilger mill state data and the electrical parameter data through self-adaptive variational mode decomposition; the data enhancement module is used for automatically removing noise in the operation condition data and the working environment data and filling missing values based on a self-adaptive preprocessing technology of time sequence analysis; and the utilization efficiency of maintenance resources is improved, and unnecessary downtime and production loss are reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of industrial equipment fault diagnosis, and more specifically, to an intelligent fault diagnosis system for a high-speed continuous cold rolling tube mill of the Internet of Things. Background Art

[0002] The existing intelligent fault diagnosis system for high-speed continuous cold rolling mill has the following main problems:

[0003] Traditional feature extraction methods usually use fixed-band decomposition or empirical rules. These methods are prone to feature loss or misjudgment when faced with complex and changeable industrial data. Traditional methods cannot adapt to the frequency characteristics of different signals, resulting in inaccurate modal decomposition results. The signals generated by high-speed continuous rolling cold rolling mills under different working conditions are diverse and complex. Fixed decomposition methods cannot flexibly deal with signals of different frequency ranges and complexities, resulting in useful information in the signals being ignored or misallocated. Most of the existing traditional feature extraction methods focus on time domain analysis and ignore the important information of the signal in the frequency domain, resulting in the failure to effectively discover potential faults. In traditional methods, the number of modal decompositions and the bandwidth are usually statically set, which may cause the system to face problems of excessive calculation and resource consumption when processing large-scale data. Excessive decomposition and calculation may lead to slow system response and low processing efficiency, and may even affect the effect of real-time fault diagnosis.

[0004] Traditional data analysis methods often only focus on the correlation of data and ignore the causal relationship between variables; they are unable to accurately identify which variables directly affect the changes in other variables, resulting in the fault diagnosis model being unable to accurately capture the true root cause; traditional methods usually use a fixed lag order, which may cause the model to not have good adaptability when facing different working conditions or equipment states, and a fixed lag order will cause overfitting or underfitting on some data; since the selection of lag order and model complexity has not been optimized, existing technologies may cause the model to be too complex during the calculation process, increasing the computational burden, especially in the case of real-time data processing and high-frequency data collection, which may cause slow processing speed or fail to meet real-time monitoring requirements; under different working conditions, the changes in equipment status and data are very complex and diverse, and systems that lack adaptive adjustment mechanisms may not be able to work continuously and stably, especially in actual operation, they will face a high failure rate.

[0005] In view of this, the present invention proposes an Internet of Things high-speed continuous rolling cold rolling tube mill intelligent fault diagnosis system to solve the above problems. Summary of the invention

[0006] In order to overcome the above-mentioned defects of the prior art and to achieve the above-mentioned purpose, the present invention provides the following technical solutions: an intelligent fault diagnosis system for an Internet of Things high-speed continuous cold rolling tube mill, comprising:

[0007] The real-time sensing module is used to collect the operation status data of the cold rolling mill in real time during the operation of the cold rolling mill; the operation status data of the cold rolling mill includes the state data of the cold rolling mill, the operation condition data, the electrical parameter data and the working environment data;

[0008] An adaptive feature extraction module is used to extract features from the cold rolling mill state data and electrical parameter data through adaptive variational mode decomposition;

[0009] Data enhancement module, based on adaptive preprocessing technology of time series analysis, automatically removes noise from operating condition data and working environment data, and fills in missing values;

[0010] The fault identification module is used to perform Granger causal analysis and integration on the processed cold rolling mill status data, operating condition data, electrical parameter data and working environment data to obtain comprehensive feature data; the comprehensive feature data is input into the trained fault prediction model to predict the probability of fault occurrence; based on the predicted probability of fault occurrence, it is determined whether the high-speed continuous cold rolling mill has a fault;

[0011] Fault source sorting module: if a high-speed continuous cold rolling mill fails, the fault data of the cold rolling mill will be automatically collected and input into the trained fault source prediction model to predict the probability of each fault source of the high-speed continuous cold rolling mill;

[0012] The intelligent response module is used to sort the occurrence probability of each fault source of the high-speed continuous rolling cold rolling tube mill in descending order to obtain a fault source probability ranking table; the intelligent fault diagnosis terminal issues a response alarm instruction according to the fault source probability ranking table.

[0013] Preferably, the cold rolling mill status data includes the working status of the cold rolling mill, the stop status of the cold rolling mill, the equipment health status and the working time of the equipment; the operating condition data includes the production speed of the cold rolling mill, the pressure parameters of the cold rolling mill and the operating load of the cold rolling mill; the electrical parameter data includes the operating current data, the operating voltage data, the power of the cold rolling mill and the operating frequency of the motor; the working environment data includes the temperature, humidity, noise level of the environment where the cold rolling mill is located and the vibration frequency of the cold rolling mill when it is working.

[0014] Preferably, the method for extracting features from the state data and electrical parameter data of the cold rolling mill by adaptive variational mode decomposition comprises:

[0015] S31. Define the cold rolling mill state data and electrical parameter data as a multidimensional time series: x(t) = [x1(t), x2(t), ..., x n (t)] T ; Among them, x(t) is a multidimensional time series; x1(t) is the first type of data in the multidimensional time series; x n (t) is the nth type of data in the multidimensional time series; n is the dimension of the time series; t is the time point index of each type of data in the multidimensional time series, t = 1, 2, ..., T; T is the total length of the multidimensional time series;

[0016] S32, taking the multidimensional time series x(t) as an input signal, decomposing the input signal into K modal components, each modal component representing a different frequency component in the signal;

[0017] S33, presetting an optimization objective function, minimizing the energy inhomogeneity and bandwidth of the modal components by optimizing the objective function, so as to maximize the energy of each mode within its corresponding frequency bandwidth; the optimization objective function is: J represents the minimum value of the optimization objective function; It means that the value of the optimization objective function is minimized by adjusting the modal components; represents the time derivative of the kth modal component; u k (t) represents the kth modal component; represents the frequency domain representation of the kth modal component; λ k represents the constraint coefficient of the kth mode; α k represents the adjustment coefficient for controlling the modal frequency concentration; K represents the number of modal decompositions; k represents the index of the modal component, ranging from 1 to K; represents the sum of the energy of the kth modal component in the frequency domain; ‖·‖ 2 The symbol represents the square norm; f represents the vibration frequency of the signal;

[0018] S34, dynamically adjusting the amount of modal decomposition through an adaptive adjustment formula, the adaptive adjustment formula is: Where B is the bandwidth of the signal; B0 is the reference bandwidth; N yp is the number of signal types; c1 is the sensitivity adjustment factor; c2 is the signal frequency complexity adjustment factor;

[0019] S35, extracting frequency domain features from each modal component obtained by decomposition, including the energy spectrum, spectrum amplitude and peak frequency of the signal; applying Fourier transform to each modal component, converting the kth modal component u k The time domain signal of (t) is converted into a frequency domain signal in the frequency domain; the calculation formula of Fourier transform is: Among them, e―i′2πft is a complex exponential function; i′ is the imaginary unit; dt is the integral symbol;

[0020] S36. In the frequency domain, calculate The spectrum amplitude of each modal component is obtained by the modal length, and the energy spectrum is obtained by calculating the square of the spectrum amplitude of each modal component. The energy spectrum is analyzed to identify the peak frequency in the energy spectrum. The state characteristic data and electrical parameter characteristic data of the cold rolling mill are finally obtained by extracting frequency domain features from all modal components.

[0021] Preferably, the method of automatically removing noise from operating condition data and working environment data and filling missing values ​​by using the adaptive preprocessing technology based on time series analysis includes:

[0022] The Kalman filter algorithm is used to reduce the noise of the collected operating condition data and working environment data, and the missing values ​​in the operating condition data and working environment data are filled by linear interpolation to obtain the operating condition characteristic data and working environment characteristic data.

[0023] Preferably, the method of performing Granger causality analysis and fusing the processed cold rolling mill state data, operating condition data, electrical parameter data and working environment data to obtain comprehensive feature data comprises:

[0024] Select any two variables a from the processed cold rolling mill status data, operating condition data, electrical parameter data and working environment data t and b t , construct a univariate regression model; the univariate regression model is: Among them, b t is the target variable at time point t; a t is the characteristic variable at time point t; p is the lag order; β0 is the intercept term; β j is the target variable b t The regression coefficient at lag order j; ∈ t is the error term of the univariate regression model; γj is the characteristic variable a t The regression coefficient at lag order j; j is the lag order; b t―j is the target variable b t The value at the time point j after the lag; a t―j is the characteristic variable a t The value at the time point j after the lag;

[0025] The lag order of the univariate regression model is determined by the Akaike information criterion formula. By calculating the Akaike information criterion values ​​under different lag orders, the lag order that minimizes the Akaike information criterion value is selected as the lag order of the univariate regression model. The Akaike information criterion formula is: Among them, AIC is the Akaike information criterion value; σ 2 is the error variance of the univariate regression model; g is the number of parameters of the univariate regression model; δ is the penalty coefficient;

[0026] The penalty coefficient δ is dynamically adjusted through the penalty coefficient adjustment formula; the penalty coefficient adjustment formula is: Among them, e1 is the coefficient for adjusting the effect of lag order and model complexity on the penalty coefficient; e2 is the coefficient for adjusting the effect of the total number of variables on the penalty coefficient; m is the number of variables in the univariate regression model;

[0027] Use the F test statistic to test whether the regression coefficient of the univariate regression model is not zero. If it is not zero, then the variable a is determined. t For variable b t There is a Granger causal relationship; through Granger causality analysis, all features with Granger causality in the processed cold rolling mill status data, operating condition data, electrical parameter data and working environment data are identified, and comprehensive feature data are formed through weighted fusion.

[0028] Preferably, the training method of the fault prediction model includes:

[0029] The data set is divided into a training set, a validation set and a test set, and the model is trained and the model performance is evaluated; a fault prediction model is constructed, where the sample set is a subset of the data set, and each sample set includes historical comprehensive feature data and the corresponding probability of fault occurrence; the fault prediction model includes an input layer, a hidden layer and an output layer; the hidden layer uses a ReLU activation function; the input layer of the model is used to input historical comprehensive feature data; the output layer of the model is used to output the probability of fault occurrence; the fault prediction model is a multimodal neural network model;

[0030] Use mean square error as the loss function to measure the error between the model's predicted value and the actual value; use the training set to train the fault prediction model, and update the model parameters through the back propagation algorithm and gradient descent method to minimize the loss function; use the validation set to evaluate the performance of the fault prediction model by calculating the coefficient of determination and tune the model's hyperparameters;

[0031] The Adam optimization algorithm is selected as the optimizer, and the hyperparameters of the model are adjusted until the performance no longer improves or the preset number of iterations is reached. The performance of the model in the prediction task is evaluated using the test set, and the trained fault prediction model is used to predict the current comprehensive feature data to obtain the probability of fault occurrence.

[0032] Preferably, the method for judging whether a high-speed continuous cold rolling mill has a fault according to the predicted probability of the fault occurring comprises:

[0033] Preset a fault probability threshold, and compare the predicted fault probability with the preset fault probability threshold;

[0034] If the predicted fault occurrence probability is less than the preset fault occurrence probability threshold, it is determined that the high-speed continuous cold rolling mill has not failed;

[0035] If the predicted fault occurrence probability is greater than or equal to the preset fault occurrence probability threshold, it is determined that the high-speed continuous cold rolling tube mill has no fault.

[0036] Preferably, the cold rolling mill fault data includes mechanical fault data, electrical fault data, process fault data and operating environment related fault data.

[0037] Preferably, the training method of the fault source prediction model includes:

[0038] The data set is divided into a training set, a validation set and a test set, and the model is trained and the model performance is evaluated; a fault source prediction model is constructed, wherein the sample set is a subset of the data set, and each sample set includes historical cold rolling mill fault data and the corresponding probability of occurrence of each fault source of the high-speed continuous cold rolling mill; the fault source prediction model is a Bayesian network model;

[0039] Define the nodes of the Bayesian network model, each node represents a variable; determine the causal relationship between the nodes, establish a directed acyclic graph, and represent the dependency relationship between the variables; calculate the conditional probability distribution of each node, and use maximum likelihood estimation to calculate the conditional probability parameters in the Bayesian network model;

[0040] The expectation maximization algorithm is used to optimize the parameters of the Bayesian network model and tune the model. The validation set is used to evaluate the model performance, and the accuracy index is calculated to evaluate the performance of the fault source prediction model. The test set is used to evaluate the generalization ability of the model, and the trained Bayesian network model is used to infer the current cold rolling mill fault data to obtain the occurrence probability of each fault source of the high-speed continuous cold rolling mill.

[0041] Preferably, the method for the intelligent fault diagnosis terminal to issue a responsive alarm instruction according to the fault source probability ranking table comprises:

[0042] Based on the fault source probability ranking table, the intelligent fault diagnosis terminal sets alarm instructions for different fault levels and indicates different fault levels with different colors; the intelligent fault diagnosis terminal adopts a graded alarm mechanism according to different fault levels and automatically issues corresponding alarm instructions.

[0043] Compared with the prior art, the present invention has the following beneficial effects:

[0044] The present invention can dynamically adjust the number and frequency bandwidth of modal decomposition according to the bandwidth of the input signal through adaptive variational modal decomposition, so that each modal component can accurately reflect the frequency domain characteristics of the cold rolling mill state data and electrical parameter data, and can effectively improve the extraction accuracy of data features, and provide more reliable input data for subsequent fault diagnosis and prediction; by optimizing the objective function, the energy of each modal component is maximized, and the number and frequency concentration of modal decomposition are dynamically adjusted, so that it can flexibly adapt to the characteristics of different types of signals. This adjustment mechanism greatly improves the adaptability to complex signals and can better adapt to various complex signals generated by the high-speed continuous cold rolling mill during actual operation; by extracting frequency domain features, such as energy spectrum, spectrum amplitude and peak frequency, etc., the abnormal modes that may occur in the operation of the cold rolling mill can be clearly revealed; these features can reflect the operating status of the equipment in real time and capture potential fault signals in advance; the introduction of adaptive adjustment formulas, based on the adjustment factors of signal type, bandwidth and complexity, enables the system to automatically optimize the number of modal decompositions according to real-time data, thereby avoiding unnecessary complexity and calculation burden;

[0045] Granger causality analysis can reveal the causal relationship between different data variables and help identify which features are the root cause of changes in other features. This enables the system to make judgments based on the correlation between more efficient features when performing fault diagnosis, thereby improving the accuracy of diagnosis. Through the Akaike information criterion and dynamic adjustment of the lag order, the optimal lag order can be selected for the characteristics of different data to avoid overfitting or underfitting problems caused by fixed lag orders. Dynamic adjustment of the penalty coefficient helps to balance the complexity and lag order of the model, ensuring the full expression of data features under different working conditions while avoiding excessive calculation and resource consumption. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 It is a structural schematic diagram of an intelligent fault diagnosis system of an Internet of Things high-speed continuous cold rolling tube mill of the present invention;

[0047] Figure 2 The present invention is a flowchart of an intelligent fault diagnosis method for an Internet of Things high-speed continuous rolling cold rolling tube mill. DETAILED DESCRIPTION

[0048] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0049] Embodiment 1

[0050] See also Figure 1 As shown, Embodiment 1 further describes an IoT high-speed continuous cold rolling mill intelligent fault diagnosis system proposed by the present invention, including:

[0051] In modern industrial production, high-speed continuous cold rolling mills are widely used in the steel industry as important pipe production equipment. The efficient operation of such equipment is crucial to the quality and efficiency of pipe production. However, equipment is prone to failure under long-term high-load operation. If it is not diagnosed and handled in time, it may cause equipment downtime, production interruption, or even more serious mechanical damage;

[0052] With the continuous advancement of industrialization, the manufacturing industry, especially the steel, machinery and other industries, has put forward higher and higher requirements on the reliability and production efficiency of equipment in the production process. Especially in the field of steel production, the high-speed continuous cold rolling mill is a key equipment, which carries a high-intensity workload during the production process. Any small failure may lead to production stagnation and cause economic losses. Therefore, real-time monitoring of the operating status of the cold rolling mill and timely detection and handling of potential failures are of great significance to ensure production safety and improve production efficiency.

[0053] At present, the fault diagnosis methods for complex equipment such as cold rolling mills mainly include methods based on vibration signal analysis, methods based on electrical parameter monitoring, and methods based on environmental data analysis such as temperature and pressure. However, these traditional methods usually face the following technical bottlenecks:

[0054] Single feature is not enough to accurately diagnose faults: Traditional fault diagnosis methods often rely on a single device status signal (such as vibration signal, electrical parameters, etc.) for analysis, while equipment failures are often accompanied by the common changes of multiple signals. Therefore, it is difficult to fully and accurately reflect the failure mode of the equipment by relying solely on the feature extraction and analysis of a single signal;

[0055] Unable to process complex signals: In actual operation, the state signals of cold rolling mills are often nonlinear and time-varying, containing noise and interference of different frequencies and amplitudes. These factors make signal feature extraction more complicated. Traditional signal processing methods (such as Fourier transform, waveform analysis, etc.) are difficult to effectively separate valuable features, resulting in a decrease in the accuracy of fault diagnosis;

[0056] It is difficult to capture hysteresis and timing characteristics: the occurrence of equipment failure is often a gradual process, and the timing relationship and hysteresis effect between characteristic variables may have been reflected before the failure occurs. However, existing fault diagnosis technologies are difficult to dynamically capture such hysteresis relationships and timing dependencies, resulting in insufficient advance time for fault prediction and diagnosis, and inability to provide early warning.

[0057] Multi-source data fusion problem: In the monitoring of complex industrial equipment, multiple data sources are involved (such as equipment status data, electrical parameter data, environmental data, etc.), and there may be complex correlations between these data sources. Existing technologies usually lack effective multi-source data fusion methods, making it difficult to extract effective comprehensive features from multiple data dimensions, further affecting the accuracy and robustness of the diagnostic model.

[0058] Model overfitting and underfitting problems: In traditional regression and prediction models, fixed parameters and lag orders usually lead to model overfitting or underfitting. Especially when faced with actual working conditions with complex and changing data features, the performance of the model may fluctuate greatly, making it difficult to provide stable and high-quality diagnostic results.

[0059] In order to effectively solve the above problems, the present invention proposes an intelligent fault diagnosis system for high-speed continuous cold rolling mill based on Internet of Things technology, including

[0060] The real-time sensing module is used to collect the operation status data of the cold rolling mill in real time during the operation of the cold rolling mill; the operation status data of the cold rolling mill includes the state data of the cold rolling mill, the operation condition data, the electrical parameter data and the working environment data;

[0061] An adaptive feature extraction module is used to extract features from the cold rolling mill state data and electrical parameter data through adaptive variational mode decomposition;

[0062] Data enhancement module, based on adaptive preprocessing technology of time series analysis, automatically removes noise from operating condition data and working environment data, and fills in missing values;

[0063] The fault identification module is used to perform Granger causal analysis and integration on the processed cold rolling mill status data, operating condition data, electrical parameter data and working environment data to obtain comprehensive feature data; the comprehensive feature data is input into the trained fault prediction model to predict the probability of fault occurrence; based on the predicted probability of fault occurrence, it is determined whether the high-speed continuous cold rolling mill has a fault;

[0064] Fault source sorting module: if a high-speed continuous cold rolling mill fails, the fault data of the cold rolling mill will be automatically collected and input into the trained fault source prediction model to predict the probability of each fault source of the high-speed continuous cold rolling mill;

[0065] The intelligent response module is used to sort the occurrence probability of each fault source of the high-speed continuous rolling cold rolling tube mill in descending order to obtain a fault source probability ranking table; the intelligent fault diagnosis terminal issues a response alarm instruction according to the fault source probability ranking table.

[0066] The status data of the cold rolling mill include the working status of the cold rolling mill, the stop status of the cold rolling mill, the equipment health status and the working time of the equipment; the operating condition data include the production speed of the cold rolling mill, the pressure parameters of the cold rolling mill and the operating load of the cold rolling mill; the electrical parameter data include the operating current data, the operating voltage data, the power of the cold rolling mill and the operating frequency of the motor; the working environment data include the temperature, humidity, noise level of the environment where the cold rolling mill is located and the vibration frequency of the cold rolling mill when it is working.

[0067] The method for extracting features of cold rolling mill state data and electrical parameter data by adaptive variational mode decomposition includes:

[0068] S31. Define the cold rolling mill state data and electrical parameter data as a multidimensional time series: x(t) = [x1(t), x2(t), ..., x n (t)] T ; Among them, x(t) is a multidimensional time series; x1(t) is the first type of data in the multidimensional time series; x n (t) is the nth type of data in the multidimensional time series; n is the dimension of the time series (i.e., the number of data types); t is the time point index of each type of data in the multidimensional time series, t = 1, 2, ..., T; T is the total length of the multidimensional time series, indicating the total time period of data collection;

[0069] Adaptive variational mode decomposition is an improved method based on variational mode decomposition technology, which can adaptively decompose the signal into multiple intrinsic mode functions. Different from the traditional variational mode decomposition method, adaptive variational mode decomposition can not only adaptively determine the frequency components in the signal, but also automatically adjust the number, center frequency and bandwidth of the modes in the decomposition process, so as to capture the multi-scale characteristics of the signal more finely.

[0070] S32, taking the multidimensional time series x(t) as an input signal, decomposing the input signal into K modal components, each modal component representing a different frequency component in the signal;

[0071] S33, presetting an optimization objective function, minimizing the energy inhomogeneity and bandwidth of the modal components by optimizing the objective function, so as to maximize the energy of each mode within its corresponding frequency bandwidth; the optimization objective function is: J represents the minimum value of the optimization objective function, which represents the final result of the objective function in the optimization process. Minimizing this optimization objective function will achieve the best modal decomposition; min uk It means that the value of the optimization objective function is minimized by adjusting the modal components; represents the time derivative of the kth modal component; u k (t) represents the kth modal component; represents the frequency domain representation of the kth modal component; λ k represents the constraint coefficient of the kth mode; α k represents the adjustment coefficient for controlling the modal frequency concentration, which measures the energy distribution of the modal component on different frequency components; K represents the number of modal decompositions; k represents the index of the modal component, ranging from 1 to K; Represents the sum of the energy of the kth modal component in the frequency domain. The goal is to control the frequency bandwidth of each modal component and avoid too wide a frequency band, because too wide a frequency band may cause the decomposed mode to have no physical meaning; ‖·‖ 2 The symbol for the square norm; f represents the vibration frequency of the signal, and the unit is usually Hertz;

[0072] S34. The number of modal decompositions is dynamically adjusted through an adaptive adjustment formula to adapt to the complexity and diversity of different signals, thereby improving the accuracy, efficiency and robustness of modal decomposition; the adaptive adjustment formula is: Where B is the bandwidth of the signal; B0 is the reference bandwidth; N yp is the number of signal types; c1 is the sensitivity adjustment factor, which is used to adjust the influence of the number of signal types on the modal decomposition number K; c2 is the signal frequency complexity adjustment factor, which is used to adjust the influence of the difference between the signal bandwidth and the reference bandwidth on the modal decomposition number K;

[0073] The number of signal types and the bandwidth are two key factors affecting signal complexity, so considering their effects simultaneously can more accurately reflect the nature of the signal and thus adjust the most suitable number of modes;

[0074] For example, the signal bandwidth B is 100 Hz, the reference bandwidth B0 is 50, and the number of signal types N is yp is 5, the sensitivity adjustment factor c1 is 0.6, and the signal frequency complexity adjustment factor c2 is 0.4, then the number of modal decompositions is

[0075] S35, extracting frequency domain features from each modal component obtained by decomposition, including the energy spectrum, spectrum amplitude and peak frequency of the signal; applying Fourier transform to each modal component, converting the kth modal component u k The time domain signal of (t) is converted into a frequency domain signal in the frequency domain; the calculation formula of Fourier transform is: Among them, e ―i′2πft is a complex exponential function, representing the basis functions of different frequency components; i′ is an imaginary unit, dt is the integral symbol;

[0076] S36. In the frequency domain, calculate The spectrum amplitude of each modal component is obtained by the modal length, and the energy spectrum is obtained by calculating the square of the spectrum amplitude of each modal component; the energy spectrum is a measure of the energy distribution of the signal at different frequencies. The frequency reflects the number of vibrations of the signal per unit time. In the frequency domain, the frequency corresponds to the spectrum of the signal; the energy spectrum is analyzed to identify the peak frequency in the energy spectrum; this frequency represents the main oscillation frequency in the modal component and is used to describe the frequency characteristics of the signal; by extracting frequency domain features from all modal components, the state characteristic data and electrical parameter characteristic data of the cold rolling mill are finally obtained.

[0077] Based on the adaptive preprocessing technology of time series analysis, the methods of automatically removing noise from the operating condition data and working environment data and filling missing values ​​include:

[0078] The Kalman filter algorithm is used to reduce the noise of the collected operating condition data and working environment data, and the missing values ​​in the operating condition data and working environment data are filled by linear interpolation to obtain the operating condition characteristic data and working environment characteristic data.

[0079] Methods for performing Granger causality analysis and fusing the processed cold rolling mill state data, operating condition data, electrical parameter data and working environment data to obtain comprehensive feature data include:

[0080] Select any two variables a from the processed cold rolling mill status data, operating condition data, electrical parameter data and working environment data t and b t , construct a univariate regression model; the univariate regression model is: Among them, b t is the target variable at time point t, for example, any variable value among the processed cold rolling mill state data, operating condition data, electrical parameter data and working environment data; a t is the characteristic variable at time point t; p is the lag order, which is the number of historical data used by the univariate regression model, that is, at the current time point t, the univariate regression model not only considers the data at the current time point, but also considers the data at p time points to predict future values; β0 is the intercept term, which is also a constant term, representing the basic level of influence of all other variables. If the influence of all autoregressive terms and lag terms is zero, then b t The value of will be equal to the constant term; β j is the target variable b t The regression coefficient at lag order j; ∈ t is the error term of the univariate regression model; γj is the characteristic variable a t The regression coefficient at lag order j; j is the lag order; b t―j is the target variable b t The value at the time point j after the lag; at―j is the characteristic variable a t The value at the time point j after the lag;

[0081] The lag order of the univariate regression model is determined by the Akaike information criterion formula. By calculating the Akaike information criterion values ​​under different lag orders, the lag order that minimizes the Akaike information criterion value is selected as the lag order of the univariate regression model. The Akaike information criterion formula is: Among them, AIC is the Akaike information criterion value; is the error variance of the univariate regression model; g is the number of parameters of the univariate regression model, including the constant term and regression coefficient of the univariate regression model; δ is the penalty coefficient, which is used to adjust the influence of the lag order and the complexity of the univariate regression model on the Akaike information criterion value;

[0082] The penalty coefficient δ is dynamically adjusted through the penalty coefficient adjustment formula to balance the complexity and fitting ability of the single variable regression model and adapt to the characteristics of different data; the penalty coefficient adjustment formula is: Among them, e1 is the coefficient for adjusting the effect of lag order and model complexity on the penalty coefficient; e2 is the coefficient for adjusting the effect of the total number of variables on the penalty coefficient; m is the number of variables in the univariate regression model;

[0083] For example, if the number of variables in a univariate regression model is 10, the number of parameters g in the univariate regression model is 5, the lag order p is 3, the coefficient e1 for adjusting the lag order and model complexity on the penalty coefficient is 0.4, and the coefficient e2 for adjusting the total number of variables on the penalty coefficient is 0.6, then the penalty coefficient δ =

[0084]

[0085] Use the F test statistic to test whether the regression coefficient of the univariate regression model is not zero. If it is not zero, then the variable a is determined. t For variable b t There is a Granger causal relationship; through Granger causality analysis, all features with Granger causality in the processed cold rolling mill status data, operating condition data, electrical parameter data and working environment data are identified, and comprehensive feature data are formed through weighted fusion.

[0086] The training methods of the fault prediction model include:

[0087] The data set is divided into training set, validation set and test set, and the model is trained and the model performance is evaluated; a fault prediction model is constructed, and the sample set is a subset of the data set, and each sample set includes historical comprehensive feature data and the corresponding fault probability; the fault prediction model includes an input layer, a hidden layer and an output layer; the hidden layer uses the ReLU activation function; the input layer of the model is used to input historical comprehensive feature data; the output layer of the model is used to output the fault probability; the fault prediction model is a multimodal neural network model;

[0088] Use mean square error as the loss function to measure the error between the model's predicted value and the actual value; use the training set to train the fault prediction model, and update the model parameters through the back propagation algorithm and gradient descent method to minimize the loss function; use the validation set to evaluate the performance of the fault prediction model by calculating the coefficient of determination and tune the model's hyperparameters;

[0089] The Adam optimization algorithm is selected as the optimizer, and the hyperparameters of the model are adjusted until the performance no longer improves or the preset number of iterations is reached. The performance of the model in the prediction task is evaluated using the test set, and the trained fault prediction model is used to predict the current comprehensive feature data to obtain the probability of fault occurrence.

[0090] According to the predicted probability of failure, the method for judging whether a high-speed continuous cold rolling mill has a failure includes:

[0091] Preset a fault probability threshold, and compare the predicted fault probability with the preset fault probability threshold;

[0092] If the predicted fault occurrence probability is less than the preset fault occurrence probability threshold, it is determined that the high-speed continuous cold rolling tube mill has not failed;

[0093] If the predicted fault occurrence probability is greater than or equal to the preset fault occurrence probability threshold, it is determined that the high-speed continuous cold rolling tube mill has no fault.

[0094] The fault data of cold rolling mill includes mechanical fault data, electrical fault data, process fault data and fault data related to the operating environment.

[0095] Mechanical failure data include abnormal roller wear, bearing damage, gear failure, hydraulic cylinder leakage, abnormal oil pressure and oil pump failure; electrical failure data include motor overload, inverter abnormality, servo motor abnormality, relay abnormality, short circuit, open circuit, poor contact, signal abnormality and sensor damage; process failure data include abnormal rolling force, plate shape defects, excessive temperature deviation of steel pipe during rolling, insufficient cooling water flow and nozzle blockage; operating environment-related failure data include lubricating oil contamination, insufficient oil supply, abnormal vibration and abnormal noise during the operation of the cold rolling mill (such as abnormal noise, screaming, etc.).

[0096] The training methods of the fault source prediction model include:

[0097] The data set is divided into a training set, a validation set and a test set, and the model is trained and the model performance is evaluated; a fault source prediction model is constructed, and the sample set is a subset of the data set. Each sample set includes historical cold rolling mill fault data and the corresponding probability of occurrence of each fault source of the high-speed continuous cold rolling mill; the fault source prediction model is a Bayesian network model;

[0098] Define the nodes of the Bayesian network model, each node represents a variable; determine the causal relationship between the nodes, establish a directed acyclic graph, and represent the dependency relationship between the variables; calculate the conditional probability distribution of each node, and use maximum likelihood estimation to calculate the conditional probability parameters in the Bayesian network model;

[0099] The expectation maximization algorithm is used to optimize the parameters of the Bayesian network model and tune the model. The validation set is used to evaluate the model performance, and the accuracy index is calculated to evaluate the performance of the fault source prediction model. The test set is used to evaluate the generalization ability of the model, and the trained Bayesian network model is used to infer the current cold rolling mill fault data to obtain the occurrence probability of each fault source of the high-speed continuous cold rolling mill.

[0100] The method for the intelligent fault diagnosis terminal to issue a response alarm instruction according to the fault source probability ranking table includes:

[0101] Based on the fault source probability ranking table, the intelligent fault diagnosis terminal sets alarm instructions for different fault levels and indicates different fault levels with different colors; the intelligent fault diagnosis terminal adopts a graded alarm mechanism according to different fault levels and automatically issues corresponding alarm instructions.

[0102] The preset fault occurrence probability threshold is set by the staff. Different fault occurrence probabilities are collected through the intelligent fault diagnosis terminal, and the average value of multiple fault occurrence probabilities is taken as the preset fault occurrence probability threshold.

[0103] In this embodiment, through adaptive variational modal decomposition, the number and frequency bandwidth of modal decomposition can be dynamically adjusted according to the bandwidth of the input signal, so that each modal component can accurately reflect the frequency domain characteristics of the cold rolling mill state data and electrical parameter data, which can effectively improve the extraction accuracy of data features and provide more reliable input data for subsequent fault diagnosis and prediction; by optimizing the objective function, the energy of each modal component is maximized, and the number and frequency concentration of modal decomposition are dynamically adjusted so that it can flexibly adapt to the characteristics of different types of signals. This adjustment mechanism greatly improves the adaptability to complex signals and can better adapt to various complex signals generated by high-speed continuous cold rolling mills during actual operation; through the extraction of frequency domain features, such as energy spectrum, spectrum amplitude and peak frequency, the abnormal modes that may occur in the operation of the cold rolling mill can be clearly revealed; these features can reflect the operating status of the equipment in real time and capture potential fault signals in advance; the introduction of the adaptive adjustment formula, based on the adjustment factors of signal type, bandwidth and complexity, enables the system to automatically optimize the number of modal decompositions according to real-time data, thereby avoiding unnecessary complexity and computational burden;

[0104] Granger causality analysis can reveal the causal relationship between different data variables and help identify which features are the root cause of changes in other features. This enables the system to make judgments based on the correlation between more efficient features when performing fault diagnosis, thereby improving the accuracy of diagnosis. Through the Akaike information criterion and dynamic adjustment of the lag order, the optimal lag order can be selected for the characteristics of different data to avoid overfitting or underfitting problems caused by fixed lag orders. Dynamic adjustment of the penalty coefficient helps to balance the complexity and lag order of the model, ensuring the full expression of data features under different working conditions while avoiding excessive calculation and resource consumption.

[0105] Embodiment 2

[0106] See also Figure 2 As shown, the part not described in detail in this embodiment is described in Example 1, which provides an intelligent fault diagnosis method for an Internet of Things high-speed continuous rolling cold rolling tube mill, including:

[0107] S1. During the operation of the cold rolling mill, the operation status data of the cold rolling mill is collected in real time; the operation status data of the cold rolling mill includes the state data of the cold rolling mill, the operation condition data, the electrical parameter data and the working environment data;

[0108] S2, extracting features from the cold rolling mill state data and electrical parameter data through adaptive variational mode decomposition;

[0109] S3, Adaptive preprocessing technology based on time series analysis, automatically removes noise from operating condition data and working environment data, and fills in missing values;

[0110] S4, performing Granger causality analysis on the processed cold rolling mill state data, operating condition data, electrical parameter data and working environment data and integrating them to obtain comprehensive feature data; inputting the comprehensive feature data into the trained fault prediction model to predict the probability of fault occurrence; judging whether the high-speed continuous cold rolling mill has faults according to the predicted probability of fault occurrence;

[0111] S5. If a high-speed continuous cold rolling mill fails, the fault data of the cold rolling mill is automatically collected and input into the trained fault source prediction model to predict the probability of each fault source of the high-speed continuous cold rolling mill;

[0112] S6. Sort the occurrence probabilities of various fault sources of the high-speed continuous cold rolling tube mill in descending order to obtain a fault source probability ranking table; the intelligent fault diagnosis terminal issues a response alarm instruction according to the fault source probability ranking table.

[0113] Since the electronic device introduced in this embodiment is an electronic device used to implement an intelligent fault diagnosis system for a high-speed continuous rolling cold rolling tube mill based on the Internet of Things in the embodiment of this application, based on the intelligent fault diagnosis system for a high-speed continuous rolling cold rolling tube mill based on the Internet of Things introduced in the embodiment of this application, the technical personnel of this field can understand the specific implementation of the electronic device of this embodiment and its various variations, so how the electronic device implements the method in the embodiment of this application is not described in detail here. As long as the technical personnel of this field implement the electronic device used in the intelligent fault diagnosis system for a high-speed continuous rolling cold rolling tube mill based on the Internet of Things in the embodiment of this application, it belongs to the scope of protection of this application.

[0114] The above formulas are all dimensionless and numerical calculations. The formula is a formula for the most recent real situation obtained by collecting a large amount of data and performing software simulation. The preset parameters and thresholds in the formula are set by technicians in this field according to actual conditions.

[0115] The above is only a preferred embodiment of the present invention, and the protection scope of the present invention is not limited to the above embodiments. All technical solutions under the concept of the present invention belong to the protection scope of the present invention. It should be pointed out that for ordinary technical users in this technical field, some improvements and modifications without departing from the principle of the present invention should also be regarded as the protection scope of the present invention.

Claims

1. An intelligent fault diagnosis system for high-speed continuous cold rolling tube mill based on the Internet of Things, characterized in that: The IoT high-speed continuous cold rolling tube mill intelligent fault diagnosis system comprises: The real-time sensing module is used to collect the operation status data of the cold rolling mill in real time during the operation of the cold rolling mill; the operation status data of the cold rolling mill includes the state data of the cold rolling mill, the operation condition data, the electrical parameter data and the working environment data; An adaptive feature extraction module is used to extract features from the cold rolling mill state data and electrical parameter data through adaptive variational mode decomposition; Data enhancement module, based on adaptive preprocessing technology of time series analysis, automatically removes noise from operating condition data and working environment data, and fills in missing values; The fault identification module is used to perform Granger causal analysis and integration on the processed cold rolling mill status data, operating condition data, electrical parameter data and working environment data to obtain comprehensive feature data; the comprehensive feature data is input into the trained fault prediction model to predict the probability of fault occurrence; based on the predicted probability of fault occurrence, it is determined whether the high-speed continuous cold rolling mill has a fault; Fault source sorting module: if a high-speed continuous cold rolling mill fails, the fault data of the cold rolling mill will be automatically collected and input into the trained fault source prediction model to predict the probability of each fault source of the high-speed continuous cold rolling mill; The intelligent response module is used to sort the occurrence probability of each fault source of the high-speed continuous rolling cold rolling tube mill in descending order to obtain a fault source probability ranking table; the intelligent fault diagnosis terminal issues a response alarm instruction according to the fault source probability ranking table.

2. The IoT high-speed continuous cold rolling mill intelligent fault diagnosis system according to claim 1 is characterized in that: The cold rolling mill status data includes the working status of the cold rolling mill, the stop status of the cold rolling mill, the equipment health status and the working time of the equipment; the operating condition data includes the production speed of the cold rolling mill, the pressure parameters of the cold rolling mill and the operating load of the cold rolling mill; the electrical parameter data includes the operating current data, the operating voltage data, the power of the cold rolling mill and the operating frequency of the motor; the working environment data includes the temperature, humidity, noise level of the environment where the cold rolling mill is located and the vibration frequency of the cold rolling mill when it is working.

3. The IoT high-speed continuous cold rolling mill intelligent fault diagnosis system according to claim 2 is characterized in that: The method for extracting features from cold rolling mill state data and electrical parameter data by adaptive variational modal decomposition comprises: S31. Define the cold rolling mill state data and electrical parameter data as a multidimensional time series: x(t) = [x1(t), x2(t), ..., x n (t)] T ; Among them, x(t) is a multidimensional time series; x1(t) is the first type of data in the multidimensional time series; x n (t) is the nth type of data in the multidimensional time series; n is the dimension of the time series; t is the time point index of each type of data in the multidimensional time series, t = 1, 2, ..., T; T is the total length of the multidimensional time series; S32, taking the multidimensional time series x(t) as an input signal, decomposing the input signal into K modal components, each modal component representing a different frequency component in the signal; S33, presetting an optimization objective function, minimizing the energy inhomogeneity and bandwidth of the modal components by optimizing the objective function, so as to maximize the energy of each mode within its corresponding frequency bandwidth; the optimization objective function is: J represents the minimum value of the optimization objective function; It means that the value of the optimization objective function is minimized by adjusting the modal components; represents the time derivative of the kth modal component; u k (t) represents the kth modal component; represents the frequency domain representation of the kth modal component; λ k represents the constraint coefficient of the kth mode; α k represents the adjustment coefficient for controlling the modal frequency concentration; K represents the number of modal decompositions; k represents the index of the modal component, ranging from 1 to K; represents the sum of the energy of the kth modal component in the frequency domain; ‖·‖ 2 The symbol represents the square norm; f represents the vibration frequency of the signal; S34, dynamically adjusting the amount of modal decomposition through an adaptive adjustment formula, the adaptive adjustment formula is: Where B is the bandwidth of the signal; B0 is the reference bandwidth; N yp is the number of signal types; c1 is the sensitivity adjustment factor; c2 is the signal frequency complexity adjustment factor; S35, extracting frequency domain features from each modal component obtained by decomposition, including the energy spectrum, spectrum amplitude and peak frequency of the signal; applying Fourier transform to each modal component, converting the kth modal component u k The time domain signal of (t) is converted into a frequency domain signal in the frequency domain; the calculation formula of Fourier transform is: Among them, e ―i′2πft is a complex exponential function; i′ is the imaginary unit; dt is the integral symbol; S36. In the frequency domain, calculate The spectrum amplitude of each modal component is obtained by the modal length, and the energy spectrum is obtained by calculating the square of the spectrum amplitude of each modal component. The energy spectrum is analyzed to identify the peak frequency in the energy spectrum. The state characteristic data and electrical parameter characteristic data of the cold rolling mill are finally obtained by extracting frequency domain features from all modal components.

4. The IoT high-speed continuous cold rolling mill intelligent fault diagnosis system according to claim 3 is characterized in that: The method of automatically removing noise from the operating condition data and the working environment data and filling missing values ​​by the adaptive preprocessing technology based on time series analysis includes: The Kalman filter algorithm is used to reduce the noise of the collected operating condition data and working environment data, and the missing values ​​in the operating condition data and working environment data are filled by linear interpolation to obtain the operating condition characteristic data and working environment characteristic data.

5. The IoT high-speed continuous cold rolling mill intelligent fault diagnosis system according to claim 4 is characterized in that: The method of performing Granger causality analysis and fusing the processed cold rolling mill state data, operating condition data, electrical parameter data and working environment data to obtain comprehensive feature data includes: Select any two variables a from the processed cold rolling mill status data, operating condition data, electrical parameter data and working environment data t and b t , construct a univariate regression model; the univariate regression model is: Among them, b t is the target variable at time point t; a t is the characteristic variable at time point t; p is the lag order; β0 is the intercept term; β j is the target variable b t The regression coefficient at lag order j; ∈ t is the error term of the univariate regression model; γj is the characteristic variable a t The regression coefficient at lag order j; j is the lag order; b t―j is the target variable b t The value at the time point j after the lag; a t―j is the characteristic variable a t The value at the time point j after the lag; The lag order of the univariate regression model is determined by the Akaike information criterion formula. By calculating the Akaike information criterion values ​​under different lag orders, the lag order that minimizes the Akaike information criterion value is selected as the lag order of the univariate regression model. The Akaike information criterion formula is: Among them, AIC is the Akaike information criterion value; is the error variance of the univariate regression model; g is the number of parameters of the univariate regression model; δ is the penalty coefficient; The penalty coefficient δ is dynamically adjusted through the penalty coefficient adjustment formula, and the penalty coefficient adjustment formula is: Among them, e1 is the coefficient for adjusting the effect of lag order and model complexity on the penalty coefficient; e2 is the coefficient for adjusting the effect of the total number of variables on the penalty coefficient; m is the number of variables in the univariate regression model; Use the F test statistic to test whether the regression coefficient of the univariate regression model is not zero. If it is not zero, then the variable a is determined. t For variable b t There is a Granger causal relationship; through Granger causality analysis, all features with Granger causality in the processed cold rolling mill status data, operating condition data, electrical parameter data and working environment data are identified, and comprehensive feature data are formed through weighted fusion.

6. The IoT high-speed continuous cold rolling mill intelligent fault diagnosis system according to claim 5 is characterized in that: The training method of the fault prediction model includes: The data set is divided into a training set, a validation set and a test set, and the model is trained and the model performance is evaluated; a fault prediction model is constructed, where the sample set is a subset of the data set, and each sample set includes historical comprehensive feature data and the corresponding probability of fault occurrence; the fault prediction model includes an input layer, a hidden layer and an output layer; the hidden layer uses a ReLU activation function; the input layer of the model is used to input historical comprehensive feature data; the output layer of the model is used to output the probability of fault occurrence; the fault prediction model is a multimodal neural network model; Use mean square error as the loss function to measure the error between the model's predicted value and the actual value; use the training set to train the fault prediction model, and update the model parameters through the back propagation algorithm and gradient descent method to minimize the loss function; use the validation set to evaluate the performance of the fault prediction model by calculating the coefficient of determination and tune the model's hyperparameters; The Adam optimization algorithm is selected as the optimizer, and the hyperparameters of the model are adjusted until the performance no longer improves or the preset number of iterations is reached. The performance of the model in the prediction task is evaluated using the test set, and the trained fault prediction model is used to predict the current comprehensive feature data to obtain the probability of fault occurrence.

7. The IoT high-speed continuous cold rolling mill intelligent fault diagnosis system according to claim 6 is characterized in that: The method for judging whether a high-speed continuous cold rolling mill has a fault according to the predicted fault occurrence probability comprises: Preset a fault probability threshold, and compare the predicted fault probability with the preset fault probability threshold; If the predicted fault occurrence probability is less than the preset fault occurrence probability threshold, it is determined that the high-speed continuous cold rolling mill has not failed; If the predicted fault occurrence probability is greater than or equal to the preset fault occurrence probability threshold, it is determined that the high-speed continuous cold rolling tube mill has no fault.

8. The IoT high-speed continuous cold rolling mill intelligent fault diagnosis system according to claim 7 is characterized in that: The cold rolling mill fault data includes mechanical fault data, electrical fault data, process fault data and operating environment related fault data.

9. The IoT high-speed continuous cold rolling mill intelligent fault diagnosis system according to claim 8 is characterized in that: The training method of the fault source prediction model includes: The data set is divided into a training set, a validation set and a test set, and the model is trained and the model performance is evaluated; a fault source prediction model is constructed, wherein the sample set is a subset of the data set, and each sample set includes historical cold rolling mill fault data and the corresponding probability of occurrence of each fault source of the high-speed continuous cold rolling mill; the fault source prediction model is a Bayesian network model; Define the nodes of the Bayesian network model, each node represents a variable; determine the causal relationship between the nodes, establish a directed acyclic graph, and represent the dependency relationship between the variables; calculate the conditional probability distribution of each node, and use maximum likelihood estimation to calculate the conditional probability parameters in the Bayesian network model; The expectation maximization algorithm is used to optimize the parameters of the Bayesian network model and tune the model. The validation set is used to evaluate the model performance, and the accuracy index is calculated to evaluate the performance of the fault source prediction model. The test set is used to evaluate the generalization ability of the model, and the trained Bayesian network model is used to infer the current cold rolling mill fault data to obtain the occurrence probability of each fault source of the high-speed continuous cold rolling mill.

10. The IoT high-speed continuous cold rolling mill intelligent fault diagnosis system according to claim 9 is characterized in that: The method for the intelligent fault diagnosis terminal to issue a response alarm instruction according to the fault source probability ranking table includes: Based on the fault source probability ranking table, the intelligent fault diagnosis terminal sets alarm instructions for different fault levels and indicates different fault levels with different colors; the intelligent fault diagnosis terminal adopts a graded alarm mechanism according to different fault levels and automatically issues corresponding alarm instructions.

Images