An online monitoring and predictive maintenance system and method for metallurgical mechanical equipment

Abstract

The application relates to the technical field of equipment monitoring control, and discloses an online monitoring and predictive maintenance system and method for metallurgical mechanical equipment. The method comprises the following steps: obtaining a preprocessed clean operation data set, determining an interaction intensity value, if the interaction intensity value exceeds a preset intensity threshold, performing fusion analysis on time period data to determine a superposition effect index; determining a temperature load combination sequence, retrieving similar records to obtain an aging speed estimation value, and determining a key combination type; extracting a significant subset, if the significant subset meets a preset condition, obtaining a maintenance timing prediction value; generating an adjustment signal and transmitting the adjustment signal to a control system of the metallurgical mechanical equipment, judging whether the change trend of the adjusted sequence data is consistent with the expected direction of the aging speed estimation value and the superposition effect index, and if the change trend is consistent, outputting a maintenance decision instruction. The application improves the precision of aging monitoring and the scientificity of maintenance decision of the metallurgical mechanical equipment.

Description

Technical Field

[0001] This application relates to the field of equipment monitoring and control technology, and in particular to an online monitoring and predictive maintenance system and method for metallurgical machinery equipment. Background Technology

[0002] Metallurgical machinery and equipment are core equipment in industrial production, often operating in extreme environments of high temperature and high load. Their operational stability directly determines the efficiency, safety, and continuity of metallurgical production, making it a key area of ​​focus for the manufacturing industry. Current maintenance methods for metallurgical equipment often focus on the impact of single environmental factors, neglecting the coupled interaction between high temperature and high load. This makes it difficult to comprehensively and accurately assess the actual damage and aging process of the equipment, easily leading to delayed maintenance or overly frequent maintenance operations. Consequently, sudden equipment failures occur frequently, severely impacting production continuity.

[0003] The combined effect of high temperature and high load creates a cumulative effect, causing equipment aging rates to far exceed those of a single factor. This is the core challenge of existing technologies: high temperatures degrade the performance of equipment materials, while high loads exacerbate wear on mechanical components. The combined effect can easily lead to problems such as component cracking and deformation. Furthermore, equipment often experiences multiple temperature peaks and load fluctuations during actual operation, making damage accumulation difficult to quantify. This further increases the difficulty of predicting aging rates, damage levels, and determining maintenance timing. Therefore, a monitoring and maintenance decision-making method that can accurately capture the impact of temperature-load coupling on equipment aging is urgently needed. Summary of the Invention

[0004] To address the aforementioned technical problems, this application provides an online monitoring and predictive maintenance system and method for metallurgical machinery and equipment, which enables accurate monitoring of the aging state of metallurgical machinery and equipment and intelligent output of maintenance decisions.

[0005] In a first aspect, this application provides an online monitoring and predictive maintenance method for metallurgical machinery and equipment, the method comprising:

[0006] Obtain the preprocessed clean operation dataset, determine the interaction intensity value between temperature and load based on the clean operation dataset, if the interaction intensity value exceeds the preset intensity threshold, extract the time period data that meets the conditions of continuous high temperature and continuous high load from the clean operation dataset, and determine the superposition effect index after performing fusion analysis on the time period data.

[0007] Based on the superposition effect index, a temperature load combination sequence is determined. Similar records are retrieved from the historical aging record database to obtain an aging rate estimate. The key combination type is determined based on the distance relationship between the aging rate estimate and the dynamic connection feature vector.

[0008] Extract a significant subset containing high temperature and high load combinations from the key combination types. If the significant subset meets the preset conditions, then fuse the superposition effect index, the real-time temperature sequence data of metallurgical machinery and equipment and the real-time load sequence data to obtain the maintenance timing prediction value.

[0009] An adjustment signal is generated based on the predicted maintenance timing and transmitted to the control system of the metallurgical machinery and equipment. The adjusted temperature sequence data and load sequence data are collected respectively, and it is determined whether the trend of the adjusted sequence data is consistent with the expected direction of the estimated aging rate and the superposition effect index. If they are consistent, a maintenance decision command is output.

[0010] Secondly, this application provides an online monitoring and predictive maintenance system for metallurgical machinery and equipment, the system comprising:

[0011] The data extraction unit is used to acquire the preprocessed clean operation dataset, determine the interaction intensity value between temperature and load based on the clean operation dataset, and if the interaction intensity value exceeds the preset intensity threshold, extract the time period data that meets the conditions of continuous high temperature and continuous high load from the clean operation dataset, and determine the superposition effect index after performing fusion analysis on the time period data.

[0012] The aging estimation unit is used to determine the temperature load combination sequence based on the superposition effect index, retrieve similar records from the historical aging record database, obtain an aging rate estimate, and determine the key combination type based on the distance relationship between the aging rate estimate and the dynamic connection feature vector.

[0013] The maintenance prediction unit is used to extract a significant subset containing high temperature and high load combinations from the key combination types. If the significant subset meets the preset conditions, the superposition effect index, the real-time temperature sequence data of the metallurgical machinery and equipment and the real-time load sequence data are fused to obtain the maintenance timing prediction value.

[0014] The maintenance decision unit is used to generate an adjustment signal based on the predicted maintenance timing and transmit it to the control system of the metallurgical machinery and equipment. It collects the adjusted temperature sequence data and load sequence data respectively, and determines whether the trend of the adjusted sequence data is consistent with the expected direction of the estimated aging rate and the superposition effect index. If they are consistent, it outputs a maintenance decision command.

[0015] Compared with the prior art, the beneficial effects of the present invention are at least as follows:

[0016] First, temperature and load sequence data are collected through a sensor network. Impulse noise is removed by sliding window midpoint filtering, and outliers are removed and interpolated using the mean-standard deviation method, ensuring the accuracy and continuity of the data source. This lays a high-quality data foundation for subsequent analysis and avoids analytical biases caused by noise in the original data. A random forest algorithm is used to mine the nonlinear correlation between temperature and load, extracting node splitting information to calculate the common frequency. The interaction strength value is obtained by combining variable importance metrics and nonlinear product functions. Furthermore, correlation coefficient corrections for load fluctuations in high-temperature ranges are incorporated to accurately characterize the nonlinear product effect of temperature and load and the accelerated degradation degree of high-temperature load fluctuations, achieving quantitative identification of the coupling effect. Simultaneously, when the interaction strength value exceeds a threshold, a time-series fusion module integrates data from continuous high-temperature, high-load periods to calculate the instantaneous impact response value of load abrupt changes and temperature peaks. The superposition effect level is adjusted using an empirical model of material grain boundary slip rate, and an amplification factor is calculated using a grain boundary slip and creep rate coupling model. Finally, a superposition effect index is obtained, comprehensively revealing the coupling amplification effect of the instantaneous impact of load abrupt changes and temperature peaks on material micro-deterioration, filling the technical gap in the quantification of coupled aging. Then, the current temperature-load combination sequence is combined with the superposition effect index, and similar records are matched from the historical aging record database through initial screening using Euclidean distance and fine screening using cosine similarity. The current aging rate is estimated using the Arrhenius equation, and the gradient of residual life decline is calculated using the Paris law, resulting in an aging rate estimate containing multi-dimensional information, enabling accurate inference of aging status based on historical similar working conditions. The aging rate estimate is combined with the superposition effect index to construct a multi-dimensional numerical vector, and features such as the mean temperature and load fluctuation variance are extracted to construct a dynamic relationship feature vector. The Euclidean distance between the two is calculated, and the equipment status is divided by a preset classification boundary trained by cluster analysis and support vector machine / decision tree algorithms. The key combination type reflecting the material micro-deterioration tracking value and the similarity distance of historical faults is determined, achieving accurate classification of high-risk aging conditions. Finally, a significant subset of high-temperature and high-load conditions is extracted from the key combination types. After determining the high-temperature duration and the number of load cycles, the joint distribution characteristics are calculated by integrating the superposition effect index and real-time temperature and load data aligned with the timestamp. The surface crack propagation rate is estimated based on the crack propagation model of similar historical records, thereby obtaining the predicted value of maintenance timing, and realizing the quantitative and accurate prediction of maintenance timing under high-risk working conditions.The temperature load adjustment range is calculated based on the maintenance timing prediction and superposition effect index, and an adjustment signal is generated. This signal is transmitted wirelessly to the equipment control system. The adjusted temperature load data is collected and preprocessed a second time. Trend verification is completed through multi-dimensional trend feature extraction, coupling trend matching degree calculation, and time series evolution analysis. Furthermore, correlation verification is performed by combining the aging rate estimate and superposition effect index. When the trends are consistent, maintenance decision instructions that correlate the high-temperature material performance degradation and temperature-vibration coupled aging response are output, forming a technical closed loop of "prediction-adjustment-verification-decision". This ensures the pertinence and effectiveness of maintenance decisions, avoids sudden failures and over-maintenance, and reduces production losses. Attached Figure Description

[0017] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a flowchart illustrating an online monitoring and predictive maintenance method for metallurgical machinery and equipment according to an embodiment of this application.

[0019] Figure 2 This is a schematic diagram illustrating the verification of crack propagation model rate prediction in an embodiment of this application.

[0020] Figure 3 This is a schematic diagram illustrating the evolution of trend matching degree and root cause localization in an embodiment of this application;

[0021] Figure 4 This is a schematic diagram of the structure of an online monitoring and predictive maintenance system for metallurgical machinery and equipment according to an embodiment of this application. Detailed Implementation

[0022] This application provides an online monitoring and predictive maintenance system and method for metallurgical machinery and equipment. The terms "first," "second," "third," "fourth," etc. (if present) in the specification, claims, and accompanying drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments described herein can be implemented in a sequence other than that illustrated or described herein. Furthermore, the terms "comprising" or "having" and any variations thereof are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or device that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or devices.

[0023] For ease of understanding, the specific process of the embodiments of this application is described below. Please refer to [link / reference]. Figure 1 One embodiment of the online monitoring and predictive maintenance method for metallurgical machinery and equipment in this application includes:

[0024] Step S1: Obtain the preprocessed clean operation dataset. Determine the interaction intensity value between temperature and load based on the clean operation dataset. If the interaction intensity value exceeds the preset intensity threshold, extract the time period data that meets the conditions of continuous high temperature and continuous high load from the clean operation dataset. After performing fusion analysis on the time period data, determine the superposition effect index.

[0025] This includes obtaining the preprocessed clean run dataset, including:

[0026] Temperature and load sequence data of metallurgical machinery and equipment are collected in real time through a sensor network; sliding window median filtering is applied to the collected temperature and load sequence data; outlier removal is performed on the filtered temperature and load sequence data, and interpolation is used to fill the missing data after outlier removal; the temperature and load sequence data after outlier removal and interpolation are time-aligned to form a clean operation dataset.

[0027] Determining the intensity of the interaction between temperature and load includes:

[0028] A random forest algorithm was used to analyze temperature and load sequence data in a clean operation dataset to construct an interaction model between temperature and load. Node splitting information for temperature values ​​and high load conditions was extracted from the interaction model, and the co-occurrence frequency of these two factors in the decision tree was calculated based on this information. According to the co-occurrence frequency, a variable importance metric was applied to quantify the impact weight of temperature increases on high load conditions. This impact weight was then combined with a nonlinear product function to calculate the interaction strength value between temperature and load. For the accelerated degradation caused by load fluctuations within the high-temperature range, a load fluctuation subsequence within the high-temperature range was extracted, and the correlation coefficient between the fluctuation amplitude and the temperature peak was calculated. The tree averaging method of the random forest was used to incorporate the correlation coefficient into the calculation of the interaction strength value, forming a comprehensive interaction strength value. This comprehensive interaction strength value characterizes the nonlinear product effect of temperature increases and load increases, as well as the degree of accelerated degradation of metallurgical machinery caused by load fluctuations within the high-temperature range.

[0029] Among them, determining the superposition effect index includes:

[0030] High-temperature and high-load thresholds for metallurgical machinery and equipment are pre-defined. Data points where the temperature continuously exceeds the high-temperature threshold and the load continuously exceeds the high-load threshold are identified and extracted from the clean operation dataset. The extracted time-series data are time-aligned, and a weighted average method is used to fuse the time-aligned data points, forming a unified high-temperature, high-load time-series dataset. The load mutation amplitude in the high-temperature, high-load time-series dataset is calculated. The instantaneous impact response value is obtained by multiplying the load mutation amplitude by the temperature peak value and combining it with the time interval. The instantaneous impact response value is judged based on a preset level threshold to determine the superposition effect level. The superposition effect level is adjusted using an empirical model of material grain boundary slip rate. A coupling model of material grain boundary slip and creep rate is introduced, and a coupling amplification factor is calculated. The instantaneous impact response value corresponding to the adjusted superposition effect level is multiplied by the coupling amplification factor to obtain the superposition effect index.

[0031] Specifically, the sensor network is deployed in key parts of metallurgical machinery and equipment, such as furnace bodies, transmission systems, and rolling mills, which are susceptible to high temperatures and high loads. Temperature sensors collect temperature change data from various key parts of the equipment in real time to form temperature sequence data, while load sensors collect load change data such as load and pressure during equipment operation in real time to form load sequence data. The acquisition frequency of the sensor network is adapted to the operating conditions of the equipment to ensure that the collected sequence data can fully reflect the real-time operating status of the equipment. Sliding window midpoint filtering is a nonlinear signal processing method. Its working principle involves selecting a fixed-size sliding window, sorting all data points within the window by value, and taking the median value as the filtered output value at the window's center. The window is then slid along the time axis, and this process is repeated to filter the entire sequence of data. The input data for this method are the original temperature and load sequences, and the output data are the filtered temperature and load sequences. Compared to traditional mean filtering, sliding window midpoint filtering effectively removes impulse noise while preserving the trend characteristics of the sequence data. For example, the default sliding window size is 5 data points. In the high-frequency vibration environment of metallurgical machinery, the sliding window size can be adjusted to 7 data points to better adapt to sudden load changes in the equipment and improve data smoothing. Outliers are extreme values ​​in sequence data that deviate from the normal data range. Their causes mainly include extreme working conditions in metallurgical environments, sensor malfunctions, and data transmission interference. This embodiment uses a mean-standard deviation method to identify outliers. First, the mean and standard deviation of the filtered sequence data are calculated. The outlier identification threshold is set to the mean plus or minus three times the standard deviation. Data points exceeding this threshold are identified as outliers. For temperature sequence data processing of high-temperature special metallurgical equipment such as steelmaking furnaces, to retain more effective high-temperature peak data, the outlier identification threshold can be adjusted to the mean plus or minus 2.5 times the standard deviation. The interpolation method is linear interpolation, which estimates the value of the missing data point by linear fitting based on the values ​​of two adjacent effective data points, ensuring the continuity of the sequence data. After outlier removal and interpolation filling, time alignment is performed to ensure that the temperature sequence data and load sequence data correspond one-to-one at the same timestamp, ultimately forming a clean operation dataset. This clean operation dataset is a two-dimensional time-series dataset containing information in three dimensions: timestamp, corresponding temperature value, and load value.

[0032] The core of the interaction strength value is to quantify the impact of the coupling effect of temperature and load on the aging of metallurgical machinery and equipment, breaking through the limitations of single-factor analysis in existing technologies. The Random Forest algorithm is a machine learning algorithm based on ensemble learning. Its working principle is to extract multiple sample sets from the original dataset using a bootstrap sampling method, construct an independent decision tree for each sample set, and finally vote on the prediction results of all decision trees to obtain the final analysis result. In this step, the input data for this algorithm are temperature and load sequence data from the clean operation dataset, and the output is a model of the interaction relationship between temperature and load. This model can uncover the nonlinear correlation characteristics between temperature and load, reflecting the coordinated change patterns of the two during equipment operation. Node splitting information refers to the feature variables and splitting thresholds used when splitting nodes in the random forest algorithm corresponding to the interaction relationship model. The high-load condition is a pre-set load threshold based on the rated operating parameters of the metallurgical machinery. When the load value exceeds this threshold, it is determined to be under high-load conditions. Based on the extracted node splitting information, the co-occurrence frequency of temperature value and high-load condition in the decision tree is calculated, i.e., the proportion of times they co-occur as splitting features in the same decision tree node out of the total number of splits. This co-occurrence frequency can preliminarily reflect the correlation between temperature and high-load conditions. The variable importance measurement method is based on the Gini index. The Gini index is used to measure the impurity of decision tree nodes; the smaller the Gini index of a node, the higher the data purity within the node. The working principle of this method is to calculate the total contribution value of each feature variable to reducing node impurity in all decision trees of the random forest, and use this contribution value as the importance weight of the feature variable. In this step, the input data of this method are the co-occurrence frequency of temperature value and high-load condition and the node impurity data of each decision tree; the output data is the weight of the impact of temperature increase on high-load conditions. The nonlinear product function is a function obtained by polynomial fitting of temperature and load data. It can reflect the multiple effect of load increase when the temperature rises by 1 degree Celsius. Its input data are temperature value, load value and the influence weights obtained above. The output data is the preliminary interaction strength value between temperature and load.To address the accelerated degradation problem caused by load fluctuations within high-temperature ranges, the initial interaction strength values ​​need to be optimized. First, the high-temperature range is determined. This range is a pre-defined temperature range based on the material properties and operating conditions of metallurgical machinery and equipment, specifically those temperatures that are prone to causing material degradation. For example, the high-temperature range for a rolling mill can be set to 450-500 degrees Celsius, and for a steelmaking furnace, it can be set to above 800 degrees Celsius. Load fluctuation subsequences within the high-temperature range are extracted from the clean operation dataset; these are subsequences composed of all load data within the high-temperature range. The Pearson correlation coefficient is used to calculate the correlation coefficient between the fluctuation amplitude of this load fluctuation subsequence and the corresponding temperature peak value. This method is a classic statistical method for quantifying the linear correlation between two continuous variables. The fluctuation amplitude is the difference between the maximum and minimum values ​​in the load fluctuation subsequence, and the temperature peak value is the maximum value of the temperature sequence data within the corresponding high-temperature range. The correlation coefficient measures the degree of linear association between the two variables, with a value range of [-1, 1]. A larger absolute value indicates a higher degree of association. The tree averaging method is a method of averaging the calculation results of all decision trees in a random forest. Its input data are the preliminary interaction strength values ​​corresponding to each decision tree and the aforementioned correlation coefficients. The output data is the comprehensive interaction strength value. This comprehensive interaction strength value can more accurately characterize the nonlinear product effect of temperature increase and load increase, as well as the degree of accelerated degradation of metallurgical machinery and equipment caused by load fluctuations in the high-temperature range. The larger the value of this strength value, the more significant the effect of the coupling effect of temperature and load on the accelerated aging of equipment.

[0033] If the comprehensive interaction strength value obtained above exceeds the preset strength threshold, it indicates that the coupling effect of temperature and load has reached the point of significantly affecting the aging of metallurgical machinery and equipment. It is necessary to continue to perform the operation of determining the superposition effect index. The superposition effect index is used to characterize the instantaneous impact response generated by the superposition of load change and temperature peak, as well as the degree of coupling amplification formed by material grain boundary slip and creep rate. It can further quantify the degree of damage to the equipment caused by the coupling effect of the two under continuous high temperature and high load conditions.

[0034] The high-temperature threshold is a critical temperature value set based on the heat resistance of the materials of key equipment components, and the high-load threshold is a critical load value set based on the rated load of the equipment, typically taken as 80% of the rated load. The time-period data refers to the operating data of metallurgical machinery equipment under continuous high-temperature and high-load conditions. The extracted time-period data is time-series aligned to ensure that temperature and load data collected from different locations remain consistent over time. The weighted average method assigns appropriate weights to different data based on their acquisition accuracy and importance, and then calculates the weighted average. In this step, the temperature sequence data and the load sequence data are given the same weight. The input data for this method is the time-series aligned continuous high-temperature and high-load time-period data, and the output data is a unified high-temperature and high-load time-period dataset. This dataset integrates temperature and load operating data under continuous high-temperature and high-load conditions, facilitating subsequent centralized analysis.

[0035] The load mutation amplitude is the rate of change from the baseline load to the load peak in the load sequence data. The baseline load is the average load during the period, and the load peak is the maximum load during the period. The calculated load mutation amplitude is multiplied by the temperature peak, which is the maximum temperature in the data during the high temperature and high load period. Combined with the time interval of the load mutation, i.e. the time from the baseline load to the load peak, the instantaneous impact response value is calculated using the formula: Instantaneous impact response value = Load mutation amplitude × Temperature peak / Time interval. This value is used to characterize the instantaneous impact on metallurgical machinery and equipment when the load mutation is superimposed on the temperature peak. The preset level threshold is a value calibrated based on historical operating data and fault data of metallurgical machinery and equipment. The superposition effect level is usually divided into three levels: low, medium, and high. To further improve the accuracy of the superposition effect level, the superposition effect level is adjusted by combining the empirical model of material grain boundary slip rate. The empirical model of material grain boundary slip rate is an empirical formula based on the material properties of key components of metallurgical machinery and equipment, which reflects the relationship between material grain boundary slip rate and temperature and load. Grain boundary slip refers to the relative displacement of material crystal boundaries at high temperatures, which is an important cause of material performance degradation. The input data of this model are the temperature and load values ​​of the dataset during high temperature and high load periods, and the output data is the material grain boundary slip rate. If the calculated grain boundary slip rate is coupled and amplified by more than 1.5 times, the superposition effect level is increased by one level.

[0036] Creep rate refers to the rate at which a material undergoes slow plastic deformation under high temperature and high load conditions. It is an important indicator for measuring the long-term aging of materials. The coupling model of grain boundary slip and creep rate is a mathematical model that reflects the synergistic effect and mutual amplification of the two. Its working principle is to multiply the grain boundary slip rate and creep rate and introduce a temperature coefficient for correction. The temperature coefficient is a coefficient set according to the material properties to reflect the influence of temperature on the coupling effect between the two. The input data of this model are the grain boundary slip rate and creep rate corresponding to the adjusted superposition effect level, and the output data is the coupling amplification factor. The larger the value of this factor, the higher the degree of coupling amplification of grain boundary slip and creep rate, and the more significant the impact on the degradation of equipment materials. The final superposition effect index integrates the instantaneous impact of the load change superimposed with the temperature peak and the coupling amplification effect of the material's grain boundary slip and creep rate. It can comprehensively reflect the degree of accelerated degradation of metallurgical machinery and equipment under continuous high temperature and high load conditions, providing a key quantitative indicator for subsequent equipment aging rate estimation and maintenance timing prediction.

[0037] Step S2: Determine the temperature load combination sequence based on the superposition effect index, retrieve similar records from the historical aging record database to obtain the aging rate estimate, and determine the key combination type based on the distance relationship between the aging rate estimate and the dynamic connection feature vector.

[0038] The estimated aging rate includes:

[0039] The current temperature and load sequence data of metallurgical machinery and equipment are integrated by timestamp, and combined with the superposition effect index to form a temperature-load combination sequence with coupling effect. The temperature-load combination sequence is then subjected to sliding window mid-range filtering. The Euclidean distance algorithm is used to calculate the similarity between the processed temperature-load combination sequence and the historical combination sequences with coupling effect in the historical aging record library, and a subset of records is initially screened based on the similarity. After normalizing the record subset, the cosine similarity algorithm is used to calculate the matching degree between the processed temperature-load combination sequence and each historical combination sequence in the record subset, and the matching similar records are determined based on the matching degree. The aging rate estimate is extracted from the matching similar records. The aging rate estimate includes the current aging rate of the metallurgical machinery and equipment, the matching degree between the current working condition and the historical similar working condition, and the decreasing gradient of the equipment's residual life under high temperature and high load alternation. Among them, the current aging rate is estimated by combining the superposition effect index using the Arrhenius equation, and the decreasing gradient of the residual life is estimated by combining the superposition effect index using the Paris law to simulate crack propagation.

[0040] Among these, identifying key combination types includes:

[0041] The aging rate estimate is combined with the superposition effect index to construct a multidimensional numerical vector. The average temperature, load fluctuation variance, and high temperature duration ratio are extracted from the real-time operation data of metallurgical machinery and equipment to construct a dynamic relationship feature vector. The Euclidean distance between the multidimensional numerical vector and the dynamic relationship feature vector is calculated. The Euclidean distance is used to classify the equipment into three states: healthy operation, early warning, and imminent failure, based on the pre-set classification boundary. The pre-set classification boundary is obtained by training the historical operation data through cluster analysis and support vector machine or decision tree algorithm. The equipment is classified into three states: healthy operation, early warning, and imminent failure. The key combination type is determined based on the classification results of the equipment states. The key combination type reflects the dynamic tracking value of the microstructure degradation parameters of the metallurgical machinery and equipment materials and the similarity distance between the historical failure mode and the current state vector.

[0042] Specifically, this step takes the superposition effect index as the core coupling feature to complete the construction of temperature load combination sequence, similar record retrieval, aging rate estimation value extraction and key combination type determination, so as to realize the progressive analysis of the aging status of metallurgical machinery and equipment from quantitative estimation to accurate determination of working condition type.

[0043] Integrating the superposition effect index into the construction process of temperature-load combination sequences overcomes the limitations of traditional single temperature-load sequence retrieval. This allows the retrieved historical records to better reflect the actual operating conditions of metallurgical equipment under high-temperature and high-load coupling. Simultaneously, combining the superposition effect index with the accurate extraction of aging rate estimates makes the calculation of aging rate and residual life reduction gradient more closely reflect the actual aging of equipment under coupling effects. In specific implementation, timestamp integration ensures a one-to-one correspondence between temperature and load values ​​at the same time point, maintaining a high degree of consistency in the time dimension of temperature-load data. The superposition effect index characterizes the instantaneous impact response of the temperature peak caused by load mutation and the degree of coupling amplification of material grain boundary slip and creep rate. It is a core indicator for quantifying the temperature-load coupling aging effect of metallurgical equipment. Integrating it into the temperature-load combination sequence ensures that the sequence not only contains time-series data of temperature and load but also carries characteristic information of coupled aging, accurately reflecting the operating status of equipment under coupled conditions. Sliding window midpoint filtering is a nonlinear signal processing method. Its working principle involves selecting a fixed-size sliding window, sorting all data points within the window by numerical value, and taking the median value as the filtered output value at the center of the window. The window is then slid along the time axis, and the above operation is repeated. In this step, the input data is a temperature load combination sequence with coupling effects, and the output data is the filtered temperature load combination sequence. This method effectively removes pulse noise caused by sudden changes in metallurgical operating conditions and sensor interference, while preserving the trend characteristics and coupling aging characteristics of the sequence. In this embodiment, the default size of the sliding window is 5 data points. In a high-frequency vibration operating environment for metallurgical machinery, the window size can be adjusted to 7 data points to better adapt to the sudden load changes of the equipment and improve the smoothness of the combination sequence.

[0044] The Euclidean distance algorithm is a classic algorithm for measuring the spatial distance between two multidimensional vectors. Its working principle is to treat two sequences as multidimensional numerical vectors, calculate the sum of squares of the differences in corresponding dimensions, and then take the square root of the sum to obtain the Euclidean distance value. The smaller the distance value, the higher the similarity between the two sequences. In this step, the input data for this algorithm consists of a filtered temperature load combination sequence and historical combination sequences with coupling effects from a historical aging record database. The output data is the Euclidean distance value between the current combination sequence and each historical combination sequence. By setting an Euclidean distance threshold, historical combination sequences with distance values ​​less than the threshold are filtered out to form a record subset, completing the initial screening of similar records, effectively narrowing the scope of subsequent precise matching, and improving retrieval efficiency. After obtaining the subset of records, it is normalized. Normalization is a numerical processing method that maps sequence data with different dimensions and numerical ranges to the [0,1] interval. Its working principle is to calculate the difference between the maximum and minimum values ​​of the sequence data, divide the difference between each data point and the minimum value by this difference, and obtain the normalized value. The input data in this step is the historical combination sequence with coupling effect in the subset of records, and the output data is the historical combination sequence contained in the normalized subset of records. It can eliminate the influence of the difference in the dimensions and numerical range of temperature and load data on the subsequent similarity calculation and ensure the accuracy of the matching degree calculation.

[0045] The cosine similarity algorithm is a method for calculating the similarity between two multidimensional vectors by measuring the cosine of the angle between them. Its working principle involves treating two sequences as multidimensional numerical vectors, calculating their dot product, and then dividing the dot product by the product of the magnitudes of the two vectors to obtain the cosine similarity value. The similarity value ranges from -1 to 1, with a value closer to 1 indicating a higher degree of matching between the two sequences. In this step, the input data for the algorithm consists of a filtered temperature-load combination sequence and a normalized subset of records. The output data is the cosine similarity value between the current combination sequence and each historical combination sequence in the record subset. In this implementation, the historical combination sequence with the highest cosine similarity value is selected as the similar record matching the current operating condition, thus achieving accurate determination of similar records. Under the operating conditions of metallurgical equipment with large load fluctuations, higher weights can be assigned to the load component and superposition effect index in the cosine similarity calculation. For example, the load weight coefficient can be set to 0.5, the superposition effect index weight coefficient to 0.1, and the temperature weight coefficient to 0.4, making the similarity calculation more consistent with the load-dominated coupled aging operating conditions and improving the accuracy of matching.

[0046] The aging rate estimate includes the current aging rate of the metallurgical machinery and equipment, the degree of matching between the current operating conditions and historically similar operating conditions, and the decreasing gradient of the equipment's residual life under alternating high temperature and high load. The current aging rate is estimated using the Arrhenius equation, a classic equation describing the quantitative relationship between reaction rate and temperature. Its working principle is that the reaction rate and temperature have an exponential relationship, and the rate constant is equal to the pre-exponential factor multiplied by the negative activation energy of the natural constant divided by the exponent of the product of the gas constant and the absolute temperature. In this step, this equation is adapted for estimating the aging rate of the metallurgical equipment. The input data includes the superposition effect index, the peak temperature in the current temperature series data, and the activation energy and gas constant of the materials of key components of the equipment (activation energy characterizes the core components of the metallurgical equipment). The physical parameter representing the minimum energy required for the thermal aging reaction of the material, the gas constant (an inherent fundamental physical constant of the Arrhenius equation, which can quantify the influence of temperature on the aging reaction rate of equipment materials), outputs the current aging rate of the metallurgical machinery. The inclusion of the superposition effect index corrects the aging rate under temperature-load coupling, making the calculation results more consistent with the actual situation of coupled aging. The decreasing gradient of residual life, combined with the superposition effect index, is calculated using the Paris Law to simulate crack propagation. The Paris Law is a classic law describing the quantitative relationship between the fatigue crack propagation rate of a material and the stress intensity factor range. Its working principle is that the crack propagation rate is proportional to a certain power of the stress intensity factor range, and the crack growth rate is equal to the material constant multiplied by the stress intensity factor range m. The power of , where m is a material constant, which takes a value of around 3 in metallurgical equipment, is used in this step to simulate the crack propagation process of key components of the equipment. The input data are the superposition effect index, the load mutation amplitude in the current load sequence data, and the range of material constants and stress intensity factors of key components of the equipment. The output data is the decreasing gradient of the equipment's residual life under high temperature and high load alternation. The inclusion of the superposition effect index can quantify the accelerating effect of coupling on crack propagation, making the calculation of the decreasing gradient of residual life more accurate. The matching degree between the current working condition and historical similar working conditions is directly represented by the similarity value calculated by the aforementioned cosine similarity algorithm. The higher the similarity value, the higher the matching degree.

[0047] Specifically, the operation of determining key combination types involves integrating the superposition effect index into the vector construction of the aging rate estimate, and combining it with the dynamic relationship feature vector to complete the classification of equipment operating status. This ultimately achieves accurate determination of key combination types under coupled aging conditions, providing a basis for subsequent screening of high-risk operating conditions. In practice, the current aging rate, the degree of matching between the current operating condition and historically similar operating conditions, the decreasing gradient of the equipment's remaining life under alternating high temperature and high load, and the values ​​of the superposition effect index are integrated to form a multi-dimensional numerical feature vector. This multi-dimensional numerical vector not only contains information on the equipment's aging status but also incorporates the characteristic information of coupled aging, comprehensively reflecting the aging and operating status of metallurgical equipment under temperature-load coupled conditions. Simultaneously, a dynamic connection feature vector is constructed, where the temperature mean is the arithmetic mean of the current real-time temperature sequence data, representing the overall temperature level of the equipment; the load fluctuation variance is the variance value of the current real-time load sequence data, representing the degree of fluctuation of the current load of the equipment; and the high temperature duration ratio is the ratio of the time in the high temperature range in the current real-time temperature sequence data to the total monitoring time, representing the duration of the equipment being in high temperature conditions. The dynamic connection feature vector formed by integrating these three feature values ​​can accurately reflect the instantaneous operating status characteristics of metallurgical equipment.

[0048] The Euclidean distance algorithm in this step takes as input a multidimensional numerical vector of aging rate estimates combined with superposition effect indicators and a dynamic correlation feature vector. The output is the Euclidean distance between the two vectors, which quantifies the correlation between the equipment's aging state and its instantaneous operating state, providing a quantitative basis for subsequent equipment state classification. The preset classification boundary is obtained by clustering historical operating data and then training using a support vector machine (SVM) or decision tree algorithm, classifying the equipment into three states: healthy operation, early warning, and impending failure. Clustering analysis is an unsupervised learning method that divides data objects into multiple clusters. Its working principle is to group data objects with high similarity into the same cluster and those with low similarity into different clusters. In this step, the input data for this analysis method is the historical operating data of the metallurgical equipment, and the output data is multiple clusters of data after clustering, providing a data foundation for training the classification boundary. The support vector machine (SVM) algorithm is a machine learning algorithm based on statistical learning theory. Its working principle is to accurately classify different categories of data by finding the optimal classification hyperplane. In this step, the input data for this algorithm is the data after clustering analysis... The system uses historical operating data of metallurgical equipment and corresponding equipment status labels. The output data is the preset classification boundary obtained through training. The decision tree algorithm is a machine learning algorithm based on a tree structure for decision-making. Its working principle is to construct a tree-like decision model by recursively dividing data features to classify the data. The input data of this algorithm in this step is the same as that of the support vector machine algorithm, and the output data is also the preset classification boundary obtained through training. In this embodiment, the support vector machine algorithm or the decision tree algorithm can be selected to train the classification boundary according to the operating data characteristics of the metallurgical equipment and the actual application requirements. The preset classification boundary obtained through training consists of multiple Euclidean distance thresholds. By comparing the currently calculated Euclidean distance value with these thresholds, the equipment status can be classified into one of healthy operation, early warning, or imminent failure.

[0049] The key combination type reflects the dynamic tracking values ​​of the microstructure degradation parameters of metallurgical machinery and equipment materials and the similarity distance between historical failure modes and the current state vector. The dynamic tracking values ​​of the microstructure degradation parameters include micro-degradation indicators such as grain boundary slip rate, dislocation density growth rate, and grain size growth rate of key equipment component materials. These indicators are directly related to temperature-load coupled operating conditions. Different equipment states correspond to different trends in the change of micro-degradation parameters, and the key combination type is the mapping of this trend to specific temperature-load operating conditions. The similarity distance between historical failure modes and the current state vector is characterized by the Euclidean distance or cosine similarity between the currently constructed multidimensional numerical vector and the state vector corresponding to the historical failure mode. Different equipment states correspond to different degrees of similarity with different historical failure modes. In this embodiment, different equipment status classifications correspond to different key combination types. For example, when the equipment status is classified as healthy operation, the corresponding key combination type is a stable operating condition with normal temperature and low load or medium temperature and low load. At this time, the tracking value of the material microstructure degradation parameter is at the baseline level, and the similarity distance with historical failure modes is relatively far. When the equipment status is classified as early warning, the corresponding key combination type is a condition with medium and high temperature accompanied by periodic high load impacts. At this time, the tracking value of the material microstructure degradation parameter shows a slight acceleration trend, and the similarity distance with historical failure modes is shortened. When the equipment status is classified as near failure, the corresponding key combination type is a condition with continuous high temperature and frequent load changes. At this time, the tracking value of the material microstructure degradation parameter shows a significant acceleration trend, and the similarity distance with historical failure modes is very close. The key combination types determined in this way not only accurately classify the temperature-load coupled operating conditions of metallurgical equipment, but also integrate the correlation information between the material microstructure degradation state and historical failures, providing a core judgment basis for the subsequent screening of high-risk, high-temperature, and high-load operating conditions.

[0050] Step S3: Extract significant subsets containing high temperature and high load combinations from key combination types. If the significant subsets meet preset conditions, the maintenance timing prediction value is obtained by fusing the superposition effect index, the real-time temperature sequence data of metallurgical machinery and equipment, and the real-time load sequence data.

[0051] Among them, the predicted maintenance timing values ​​include:

[0052] Iterate through all elements of key combination types, filter out element combinations with temperature values ​​exceeding a preset high-temperature threshold and load values ​​exceeding a preset high-load threshold, forming a significant subset containing high-temperature and high-load combinations; calculate the continuous duration of the high-temperature interval within the significant subset, count the number of cyclic changes in load from low to high within the significant subset, and determine whether the high-temperature duration exceeds the limit and whether the load cycle count reaches the trigger condition based on the continuous duration and the number of cyclic changes; when both conditions are met, align the real-time temperature sequence data and real-time load sequence data of the metallurgical machinery and equipment by timestamp, and calculate the joint distribution characteristics of the three by fusing the superposition effect index; based on the crack propagation model of similar records in the historical aging record library, combine the joint distribution characteristics to estimate the surface crack propagation rate of equipment components, and obtain the predicted value of maintenance timing for the metallurgical machinery and equipment based on the surface crack propagation rate.

[0053] Specifically, this step takes key combination types as the basis for analysis. By extracting significant subsets of high temperature and high load and determining preset conditions, the superposition effect index is integrated into the fusion analysis of real-time temperature and load data. Combined with the crack propagation model, the prediction value of maintenance timing is calculated. This process accurately focuses on the high-risk working conditions of high temperature and high load coupling in metallurgical machinery and equipment, realizing the transformation of maintenance timing from qualitative judgment to quantitative prediction.

[0054] The key combination type is obtained by classifying equipment states based on the Euclidean distance between the estimated aging rate and the dynamic relationship feature vector. It reflects the dynamic tracking value of the microstructure degradation parameters of the metallurgical machinery and equipment materials and the similarity distance between the historical failure mode and the current state vector. Each element is a temperature-load operating condition combination carrying aging state information. The preset high temperature threshold is a critical temperature value pre-defined based on the heat resistance performance of the key components of the metallurgical machinery and equipment and the operating conditions. The setting of this threshold combines the rated operating temperature of the equipment and the critical temperature of high temperature degradation of the material. For example, the preset high temperature threshold of the rolling mill can be set to 450 degrees Celsius, and the preset high temperature threshold of the steelmaking furnace can be set to 800 degrees Celsius. The preset high load threshold is a critical load value pre-defined based on the rated load parameters of the metallurgical machinery and equipment, usually taken as 80% of the rated load of the equipment. The setting of this threshold takes into account the load-bearing capacity of the equipment's mechanical structure and the fatigue damage characteristics under load fluctuations. In practice, all temperature-load condition combination elements in the key combination type are iterated one by one. For each element, its temperature characteristic value and load characteristic value are extracted and compared with the preset high temperature threshold and preset high load threshold, respectively. The element combination with the temperature characteristic value continuously exceeding the preset high temperature threshold and the load characteristic value continuously exceeding the preset high load threshold is selected. These high-risk condition combinations are integrated to form a significant subset containing high temperature and high load combinations. The extraction of this significant subset enables precise focusing on high aging risk conditions in the key combination type, and eliminates low-risk normal temperature and low load condition combinations, making subsequent analysis and calculation more targeted and effectively improving the accuracy and efficiency of maintenance timing prediction.

[0055] The high-temperature range is the temperature range above a preset high-temperature threshold, i.e., the temperature operating range corresponding to the combination of elements within a significant subset. The continuous duration of the high-temperature range is the length of time that the metallurgical machinery and equipment continuously operates within this high-temperature range. Its calculation is based on timestamps, performing time-span statistics on the time series data continuously operating within the high-temperature range within the significant subset to obtain the cumulative continuous operating time. The number of load cycles from low to high is the cumulative number of times the load value of the metallurgical machinery and equipment rises from below the preset high-load threshold to above the preset high-load threshold level within the operating segment corresponding to the significant subset; that is, the number of times the load completes one cycle of fluctuation from low to high load. Its statistical process is based on the fluctuation characteristics of the load sequence data, identifying and counting each complete fluctuation process from low to high load. The high-temperature duration limit is the critical duration of continuous high-temperature operation preset based on the historical aging and failure data of the metallurgical machinery and equipment. This limit is set in conjunction with the variation patterns of degradation parameters such as grain boundary slip rate and creep rate of materials at high temperatures. For example, the high-temperature duration limit for a metallurgical furnace can be set to 5 hours, and the high-temperature duration limit for a rolling mill can be set to 2 hours. The load cycle trigger condition is a pre-set critical number of load cycles based on historical aging and failure data of metallurgical machinery and equipment. The setting of this trigger condition takes into account the cumulative fatigue damage of the equipment's mechanical components under load fluctuations. For example, the load cycle trigger condition for metallurgical equipment can be set to 5 times per hour or 30 times cumulatively. In practice, the process begins by using the time-series data corresponding to the significant subset to count the continuous duration of the high-temperature range, with timestamps as the axis. Then, by identifying the fluctuation characteristics of the load sequence data, the number of cyclic changes in the load from low to high is counted. Subsequently, the counted continuous duration of high temperature is compared with a preset limit value for the duration of high temperature, and the counted number of load cyclic changes is compared with a preset trigger condition for the number of load cyclic changes. Only when the continuous duration of high temperature exceeds the limit value and the number of load cyclic changes reaches the trigger condition is the significant subset deemed to meet the preset condition. This indicates that the metallurgical machinery and equipment are in a high-risk operating condition of continuous high temperature and frequent load fluctuations, and material deterioration and component fatigue damage are entering an accelerated accumulation stage. Further data fusion is needed to estimate the maintenance timing. If both conditions are not met simultaneously, the significant subset is deemed not to meet the preset condition, and no further maintenance timing estimation is required. This determination of the preset condition enables accurate identification of high-risk aging conditions of metallurgical equipment, avoiding over-maintenance under low-risk conditions and delayed maintenance under high-risk conditions.

[0056] When a significant subset simultaneously meets the preset conditions of high temperature duration exceeding the limit and load cycle count reaching the trigger condition, the operation of aligning the real-time temperature sequence data and real-time load sequence data of the metallurgical machinery equipment by timestamp and calculating the joint distribution characteristics of the three by fusing the superposition effect index is performed. The real-time temperature sequence data and real-time load sequence data are the temperature and load time series data of the metallurgical machinery equipment currently in operation, which are collected in real time by the sensor network and preprocessed by sliding window mid-value filtering and outlier removal. They can accurately reflect the real-time operating status of the equipment. Timestamp alignment means using the timestamp of data collection as the benchmark to make the real-time temperature sequence data and real-time load sequence data correspond one-to-one at the same time node, ensuring that the temperature value and load value at the same timestamp form an accurate matching relationship, avoiding errors in the fusion analysis results due to deviations in the time dimension. The superposition effect index is a quantitative index calculated based on continuous high temperature and high load period data when the interaction intensity value exceeds the preset threshold in the aforementioned steps. It characterizes the instantaneous impact response generated by the load change superimposed temperature peak and the degree of coupling amplification formed by material grain boundary slip and creep rate. It is the core index for quantifying the high temperature and high load coupled aging effect of metallurgical equipment. In practice, it is essential to ensure a high degree of consistency between temperature and load data over time. Then, the aligned real-time temperature and load sequences are fused with the superposition effect index. The superposition effect index, as a coupled aging characteristic value, is integrated into the fusion analysis of the temperature and load data to correct the aging characteristics of the real-time temperature and load data. Subsequently, mathematical statistical analysis methods are used to calculate the joint distribution characteristics of the three. The joint distribution characteristics are a set of features reflecting the statistical correlation between the real-time temperature sequence data, real-time load sequence data, and the superposition effect index, including statistical characteristics such as the mean, variance, covariance, and joint probability density. This joint distribution characteristic can comprehensively reflect the statistical correlation between the real-time temperature and load operating status and the coupled aging effect of metallurgical equipment under continuous high-temperature superposition load fluctuations.

[0057] The historical aging record database stores aging data, fault data, and crack propagation data of metallurgical machinery and equipment under different temperature and load conditions. It contains a large amount of historical operating records and crack propagation data under high-temperature and high-load coupled conditions similar to the current operating conditions. This data forms the foundation for the construction and calculation of crack propagation models. The crack propagation model is a material crack propagation calculation model based on Paris's Law, applicable to high-temperature and high-load coupled conditions in metallurgical machinery and equipment. Paris's Law is a classic law describing the quantitative relationship between the fatigue crack propagation rate and the stress intensity factor range of materials. The core expression of Paris's Law is: ,in, Where C is the crack propagation rate, and C and m are intrinsic material constants. The stress intensity factor range is defined by the principle that the crack propagation rate of a material is proportional to a power of the stress intensity factor range. This embodiment adapts and optimizes the Paris Law, incorporating the coupled influence of temperature and load to make it suitable for the high-temperature, high-load operating characteristics of metallurgical equipment. The final model acquisition requires parameter calibration and verification using a historical aging record database. This database retrieves complete operating records of key components of the same material and type from the metallurgical machinery under high-temperature, high-load coupled conditions, including temperature-load time-series data, crack propagation monitoring data, and fault records. This historical data is used to fit and train the optimized model, calibrating the correction coefficients for the metallurgical operating conditions and the correction values ​​of the material constants C and m under different temperature-load coupled scenarios. Simultaneously, verification samples from different operating conditions in the historical record database are selected, and the model calculation results are compared with actual crack propagation monitoring data. Model parameters are adjusted until the calculation error meets engineering requirements, thus verifying the model's effectiveness and ultimately obtaining a crack propagation model suitable for high-temperature, high-load coupled conditions in metallurgical machinery. The input data for this crack propagation model consists of crack propagation data from similar records in a historical aging database and the currently calculated joint distribution characteristics. The output data is the crack propagation rate on the surface of metallurgical machinery and equipment components. The stress intensity factor range is a physical quantity that reflects the stress field intensity at the crack tip of the material. Its calculation incorporates the effects of real-time temperature, real-time load, and superposition effect index. Increased temperature reduces the fracture toughness of the material, thus increasing the stress intensity factor range. Fluctuations in load increase the fatigue stress of the material, further amplifying the stress intensity factor range. The superposition effect index corrects the stress intensity factor range under high temperature and high load coupling, accurately reflecting the accelerating effect of coupled aging on crack propagation. In practice, the process begins by retrieving similar operating condition records from the historical aging record database that match the current joint distribution characteristics. Crack propagation model parameters, including material constants and stress intensity factor correction coefficients, are extracted from these similar records. The currently calculated joint distribution characteristics are then substituted into the adapted crack propagation model. The model calculates the crack propagation rate on the surface of the metallurgical machinery equipment components. This crack propagation rate is the length of crack propagation on the surface of the equipment components per unit time or per unit load cycle, accurately reflecting the rate of crack propagation under high temperature and high load coupling conditions. Subsequently, the critical crack size of the equipment components, i.e., the critical length of the crack when the equipment components fail, is combined with the crack propagation rate and the current crack size to calculate the remaining time from the current moment until the crack reaches the critical size. This remaining time is the predicted maintenance time for the metallurgical machinery equipment. The current crack size can be obtained in real time, including but not limited to non-destructive testing. The critical crack size is predetermined based on the material properties, mechanical structure, and operating conditions of the key components of the metallurgical machinery equipment and is an important size threshold for determining whether the equipment needs maintenance.

[0058] For example, Figure 2 The graph serves as a verification of the crack propagation model's rate prediction, accurately reflecting the crack propagation behavior of the material under coupled operating conditions. This verification result confirms that the method of estimating maintenance timing based on the crack propagation model by fusing the superposition effect index with real-time temperature and load data possesses high accuracy and reliability.

[0059] In this step, the calculation of maintenance timing predictions revolves around the aging characteristics of metallurgical machinery and equipment under high temperature and high load coupling. The core coupled aging characteristic, the superposition effect index, is integrated into the fusion analysis of real-time temperature and load data and the calculation of crack propagation rate. This breaks through the limitation of traditional crack propagation models that only consider a single load factor, making the calculated crack propagation rate more consistent with the actual aging condition of metallurgical equipment. Consequently, the obtained maintenance timing predictions are more accurate and practical, providing precise timing basis for subsequent maintenance decisions. This effectively avoids sudden failures caused by crack propagation under high temperature and high load coupling conditions, and realizes predictive maintenance of the aging state of metallurgical machinery and equipment.

[0060] Step S4: Generate an adjustment signal based on the maintenance timing prediction and transmit it to the control system of the metallurgical machinery and equipment. Collect the adjusted temperature sequence data and load sequence data respectively, and determine whether the trend of the adjusted sequence data is consistent with the expected direction of the aging rate estimate and the superposition effect index. If they are consistent, output the maintenance decision command.

[0061] The output maintenance decision instructions include:

[0062] Based on the predicted maintenance timing and combined with the superposition effect index, the adjustment range of temperature and load for metallurgical machinery and equipment is calculated, and a corresponding adjustment signal is generated. The adjustment signal is transmitted to the control system of the metallurgical machinery and equipment through a wireless communication interface. The adjusted temperature sequence data and load sequence data of the metallurgical machinery and equipment are collected in real time through a sensor network and preprocessed. The changing trend of the preprocessed temperature sequence data and load sequence data is analyzed. When the changing trend is consistent with the expected direction, a maintenance decision command is output. The maintenance decision command, combined with the superposition effect index, correlates the increase in load sensitivity due to material performance degradation at high temperature and the coupled aging response of vibration amplitude and temperature strain.

[0063] The analysis of the changing trends of the preprocessed temperature and load sequence data includes:

[0064] The preprocessed temperature and load sequences are timestamped and aligned using the time the control system receives the adjustment signal to form adjusted temperature and load sequences. Multi-dimensional trend features are extracted from the adjusted temperature and load sequences. The adjusted time period is divided into several continuous time slices. Multi-dimensional trend features are extracted from the adjusted temperature and load sequences within each time slice, and the coupling trend matching degree is calculated. The temporal evolution slope of the coupling trend matching degree is fitted, and the number of time slices that continuously meet the preset matching degree threshold is counted to determine the stability and evolution trend of the adjusted trend. For time slices where the trend deteriorates, root cause localization is performed using local fluctuation and abrupt change features. Simultaneously, the multi-dimensional trend features, coupling trend matching degree, temporal evolution slope, aging rate estimate, and superposition effect index are correlated and verified. Based on the verification results, the changing trends of the adjusted temperature and load sequences are output. The multi-dimensional trend features include trend slope, local fluctuation variance, peak percentage, number of abrupt changes, and abrupt change amplitude.

[0065] Specifically, this step uses the maintenance timing prediction as the core basis, combined with the superposition effect index to complete the generation and transmission of adjustment signals. By collecting, preprocessing and trend analysis of the adjusted temperature and load sequence data, the consistency between the data change trend and the estimated aging rate and the expected direction of the superposition effect index is verified. Finally, maintenance decision instructions that fit the actual aging state of metallurgical machinery and equipment are output. This process realizes closed-loop management from maintenance timing prediction to operating condition adjustment, trend verification and decision output. The data processing and trend analysis methods involved are all adapted to the high temperature and high load coupling operation characteristics of metallurgical equipment, ensuring the accuracy and practicality of maintenance decision instructions.

[0066] The predicted maintenance timing value is calculated based on the crack propagation model by integrating the superposition effect index, real-time temperature sequence data, and real-time load sequence data in the above steps. It reflects the remaining maintenance time of metallurgical machinery and equipment from the current moment until the crack in the equipment component reaches the critical size. It is the core quantitative indicator for determining the maintenance timing of equipment. The superposition effect index characterizes the instantaneous impact response caused by the superposition of load change and temperature peak, as well as the degree of coupling amplification formed by material grain boundary slip and creep rate. It can accurately reflect the accelerated aging of equipment under the coupling effect of high temperature and high load. The larger the value of this index, the more significant the acceleration effect of coupling effect on equipment aging. In practice, the adjustment range of temperature and load for metallurgical machinery and equipment is comprehensively calculated by combining the time frame of the maintenance opportunity prediction value and the magnitude of the superposition effect index. For high-risk operating conditions with short maintenance opportunity prediction values ​​and large superposition effect index values, a larger adjustment range is adopted to quickly reduce the aging rate of the equipment. For general operating conditions with long maintenance opportunity prediction values ​​and small superposition effect index values, a moderate adjustment range is adopted to balance equipment operating efficiency and aging control. Based on the calculated temperature and load adjustment ranges, corresponding adjustment signals are generated. These adjustment signals are control instructions that include the temperature adjustment target value, the load adjustment target value, and the adjustment execution sequence, which can directly guide the metallurgical machinery and equipment control system to complete the operating condition adjustment.

[0067] The wireless communication interface is a wireless data transmission interface adapted to industrial production environments. It has the characteristics of anti-interference and high transmission rate, and can realize the stable and fast transmission of adjustment signals from the monitoring and analysis end to the equipment control end. The control system of metallurgical machinery equipment is the core control unit of the equipment. It is responsible for receiving and executing various control commands to achieve precise control of equipment operating parameters. After receiving the adjustment signal, the control system will adjust the target value according to the temperature and load in the signal, and regulate the actuators such as the heating system and transmission system of the equipment, thereby changing the operating conditions of the equipment, achieving precise adjustment of temperature and load, and reducing the aging effect of high temperature and high load coupling on the equipment.

[0068] The sensor network is a data acquisition network composed of various types of sensors deployed in key parts of metallurgical machinery and equipment, including temperature sensors and load sensors. It can collect temperature and load change data of key parts of the equipment in real time after adjustment, and form adjusted temperature sequence data and load sequence data. This acquisition process uses the same acquisition frequency as the original data acquisition to ensure that the data can fully reflect the real-time operating status of the equipment after adjustment. The preprocessing operations performed on the collected adjusted temperature and load sequence data are consistent with the preprocessing method for the clean operation dataset in step S1 above. Sliding window mid-value filtering and outlier removal are performed sequentially. Sliding window mid-value filtering is a non-linear signal processing method. Its working principle involves selecting a fixed-size sliding window, sorting all data points within the window by numerical value, and taking the median value as the filtered output value at the center of the window. The window is then slid along the time axis, and the above operation is repeated. In this step, the input data are the collected original adjusted temperature and load sequence data, and the output data is the filtered temperature and load sequence data. This method effectively removes pulse noise caused by instantaneous impacts on the sensor and on-site interference during the adjustment of operating conditions, while preserving the trend characteristics of the sequence data. In this embodiment, the default size of the sliding window is 5 data points. In the high-frequency vibration operating environment of metallurgical machinery, the window size can be adjusted to 7 data points to better adapt to the sudden load changes of the equipment. Outlier removal adopts a mean-based method. The standard deviation method first calculates the mean and standard deviation of the filtered sequence data. An outlier identification threshold is set to the mean plus or minus three times the standard deviation. Data points exceeding this threshold are identified as outliers. For temperature sequence data processing of high-temperature special metallurgical equipment such as steelmaking furnaces and blast furnaces, to retain more effective data of high-temperature peaks, the outlier identification threshold can be adjusted to the mean plus or minus 2.5 times the standard deviation. For identified outliers, linear interpolation is used to fill the missing data after removal, ensuring the continuity of the sequence data. The input data for this preprocessing operation is the temperature and load sequence data filtered through a sliding window, and the output data is the clean and adjusted sequence data after outlier removal and interpolation filling, ensuring the accuracy of subsequent trend analysis.

[0069] The maintenance decision-making instructions, combined with the superposition effect index, correlate the increased load sensitivity of material performance degradation under high temperature and the coupled aging response of vibration amplitude and temperature strain. The increased load sensitivity of material performance degradation under high temperature refers to the significant decline in the performance of metallurgical equipment materials under high temperature environments. At this time, the material's response to load changes becomes more sensitive, and even small load fluctuations can lead to a significant increase in the aging rate of the material. The coupled aging response of vibration amplitude and temperature strain refers to the coupling between the strain generated by vibration during equipment operation and the thermal strain under high temperature environments, forming a superposition effect that further accelerates grain boundary slip and creep of the material, exacerbating the aging process of the equipment. The integration of the superposition effect index allows the maintenance decision-making instructions to more accurately match the coupled aging characteristics, ensuring that the instructions can effectively alleviate equipment aging under the coupled effects of high temperature and high load.

[0070] Specifically, timestamp alignment uses the time when the metallurgical machinery equipment control system receives the adjustment signal as the reference time point to calibrate the time axis of the preprocessed temperature sequence data and load sequence data, ensuring that the temperature value and load value at the same timestamp correspond one-to-one, eliminating time dimension deviations caused by sensor acquisition delays and data transmission time differences. The input data for this operation are the preprocessed clean and adjusted temperature sequence data and load sequence data, and the output data are the adjusted temperature and load sequences after accurate timestamp alignment, providing a time dimension consistent data source for the extraction of multi-dimensional trend features.

[0071] Multi-dimensional trend features include trend slope, local fluctuation variance, peak percentage, number of mutations, and mutation amplitude. This provides a comprehensive quantitative representation of the changing trends of the adjusted temperature and load sequences from three dimensions: global change, local fluctuation, and instantaneous mutation. The trend slope is obtained by linearly fitting the adjusted temperature and load sequences using the least squares method, representing the global direction and rate of change. A positive trend slope indicates an upward trend, while a negative slope indicates a downward trend; a larger absolute value indicates a faster rate of change. Local fluctuation variance is calculated by traversing the sequence within a sliding time window, representing the degree of fluctuation within a local time period; a larger variance value indicates more severe local fluctuations. Peak percentage is the proportion of peak data points exceeding the adjustment target value within the sliding time window, representing the degree of deviation between the local trend and the adjustment target. The number of mutations and mutation amplitude are calculated using the first-order difference method, which calculates the difference between adjacent data points in the sequence to obtain a difference sequence. The mutation threshold is set to the mean of the difference sequence plus or minus 3. The standard deviation is multiplied by 1, and the difference results exceeding the threshold are judged as mutations. The mutation number is the total number of mutations in the adjusted time period, and the mutation amplitude is the numerical difference of each mutation, which can characterize the instantaneous abnormal changes in the sequence. In this embodiment, the overall running period after equipment adjustment is divided into several continuous time slices. The duration of the time slice is adapted to the operating conditions of the metallurgical machinery and equipment. For example, the time slice duration of the metallurgical furnace is set to 5 minutes, and the time slice duration of the rolling mill is set to 1 minute. Multi-dimensional trend features are extracted from the adjusted temperature and load sequences in each time slice, and then the coupling trend matching degree is calculated based on the extracted features. The coupling trend matching degree is a quantitative indicator that characterizes the degree of matching between the coordinated change trend of temperature and load sequences and the adjustment target. Its calculation process is based on the interaction strength value of temperature and load and the superposition effect index to determine the weight coefficient of temperature and load. The matching degree between the single sequence trend of temperature and load and the adjustment target is calculated separately. Then, a fluctuation penalty coefficient is introduced to penalize sequences with excessive local fluctuations. Finally, a comprehensive coupling trend matching degree is obtained. The value range of this indicator is [0,1]. The larger the value, the higher the degree of matching between the coupling trend and the adjustment target.After calculating the coupling trend matching degree for each time slice, the least squares method is used to linearly fit the coupling trend matching degree of all time slices to obtain the temporal evolution slope of the coupling trend matching degree. This slope can characterize the overall change trend of the coupling trend matching degree. A positive slope indicates that the coupling trend gradually optimizes over time, while a negative slope indicates that the coupling trend gradually deteriorates over time. At the same time, the number of time slices that continuously meet the preset matching degree threshold is counted. The preset matching degree threshold is set according to the historical operating data and aging control requirements of the metallurgical equipment, and is usually taken as 0.6. The higher the proportion of the number of time slices that continuously meet this threshold to the total number of time slices, the better the stability of the adjusted trend. By combining the temporal evolution slope and the number of consecutive effective matching time slices, a comprehensive judgment on the stability and evolution trend of the adjusted trend can be made.

[0072] The time slice for trend deterioration is the time slice where the coupling trend matching degree is lower than the preset matching degree threshold. Root cause localization is performed by combining the local fluctuation and abrupt change characteristics within this time slice, i.e., determining whether the trend deterioration is caused by excessive local fluctuations in the temperature sequence, instantaneous abrupt changes in the load sequence, or insufficient coordination between the temperature and load sequences. This provides a specific basis for optimizing subsequent maintenance decision-making instructions. The aging rate estimate includes the current aging rate of the metallurgical machinery and equipment, the matching degree between the current operating conditions and similar historical operating conditions, and the decreasing gradient of the equipment's remaining life under alternating high temperature and high load. It is a core indicator characterizing the overall aging state of the equipment. Its expected direction is a slowdown in the aging rate and a reduction in the decreasing gradient of the remaining life. The expected direction of the superposition effect index is a decrease in value, i.e., a weakening of the instantaneous impact response of load abrupt changes and temperature peaks, and a reduction in the coupling amplification degree of material grain boundary slip and creep rate. The extracted multi-dimensional trend features, coupling trend matching degree, temporal evolution slope, aging rate estimate, and superposition effect index are correlated and verified. Specifically, it verifies whether the rate of change and stability of the adjusted temperature and load sequences match the decreasing gradient of the aging rate estimate, and whether the evolution trend of the coupling trend is consistent with the expected change direction of the superposition effect index. If the adjusted temperature and load sequences show a downward trend with small fluctuations, the coupling trend matching degree is high and shows an upward trend, and it is consistent with the expected direction of the aging rate estimate and the superposition effect index, then the adjusted trend is determined to be consistent with the expected direction; otherwise, it is determined to be inconsistent. Based on the results of this correlation verification, the final output is the trend analysis results of the adjusted temperature sequence data and load sequence data.

[0073] When the trend analysis results show that the changing trends of the adjusted temperature sequence data and load sequence data are consistent with the expected directions of the aging rate estimate and the superposition effect index, it indicates that the adjustment signal generated based on the maintenance timing prediction and the superposition effect index can effectively reduce the aging rate of metallurgical machinery and equipment, and alleviate the accelerated aging effect under the coupling effect of high temperature and high load. At this time, a maintenance decision instruction is output. This maintenance decision instruction is a specific maintenance instruction that fits the actual aging state and operating conditions of the equipment. It includes operation requirements such as load reduction, shutdown for maintenance, and continuous adjustment of operating conditions. Moreover, the instruction, combined with the superposition effect index, accurately correlates the increase in load sensitivity due to material performance degradation under high temperature and the coupled aging response of vibration amplitude and temperature strain. It can guide on-site operation and maintenance personnel to complete targeted maintenance operations, realize predictive maintenance of metallurgical machinery and equipment, effectively avoid sudden failures caused by accelerated aging of equipment under high temperature and high load coupling conditions, and improve the operational reliability and service life of the equipment.

[0074] For example, Figure 3 This is a trend matching degree evolution and root cause localization diagram. The diagram is divided into two parts: the upper subplot shows the curve of coupling trend matching degree changing over time, and the lower subplot shows the corresponding temperature and load sequence curves. Figure 3 Taking a 60-minute monitoring period as an example, the time slices of trend deterioration (areas with a matching degree below 0.7) were clearly marked, and the key root causes leading to trend deterioration—temperature peaks (around 20 minutes) and load mutations (around 21 minutes)—were located within this interval. Through the linkage analysis of the upper and lower subplots, the strong correlation between the decrease in trend matching degree and abnormal temperature and load events can be intuitively verified, confirming the accuracy and effectiveness of this method in post-adjustment trend monitoring and root cause tracing.

[0075] The above describes an online monitoring and predictive maintenance method for metallurgical machinery and equipment according to an embodiment of this application. The following describes an online monitoring and predictive maintenance system for metallurgical machinery and equipment according to an embodiment of this application. Please refer to [link / reference]. Figure 4 One embodiment of the online monitoring and predictive maintenance system for metallurgical machinery and equipment in this application includes:

[0076] The data extraction unit is used to acquire the preprocessed clean operation dataset, determine the interaction intensity value between temperature and load based on the clean operation dataset, and if the interaction intensity value exceeds the preset intensity threshold, extract the time period data that meets the conditions of continuous high temperature and continuous high load from the clean operation dataset, and determine the superposition effect index after performing fusion analysis on the time period data.

[0077] The aging estimation unit is used to determine the temperature load combination sequence based on the superposition effect index, retrieve similar records from the historical aging record database to obtain the aging rate estimate, and determine the key combination type based on the distance relationship between the aging rate estimate and the dynamic relationship feature vector.

[0078] The maintenance prediction unit is used to extract a significant subset containing high temperature and high load combinations from key combination types. If the significant subset meets the preset conditions, the maintenance timing prediction value is obtained by fusing the superposition effect index, the real-time temperature sequence data of metallurgical machinery and equipment and the real-time load sequence data.

[0079] The maintenance decision unit is used to generate adjustment signals based on the predicted maintenance timing and transmit them to the control system of the metallurgical machinery and equipment. It collects the adjusted temperature sequence data and load sequence data respectively, and judges whether the changing trend of the adjusted sequence data is consistent with the expected direction of the aging rate estimate and the superposition effect index. If they are consistent, it outputs the maintenance decision command.

[0080] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0081] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0082] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A method for online monitoring and predictive maintenance of metallurgical machinery and equipment, characterized in that, The method includes: Obtain the preprocessed clean operation dataset, determine the interaction intensity value between temperature and load based on the clean operation dataset, if the interaction intensity value exceeds the preset intensity threshold, extract the time period data that meets the conditions of continuous high temperature and continuous high load from the clean operation dataset, and determine the superposition effect index after performing fusion analysis on the time period data. Based on the superposition effect index, a temperature load combination sequence is determined. Similar records are retrieved from the historical aging record database to obtain an aging rate estimate. The key combination type is determined based on the distance relationship between the aging rate estimate and the dynamic connection feature vector. Extract a significant subset containing high temperature and high load combinations from the key combination types. If the significant subset meets the preset conditions, then fuse the superposition effect index, the real-time temperature sequence data of metallurgical machinery and equipment and the real-time load sequence data to obtain the maintenance timing prediction value. An adjustment signal is generated based on the predicted maintenance timing and transmitted to the control system of the metallurgical machinery and equipment. The adjusted temperature sequence data and load sequence data are collected respectively, and it is determined whether the trend of the adjusted sequence data is consistent with the expected direction of the estimated aging rate and the superposition effect index. If they are consistent, a maintenance decision command is output.

2. The method according to claim 1, characterized in that, Obtain the preprocessed clean run dataset, including: Temperature and load sequence data of metallurgical machinery and equipment are collected in real time through a sensor network; sliding window median filtering is applied to the collected temperature and load sequence data; outlier removal is performed on the filtered temperature and load sequence data, and interpolation is used to fill the missing data after outlier removal; the temperature and load sequence data after outlier removal and interpolation are time-aligned to form the clean operation dataset.

3. The method according to claim 2, characterized in that, Determine the intensity of the interaction between temperature and load, including: A random forest algorithm is used to analyze the temperature and load sequence data in the clean operation dataset to construct an interaction model between temperature and load. Node splitting information for temperature values ​​and high load conditions is extracted from this model, and the co-occurrence frequency of both in the decision tree is calculated based on this information. According to the co-occurrence frequency, a variable importance metric is applied to quantify the impact weight of temperature rise on high load conditions. This impact weight is then combined with a nonlinear product function to calculate the interaction strength value between temperature and load. To address the accelerated degradation of load fluctuations within the high-temperature range, a load fluctuation subsequence within the high-temperature range is extracted, and the correlation coefficient between the fluctuation amplitude and the temperature peak is calculated. The tree averaging method of the random forest is used to incorporate this correlation coefficient into the calculation of the interaction strength value, forming a comprehensive interaction strength value.

4. The method according to claim 1, characterized in that, Determine the indices of cumulative effects, including: A high-temperature threshold and a high-load threshold are preset for metallurgical machinery and equipment. Time periods where the temperature continuously exceeds the high-temperature threshold and the load continuously exceeds the high-load threshold are identified and extracted from the clean operation dataset. The extracted time period data are time-series aligned, and a weighted average method is used to fuse the time-series aligned time period data to form a unified high-temperature, high-load time period dataset. The load mutation amplitude in the high-temperature, high-load time period dataset is calculated. The load mutation amplitude is multiplied by the temperature peak value, and combined with the time interval to obtain the instantaneous impact response value. The instantaneous impact response value is judged according to a preset level threshold to determine the superposition effect level. The superposition effect level is adjusted based on an empirical model of material grain boundary slip rate. A coupling model of material grain boundary slip and creep rate is introduced, and a coupling amplification factor is calculated. The instantaneous impact response value corresponding to the adjusted superposition effect level is multiplied by the coupling amplification factor to obtain the superposition effect index.

5. The method according to claim 1, characterized in that, The estimated aging rate is obtained, including: The current temperature sequence data and load sequence data of the metallurgical machinery and equipment are integrated by timestamp, and combined with the superposition effect index to form a temperature-load combination sequence with coupling effect. The temperature-load combination sequence is then subjected to sliding window mid-range filtering. The similarity between the processed temperature-load combination sequence and the historical combination sequences with coupling effect in the historical aging record library is calculated using the Euclidean distance algorithm, and a subset of records is initially screened based on the similarity. After normalizing the subset of records, the cosine similarity algorithm is used to calculate the matching degree between the processed temperature-load combination sequence and each historical combination sequence in the subset of records, and the matching similar records are determined based on the matching degree. An aging rate estimate is extracted from the matching similar records. The aging rate estimate includes the current aging rate of the metallurgical machinery and equipment, the matching degree between the current working condition and the historical similar working condition, and the decreasing gradient of the equipment's residual life under high temperature and high load alternation.

6. The method according to claim 1, characterized in that, Identify key combination types, including: The estimated aging rate is combined with the superposition effect index to construct a multidimensional numerical vector. The average temperature, load fluctuation variance, and high temperature duration ratio are extracted from the real-time operation data of metallurgical machinery and equipment to construct a dynamic relationship feature vector. The Euclidean distance between the multidimensional numerical vector and the dynamic relationship feature vector is calculated. The Euclidean distance is used to classify the equipment into three states: healthy operation, early warning, and imminent failure, based on the preset classification boundary. The key combination type is determined according to the classification results of the equipment states.

7. The method according to claim 1, characterized in that, The predicted maintenance timing values ​​include: Iterate through all elements of the key combination type, and filter out element combinations whose temperature values ​​exceed a preset high-temperature threshold and whose load values ​​exceed a preset high-load threshold to form a significant subset containing high-temperature and high-load combinations; calculate the continuous duration of the high-temperature interval within the significant subset, count the number of cyclic changes in load from low to high within the significant subset, and determine whether the high-temperature duration exceeds the limit value and whether the load cycle count reaches the trigger condition based on the continuous duration and the number of cyclic changes; when both conditions are met, align the real-time temperature sequence data and real-time load sequence data of the metallurgical machinery and equipment by timestamp, and calculate the joint distribution characteristics of the three by fusing the superposition effect index; based on the crack propagation model of similar records in the historical aging record library, combine the joint distribution characteristics to estimate the surface crack propagation rate of equipment components, and obtain the predicted value of the maintenance timing of the metallurgical machinery and equipment based on the surface crack propagation rate.

8. The method according to claim 1, characterized in that, Output maintenance decision instructions, including: Based on the predicted maintenance timing and combined with the superposition effect index, the adjustment range of temperature and load of the metallurgical machinery and equipment is calculated, and a corresponding adjustment signal is generated. The adjustment signal is transmitted to the control system of the metallurgical machinery and equipment through a wireless communication interface. The adjusted temperature sequence data and load sequence data of the metallurgical machinery and equipment are collected in real time through a sensor network and preprocessed. The changing trend of the preprocessed temperature sequence data and load sequence data is analyzed. When the changing trend is consistent with the expected direction, a maintenance decision command is output. The maintenance decision command is combined with the superposition effect index to correlate the increase in load sensitivity due to material performance degradation at high temperature and the coupled aging response of vibration amplitude and temperature strain.

9. The method according to claim 8, characterized in that, Analyze the changing trends of the preprocessed temperature and load series data, including: The preprocessed temperature and load sequence data are timestamped and aligned based on the time when the control system receives the adjustment signal, forming adjusted temperature and load sequences. Multi-dimensional trend features are extracted from the adjusted temperature and load sequences. The adjusted time period is divided into several continuous time slices. The multi-dimensional trend features are extracted from the adjusted temperature and load sequences within each time slice, and the coupling trend matching degree is calculated. The temporal evolution slope of the coupling trend matching degree is fitted, and the number of time slices that continuously meet the preset matching degree threshold is counted to determine the stability and evolution trend of the adjusted trend. For time slices where the trend deteriorates, root cause localization is performed by combining local fluctuation features and abrupt change features. Simultaneously, the multi-dimensional trend features, coupling trend matching degree, temporal evolution slope, aging rate estimate, and superposition effect index are correlated and verified. Based on the verification results, the changing trends of the adjusted temperature and load sequence data are output.

10. An online monitoring and predictive maintenance system for metallurgical machinery and equipment, used to implement the online monitoring and predictive maintenance method for metallurgical machinery and equipment as described in any one of claims 1-9, characterized in that, The system includes: The data extraction unit is used to acquire the preprocessed clean operation dataset, determine the interaction intensity value between temperature and load based on the clean operation dataset, and if the interaction intensity value exceeds the preset intensity threshold, extract the time period data that meets the conditions of continuous high temperature and continuous high load from the clean operation dataset, and determine the superposition effect index after performing fusion analysis on the time period data. The aging estimation unit is used to determine the temperature load combination sequence based on the superposition effect index, retrieve similar records from the historical aging record database, obtain an aging rate estimate, and determine the key combination type based on the distance relationship between the aging rate estimate and the dynamic connection feature vector. The maintenance prediction unit is used to extract a significant subset containing high temperature and high load combinations from the key combination types. If the significant subset meets the preset conditions, the superposition effect index, the real-time temperature sequence data of the metallurgical machinery and equipment and the real-time load sequence data are fused to obtain the maintenance timing prediction value. The maintenance decision unit is used to generate an adjustment signal based on the predicted maintenance timing and transmit it to the control system of the metallurgical machinery and equipment. It collects the adjusted temperature sequence data and load sequence data respectively, and determines whether the trend of the adjusted sequence data is consistent with the expected direction of the estimated aging rate and the superposition effect index. If they are consistent, it outputs a maintenance decision command.

Images