Predictive maintenance system for graphite electrode production equipment

Abstract

The invention discloses a predictive maintenance system for graphite electrode production equipment, and relates to the technical field of industrial automation control, and the system comprises a cloud management center which is in communication connection with the following modules: a sensor monitoring configuration module which is used for deploying a main thermocouple and a slave acoustic / infrared node in each high-temperature area of a graphitization furnace, a high-temperature area dual-redundancy sensing network is formed, and equipment operation data in the graphite electrode production process are collected; and the temperature measurement error compensation module is used for calibrating the acoustic temperature measurement error when the acoustic temperature measurement is switched to. Through combination of decentralized control and the ant colony algorithm, acoustic temperature measurement errors can be analyzed, a minimum error path is dynamically searched among multiple sensors, the problem of observation blind areas caused by failure of a single sensor is effectively solved, the accuracy and reliability of temperature measurement are remarkably improved, and the method is suitable for popularization and application. Therefore, the prediction precision of the potential fault of the graphite electrode production equipment is improved, and the false alarm and the missing alarm are reduced.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of industrial automation control, and in particular to a predictive maintenance system for graphite electrode production equipment. BACKGROUND

[0002] Graphite electrodes are key materials for transmitting electric current to generate high temperature to melt iron ore. Graphite electrodes are mainly used for electric arc furnace steelmaking, especially in the recycling process of scrap steel, which is melted into liquid steel by the high temperature of the electric arc furnace. Therefore, graphite electrode production equipment plays an important role in the steel and metallurgical industries, especially in the process of electric arc furnace steelmaking. The quality of graphite electrodes directly affects the efficiency of steelmaking, energy consumption and the quality of the final product. With the increase of production capacity, the running intensity of the equipment increases, and the probability of failure increases. If the abnormality of the equipment cannot be detected in time, it may cause production line shutdown, equipment damage or safety accidents. Through predictive maintenance, potential failures of graphite electrode production equipment can be predicted, and appropriate measures can be taken in advance to avoid sudden shutdown or major failures.

[0003] In the prior art, the high temperature of the graphite electrode production equipment (graphitization furnace) can greatly reduce the service life of the thermocouple and cause sensor failure. Once the system detects that the sensor is offline, it can only give up the prediction of the area and default to health, which is easy to cause observation blind area and has major safety problems. Therefore, how to analyze the acoustic temperature measurement error by combining decentralized control with ant colony algorithm, and use ants to find the minimum error path among multiple sensors to switch the sensor failure in a short time to avoid electrode burnout accidents caused by observation blind area and improve the accuracy of predicting potential failures of graphite electrode production equipment is a problem to be solved by the present application. Therefore, a predictive maintenance system for graphite electrode production equipment is proposed. SUMMARY

[0004] The present application aims to provide a predictive maintenance system for graphite electrode production equipment to solve the problems raised in the background art.

[0005] To solve the above technical problems, the technical solution adopted by the present application is: A predictive maintenance system for graphite electrode production equipment, comprising a cloud management center, the cloud management center being communicatively connected with the following modules, wherein: A sensor monitoring configuration module is used to deploy master thermocouples and slave acoustic / infrared nodes in each high temperature area of the graphitization furnace to form a double-redundancy sensing network in the high temperature area, collect equipment operation data during the graphite electrode production process, and eliminate the observation blind area caused by single sensor failure. A temperature measurement error compensation module is configured to calibrate acoustic temperature measurement errors when switching to acoustic temperature measurement, improve the reliability of redundant sensor data, and convert acoustic temperature measurement errors into pheromone evaporation factors of the ant colony algorithm. An ant colony path search engine is configured to simulate the foraging behavior of ants to find the minimum error path among multiple sensors based on the analysis results of acoustic temperature measurement errors, so as to achieve the search goal of minimizing the root mean square error of fused temperature and real temperature. A real-time weighted fusion module is configured to perform weighted averaging on multi-sensor data including thermocouples, acoustics and infrared based on the minimum error path output by the ant colony algorithm, to generate an optimal soft measurement temperature, realize seamless switching when the sensor fails, and meet the process control requirements. A device fault prediction engine is configured to input the obtained optimal soft measurement temperature into a pre-trained XGBoost fault prediction model, to identify potential fault risks including electrode overheating and cracking in advance, and provide decision support for maintenance personnel according to the fault prediction results.

[0006] Further improvements of the technical scheme of the present application are that the sensor monitoring configuration module comprises a sensor deployment monitoring unit and a distributed edge computing node unit. The sensor deployment monitoring unit is configured to deploy master thermocouples (traditional temperature measurement) and slave acoustic / infrared nodes (redundant temperature measurement) in each high-temperature zone of the graphitization furnace synchronously, form a dual-redundancy perception network in the high-temperature zone, and collect device operation data of the graphitization furnace during the production process of the graphite electrode. The distributed edge computing node unit is configured to deploy edge computing devices locally in each high-temperature zone, process device operation data in real time, and perform preliminary fault diagnosis.

[0007] Further improvements of the technical scheme of the present application are that the sensor deployment monitoring unit specifically comprises: The master thermocouples and the slave acoustic / infrared nodes are configured synchronously in each high-temperature zone of the graphitization furnace, forming a dual-redundancy perception network in the high-temperature zone, and realizing millisecond-level response and high-temperature resistance complementation through the redundant temperature measurement system. The deployed multi-modal sensors are used to collect device operation data of the graphitization furnace during the production process of the graphite electrode, wherein the master thermocouples convert thermoelectric potential into temperature signals, the acoustic nodes calculate temperature through a sound speed model, and the infrared nodes scan the temperature field, and each sensor locally completes the preprocessing operation of filtering and amplification. The three-sensor data are synchronized to the distributed edge computing node unit in real time at a period of 100 ms through an industrial Ethernet, ensuring time alignment and supporting data continuity and redundant switching during fault switching.

[0008] Further improvements of the technical scheme of the present application are that the distributed edge computing node unit specifically comprises: The edge computing device receives the multi-sensor collected device operation data, checks the integrity and eliminates abnormal frames, performs pre-processing operations of cold end compensation, filtering and bad point repair, and aligns the timestamps to generate unified temperature measurement data packets; Based on the time synchronization data of the temperature measurement data packet, weighted fusion is performed, sensor anomalies are identified through threshold detection and trend analysis, and a redundancy switching mechanism is triggered to ensure monitoring continuity; The device operation data and diagnostic results are stored in time sequence to a local solid state disk, and when the device state is abnormal, an alarm signal is sent to the cloud management center through an industrial Ethernet, and a local protection action is synchronously performed.

[0009] The further improvement of the technical scheme of the application is that the temperature measurement error compensation module comprises an acoustic temperature measurement error analysis unit and an error pheromone mapping unit. The acoustic temperature measurement error analysis unit is used to analyze the error of acoustic temperature measurement when the main thermocouple fails and the system automatically switches to acoustic temperature measurement, compare the acoustic temperature measurement data with the historical thermocouple data, calculate the range of acoustic temperature measurement error, compress the acoustic temperature measurement error from ±25℃ to within ±10℃, and improve the availability of redundant data. The error pheromone mapping unit is used to use the ant colony algorithm to regard the acoustic temperature measurement error as an information pheromone volatilization factor, guide path optimization, and determine a mapping rule, that is, the greater the acoustic temperature measurement error, the faster the information pheromone volatilization speed (indicating that the sensor data has low reliability), and then adjust the information pheromone concentration according to the real-time acoustic temperature measurement error.

[0010] The further improvement of the technical scheme of the application is that the acoustic temperature measurement error analysis unit specifically comprises: After the system detects that the main thermocouple fails, the acoustic temperature measurement mode is automatically switched, the acoustic node collects the acoustic signal and calculates the propagation speed, the historical main thermocouple data is synchronously retrieved, and after ensuring that the two are time-aligned, they are synchronized to the edge computing device. The edge computing device compares the acoustic temperature measurement data with the historical main thermocouple data, calculates the temperature difference and the range of acoustic temperature measurement error, analyzes the influence of the pressure in the furnace and the fluctuation of the gas composition on the acoustic velocity model, and determines the main source of error. Based on the range of acoustic temperature measurement error, the parameters of the acoustic velocity-temperature model are adjusted, the error correction program is started, and through the triple verification of time consistency, spatial consistency and physical boundary constraints, the error range is compressed to within ±10℃, and is marked as reliable data for subsequent monitoring.

[0011] The further improvement of the technical scheme of the application is that the error pheromone mapping unit specifically comprises: The acoustic temperature measurement error is converted into pheromone evaporation rate through a nonlinear function, the larger the acoustic temperature measurement error is, the faster the evaporation is, and a pheromone deposition mechanism is introduced, when the acoustic temperature measurement error is lower than a preset error threshold, the concentration is supplemented, the pheromone concentration and the evaporation rate are updated in real time, and a self-adaptive feedback loop is formed; The weighted decision network is constructed on the edge computing device, the pheromone concentration is used as the weight, the ants preferentially select the path with high pheromone concentration, the exploration probability is adjusted according to the current error, the local optimum is avoided, the path weight is updated through iteration, the proportion of high-error sensors is reduced, and the contribution degree of low-error channels is improved; The corrected acoustic temperature measurement error distribution of each channel is periodically counted and fed back to the dynamic mapping model, if the system has long-term high error, the parameter adaptive adjustment is triggered, the adjustment parameter is optimized by the gradient descent method, overfitting is prevented, and a complete closed loop of 'error awareness-pheromone update-path optimization-parameter iteration' is formed.

[0012] The ant colony path search engine further comprises: When the system is started, the acoustic temperature measurement error data of each sensor in the high-temperature zone double-redundancy sensing network is used to convert the initial error data of each sensor into a pheromone basic value, the path with small error is given higher initial concentration, and the pheromone evaporation coefficient and the number of ants are set to build an initial search environment based on pheromone concentration; The ant starts from the starting sensor, selects the next node according to the path pheromone concentration, preferentially selects the path with small error, calculates the fusion temperature root mean square error of the path after traversal, the path with small error releases more pheromone, and all path pheromones are attenuated according to the evaporation coefficient, and low-quality paths are gradually eliminated; After multiple rounds of ant search and dynamic pheromone update, the pheromone concentration of the path with small error is continuously accumulated and enhanced, and the path with large error is gradually eliminated, and when the preset iteration number or convergence condition is reached, the minimum error path with the highest pheromone concentration and related parameters are output.

[0013] The real-time weighted fusion module further comprises: The minimum error path output by the ant colony algorithm is analyzed, the pheromone concentration is normalized into a dynamic weight through a Softmax function, the weight of the sensor with small error is high, the weight of the sensor with large error is low, the state of the sensor is monitored in real time, if a failed node is detected, path reconstruction is triggered immediately, the failed node is removed and the weight is redistributed to the effective sensors on the redundant path, and the real-time and robustness of weight distribution are ensured; The raw data of the main thermocouple, acoustic temperature measurement and infrared temperature measurement are time-aligned (interpolation or sliding window synchronization sampling time), space-aligned (mapping to a unified coordinate system to correct deviation), and preprocessed by outlier rejection (3 sigma criterion), noise filtering (Kalman or wavelet denoising) and dimensionless normalization to ensure data comparability; The preprocessed multi-sensor data is weighted and averaged according to dynamic weights to generate an optimal soft measurement temperature value, and the root mean square error of the optimal soft measurement temperature value is calculated in real time and compared with a preset threshold, and when the error exceeds the threshold, a feedback mechanism is triggered to adjust the ant colony algorithm parameters or reinitialize the search until the error meets the process requirements.

[0014] The further improvement of the technical scheme of the present application is that the device fault prediction engine specifically comprises: The optimal soft measurement temperature output by the real-time weighted fusion module is input into the pre-trained XGBoost fault prediction model, the non-linear relationship between temperature anomaly and fault mode is learned according to the gradient boosting tree, and the potential fault mode is identified, and the probability of occurrence of each potential fault is output in real time; According to the probability of occurrence of potential faults output by the XGBoost fault prediction model, decision support information is automatically generated, including the predicted fault location, type and targeted treatment guide, which is pushed to the maintenance personnel terminal in synchronization, and visual inspection suggestions are provided to clearly identify the key components that need to be checked; Based on the fault prediction result, a maintenance work order is automatically generated, which records the fault type, location and treatment measures in detail, and is pushed to the work order management system, and the maintenance personnel perform inspection and repair according to the instructions, and the processing result is fed back to update the model data, forming a closed loop of prediction-decision-execution-optimization, reducing unplanned downtime and prolonging the service life of the equipment.

[0015] Due to the adoption of the above technical scheme, the present application has the following technical progress compared with the prior art: 1、The present application provides a predictive maintenance system for graphite electrode production equipment, which can analyze acoustic temperature measurement errors and dynamically find the minimum error path among multiple sensors by combining decentralized control with ant colony algorithm, effectively solving the problem of observation blind area caused by single sensor failure, significantly improving the accuracy and reliability of temperature measurement, and thereby improving the prediction accuracy of potential faults of graphite electrode production equipment and reducing false positives and false negatives.

[0016] 2、The present application provides a predictive maintenance system for graphite electrode production equipment, which synchronously deploys a main thermocouple and a slave acoustic / infrared node in each high-temperature zone of a graphitization furnace to form a dual-redundancy sensing network in the high-temperature zone, and when the main thermocouple fails, the system can automatically switch to the slave node to ensure disturbance-free switching within 1 second, avoiding monitoring interruption caused by sensor failure and ensuring the continuity and stability of the production process, thereby reducing the production risk caused by equipment failure. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0018] Figure 1 This is a schematic diagram of the workflow of a predictive maintenance system for graphite electrode production equipment according to the present invention. Figure 2 This is a data flow diagram of a predictive maintenance system for graphite electrode production equipment according to the present invention. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] Example 1, as Figure 1 , Figure 2 As shown, this invention provides a predictive maintenance system for graphite electrode production equipment, including a cloud management center. The cloud management center is communicatively connected to the following modules, wherein: The sensor monitoring configuration module is used to deploy main thermocouples and slave acoustic / infrared nodes in each high-temperature zone of the graphitization furnace to form a dual-redundant sensing network in the high-temperature zone, collect equipment operation data of the graphite electrode production process, eliminate the observation blind spot caused by the failure of a single sensor, and deploy edge computing devices to perform preliminary fault diagnosis. The sensor monitoring configuration module includes a sensor deployment monitoring unit and a distributed edge computing node unit. The sensor deployment monitoring unit is used to synchronously deploy the main thermocouple (traditional temperature measurement) and the slave acoustic / infrared node (redundant temperature measurement) in each high-temperature zone of the graphitization furnace to form a high-temperature zone double-redundancy sensing network, and collect the equipment operation data of the graphitization furnace in the production process of the graphite electrode. The thermocouple directly contacts the temperature measurement, has a fast response but is easy to be oxidized and fail at high temperature. The acoustic node indirectly measures the temperature by analyzing the sound wave propagation speed in the furnace (which is strongly related to the temperature), is resistant to high temperature but needs to be calibrated. The infrared node non-contact temperature measurement has a wide coverage range but is easy to be disturbed by the dust in the furnace. The physical redundancy of the sensor is realized. When the main thermocouple fails, the system automatically switches to the slave node to ensure that the switching is not disturbed within 1 second. The main thermocouple and the slave acoustic / infrared node are synchronously configured in each high-temperature zone of the graphitization furnace to form a high-temperature zone double-redundancy sensing network. Through the redundant temperature measurement system, the millisecond-level response and the high-temperature resistance are complementary. The multi-modal sensors deployed are used to collect the equipment operation data of the graphitization furnace in the production process of the graphite electrode. The main thermocouple converts the thermoelectric potential into a temperature signal. The acoustic node calculates the temperature through the sound speed model. The infrared node scans the temperature field. Each sensor locally completes the preprocessing operation of filtering and amplification. The data of the three sensors are real-time synchronized to the distributed edge computing node unit at a period of 100 ms through the industrial Ethernet to ensure time alignment, support data continuity and redundant switching when the fault switching occurs. The sensor deployment monitoring unit specifically works as follows: in each high-temperature zone of the graphitization furnace, a main thermocouple (contact temperature measurement) and a slave acoustic / infrared node (non-contact redundant temperature measurement) are deployed synchronously to form a dual-redundant sensing network in the high-temperature zone, wherein the main thermocouple directly contacts the electrode or the furnace wall and realizes millisecond-level response through thermoelectric potential-temperature conversion, but is prone to oxidation failure due to long-term exposure to a high-temperature environment of 1500-3000°C and needs to be replaced regularly; the slave acoustic node collects sound wave signals in the furnace through a microphone array and indirectly calculates the temperature based on a sound speed-temperature model (v = 331.4 + 0.6T), which has high-temperature resistance but needs to be calibrated to eliminate the interference of furnace pressure and gas composition; the infrared node covers the whole scene through non-contact radiation temperature measurement, but needs to solve the problem of signal attenuation caused by dust shielding; the data of the three sensors are synchronously transmitted to an edge computing device in real time through an industrial Ethernet, and the sampling period is uniformly set to 100 ms to ensure time alignment and data continuity during fault switching; the main thermocouple is installed at the electrode connection and the key temperature measurement point of the furnace wall (4-6 units are deployed per high-temperature zone), and a B-type or R-type platinum-rhodium thermocouple (temperature resistance ≥ 3000°C) is selected, which has a response time of <50 ms (to reach 95% steady-state value) and adopts an alumina ceramic coating to slow down the oxidation rate, and its failure modes include short-term failure (thermal fuse of thermocouple wire) and long-term failure (thermal potential drift caused by oxidation (error > ±50°C)); the acoustic node adopts a microphone array: 4-8 channel MEMS microphone (frequency response 20 Hz-20 kHz), which adopts a multi-layer metal mesh structure for anti-interference design, and attenuates external noise by >20 dB; the detector of the infrared node is a mid-wave infrared (3-5 μm) refrigeration-type focal plane array (640×512 pixels), which has a horizontal field of view of 60° and a vertical field of view of 45°, covers the whole scene of a single high-temperature zone, and identifies dust shielding areas by comparing the changes in radiation intensity of consecutive frames to perform dust compensation, and performs two-point calibration (500°C / 2500°C) every 8 hours; the sensors deployed are used to collect the equipment operation data of the graphitization furnace during the production of the graphite electrode, wherein the main thermocouple monitors the furnace temperature in real time and converts the temperature signal into an electrical signal output; the acoustic node receives the sound wave signals in the furnace, analyzes the sound wave propagation speed through signal processing, and then calculates the temperature information; the infrared node performs non-contact temperature scanning on the furnace body to obtain temperature distribution data, and performs local preliminary processing on the collected sensor data, including filtering and amplification, to improve data quality; The distributed edge computing node unit is used for locally deploying an edge computing device in each high-temperature zone, processing device operation data in real time and performing preliminary fault diagnosis, locally storing historical device operation data, running a threshold detection algorithm, quickly identifying that a thermocouple is offline or data is abnormal, reducing communication delay, ensuring that the system can immediately trigger redundancy switching logic when a fault occurs, the edge computing device receives device operation data collected by multiple sensors, checks the integrity and removes abnormal frames, performs preprocessing operations such as cold-end compensation, filtering and bad point repair, and aligns timestamps to generate unified temperature measurement data packets, based on the temperature measurement data packets of time-synchronized data, performs weighted fusion, identifies sensor abnormalities through threshold detection and trend analysis, triggers a redundancy switching mechanism to ensure monitoring continuity, stores device operation data and diagnosis results in time sequence to a local solid-state hard disk, and sends an alarm signal to a cloud management center through an industrial Ethernet when it is confirmed that the device state is abnormal, and synchronously performs a local protection action; The specific working content of the distributed edge computing node unit is as follows: through the edge computing device locally deployed in each high-temperature zone, device operation data including a main thermocouple, an acoustic node and an infrared node are received, the device operation data is checked for integrity, data frames with transmission packet loss or format errors are removed, and local preprocessing is performed, including cold-end compensation and nonlinear correction of thermocouple signals, band-pass filtering and mutual power spectrum calculation of acoustic signals, and bad point repair and dynamic partitioning of infrared images, the preprocessed data is aligned according to timestamps to generate temperature measurement data packets in a unified format, reducing communication bandwidth occupancy; the edge computing device performs a weighted fusion algorithm to fuse temperature values based on the temperature measurement data packets of time synchronization, wherein the main thermocouple weight is 0.7, the acoustic weight is 0.2, and the infrared weight is 0.1, the fused temperature values are preliminarily judged for device state through a threshold detection algorithm, at the same time, abnormality of main thermocouple oxidation drift, acoustic signal interference or infrared emissivity mismatch is identified by comparing historical data trends, if the main thermocouple is detected to be offline or data is invalid, redundancy switching logic is immediately triggered to promote acoustic or infrared data to a main temperature measurement source, ensuring temperature monitoring continuity; the edge computing device stores device operation data and diagnosis results in time sequence to a local solid-state hard disk, supports at least 72 hours of historical data backtracking, meets fault tracing requirements, when it is confirmed that the device state is abnormal, an alarm signal is sent to a cloud management center through an industrial Ethernet when the temperature exceeds 3100℃ for 10 seconds, sensor failure: all temperature measurement source data is invalid, and a local protection action is synchronously performed, the local protection action includes cutting off electrode power supply (through a relay output), starting nitrogen purging (controlling a solenoid valve) and recording decision logs (including timestamps, trigger conditions and action types), wherein the whole process is completed in a closed loop on the edge side, communication delay is controlled within 10 ms, meeting real-time control requirements in high-temperature zones; The temperature measurement error compensation module is used for calibrating the acoustic temperature measurement error when switching to acoustic temperature measurement, improving the reliability of redundant sensor data, and converting the acoustic temperature measurement error into the pheromone evaporation factor of the ant colony algorithm. The temperature measurement error compensation module includes an acoustic temperature measurement error analysis unit and an error pheromone mapping unit. The acoustic temperature measurement error analysis unit is used for analyzing the error of acoustic temperature measurement when the main thermocouple fails and the system automatically switches to acoustic temperature measurement. By comparing the acoustic temperature measurement data with the historical thermocouple data, the range of acoustic temperature measurement error is calculated, and the acoustic temperature measurement error is compressed from ±25℃ to within ±10℃ to improve the availability of redundant data. After the system detects the failure of the main thermocouple, it automatically switches to the acoustic temperature measurement mode. The acoustic node collects acoustic signals and calculates the propagation speed. The historical main thermocouple data is synchronously retrieved, and after ensuring that the two are time-aligned, they are synchronized to the edge computing device. The edge computing device compares the acoustic temperature measurement data with the historical main thermocouple data, calculates the temperature difference, and calculates the range of acoustic temperature measurement error. The influence of the pressure and gas composition fluctuations in the furnace on the sound speed model is analyzed to determine the main source of error. Based on the range of acoustic temperature measurement error, the parameters of the sound speed-temperature model are adjusted, the error correction program is started, and through the triple verification of time consistency, spatial consistency, and physical boundary constraints, the error range is compressed to within ±10℃, and it is marked as reliable data for subsequent monitoring. The specific working content of the acoustic temperature measurement error analysis unit is as follows: when the main thermocouple fails, the system automatically switches to the acoustic temperature measurement mode. At this time, the acoustic node starts to collect the sound wave signals in the furnace and obtains the sound wave propagation speed data through the microphone array. At the same time, the historical main thermocouple temperature measurement data is called from the local storage to ensure that the acoustic temperature measurement data and the historical main thermocouple data are aligned in time and synchronized to the edge computing device through the industrial Ethernet, ensuring the integrity and time consistency of the data. In the edge computing device, the system compares and analyzes the acoustic temperature measurement data and the historical main thermocouple data collected, calculates the temperature difference at the same time point, determines the error value of the acoustic temperature measurement, and statistically analyzes the error values at multiple time points to calculate the error range of the acoustic temperature measurement, wherein the initial error range is usually about ±25℃. Through comparative analysis, the main sources of error are identified, including the influence of furnace pressure change and gas composition fluctuation on the sound velocity model. Based on the determined error range of the acoustic temperature measurement, the system starts the error correction program, adjusts the parameters in the sound velocity-temperature model, corrects and multi-dimensionally verifies the acoustic temperature measurement data, verifies from three dimensions including time consistency, spatial consistency and physical boundary constraint, compares the deviation of the compensated data from the historical trend through time consistency verification, and if the deviation of the compensated data from the historical trend through time consistency verification exceeds ±2σ (σ is the historical residual standard deviation) for 5 consecutive sampling points, the recalibration program is triggered. The spatial consistency verification cross- verifies the spatial uniformity of the acoustic temperature measurement through the partition temperature data of the infrared thermal imager (calibrated to ±5℃ accuracy), and if the regional temperature difference exceeds twice the acoustic data after compensation, it is marked as suspicious data. The physical boundary constraint combines the graphitization furnace process knowledge (the maximum allowed temperature is 3200℃), and applies a hard boundary limit to the compensated data. Finally, the acoustic temperature measurement error is compressed to within ±10℃, meeting the redundancy data availability requirement, reducing the influence of external factors on the acoustic temperature measurement, improving the accuracy and reliability of the acoustic temperature measurement, and then re-labeling the corrected acoustic temperature measurement data as reliable data for subsequent temperature monitoring and fault prediction, ensuring that the acoustic temperature measurement can provide sufficiently accurate temperature data when the main thermocouple fails, and ensuring the stable operation of the system. The error pheromone mapping unit is used to guide path optimization by taking the acoustic temperature measurement error as a pheromone evaporation factor by using an ant colony algorithm, and to determine a mapping rule, i.e., the greater the acoustic temperature measurement error, the faster the pheromone evaporation speed (indicating that the sensor data has low reliability), and then to adjust the pheromone concentration according to the real-time acoustic temperature measurement error, to quickly adapt to the sensor state change, to convert the acoustic temperature measurement error into the pheromone evaporation rate by using a nonlinear function, to make the greater the acoustic temperature measurement error, the faster the evaporation, and to introduce a pheromone deposition mechanism to supplement the concentration when the acoustic temperature measurement error is lower than a preset error threshold, to update the pheromone concentration and evaporation rate in real time, to form an adaptive feedback loop, to construct a weighted decision network on the edge computing device, to select a path with high pheromone concentration by the ant according to the pheromone concentration as the weight, and to adjust the exploration probability according to the current error to avoid local optimization, to update the path weight by iteration, to reduce the proportion of high-error sensors, to improve the contribution of low-error channels, to periodically count the acoustic temperature measurement error distribution of each channel after correction, and to feed back to the dynamic mapping model, and if the system has long-term high error, to trigger parameter adaptive adjustment, to optimize and adjust the parameters by using the gradient descent method, to prevent overfitting, and to form a complete closed loop of “error awareness-pheromone update-path optimization-parameter iteration”. The specific work of the error pheromone mapping unit is as follows: Based on the distributed optimization characteristics of the ant colony algorithm, the acoustic temperature measurement error is quantified into a pheromone evaporation factor, and a dynamic mapping model is constructed. Here, pheromone concentration is defined as a quantitative indicator of sensor data reliability, with its initial value determined by historical calibration data. The acoustic temperature measurement error is converted into an evaporation rate through a nonlinear function, ensuring that the larger the acoustic temperature measurement error, the faster the pheromone evaporates. Simultaneously, a pheromone deposition mechanism is introduced; when the acoustic temperature measurement error is lower than a preset error threshold, the concentration is replenished at a fixed rate to maintain system stability. The dynamic mapping model forms an adaptive feedback loop driven by the acoustic temperature measurement error by updating the pheromone concentration and evaporation rate in real time. Using pheromone concentration as the weight, the multi-sensor data fusion path is optimized. A weighted decision network is constructed on the edge computing device, where each node represents an acoustic temperature measurement channel. The edge weights are dynamically determined by the pheromone concentration, and the path selection rules of the ant colony algorithm are adopted. The ant (data stream) prioritizes paths with high pheromone concentrations and adjusts the exploration probability based on the current acoustic temperature measurement error to avoid getting trapped in local optima. By iteratively updating path weights, it automatically reduces the data proportion of high-error sensors and increases the contribution of low-error channels, achieving real-time adaptive adjustment of sensor states. An error-pheromone closed-loop feedback mechanism is established to continuously optimize mapping rules. Edge computing devices periodically collect the corrected acoustic temperature measurement error distribution of each channel and feed it back to the dynamic mapping model. If the system is in a high-error state for a long time, it triggers adaptive parameter adjustment. The gradient descent method is used to optimize the adjustment parameters, improving the matching degree between the evaporation rate and the acoustic temperature measurement error. At the same time, a historical data backtracking mechanism is introduced to compare the model performance under different operating conditions to prevent parameter overfitting. Finally, a complete closed loop of "error perception - pheromone update - path optimization - parameter iteration" is formed to ensure that the system can maintain stable temperature measurement accuracy when the sensor state changes dynamically. The expression for pheromone concentration is as follows: ; In the formula: For the current moment pheromone concentration, For the previous moment pheromone concentration, For the previous moment pheromone evaporation rate For the current moment The pheromone deposition rate is adjusted according to the acoustic temperature measurement error. When the acoustic temperature measurement error is below a preset error threshold, the concentration is replenished at a fixed rate; when the acoustic temperature measurement error is below a preset error threshold, the concentration is replenished at a fixed rate. As the concentration increases, the evaporation rate of pheromones increases. Enlargement, leading to Reduce; when acoustic temperature measurement error When the rate of pheromone evaporation decreases, decrease, and pheromone deposition supplement, resulting in increase; The expression of the evaporation rate is as follows: ; In the formula: is the evaporation rate of pheromone at the current moment , is a first adjustment parameter for controlling the size of the evaporation rate, is the acoustic thermometry error at the current moment , is a second adjustment parameter for controlling the sensitivity of the evaporation rate to the error; as the acoustic thermometry error increases, increases, indicating that the pheromone evaporates faster; as the acoustic thermometry error decreases, decreases, indicating that the pheromone evaporates slower; is limited to , indicating the difference between the thermometry value and the true value, the greater the error, the greater the pheromone evaporation rate ; the smaller the error, the smaller the evaporation rate of the pheromone ; The ant colony path search engine simulates the foraging behavior of ants among multiple sensors to find the minimum error path based on the analysis result of the acoustic thermometry error, wherein the sensor path with high pheromone concentration (small error) is preferentially selected, and the pheromone on the path evaporates over time (the path with large error is gradually eliminated), so as to achieve the search goal of minimizing the root mean square error of the fusion temperature and the true temperature; The real-time weighted fusion module is used for weighted averaging of the multi-sensor data including thermocouples, acoustics and infrared according to the minimum error path output by the ant colony algorithm, to generate an optimal soft measurement temperature, realize seamless switching when the sensor fails, and meet the process control requirements. The device fault prediction engine is used for inputting the obtained optimal soft measurement temperature into a pre-trained XGBoost fault prediction model, to identify potential fault risks including electrode overheating and cracking in advance, and provide decision support for maintenance personnel according to the fault prediction result.

[0021] In the embodiment 2, as shown in Figure 1 , Figure 2 the ant colony path search engine specifically comprises: When the system starts, the initial error data of each sensor is converted into the pheromone base value according to the acoustic temperature measurement error data of each sensor in the high-temperature zone double-redundancy sensing network, the path with small error is given a higher initial concentration, and the pheromone evaporation coefficient and the number of ants are set to build an initial search environment based on the pheromone concentration. The ants start from the starting sensor, select the next node according to the path pheromone concentration, preferentially select the path with small error, and calculate the root mean square error of the fusion temperature of the path after traversal. The path with small error releases more pheromone, and all path pheromones decay according to the evaporation coefficient, gradually eliminating low-quality paths. After multiple rounds of ant search and dynamic pheromone update, the pheromone concentration of the path with small error continuously accumulates and increases, and the path with large error is gradually eliminated. When the preset iteration number or convergence condition is reached, the minimum error path with the highest pheromone concentration and related parameters are output. The specific working content of the ant colony path search engine is: when the system starts, the pheromone concentration of each path connecting the sensors in the high-temperature zone double-redundancy sensing network is initialized, the initial error data of each sensor is converted into the pheromone base value according to the acoustic temperature measurement error analysis result, the path of the sensor with small error is given a higher initial pheromone concentration, and the path of the sensor with large error is given a lower initial pheromone concentration. At the same time, the pheromone evaporation coefficient and the number of ants are set, wherein the ant represents a search agent, simulates foraging behavior in the high-temperature zone double-redundancy sensing network, selects a path according to the pheromone concentration, and builds an initial search environment based on the pheromone concentration. The ant starts from the starting sensor, selects the next sensor node according to the pheromone concentration on the path, and the higher the concentration, the greater the probability of being selected, that is, the sensor path with small error is preferentially selected. After the ant traverses the high-temperature zone double-redundancy sensing network to form a complete path, the root mean square error of the fusion temperature corresponding to the path and the true temperature is calculated. If the root mean square error is small, the ant releases more pheromone on the path it passes through, enhancing the attractiveness of the path. Otherwise, it releases less. At the same time, all path pheromones gradually decrease over time according to the evaporation coefficient, gradually eliminating paths with large error and low pheromone concentration, guiding subsequent ants to preferentially select paths with small error and continuously optimizing the search direction. After multiple rounds of ant path search and pheromone update iterations, the pheromone concentration of each path is continuously adjusted. With the increase of the number of iterations, the pheromone concentration continuously accumulates and increases on the path with small error, and gradually decreases on the path with large error. When the preset iteration number or convergence condition is reached, the search process ends, the path with the highest pheromone concentration is determined as the minimum error path, and the minimum error path and related parameters are output. The real-time weighted fusion module specifically includes: The algorithm analyzes the minimum error path output by the ant colony algorithm and normalizes the pheromone concentration into dynamic weights using the Softmax function. Sensors with smaller errors have higher weights, while those with larger errors have lower weights. The sensor status is monitored in real time. If a failure node is detected, path reconstruction is immediately triggered to remove the failure node and reallocate weights to valid sensors on redundant paths, ensuring the real-time performance and robustness of weight allocation. The raw data from the main thermocouple, acoustic temperature measurement, and infrared temperature measurement are time-aligned (interpolation or sliding window synchronous sampling time) and spatially aligned (mapped to a unified coordinate system to correct deviations). Preprocessing operations such as outlier removal (3σ criterion), noise filtering (Kalman or wavelet denoising), and dimension normalization are performed to ensure data comparability. The preprocessed multi-sensor data are weighted and averaged according to the dynamic weights to generate the optimal soft measurement temperature value. The root mean square error of the optimal soft measurement temperature value is calculated in real time and compared with a preset threshold. If the threshold is exceeded, a feedback mechanism is triggered to adjust the ant colony algorithm parameters or re-initialize the search until the error meets the process requirements. The specific tasks of the real-time weighted fusion module are as follows: Based on the minimum error path output by the ant colony algorithm, it analyzes the sensor connection relationships and corresponding pheromone concentration distributions, normalizes the pheromone concentrations into dynamic weights (through the Softmax function), ensuring that sensors with smaller errors receive higher weights and sensors with larger errors have their weights suppressed. Simultaneously, it continuously monitors the sensor status; if a sensor failure is detected (data anomaly or communication interruption), it immediately triggers a path reconstruction mechanism, automatically removing the failed node and reallocating weights to valid sensors on redundant paths, ensuring the real-time performance and robustness of weight allocation. After weight allocation, it performs spatiotemporal alignment and preprocessing on the raw data from the main thermocouple, acoustic temperature measurement, and infrared temperature measurement. Alignment synchronizes the sampling times of different sensors through interpolation or sliding windows to eliminate time delay differences; spatial alignment maps the measurement positions of each sensor to a unified coordinate system to correct spatial distribution deviations. The preprocessing stage includes outlier removal (based on the 3σ criterion), noise filtering (Kalman filtering or wavelet denoising), and dimensional normalization to ensure that different physical quantities are comparable before fusion; the preprocessed multi-sensor data is weighted and averaged according to dynamic weights to generate the optimal soft measurement temperature value. During the fusion process, the root mean square error (RMSE) of the optimal soft measurement temperature value is calculated in real time and compared with a preset threshold. If the RMSE exceeds the limit, a feedback mechanism is triggered to adjust the ant colony algorithm parameters or re-initialize the search until the error meets the process requirements. The expression for the optimal soft-sensor temperature is as follows: ; In the formula: For optimal soft measurement temperature, For the first Dynamic weights of each sensor, For the first the pre-processed temperature value of the sensor; the expression of the root mean square error is as follows: ; wherein: the root mean square error, the number of data points, the optimal soft measurement temperature of the i-th data point, the true temperature of the i-th data point; the true temperature of the i-th data point; the true temperature of the i-th data point; The equipment fault prediction engine specifically comprises: The equipment fault prediction engine specifically comprises: The equipment fault prediction engine specifically comprises: The expression of the probability of the potential fault occurrence is as follows: ; ; In the formula: For the first The probability of a potential failure mode occurring. The optimal soft-measurement temperature is the temperature value output by the real-time weighted fusion module. For the XGBoost fault prediction model, the first Prediction scores for various failure modes This represents the total number of potential failure modes. This is an index variable used to iterate through all potential failure modes. The number of trees in the model. For the first tree to the first Weights of different failure modes For the first Among the trees The leaf node region where it is located For indicator functions, when Falling in the leaf node region The value is 1 if it is true, and 0 otherwise. The closer the value is to 1, the higher the probability that the failure mode will occur; The closer the value is to 0, the lower the probability.

[0022] The above are merely specific embodiments of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A graphite electrode production plant predictive maintenance system comprising a cloud management center, characterized by, The cloud management center is communicatively connected with the following modules, wherein: A sensor monitoring configuration module is configured to deploy a main thermocouple and a slave acoustic / infrared node in each high-temperature zone of the graphitization furnace to form a dual-redundancy sensing network in the high-temperature zone and collect device operation data in the production process of the graphite electrode; A temperature measurement error compensation module is configured to calibrate the acoustic temperature measurement error when switching to acoustic temperature measurement and convert the acoustic temperature measurement error into a pheromone evaporation factor of the ant colony algorithm; An ant colony path search engine is configured to simulate the foraging behavior of ants to find a minimum error path among multiple sensors based on the analysis result of the acoustic temperature measurement error, so as to achieve the search goal of minimizing the root mean square error of the fused temperature and the true temperature; A real-time weighted fusion module is configured to perform weighted averaging on the multi-sensor data including the thermocouple, the acoustic sensor and the infrared sensor according to the minimum error path output by the ant colony algorithm to generate an optimal soft measurement temperature; A device fault prediction engine is configured to input the optimal soft measurement temperature into a pre-trained XGBoost fault prediction model to identify potential fault risks in advance and provide decision support for maintenance personnel according to the fault prediction result.

2. A graphite electrode production plant predictive maintenance system according to claim 1, characterized in that: The sensor monitoring configuration module includes a sensor deployment monitoring unit and a distributed edge computing node unit; The sensor deployment monitoring unit is configured to deploy a main thermocouple and a slave acoustic / infrared node in each high-temperature zone of the graphitization furnace synchronously to form a dual-redundancy sensing network in the high-temperature zone and collect device operation data of the graphitization furnace in the production process of the graphite electrode; The distributed edge computing node unit is configured to deploy an edge computing device locally in each high-temperature zone to process the device operation data in real time and perform preliminary fault diagnosis.

3. A graphite electrode production plant predictive maintenance system according to claim 2, characterized in that: The sensor deployment monitoring unit specifically includes: A contact-type main thermocouple and a non-contact-type slave acoustic / infrared node are configured in each high-temperature zone of the graphitization furnace synchronously to form a dual-redundancy sensing network in the high-temperature zone; The deployed multi-modal sensors are used to collect device operation data of the graphitization furnace in the production process of the graphite electrode, wherein the main thermocouple converts thermoelectric potential into a temperature signal, the acoustic node calculates temperature through a sound speed model, and the infrared node scans a temperature field, and each sensor locally completes a preprocessing operation of filtering and amplification; The three-sensor data is synchronized to the distributed edge computing node unit in real time at a period of 100 ms through an industrial Ethernet.

4. The graphite electrode production plant predictive maintenance system of claim 2, wherein: The distributed edge computing node unit specifically includes: The edge computing device receives device operation data collected by the multi-sensor, checks the integrity and eliminates abnormal frames, performs preprocessing operations of cold end compensation, filtering and bad point repair, and aligns time stamps to generate a unified temperature measurement data packet; Based on the temperature measurement data packet of the time synchronization data, weighted fusion is performed, sensor abnormalities are identified through threshold detection and trend analysis, and a redundancy switching mechanism is triggered; The device operation data and the diagnosis result are stored in a local solid state disk in time sequence, and when the device state is abnormal, an alarm signal is sent to the cloud management center through an industrial Ethernet to perform a local protection action synchronously.

5. A graphite electrode production plant predictive maintenance system according to claim 2, characterized in that: The temperature measurement error compensation module includes an acoustic temperature measurement error analysis unit and an error pheromone mapping unit; The acoustic temperature measurement error analysis unit is configured to analyze the error of acoustic temperature measurement when the main thermocouple fails and the system automatically switches to acoustic temperature measurement, and calculate the range of acoustic temperature measurement error by comparing the acoustic temperature measurement data with historical thermocouple data. The error pheromone mapping unit is configured to use an ant colony algorithm to take the acoustic temperature measurement error as a pheromone evaporation factor to guide path optimization and determine a mapping rule, and adjust the pheromone concentration according to real-time acoustic temperature measurement error.

6. A graphite electrode production plant predictive maintenance system according to claim 5, characterized in that: The acoustic temperature measurement error analysis unit specifically includes: After the system detects that the main thermocouple fails, the system automatically switches to an acoustic temperature measurement mode, an acoustic node collects acoustic wave signals and calculates a propagation speed, and historical main thermocouple data is synchronously retrieved to ensure that the two are time-aligned and then synchronized to an edge computing device. The edge computing device compares acoustic temperature measurement data with historical main thermocouple data, calculates temperature differences and statistics the range of acoustic temperature measurement error, analyzes the influence of in-furnace pressure and gas composition fluctuation on the acoustic velocity model, and determines the main source of error. Based on the range of acoustic temperature measurement error, the parameters of the acoustic velocity-temperature model are adjusted, an error correction program is started, and the error range is compressed to within ±10℃ through three verifications of time consistency, spatial consistency and physical boundary constraints, and marked as reliable data.

7. A graphite electrode production plant predictive maintenance system according to claim 5, characterized in that: The error pheromone mapping unit specifically includes: The acoustic temperature measurement error is converted into a pheromone evaporation rate through a nonlinear function, and a pheromone deposition mechanism is introduced, the concentration is supplemented when the acoustic temperature measurement error is lower than a preset error threshold, the pheromone concentration and evaporation rate are updated in real time, and a self-adaptive feedback loop is formed. A weighted decision network is constructed on the edge computing device, the pheromone concentration is used as a weight basis, ants preferentially select paths with high pheromone concentration, and the exploration probability is adjusted according to the current error, the path weight is updated through iteration, and the proportion of high-error sensors is reduced. The acoustic temperature measurement error distribution of each channel after correction is periodically counted and fed back to a dynamic mapping model, if the system has a long-term high error, a parameter self-adaptive adjustment is triggered, the adjustment parameters are optimized through a gradient descent method, and a complete closed loop of "error awareness-pheromone update-path optimization-parameter iteration" is formed.

8. A graphite electrode production plant predictive maintenance system according to claim 5, characterized in that: The ant colony path search engine specifically includes: When the system starts, the acoustic temperature measurement error data of each sensor in the high-temperature zone double-redundancy sensing network is used to convert the initial error data of each sensor into a pheromone basic value, and a pheromone evaporation coefficient and the number of ants are set to construct an initial search environment based on pheromone concentration. The ants start from the starting sensor, select the next node according to the path pheromone concentration, preferentially select paths with small errors, calculate the fused temperature root mean square error of the path after traversal, and gradually eliminate low-quality paths by attenuating all path pheromones according to the evaporation coefficient. After multiple rounds of ant search and dynamic pheromone update, the pheromone concentration of the path with small error continuously accumulates and strengthens, and the path with large error is gradually eliminated, and when the preset iteration number or convergence condition is reached, the minimum error path with the highest pheromone concentration and related parameters are output.

9. A graphite electrode production plant predictive maintenance system according to claim 8, characterized in that: The real-time weighted fusion module specifically includes: The minimum error path output by the ant colony algorithm is analyzed, the pheromone concentration is normalized to a dynamic weight through a Softmax function, the sensor state is monitored in real time, if a failed node is detected, path reconstruction is triggered immediately, the failed node is removed and the weight is redistributed to the effective sensors on the redundant path; The raw data of the main thermocouple, acoustic temperature measurement and infrared temperature measurement are time-aligned and space-aligned, and are subjected to preprocessing operations of outlier rejection, noise filtering and dimensionless normalization; The preprocessed multi-sensor data is weighted and averaged according to the dynamic weight to generate an optimal soft measurement temperature value, the root mean square error of the optimal soft measurement temperature value is calculated in real time and compared with a preset threshold, and when the threshold is exceeded, a feedback mechanism is triggered to adjust the ant colony algorithm parameters or reinitialize the search until the error meets the process requirements.

10. A graphite electrode production plant predictive maintenance system according to claim 9, characterized in that: The device fault prediction engine specifically comprises: The optimal soft measurement temperature output by the real-time weighted fusion module is input into a pre-trained XGBoost fault prediction model, the nonlinear relationship between temperature anomalies and fault modes is learned according to gradient boosting trees, potential fault modes are identified, and the probability of occurrence of each potential fault is output in real time; According to the probability of occurrence of potential faults output by the XGBoost fault prediction model, decision support information is automatically generated, including the predicted fault location, type and targeted treatment guidelines, which is synchronously pushed to the maintenance personnel terminal and visual inspection suggestions are provided to clearly identify the key components that need to be checked; Based on the fault prediction results, a maintenance work order is automatically generated, recording the fault type, location and treatment measures in detail, and is pushed to the work order management system, maintenance personnel perform inspection and repair according to the instructions, feedback the processing results to update the model data, forming a closed loop of prediction-decision-execution-optimization.

Images