State monitoring system suitable for vacuum electric furnace

Abstract

The invention relates to the technical field of vacuum electric furnace monitoring, and discloses a state monitoring system suitable for a vacuum electric furnace. A multi-source sensor array of the system collects multi-dimensional physical signals such as temperature distribution, pressure change and vacuum degree fluctuation in a furnace in real time; a furnace cavity feature reconstruction module extracts sampling point feature parameters and correlates coordinates to construct a three-dimensional dynamic feature field; the process anomaly analysis module calculates a process deviation degree in combination with a preset reference parameter, and marks an anomaly coordinate area; the state transition evaluation module analyzes an abnormal trend according to historical records and predicts a state transition path and rate; the collaborative regulation and control decision-making module generates a multi-stage vacuum maintenance compensation strategy and a heating power regulation gradient scheme according to the multi-stage vacuum maintenance compensation strategy; and the running log feedback module records a strategy execution process, and associates the three-dimensional feature field data to generate a state tracing log. The system can realize comprehensive monitoring of the state of the vacuum electric furnace, accurate abnormity identification, trend prediction, cooperative regulation and control and state tracing, and helps to improve the operation management level of the vacuum electric furnace.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of vacuum electric furnace monitoring, in particular to a state monitoring system suitable for a vacuum electric furnace. BACKGROUND

[0002] As a key equipment in the field of metal material precision machining, special alloy preparation and high-temperature heat treatment, the running state of the vacuum electric furnace is directly related to the microstructure, mechanical properties and product qualification rate. In actual production process, the stability of physical parameters such as temperature distribution, pressure change and vacuum fluctuation in the furnace cavity has a significant impact on material phase transition, composition uniformity and process repeatability. However, the existing vacuum electric furnace monitoring technology still has many limitations, which is difficult to meet the high-precision production demand.

[0003] The current mainstream monitoring method mainly uses single type sensor, which can only collect local temperature or vacuum data, and cannot realize the synchronous acquisition of multi-dimensional physical signals in the furnace cavity. For example, some systems only monitor the temperature of limited points through the pre-embedded thermocouple of the furnace wall, which is difficult to reflect the temperature gradient difference between the center and the edge of the furnace cavity, resulting in deviation in the judgment of material heating uniformity; another system focuses on real-time reading of vacuum degree, but ignores the coupling relationship between pressure change and temperature field, which cannot timely capture abnormal signs when micro-leakage or gas composition change occurs in the furnace.

[0004] In terms of data processing and feature presentation, the existing technology lacks effective furnace cavity state reconstruction capability. Most systems only display the collected parameters in numerical form in real time, and it is difficult for the operator to intuitively master the overall working condition distribution in the furnace cavity. When the parameters of the local area are abnormal, it is difficult to quickly locate the abnormal position and the influence range. At the same time, the process abnormality judgment mainly depends on the comparison of fixed threshold value, without considering the dynamic change trend of the parameters with time, which is easy to cause misjudgment or omission. For example, if a sampling point temperature temporarily exceeds the threshold value, it may be a normal process fluctuation, but if the abnormality is determined only according to the static threshold value and the regulation is started, it will damage the process stability; while the abnormal parameters slowly drift, the fixed threshold value is difficult to trigger the early warning in time, which leads to the expansion of the abnormality.

[0005] The existing system has obvious short board in the aspects of abnormality processing and state tracing. After the abnormality occurs, the regulation strategy is mainly single parameter compensation, such as only adjusting the vacuum pump power to maintain the vacuum degree, without considering the indirect influence of the operation on the temperature field, which is easy to cause new process imbalance. At the same time, the operation log only records the historical values of the key parameters, without associating the furnace cavity feature distribution data, which makes it difficult to trace the overall state of the furnace cavity when the abnormality occurs when the subsequent product quality problem occurs, resulting in low problem troubleshooting efficiency and increasing production loss. SUMMARY

[0006] The present application aims to provide a state monitoring system suitable for a vacuum electric furnace to solve the problems presented in the background.

[0007] To achieve the above-mentioned purpose, the present application provides a state monitoring system suitable for a vacuum electric furnace, which comprises: a multi-source sensor array, which collects multi-dimensional physical signals inside the vacuum electric furnace in real time, including temperature distribution signals, pressure change signals and vacuum degree fluctuation signals; a furnace cavity feature reconstruction module, which extracts feature parameters of spatially discrete sampling points based on the multi-dimensional physical signals, and constructs a three-dimensional dynamic feature field by correlating the furnace cavity coordinate positions; a process abnormality analysis module, which calls a preset process reference parameter set, combines real-time feature parameters in the three-dimensional dynamic feature field, calculates process deviation degrees of each sampling point, and marks abnormal coordinate regions exceeding the deviation threshold; a state migration evaluation module, which analyzes the dynamic change trend of the abnormal coordinate regions based on historical process operation records, and predicts the state migration path and migration rate of the abnormal regions; a collaborative control decision module, which generates a multi-level vacuum maintenance compensation strategy and a heating power adjustment gradient scheme according to the state migration path and migration rate; a running log feedback module, which records the execution process of the multi-level vacuum maintenance compensation strategy, and generates a state traceability log by correlating the three-dimensional dynamic feature field data at the corresponding time.

[0008] Preferably, the furnace cavity feature reconstruction module comprises: a discrete feature extraction unit, which performs spatio-temporal alignment processing on the multi-dimensional physical signals, and extracts time series feature values and spatial correlation feature values of each sampling point; a dynamic field construction unit, which maps the time series feature values to furnace cavity three-dimensional grid nodes, superimposes the spatial correlation feature values to generate dynamic feature change vectors, and integrates all grid node vectors to form a three-dimensional dynamic feature field.

[0009] Preferably, the process abnormality analysis module comprises: a reference comparison unit, which calls temperature reference ranges, pressure reference curves and vacuum degree tolerance thresholds of corresponding process stages in the preset process reference parameter set; a deviation degree calculation unit configured to compare, node by node, a difference between a real-time characteristic value in the three-dimensional dynamic characteristic field and the temperature reference range, the pressure reference curve and the vacuum degree tolerance threshold, and to calculate a comprehensive process deviation degree of each grid node by weighting; an anomaly marking unit configured to filter grid nodes with a comprehensive process deviation degree exceeding a preset deviation threshold, and to cluster adjacent anomaly nodes to generate an anomaly coordinate region boundary.

[0010] Preferably, the state transition evaluation module comprises: a historical trajectory analysis unit configured to retrieve a historical anomaly record similar to a current anomaly coordinate region feature, and to extract a spatial expansion trajectory and an evolution time interval of a historical anomaly region; a migration prediction unit configured to calculate a migration path and a migration rate of the anomaly region to a furnace cavity key component based on the spatial expansion trajectory and the evolution time interval, and in combination with a gradient change direction of the current three-dimensional dynamic characteristic field.

[0011] Preferably, the coordinated regulation decision module comprises: a vacuum compensation strategy unit configured to divide different vacuum leakage risk levels according to a furnace cavity partition through which the migration path passes, and to generate a multi-level vacuum maintenance compensation strategy corresponding to the partition; a power adjustment unit configured to identify a heating region involved in the migration path, to calculate a heating power decay gradient based on the migration rate, and to generate a heating power adjustment gradient scheme in time segments.

[0012] Preferably, the system further comprises a process characteristic fusion module configured to associate a current process stage identifier, to extract a feature subset in the three-dimensional dynamic characteristic field that is strongly related to a process, to fuse an execution state of the multi-level vacuum maintenance compensation strategy, and to generate a process characteristic fusion dataset.

[0013] Preferably, the system further comprises a fault tracing module configured to match a case with the same feature mode in a historical fault case library based on the process characteristic fusion dataset, and to output a potential fault inducement and an associated anomaly coordinate region number.

[0014] Preferably, the system further comprises a strategy optimization module configured to correct a compensation intensity parameter in the multi-level vacuum maintenance compensation strategy according to the potential fault inducement, and to update a decay step in the heating power adjustment gradient scheme.

[0015] Preferably, the operation log feedback module comprises: A policy execution recording unit, which stores actual execution parameters of the multi-stage vacuum maintenance compensation policy and actual output values of the heating power adjustment gradient scheme by time stamp; A log association unit, which extracts the process feature fusion data set and the abnormal coordinate region boundary data at the policy execution time, and generates a state trace log with a time and space marker.

[0016] Preferably, the system further comprises a real-time warning module, which monitors the incremental change of the state transition rate, and when the increment exceeds a preset acceleration threshold, triggers an emergency shutdown instruction and synchronously updates the warning marker of the state trace log.

[0017] Compared with the prior art, the present application has the following advantages: The state monitoring system suitable for a vacuum electric furnace breaks the limitation of a single sensor in traditional monitoring methods that can only obtain local parameters, realizes comprehensive coverage and synchronous acquisition of multi-dimensional physical signals in the furnace cavity, enables an operator to obtain more complete basic data of the furnace conditions, and avoids one-sided judgment of the furnace conditions due to missing data.

[0018] The furnace cavity feature reconstruction module extracts feature parameters of spatial discrete sampling points based on multi-dimensional physical signals, and constructs a three-dimensional dynamic feature field by associating the furnace cavity coordinate position, thereby converting the originally dispersed numerical parameters into intuitive spatial distribution images. The operator can clearly understand the temperature, pressure and vacuum degree distribution in different coordinate regions in the furnace cavity through the three-dimensional dynamic feature field, quickly identify the regions with uneven parameter distribution, and no longer need to speculate the furnace conditions through tedious numerical comparison, which greatly improves the perception efficiency and accuracy of the overall working conditions of the furnace cavity.

[0019] The process abnormality analysis module calls a preset process reference parameter set, calculates the process deviation degree of each sampling point in combination with the real-time feature parameters in the three-dimensional dynamic feature field, can accurately quantify the difference degree of each sampling point parameter and the standard reference, and through marking the abnormal coordinate regions exceeding the deviation threshold, can accurately locate the abnormal positions, avoid the misjudgment and omission problems in the traditional fixed threshold comparison method without considering the dynamic change of parameters, make the abnormal identification more targeted and reliable, and provide a clear target direction for subsequent abnormal treatment.

[0020] The state migration evaluation module analyzes the dynamic change trend of the abnormal coordinate area and predicts the state migration path and migration rate of the abnormal area based on the historical process operation record, which changes the situation that the traditional system can only passively deal with the problems that have occurred in the abnormal processing. By predicting the development direction and speed of the abnormality in advance, the operator can take intervention measures before the abnormality expands, effectively curb the spread of the abnormality, reduce the influence range and degree of the abnormality on the process stability and product quality, and improve the active control ability of the furnace cavity state.

[0021] The synergistic control decision module generates a multi-stage vacuum maintenance compensation strategy and a heating power adjustment gradient scheme according to the state migration path and migration rate, which realizes the synergistic control of the vacuum degree and the heating power. Compared with the traditional single parameter compensation mode, the control scheme generated by the module fully considers the mutual influence between the vacuum degree adjustment and the heating power change, avoids the new process imbalance caused by single control, ensures that each parameter in the furnace cavity can be coordinated with each other during the adjustment process, and gradually recovers to a stable state, which improves the control effect while ensuring the process continuity.

[0022] The operation log feedback module records the execution process of the multi-stage vacuum maintenance compensation strategy, and generates a state trace log by associating the three-dimensional dynamic characteristic field data at the corresponding time, which perfects the state trace system of the system. When subsequent product quality problems or process review are needed, the operator can check the execution of the control strategy in the abnormal processing process and the three-dimensional feature distribution of the furnace cavity at the corresponding time through the state trace log, clearly restore the furnace working condition at that time, provide complete historical data support for problem troubleshooting and process optimization, help to summarize experience and improve the subsequent production process, and reduce the repeated occurrence of similar problems. BRIEF DESCRIPTION OF DRAWINGS

[0023] Figure 1 The timing diagram of the state monitoring system suitable for the vacuum electric furnace described in the present application; Figure 2 The working flowchart of the furnace cavity feature reconstruction module; Figure 3 The working flowchart of the synergistic control decision module; Figure 4 The working flowchart of the fault tracing module. DETAILED DESCRIPTION

[0024] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0025] Please refer toFigure 1 The application provides a state monitoring system suitable for a vacuum electric furnace, which comprises a technical framework of multi-source sensing, three-dimensional field reconstruction, abnormality analysis, migration prediction and collaborative regulation.

[0026] A multi-source sensor array is arranged at key positions inside a furnace cavity of the vacuum electric furnace, including high-temperature-resistant distributed temperature sensors, high-frequency pressure sensors and vacuum gauge probes, to synchronously collect temperature distribution signals, pressure change signals and vacuum degree fluctuation signals at a sampling frequency of no less than 10 Hz. A furnace cavity feature reconstruction module receives the above multi-dimensional physical signals, performs filtering and amplification through a signal conditioning circuit, and then transmits the signals to an embedded processor for spatio-temporal alignment processing, extraction of feature parameters of spatially discrete sampling points, and association with a three-dimensional coordinate of the furnace cavity to construct a dynamically updated three-dimensional feature field. An abnormality analysis module has a preset process reference parameter set stored therein, which stores standard temperature ranges, pressure curves and vacuum degree thresholds according to process stages, calculates process deviation degrees of each spatial position by comparing the feature parameters in the three-dimensional dynamic feature field in real time, and marks abnormal areas. A state migration evaluation module accesses a historical process running database, analyzes dynamic change trajectories of the abnormal areas by using a pattern matching algorithm, and predicts migration paths and rates of the abnormal areas to furnace walls, electrodes or observation windows and other key components. A collaborative regulation decision module outputs multi-stage vacuum maintenance compensation strategies and heating power adjustment gradient schemes according to the migration prediction results, wherein the vacuum compensation strategies are realized by controlling the opening degrees of electromagnetic valves and the rotating speeds of molecular pumps, and the power adjustment schemes are realized by adjusting the output of thyristor power regulators. A running log feedback module records the execution parameters of the regulation instructions and the corresponding three-dimensional field data with time stamps as indexes, and forms traceable state log files.

[0027] Embodiment 1: see Figure 2 , which relates to the specific implementation of a furnace cavity feature reconstruction module and an abnormality analysis module in a vacuum electric furnace state monitoring system. The system collects temperature distribution signals, pressure change signals and vacuum degree fluctuation signals inside the vacuum electric furnace in real time through a multi-source sensor array. The temperature distribution signals are collected by a K-type thermocouple array arranged at the top, side wall and bottom of the furnace chamber, with a measurement range of room temperature to 1800℃ and a sampling frequency of 20 Hz. The pressure change signals are obtained by high-frequency resonant pressure sensors installed in multiple partitions of the furnace body, with a range of 0-10 MPa and a sampling frequency of 50 Hz. The vacuum degree fluctuation signals are measured by a combination of a cold cathode ionization vacuum gauge and a capacitive diaphragm vacuum gauge, with a sampling frequency of 10 Hz. All sensor signals are transmitted to a signal conditioning unit through shielded cables for preliminary filtering and amplification.

[0028] ​​The discrete feature extraction unit of the furnace cavity feature reconstruction module receives the above-mentioned pre-processed multi-modal signals. First, the interpolation synchronization algorithm based on hardware timestamps is used to perform space-time alignment processing on all sensor data streams, eliminating millisecond-level timing deviations caused by differences in sensor response times. The aligned data streams are organized by spatial sampling points, and each sampling point is associated with its unique three-dimensional coordinate identifier. For temperature signals, the instantaneous value, moving average trend value, first-order difference gradient value, and low-frequency energy proportion feature obtained by Fourier transform within a 5-second time window are extracted for each sampling point. For pressure signals, the root mean square value of the dynamic fluctuation range, the pressure difference value with adjacent sampling points, and the main frequency amplitude in the frequency spectrum are calculated. For vacuum degree signals, the instantaneous value, 10-second sliding variance value, and deviation integral value from the ideal vacuum curve are extracted. The calculation of spatial correlation features uses the Kriging spatial interpolation algorithm to estimate the values of physical quantities in areas not directly sampled, and calculates the temperature gradient vector, pressure covariance coefficient, and vacuum degree spatial differential value of each sampling point and its eight adjacent points.

[0029] The dynamic field construction unit divides the furnace cavity space into a cubic grid with a size of 5 cm x 5 cm x 5 cm, with a total of approximately 8000 grid nodes. Each node is associated with its three-dimensional coordinates (x, y, z). The time series feature values of each sampling point output by the discrete feature extraction unit are mapped to the nearest grid node. For cases where multiple sampling points are mapped to the same node, a fusion algorithm based on sensor accuracy weighting is used: the weight coefficient of a sensor with an accuracy level of 0.1 is 0.7, and the weight coefficient of a sensor with an accuracy level of 0.5 is 0.3. The spatial correlation feature values are superimposed on the node data in the form of additional vectors, forming a dynamic feature change vector that includes the temperature spatial gradient, pressure propagation direction, and vacuum degree diffusion vector. The vector data of all grid nodes are organized in a three-dimensional matrix structure, with the matrix dimensions consistent with the physical furnace cavity space. The data is updated every second, forming a continuously updated three-dimensional dynamic feature field. This field data is transmitted to the central processing server for storage via a gigabit Ethernet network.

[0030] The reference comparison unit of the process anomaly analysis module has a pre-set process reference parameter set, which is indexed by process recipe number and stores the standard parameters for each process stage. Taking the aluminum alloy solid solution treatment process as an example, the standard temperature reference range for the heating stage is room temperature to 520°C ± 5°C, and the pressure reference curve is a step-down mode: from atmospheric pressure to The pressure log decrease value per minute is required to be no less than 0.5, and the vacuum degree tolerance threshold is that the deviation between the actual measured value and the theoretical value is not more than 15%. The temperature reference range for the holding stage is 520°C ± 2°C, the pressure fluctuation tolerance is ± 0.05 Pa, and the vacuum degree stability threshold requires that the fluctuation amplitude be less than 5%. The temperature drop rate reference for the cooling stage is 4-6°C per minute, and the vacuum degree recovery threshold requires that the vacuum degree reach The reference parameters are set differently according to different zones of the furnace chamber. For example, the area close to the furnace door is allowed to have a wide temperature range of ±3℃, and the area near the electrode introduction end requires a vacuum degree deviation of less than 10%.

[0031] The deviation calculation unit traverses each grid node of the three-dimensional dynamic characteristic field and reads its real-time characteristic value. The temperature deviation calculation uses a piecewise weighting algorithm: when the real-time temperature is lower than the lower limit of the reference, the deviation is (lower limit value - measured value) / lower limit value; when it is higher than the upper limit of the reference, the deviation is (measured value - upper limit value) / upper limit value. The pressure deviation calculation calculates the shape difference between the real-time pressure curve and the reference curve by the dynamic time warping algorithm, and then multiplies by the amplitude scaling factor. The vacuum degree deviation calculation uses the relative error percentage algorithm. The weight coefficients of each physical quantity are dynamically determined by the entropy weight method: the temperature weight usually accounts for 0.5, the pressure weight accounts for 0.3, and the vacuum degree weight accounts for 0.2. The node comprehensive process deviation is the weighted sum of each sub-item deviation, normalized to the range of 0-1.

[0032] The abnormality marking unit sets the comprehensive process deviation threshold to 0.7, and all grid nodes with a deviation exceeding this threshold are marked as abnormal nodes. A density-based spatial clustering algorithm is used to group the abnormal nodes, with a clustering radius of 3 grid spacings and a minimum clustering node number of 5. For each abnormal region formed by clustering, the minimum circumscribed cube boundary is calculated, and the coordinates of the eight vertices of the boundary are recorded. At the same time, the characteristic statistics of each abnormal region are calculated, including the average deviation, the maximum deviation node position, and the volume change rate of the region. All abnormal region data are updated synchronously with the three-dimensional dynamic characteristic field, with an update frequency of 1 Hz. The abnormal region boundary data are transmitted to the monitoring terminal display through the OPCUA protocol, and stored in the real-time database in binary format for reference.

[0033] Example 2: refer to Figure 3 The state transition evaluation module in the vacuum electric furnace state monitoring system is based on the abnormal coordinate region data output by the process abnormality analysis module, and predicts the dynamic change trend of the abnormal region by analyzing similar patterns in the historical process operation records. The historical trajectory analysis unit of this module accesses the historical process operation database, which uses a time series database structure to store three-dimensional dynamic characteristic field data, abnormal region boundaries, and key event logs from the past three years. The database index is classified by process type, material batch, and equipment number, supporting multi-dimensional fast retrieval.

[0034] The historical trajectory analysis unit takes the feature vector of the current abnormal coordinate region as the query condition, which contains the region temperature distribution pattern (the position of the highest temperature point, the direction of the temperature gradient), the pressure anomaly characteristics (the fluctuation frequency, the amplitude) and the vacuum degree decline rate. The retrieval algorithm uses similarity matching based on dynamic time warping to find historical abnormal records with a feature similarity of more than 85% from the database. During the matching process, records of the same process stage (such as heating, holding or cooling) and the same furnace cavity partition are given priority. For each matching record, the spatial expansion trajectory of its abnormal region is extracted, which contains the sequence value of the change of the center coordinates of the abnormal region over time, and the time function of the volume expansion of the region. The evolution time interval of the abnormal region in the adjacent time interval is also recorded, and its expansion rate vector is calculated. These historical trajectory data are organized in time sequence to form the abnormal region behavior pattern library.

[0035] The migration prediction unit receives the matching record set output by the historical trajectory analysis unit, first smooths the historical spatial expansion trajectory, and uses the cubic spline interpolation algorithm to reconstruct the complete motion path. Based on the statistical analysis of multiple similar trajectories, the direction probability distribution function and the rate distribution function of the abnormal region movement are derived. Combined with the real-time data calculated in the current three-dimensional dynamic feature field, including the gradient change direction of the temperature field, the isobaric surface propagation vector of the pressure field and the leakage diffusion direction of the vacuum degree, the historical derived function is real-time corrected. The corrected prediction model outputs the migration path of the abnormal region to the key components of the furnace cavity (such as the furnace door sealing ring, the electrode introduction end, the water-cooled coil or the observation window), and the path is represented as a three-dimensional space polyline, each path point contains coordinate value and the probability of reaching the point. The migration rate calculation integrates the historical average expansion rate and the current field gradient correction factor to output the rate value and the confidence interval. The prediction result is updated every 10 seconds and transmitted to the collaborative control decision module through the data bus.

[0036] The vacuum compensation strategy unit of the synergistic control decision module receives the migration path data output by the state transition evaluation module. The unit is built-in with a furnace cavity partition risk level atlas, which is divided according to the structural characteristics of the vacuum system: the high-risk area includes the dynamic sealing part (furnace door rotary seal, electrode lead-in end seal), vacuum valve group interface area and sensor through-piece area; the medium-risk area includes the static flange connection surface, observation window sealing ring and water cooling jacket interface; the low-risk area includes the overall furnace wall, cooling pipeline coverage area and internal support area. Analyzing the specific partitions passed by the migration path, a hierarchical compensation strategy is generated. For the case where the path passes through the high-risk area, a first-level vacuum maintenance compensation strategy is triggered: instructing the molecular pump group to increase the speed to 115%-125% of the rated value, starting the standby diffusion pump, and adjusting the main valve opening to increase the pumping speed. For the case where the path passes through the medium-risk area, a second-level compensation strategy is executed: adjusting the inlet compensation solenoid valve opening to increase the inert gas flow by 8%-12%, maintaining pressure balance. For the case where the path only passes through the low-risk area, a third-level compensation strategy is adopted: activating the vacuum gauge automatic calibration program to improve monitoring accuracy. All compensation strategy parameters are dynamically adjusted according to the confidence and rate values of the migration path.

[0037] The power adjustment unit synchronously analyzes the heating area involved in the migration path, and the unit accesses the heating system layout diagram, which marks the spatial position and power control partition of all heating elements. Identify the heating area passed or possibly affected by the migration path, calculate the heating power decay gradient based on the migration rate. When the migration rate is higher than 0.4 cm / s, it is determined as fast migration mode, generating a power step-down scheme: reducing the rated power by 4%-6% every 20-30 seconds, a total of 5-8 steps to complete the adjustment. When the migration rate is between 0.2-0.4 cm / s, a moderate decay scheme is adopted: reducing the rated power by 2%-3% per minute. When the migration rate is less than 0.2 cm / s, a slow linear decay scheme is adopted: reducing the rated power by 1% every 2 minutes. All power adjustment schemes are time-coded and output to the thyristor power regulator for execution. The power regulator accurately controls the trigger angle of the heating element according to the scheme to achieve smooth transition of power. At the same time, the power adjustment unit monitors the temperature stability during the adjustment process, and automatically inserts a power holding segment when detecting that the temperature fluctuation exceeds the allowed range, ensuring process safety.

[0038] During the entire control process, the vacuum compensation strategy unit and the power adjustment unit maintain real-time data interaction. The pressure changes caused by vacuum compensation operations will be input as feedback to the power adjustment algorithm, and the temperature changes caused by power adjustment will also affect the optimization of vacuum compensation parameters. This cross-module collaborative control ensures the integrity and consistency of system response. All generated strategy schemes are time-stamped and version-identified, transmitted to the actuator through industrial Ethernet, and stored in the strategy log database for subsequent analysis.

[0039] Embodiment 3: Refer to Figure 4 , which relates to the specific implementation of the process feature fusion module and the fault tracing module in the vacuum electric furnace state monitoring system. The process feature fusion module receives the current process stage identifier from the process control unit, which is represented by a four-digit code, defining the process type, stage number, temperature interval code, and pressure mode code, respectively. For example, the code T203 represents the second stage of the aluminum alloy solid solution treatment process, with a temperature interval of 500-550°C and a pressure mode of stepwise pressure reduction. The module simultaneously receives real-time data streams of the three-dimensional dynamic feature field and execution state data of the multi-level vacuum maintenance compensation strategy.

[0040] The process feature fusion module dynamically adjusts the feature extraction strategy according to the process stage identifier. For the heating stage, the module extracts all grid node data in the three-dimensional dynamic feature field with a temperature rise rate exceeding 2.5°C / s, recording its spatial coordinates, instantaneous temperature value, and heating acceleration. At the same time, the pressure change characteristics are monitored, and the area data with a pressure drop rate between 0.1-0.5 Pa / s are extracted. For the holding stage, the focus is on extracting temperature stability features, selecting grid nodes with a temperature fluctuation standard deviation below 0.8°C, recording their temperature control accuracy and spatial distribution pattern. In terms of pressure characteristics, stable regions with a fluctuation amplitude within ±0.08 Pa are extracted. For the cooling stage, the vacuum degree recovery characteristic data is extracted, recording the recovery time constant when the vacuum degree reaches the order of magnitude, and the cooling uniformity data with a temperature drop rate in the range of 3-7°C / min.

[0041] The execution state data of the multi-level vacuum maintenance compensation strategy includes the real-time rotational speed value of the molecular pump group, which is obtained through encoder feedback with an accuracy of ±5 rpm; the opening percentage of the electromagnetic valve is measured by a position sensor with a resolution of 0.1%; and the activation state of the diffusion pump is a Boolean quantity, recording its start and stop timestamps and running duration. These execution state data are collected with a sampling period of 100 ms and are time-synchronized with the process feature data.

[0042] The feature fusion process uses a time series-based alignment algorithm to integrate the process feature data and the execution state data on a unified time axis. The fused data set is stored in a multi-dimensional tensor structure, with each data point containing a timestamp, process stage identifier, spatial coordinates, temperature feature value, pressure feature value, vacuum degree feature value, and corresponding vacuum compensation strategy parameters. The update frequency of the data set is consistent with the three-dimensional dynamic feature field, which is once per second. To quantify the correlation strength between features, a feature coupling coefficient is introduced:

[0043] where: represents the feature coupling coefficient, the number of feature sampling points, a process feature value (such as temperature or pressure) representing the i-th sampling point, a process feature value (such as temperature or pressure) representing the i-th sampling point, a compensation strategy parameter (such as molecular pump speed or electromagnetic valve opening) representing the i-th sampling point, a compensation strategy parameter (such as molecular pump speed or electromagnetic valve opening) representing the i-th sampling point, a spatial weight factor reflecting the spatial distance relationship between the sampling point and the actuator. This coefficient is used to evaluate the degree of influence of the compensation strategy on the process feature.

[0044] The fault tracing module accesses the historical fault case library, which is stored in a graph database structure. The nodes represent fault events, and the edges represent the association between faults. Each fault event node contains fields such as fault type code, occurrence time, process stage, feature data pattern, and treatment measures. The fault type code uses a hierarchical structure, such as VL-101 representing flange seal failure in vacuum leakage class faults, and HE-203 representing electrode arc discharge in heating system faults.

[0045] The matching algorithm uses a combination of dynamic time warping and feature pattern recognition. First, the current process feature fusion data set is sliced by time window, with each window length of 30 seconds and an overlap rate of 50%. Feature pattern vectors are extracted from each window data, including temperature distribution skewness, pressure fluctuation kurtosis, and vacuum degree change gradient. Similarity calculation is performed between these feature pattern vectors and historical patterns in the fault case library. The similarity metric uses an improved cosine similarity algorithm, which considers the weight factor of the process stage and the dimensionless normalization of the feature value.

[0046] When a matching case with a similarity greater than 0.92 is found, the fault tracing module outputs a potential fault cause report. The report includes fault type description, possible occurrence mechanism, historical occurrence frequency, and corresponding abnormal coordinate region number. For example, when the temperature distribution skewness in a specific region is greater than 2.5 and the vacuum degree change gradient is abnormal, a vacuum leakage fault case may be matched, and the system will output an analysis of the cause of the seal ring thermal degradation and label the corresponding abnormal region number A-107. The matching process also considers multiple fault concurrent situations and uses a probabilistic graph model to calculate the likelihood of different fault combinations, outputting a fault cause list sorted by confidence.

[0047] All matching results are timestamped and confidence-indexed, and are transmitted to the monitoring terminal through a data interface. At the same time, successful matching records are fed back to the historical fault case library for enriching case data and optimizing the matching algorithm. The case library automatically performs clustering analysis every week, merging similar fault patterns and updating the feature pattern template library. The entire fault tracing process uses a pipeline architecture, supporting multi-channel parallel matching to ensure that fault pattern analysis for a single time window is completed within 1 second.

[0048] The process feature fusion dataset and the fault tracing result are stored in a standardized format in the central database, and the data retention strategy is to store complete data for the last 90 days. The database establishes feature index and fault coding index, supports multi-dimensional query analysis according to time range, process type, fault category, etc. The data export format supports CSV, JSON and binary stream, etc. to meet the data access needs of different analysis tools.

[0049] Embodiment 4: relates to the specific implementation of the strategy optimization module and the real-time early warning module in the vacuum electric furnace state monitoring system. The strategy optimization module receives the potential fault inducement report output by the fault tracing module, which adopts a structured data format and includes fault type code, confidence score, impact range evaluation and associated abnormal area number list. The fault type code is formulated in accordance with the international vacuum metallurgical equipment fault classification standard, such as VL-201 representing "furnace door sealing ring heat-induced creep" and HE-305 representing "electrode surface local arc erosion".

[0050] The strategy optimization module analyzes the fault inducement report and adjusts the control parameters according to the historical treatment experience library of the fault type. For vacuum leakage type faults, the strength parameters of the multi-stage vacuum maintenance compensation strategy are modified. When VL-201 type fault is identified, the parameter adjustment rule library is queried to know that the compensation strength of the dynamic sealing area needs to be enhanced. The molecular pump speed of the high-risk area is increased from the baseline value of 120% to the range of 130-140%, and the specific value is determined by linear interpolation according to the leakage confidence score. The electromagnetic valve opening increment of the medium-risk area is adjusted from 10% of the baseline value to 12-18%, and the adjustment amplitude is proportional to the abnormal area volume growth rate. At the same time, the auxiliary sealing detection program is activated, and the sampling frequency of the helium mass spectrometer leak detector is increased.

[0051] For heating system related faults, the strategy optimization module updates the heating power adjustment gradient scheme. When HE-305 type fault is detected, the power adjustment rule library is queried to obtain the correction parameters. The step size of the power step-down scheme is increased from 5% of the baseline value to 6-8%, and the specific value is calculated according to the temperature gradient change rate of the electrode area. The step size of the linear decay scheme is adjusted from 2% to 3-4%, and the adjustment amount is associated with the confidence score of the arc erosion risk. At the same time, the electrode surface infrared monitoring compensation program is enabled, and the sampling point density of the thermal imager is increased.

[0052] All parameter adjustment processes follow the principle of gradual optimization, and small step multiple adjustment is adopted to avoid system disturbance. Each parameter modification records the pre-adjustment value, post-adjustment value, adjustment basis and adjustment timestamp to form a complete parameter change log. The optimized strategy parameters are verified for rationality by a security verification algorithm before they can be issued for execution.

[0053] The real-time warning module continuously monitors the migration rate data stream output by the state transition evaluation module, which is updated at a frequency of 10 Hz and contains multiple dimensions such as the center coordinate movement rate of the abnormal region, the volume expansion rate, and the morphological change rate. The warning algorithm calculates the rate increment in adjacent sampling periods and uses moving average filtering to eliminate transient fluctuation noise. When the rate increment is detected to be continuously greater than 0.15 cm / s² for 3 periods, a warning state is triggered. After the warning is triggered, the system starts a multi-level response mechanism. First, an audible and visual warning signal is sent to the operator to prompt the abnormal acceleration state. At the same time, the data recording frequency is automatically increased, and the sampling frequency of the relevant sensors is increased from the regular 10 Hz to 50 Hz. If the rate increment continues to increase to more than 0.2 cm / s² in the next 2 periods, the system triggers an emergency shutdown command sequence.

[0054] The emergency shutdown command sequence adopts a redundant design and sends control commands to the actuators through hard-wired connections and industrial Ethernet. The command sequence includes: immediately cutting off the main heating power supply, sequentially closing each partition heater; quickly closing all vacuum valve groups, including high vacuum valves, roughing valves, and gas inlet valves; activating the nitrogen filling program to inject high-purity nitrogen into the furnace chamber to a normal pressure state; starting the emergency cooling system to increase the water cooling circulation flow. All commands have a timestamp and priority marker to ensure the correctness of the execution sequence.

[0055] Referring to Table 1, the system maintains a warning event record table that details the relevant data of each warning event.

[0056] Table 1: The structure of the warning event record table is as follows.

[0057]

[0058] The warning event record table uses a ring buffer storage mechanism to retain the last 1000 event records. Each record is automatically associated with the corresponding process feature fusion data set and state trace log index at the corresponding time, forming a complete event context data chain. When a warning event is confirmed, the system automatically generates an event analysis report, including the acceleration change curve, the associated abnormal region evolution trend graph, and the effect evaluation of the executed control measures. The real-time warning module also includes a self-learning mechanism that dynamically adjusts the acceleration threshold and warning sensitivity by analyzing the data patterns of historical warning events. The system periodically performs cluster analysis on warning events to identify frequently occurring warning patterns and optimize warning parameter settings. All warning parameter adjustments are simulated and verified before being applied to the actual system to ensure the accuracy and reliability of the warning.

[0059] The system provides a pre-warning board interface to display the current migration rate, acceleration value and pre-warning state in real time. The interface uses color coding to distinguish between normal, pre-warning and emergency states, and provides a visual query function for historical pre-warning events. The operator can confirm the pre-warning event through the board, enter the processing opinion, and view the historical processing scheme of similar events. All interactive operations are recorded in the audit log to meet the traceability requirements of the quality management system.

[0060] Embodiment 5: relates to the specific implementation of the running log feedback module in the vacuum electric furnace state monitoring system. The core function of this module is to comprehensively record various state data and control instructions during system operation, forming a state traceability log with complete space-time correlation. The running log feedback module contains two main components: a strategy execution recording unit and a log association unit, which work together to achieve data collection, integration and storage.

[0061] The strategy execution recording unit is responsible for real-time collection and storage of control instructions output by the collaborative control decision module and their actual execution. This unit communicates with the actuator controller through the industrial bus interface to record the actual execution parameters of the multi-stage vacuum maintenance compensation strategy with millisecond-level time accuracy. For the vacuum maintenance compensation strategy, the recorded data includes the real-time speed set value of the molecular pump group, the actual feedback speed and the deviation value between the two; the opening instruction value of the electromagnetic valve, the actual opening sensor feedback value and the response delay time; the start instruction time of the diffusion pump, the actual start confirmation signal and the preheating state parameter. For the heating power adjustment gradient scheme, the recorded data includes the power set value of the thyristor power regulator, the actual output power measurement value, the trigger angle change curve and the cooling water temperature influence coefficient. All data are timestamped with high precision, and the time synchronization signal comes from the Beidou / GPS dual-mode clock source of the system, ensuring the accuracy and consistency of the timestamp. The recorded data uses a segmented storage strategy, with recent data saved in the cache for real-time query, and historical data compressed and stored in the time series database.

[0062] The log association unit works based on the data collected by the strategy execution recording unit to further extract and associate more extensive system state information. This unit automatically triggers the data collection program at each control strategy execution time to obtain the complete process feature data set at that time from the process feature fusion module. These data include the current process stage identifier, the feature subset vector extracted from the three-dimensional dynamic feature field, the statistical feature value of the multi-source sensor raw data, and the execution state code of the vacuum compensation strategy. At the same time, the unit accesses the process anomaly analysis module to obtain the abnormal coordinate region boundary data at the same time, including the abnormal region number list, the geometric center coordinates of each region, the boundary box size volume, the average deviation index and the region shape feature descriptor.

[0063] The data association process adopts a fusion algorithm based on space-time matching. First, a unified time index system is established to align all data streams to the same time coordinate system. Space matching is achieved through coordinate transformation to unify the spatial data from different sources into the furnace cavity reference coordinate system. For each recording time, the system generates a complete state snapshot, including time identification, spatial information, process characteristics, abnormal area description, and control instruction execution details. These snapshots are organized in chronological order to form a continuous state trace log.

[0064] The state trace log is designed with a hierarchical storage structure. The bottom layer is the raw data layer, which stores the raw sensor readings and control instruction byte streams without processing. The middle layer is the feature data layer, which stores the processed feature vectors and abnormal area description data. The top layer is the application data layer, which stores the fully associated state snapshot data. Each layer of data has a corresponding index mechanism to support fast retrieval by time range, spatial area, process stage, or abnormal type. Data compression algorithms are applied differently based on data type characteristics. Lossy compression is used for high-frequency sensor data, and lossless compression is used for control instructions and state identification data, optimizing storage efficiency while ensuring data accuracy.

[0065] The log management system provides a complete set of data access interfaces. The basic query interface supports browsing the state log in chronological order, and the advanced query interface supports multi-condition combined retrieval. The data export interface supports converting selected log records into standard format files for offline analysis or third-party system integration. The system security mechanism ensures the integrity and confidentiality of log data, including data checksum, access permission control, and operation audit tracking.

[0066] The state trace log also implements automatic archiving and life cycle management functions. The system automatically executes archiving strategies based on the timestamp of log data. Recent data is retained in online storage for real-time access, early data is migrated to nearline storage devices, and historical data is eventually archived to offline storage media. The archiving process maintains data integrity and readability, ensuring that historical states can be accurately restored even after long-term storage.

[0067] The visualization of log data is achieved through a dedicated monitoring interface. This interface provides a timeline navigation function, allowing operators to browse the system state change process along the timeline. The spatial display component superimposes abnormal area boundary data on the furnace cavity three-dimensional model, intuitively displaying the spatiotemporal evolution trajectory of abnormal areas. Control instruction execution is displayed in the form of time series charts, clearly presenting the correspondence between control strategies and actual execution. All visualization components support interactive operations, allowing detailed data information to be obtained through clicking or dragging.

[0068] The operation log feedback module also contains a self-monitoring function that monitors the integrity and accuracy of the log recording process in real time. The module performs data integrity checks regularly and automatically triggers a compensation recording program when it detects missing records or data anomalies. The performance monitoring component tracks the storage usage, query response time, and data throughput of the log system to ensure the reliability of the log system itself. When it detects insufficient storage space or a decline in system performance, it automatically issues a maintenance alert and initiates the corresponding cleaning or optimization program.

[0069] The entire log system is designed with reliability requirements in mind for industrial environments, using redundant storage architecture and error recovery mechanisms. Important data is stored in dual backup, and critical index data is additionally saved in triplicate. The system has a power failure protection function that can maintain data integrity in the event of accidental power failure and automatically continue recording work after power is restored. All these features together ensure the reliability, integrity, and availability of the state trace log, providing a solid data foundation for the analysis and traceability of system operating states.

[0070] It should be noted that the relational terms herein such as first and second and the like are used solely to distinguish one from another entity or action without necessarily requiring or implying that the entities or actions are in any way actually related or ordered other than by the context of use. Moreover, the terms "comprises", "comprising", or any other variations thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but can also include other elements not expressly listed or inherent to such process, method, article, or apparatus.

[0071] While the embodiments of the application have been illustrated and described, it will be understood by those skilled in the art that various changes, modifications, substitutions, and alterations can be made therein without departing from the spirit and scope of the application, which is defined by the appended claims and their equivalents.

Claims

1. A condition monitoring system suitable for use in a vacuum electric furnace, characterized in that, The system comprises: A multi-source sensor array that collects multi-dimensional physical signals inside a vacuum electric furnace in real time, the multi-dimensional physical signals including temperature distribution signals, pressure change signals, and vacuum degree fluctuation signals; A furnace cavity feature reconstruction module that extracts feature parameters of spatially discrete sampling points based on the multi-dimensional physical signals, correlates furnace cavity coordinate positions to construct a three-dimensional dynamic feature field; A process anomaly analysis module that calls a preset process reference parameter set, combines real-time feature parameters in the three-dimensional dynamic feature field, calculates process deviation degrees of each sampling point, and marks abnormal coordinate regions that exceed a deviation threshold; A state migration evaluation module that analyzes dynamic change trends of the abnormal coordinate regions based on historical process operation records, predicts state migration paths and migration rates of abnormal regions; A collaborative control decision module that generates a multi-level vacuum maintenance compensation strategy and a heating power adjustment gradient scheme according to the state migration paths and migration rates; An operation log feedback module that records an execution process of the multi-level vacuum maintenance compensation strategy, correlates the three-dimensional dynamic feature field data at corresponding time points to generate a state trace log.

2. A condition monitoring system for a vacuum electric furnace according to claim 1, wherein The furnace cavity feature reconstruction module comprises: A discrete feature extraction unit that performs spatio-temporal alignment processing on the multi-dimensional physical signals, extracts time series feature values and spatial correlation feature values of each sampling point; A dynamic field construction unit that maps the time series feature values to furnace cavity three-dimensional grid nodes, superimposes the spatial correlation feature values to generate dynamic feature change vectors, and integrates all grid node vectors to form a three-dimensional dynamic feature field.

3. A condition monitoring system for a vacuum electric furnace as claimed in claim 2, wherein, The process anomaly analysis module comprises: A reference comparison unit that calls temperature reference ranges, pressure reference curves, and vacuum degree tolerance thresholds of corresponding process stages in the preset process reference parameter set; A deviation degree calculation unit that compares real-time feature values in the three-dimensional dynamic feature field with differences between the temperature reference ranges, the pressure reference curves, and the vacuum degree tolerance thresholds, and weightedly calculates comprehensive process deviation degrees of each grid node; An abnormal marking unit that screens grid nodes with comprehensive process deviation degrees exceeding a preset deviation threshold, clusters adjacent abnormal nodes to generate an abnormal coordinate region boundary.

4. A condition monitoring system for a vacuum electric furnace as claimed in claim 3, wherein, The state migration evaluation module comprises: A historical trajectory analysis unit that retrieves historical abnormal records similar to features of a current abnormal coordinate region, extracts spatial expansion trajectories and evolution time intervals of historical abnormal regions; A migration prediction unit that calculates migration paths and migration rates of abnormal regions to furnace cavity key components based on the spatial expansion trajectories and evolution time intervals, in combination with gradient change directions of a current three-dimensional dynamic feature field.

5. A condition monitoring system for a vacuum electric furnace as claimed in claim 4, wherein, The collaborative control decision module comprises: A vacuum compensation strategy unit, which divides different vacuum leakage risk levels according to the furnace cavity partitions passed by the migration path, and generates a multi-level vacuum maintenance compensation strategy corresponding to the partitions; A power adjustment unit, which identifies the heating area involved in the migration path, calculates the heating power decay gradient based on the migration rate, and generates a heating power adjustment gradient scheme for each time period.

6. A condition monitoring system for a vacuum electric furnace as claimed in claim 5, wherein, The system further comprises a process feature fusion module, which associates with the current process stage identifier, extracts a feature subset in the three-dimensional dynamic feature field that is strongly related to the process, fuses the execution state of the multi-level vacuum maintenance compensation strategy, and generates a process feature fusion dataset.

7. A condition monitoring system for a vacuum electric furnace as claimed in claim 6, wherein, The system further comprises a fault tracing module, which matches the cases with the same feature mode in the historical fault case library based on the process feature fusion dataset, and outputs the potential fault inducement and the associated abnormal coordinate region number.

8. A condition monitoring system for a vacuum electric furnace as claimed in claim 7, wherein, The system further comprises a strategy optimization module, which corrects the compensation intensity parameter in the multi-level vacuum maintenance compensation strategy according to the potential fault inducement, and updates the decay step in the heating power adjustment gradient scheme.

9. A condition monitoring system for a vacuum electric furnace as claimed in claim 8, wherein, The operation log feedback module comprises: A strategy execution record unit, which stores the actual execution parameters of the multi-level vacuum maintenance compensation strategy and the actual output values of the heating power adjustment gradient scheme according to the time stamp; A log association unit, which extracts the process feature fusion dataset and the abnormal coordinate region boundary data at the time of strategy execution, and generates a state tracing log with time and space markers.

10. A condition monitoring system for a vacuum electric furnace as claimed in claim 9, wherein, The system further comprises a real-time warning module, which monitors the incremental change of the state migration rate, triggers an emergency stop command when the increment exceeds a preset acceleration threshold, and synchronously updates the warning marker of the state tracing log.

Images