A silicon-manganese ore hot furnace eye lining abnormal loss identification method and system based on eye orientation heat field and stable stream offset

Abstract

The application relates to the technical field of data processing, and discloses a silicon-manganese ore hot stove eye lining abnormal loss identification method and system based on a stove eye direction heat field and stable flow offset, wherein the method comprises the following steps: acquiring a stove eye periphery thermal image sequence and a tapping flow video, adopting a thermal response dynamic baseline compensation output direction heat flow representation data; outputting heat flow coupling observation sequences based on asymmetric scouring efficiency; outputting direction collaborative feature sequences through lag convolution matching; outputting abnormal loss direction cluster data through space-time density clustering; and outputting stove eye maintenance treatment data through cluster evolution analysis. In view of the problem that single thermal image or flow offset is easily interfered by slag hanging, temperature drift and occasional swing, the application forms a closed loop verification of heat field abnormality, stable flow offset and cross-furnace evolution evidence, improves the reliability of stove eye lining abnormal loss identification and maintenance treatment in a multi-furnace production scene, and provides a data basis for determination of stove eye maintenance priority and repair recommendation.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to a method and system for identifying abnormal losses in the borehole lining of a ferrosilicon manganese submerged arc furnace based on the borehole orientation thermal field and stable flow stream offset. Background Technology

[0002] Currently, in the continuous production process of ferrosilicon manganese submerged arc furnaces, the monitoring of abnormal wear of the furnace eye lining mainly relies on two independent sensing methods: on the one hand, using an infrared thermal imager installed in front of the furnace eye to acquire the temperature distribution sequence of the furnace eye periphery, and using fixed thresholds or historical temperature rise statistics to identify directional hot spots, thereby indirectly inferring the location of lining thinning; on the other hand, using visible light or near-infrared cameras to capture the morphology of the iron flow stream, extracting the offset direction and amplitude of the flow stream's geometric center, and using flow stream offset statistics to quantify the changes in the furnace eye's internal shape. Existing processing paths are mostly focused on a single mode, such as detecting anomalies in the highest sector temperature within the thermal image sequence, or performing temporal clustering of the flow stream offset angle to determine the dominant offset direction. However, these methods all treat thermal field information and flow stream offset as independent signal sources, without establishing a cross-modal correlation analysis mechanism between the two in the directional and temporal dimensions.

[0003] For example, in multi-heat continuous tapping scenarios, single-heat infrared thermography is easily affected by slag buildup at the furnace mouth, fluctuations in environmental radiation, and the opening and closing of tap holes, resulting in localized false temperature rises. If detection is based solely on the thermal field anomaly sequence, such instantaneous interference can easily be misjudged as abnormal lining wear. Simultaneously, stream deviation itself is affected by multiple factors such as the instantaneous shape of the tap hole, differences in molten iron temperature, and changes in the furnace charge level. Even if the stream shows significant deviation in a certain direction, that direction may not strictly correspond to the lining erosion location, leading to contradictory or insufficiently reliable anomaly indications from the two modes. Particularly noteworthy is that as heat cycles accumulate, the thermal field baseline undergoes systematic drift, the absolute temperature threshold of the thermographic sequence cannot be effectively calibrated, and the statistical distribution of the stream deviation sequence changes accordingly, further diminishing the reliability of single-mode information in characterizing the lining condition. In this situation, it is impossible to verify the evolution trend of the directional thermal field across furnace cycles with the deviation direction of the stable tapping iron stream in time and space. On-site operators often can only shut down the furnace for maintenance after obvious signs of iron leakage appear in the furnace hole, which increases safety hazards and causes unplanned shutdown losses.

[0004] Therefore, there is an urgent need for an analytical method that can automatically construct a cross-modal correlation model between the thermal field change sequence of the furnace periphery and the stable tapping stream offset sequence during multiple continuous tapping processes. This method would enable the system to significantly enhance the reliability of lining abnormal wear determination by detecting the temporal and spatial consistency between the hot spot orientation and the stream offset direction, even when the single-mode signal is subject to noise, drift, or artifact interference. This would improve the anti-interference capability of the furnace condition monitoring system and provide a reliable basis for the scientific determination of furnace maintenance priorities and accurate repair recommendations. Summary of the Invention

[0005] To address the aforementioned technical shortcomings, the purpose of this invention is to propose a method for identifying abnormal wear of the furnace eye lining in a ferrosilicon manganese submerged arc furnace based on the furnace eye orientation thermal field and the stable flow stream offset. This method aims to solve the technical problem that existing technologies lack a mechanism for cross-modal time-series correlation analysis of changes in the furnace eye periphery thermal field orientation and the flow stream offset direction. In particular, under the condition that single-modal information is easily interfered with and difficult to verify during multiple continuous iron tapping processes, it is impossible to reliably identify abnormal wear of the lining.

[0006] To solve the above-mentioned technical problems, the present invention adopts the following technical solution: The present invention provides a method for identifying abnormal losses of the borehole lining of a ferrosilicon manganese submerged arc furnace based on the borehole orientation thermal field and stable flow stream offset.

[0007] The method for identifying abnormal wear of the borehole lining in a ferrosilicon manganese submerged arc furnace based on the borehole orientation thermal field and stable flow stream offset includes:

[0008] Step S10: Acquire synchronous observation data of the furnace eye consisting of thermal imaging sequences around the furnace eye and videos of the iron tapping stream. Based on the synchronous observation data of the furnace eye, perform the task of constructing the azimuth thermal field anomaly using a thermal response dynamic baseline compensation mechanism, and output azimuth thermal flow characterization data.

[0009] Step S20: Based on the azimuth heat flow characterization data, an asymmetric scour efficiency mechanism is used to perform the flow stream migration extraction task, and a heat flow coupled observation sequence is output;

[0010] Step S30: Based on the heat flow coupled observation sequence, a hysteresis convolution matching mechanism is used to perform a spatiotemporal correlation modeling task, and the orientation cooperative feature sequence is output;

[0011] Step S40: Based on the azimuth collaborative feature sequence, a spatiotemporal density clustering mechanism is used to perform an anomaly verification task and output anomaly loss azimuth cluster data;

[0012] Step S50: Based on the abnormal loss location cluster data, the cluster evolution analysis mechanism is used to perform the loss pattern inference task and output the furnace eye maintenance and disposal data;

[0013] Preferably, step S10, which involves acquiring synchronous observation data of the furnace eye consisting of a thermal imaging sequence of the furnace eye periphery and a video of the tapping iron stream, and using a dynamic baseline compensation mechanism based on the synchronous observation data to perform a task of constructing the azimuth thermal field anomaly and outputting azimuth thermal flow characterization data, specifically includes:

[0014] Step S101: Set up an infrared thermal imager in front of the furnace eye, and set up a visible light camera or near-infrared camera that is aligned with the calibration of the infrared thermal imager. Synchronously collect thermal image sequences and iron flow videos of the furnace eye periphery area during each tapping furnace from opening to closing. Associate the thermal image sequences and the iron flow videos according to the collection timestamp to obtain the synchronous observation data of the furnace eye.

[0015] Step S102: Extract the thermal image sequence from the synchronous observation data of the furnace eye, set an annular region of interest with the geometric center of the furnace eye outlet section as the origin, set the inner diameter of the annular region of interest to 0.7 to 0.9 times the design diameter of the furnace eye, set the outer diameter of the annular region of interest to 2.5 to 3.5 times the design diameter of the furnace eye, and divide the annular region of interest into 72 sector areas at equal angles along the circumference to obtain the annular sector division;

[0016] Step S103: For each frame of thermal image of each tapping furnace, extract the highest temperature and average temperature of each sector area, and perform moving median filtering on the highest temperature and average temperature arranged according to the acquisition time to obtain a time-regular azimuth thermal field sequence.

[0017] Step S104: For each sector, divide the difference between the current highest temperature and the dynamic thermal response baseline by the standard deviation of the highest temperature during the stable tapping stage of the most recent preset number of normal furnaces M in that sector to obtain the azimuth thermal field anomaly sequence. The dynamic thermal response baseline consists of the reference temperature before the furnace opening, the historical empirical equilibrium temperature rise of the sector, and an exponential approach term determined by the time after the furnace opening and the thermal response time constant of the sector. Extract the tapping stream video from the furnace eye synchronous observation data, and combine the azimuth thermal field anomaly sequence, the annular sector division, and the tapping stream video into the azimuth thermal flow characterization data.

[0018] Preferably, step S20, which involves using an asymmetric scouring efficiency mechanism to perform stream migration extraction based on the azimuth heat flow characterization data and outputting a heat flow coupled observation sequence, specifically includes:

[0019] Step S201: Extract the iron-producing stream video from the azimuth heat flow characterization data, and extract the stream foreground region from the iron-producing stream video using a stream instance segmentation model;

[0020] Step S202: Calculate the offset of the geometric center of the flow path area relative to the design center axis of the furnace eye on the detection cross section at a preset detection distance from the furnace eye outlet, and obtain the offset distance and offset orientation;

[0021] Step S203: Determine the stability index based on the cross-sectional area change rate over time of the foreground region of the stream and the angular change rate of the offset orientation. When the stability index is greater than a preset stability threshold, add the offset distance and offset orientation at the corresponding time to the effective offset sequence.

[0022] Step S204: Extract the annular sector division and the azimuth thermal field anomaly sequence from the azimuth thermal flux characterization data; discretize the continuous offset azimuth in the effective offset sequence into sector indices according to the annular sector division; calculate the product of the offset distance and the absolute value of the sine of the angle between the offset azimuth and the radial normal of the corresponding sector to obtain the offset sequence with asymmetric scour effect; combine the azimuth thermal field anomaly sequence and the offset sequence with asymmetric scour effect into the thermal flux coupling observation sequence.

[0023] Preferably, in step S202, the preset detection distance is a preset multiple L of the furnace eye design diameter; the offset distance and the offset orientation are obtained by determining the geometric center of the flow foreground area on the detection cross section and calculating the offset of the geometric center relative to the furnace eye design center axis.

[0024] Preferably, step S30, which involves performing a spatiotemporal correlation modeling task based on the heat-fluid coupling observation sequence using a hysteresis convolution matching mechanism and outputting a directional collaborative feature sequence, specifically includes:

[0025] Step S301: Extract the azimuth thermal field anomaly sequence and the offset sequence with asymmetric scour effect from the thermal flux coupling observation sequence, and perform time axis alignment resampling on the azimuth thermal field anomaly sequence and the offset sequence with asymmetric scour effect according to a preset sampling interval.

[0026] Step S302: Perform hysteresis convolution matching on the time-axis aligned azimuth thermal anomaly sequence and the offset sequence with asymmetric scour effect to construct a spatiotemporal correlation intensity map, which is calculated according to the following formula:

[0027]

[0028] Step S303: Extract the global peak position and global peak intensity of the spatiotemporal correlation intensity map. Determine the azimuth coordination index based on the total scour efficiency accumulation of the sector corresponding to the global peak intensity and the global peak position. The azimuth coordination index is calculated according to the following formula:

[0029]

[0030] In the formula, Indicates the first The blast furnace tapping time is delayed. and sector area The spatiotemporal correlation strength at that location; Indicates the first The fan-shaped area of ​​each tapping furnace At any moment The directional thermal field anomaly; Indicates the first The fan-shaped area of ​​each tapping furnace At any moment Asymmetric scouring efficiency; Represents a time variable. Indicated by time For integration variables; Indicates the lag time. The value range is from 0 to the preset maximum lag time. ; Indicates the first The global peak intensity of the spatiotemporal correlation intensity map of each tapping furnace; This indicates the peak lag time corresponding to the global peak position; This represents the sector index corresponding to the global peak position; Indicates the first The directional coordination index of each tapping furnace is obtained; the lower limit of integration, "stable tapping segment", represents the stable tapping time interval of the corresponding tapping furnace; the global peak position, the global peak intensity and the directional coordination index are arranged in the order of tapping furnaces to obtain the directional coordination feature sequence.

[0031] Preferably, step S40, which involves performing anomaly verification based on the azimuth collaborative feature sequence using a spatiotemporal density clustering mechanism and outputting anomalous loss azimuth cluster data, specifically includes:

[0032] Step S401: Calculate the historical average value and historical standard deviation of the azimuth coordination index according to the order of tapping furnaces, and determine the adaptive threshold by multiplying the historical average value by the preset threshold coefficient K and the historical standard deviation; when the azimuth coordination index of the current tapping furnace exceeds the adaptive threshold, record the global peak position, azimuth coordination index and global peak intensity of the current tapping furnace as a valid bimodal abnormal pointing event.

[0033] Step S402: Select the most recent preset number N effective bimodal anomaly pointing events from the azimuth coordination feature sequence, and use the peak sector angle, peak lag time and azimuth coordination index of each effective bimodal anomaly pointing event as three-dimensional features. Determine the spatial proximity distance based on the angle difference, lag time difference and corresponding preset mapping coefficients between two effective bimodal anomaly pointing events, and perform spatiotemporal density clustering based on the spatial proximity distance.

[0034] Step S403: For candidate clusters formed by spatiotemporal density clustering, when the number of sample points of the candidate cluster is greater than the preset minimum number of samples MinPts, the sector corresponding to the candidate cluster is determined as an abnormal loss orientation cluster, and the confidence level is determined according to the center orientation, average lag time, cluster density and total weight of the candidate cluster. The abnormal loss orientation cluster, the confidence level, the center orientation, the average lag time, the cluster density and the total weight are combined into the abnormal loss orientation cluster data.

[0035] Preferably, step S50, which involves using a cluster evolution analysis mechanism to perform a loss pattern inference task based on the abnormal loss location cluster data and outputting furnace eye maintenance and disposal data, specifically includes:

[0036] Step S501: Extract multiple consecutive abnormal loss azimuth cluster data according to the order of tapping furnaces, calculate the change in center azimuth and the change in average lag time of adjacent abnormal loss azimuth clusters respectively, and divide the change in center azimuth and the change in average lag time by the corresponding tapping furnace interval to obtain the azimuth evolution rate and the lag time evolution rate, and combine the azimuth evolution rate and the lag time evolution rate into a cluster evolution vector.

[0037] Step S502: Generate a total weight change curve based on the total weight in the continuous multiple abnormal loss azimuth cluster data; when the lag time evolution rate is less than a preset negative lag threshold and the growth rate of the total weight change curve is greater than a preset growth threshold, determine the loss mode as an accelerated erosion mode; when the azimuth evolution rate is greater than a preset azimuth change threshold, the average lag time is within a preset stable range, and the cluster density decreases, determine the loss mode as a stable scouring expansion mode; when the absolute values ​​of the azimuth evolution rate and the lag time evolution rate are both not greater than a preset stable evolution threshold and the growth rate of the total weight change curve is not greater than a preset smooth growth threshold, determine the loss mode as a static stable loss mode.

[0038] Step S503: Determine the furnace eye maintenance priority and repair recommendations based on the loss pattern, the abnormal loss location cluster, and the confidence level, and combine the furnace eye maintenance priority and the repair recommendations into the furnace eye maintenance and disposal data.

[0039] This invention also provides a system for identifying abnormal wear of the borehole lining in a ferrosilicon manganese submerged arc furnace based on the borehole orientation thermal field and stable flow stream offset, the system comprising:

[0040] The azimuth heat flow characterization module is used to acquire synchronous observation data of the furnace eye consisting of thermal image sequences of the furnace eye periphery and videos of the iron tapping stream. Based on the synchronous observation data of the furnace eye, the azimuth thermal field anomaly degree construction task is performed using a thermal response dynamic baseline compensation mechanism, and the azimuth heat flow characterization data is output.

[0041] The scour efficiency extraction module is used to perform the stream offset extraction task based on the azimuth heat flow characterization data using an asymmetric scour efficiency mechanism, and output the heat flow coupled observation sequence.

[0042] The spatiotemporal correlation matching module is used to perform spatiotemporal correlation modeling tasks based on the heat flow coupled observation sequence using a hysteresis convolution matching mechanism, and output the orientation cooperative feature sequence.

[0043] The abnormal loss confirmation module is used to perform an abnormal verification task based on the azimuth collaborative feature sequence using a spatiotemporal density clustering mechanism, and output abnormal loss azimuth cluster data.

[0044] The loss pattern inference module is used to perform loss pattern inference tasks based on the abnormal loss location cluster data using a cluster evolution analysis mechanism, and output furnace eye maintenance and disposal data.

[0045] This invention also provides a device for identifying abnormal wear of the borehole lining of a ferrosilicon manganese submerged arc furnace based on the borehole orientation thermal field and stable flow stream offset. The device includes a memory, a processor, and a program for identifying abnormal wear of the borehole lining based on the borehole orientation thermal field and stable flow stream offset stored in the memory and executable on the processor. When the program for identifying abnormal wear of the borehole lining based on the borehole orientation thermal field and stable flow stream offset is executed by the processor, the above method is implemented.

[0046] The present invention also provides a computer program product, the computer program product including a program for identifying abnormal wear of the lining of a ferrosilicon manganese submerged arc furnace based on the thermal field of the furnace eye orientation and the offset of the stable flow stream, the program for identifying abnormal wear of the lining of a ferrosilicon manganese submerged arc furnace based on the thermal field of the furnace eye orientation and the offset of the stable flow stream, the above method is implemented when the processor executes the program.

[0047] The beneficial effects of this invention are as follows:

[0048] This invention establishes individualized baselines for the differences in thermal response time constants of each annular sector through a dynamic baseline compensation mechanism for thermal response. This effectively separates the normal temperature rise during the tapping process from the temperature rise component caused by lining anomalies, enhancing the characterization ability of the azimuth thermal field anomaly sequence for lining damage. Simultaneously, an asymmetric scour efficiency factor is used to transform the geometric information of the flow stream offset into the actual scour intensity experienced by each sector, achieving a structured mapping from flow stream morphology to lining damage and providing accurate azimuth input for cross-modal correlation.

[0049] This invention utilizes hysteresis convolution matching combined with spatiotemporal density clustering to collaboratively verify the thermal field and flow stream modal data across multiple furnace dimensions. It automatically discovers the optimal spatiotemporal correspondence between scour orientation and hot spot orientation, and eliminates single-instance random interference through cluster confirmation, improving the reliability and confidence of identifying abnormal wear in the furnace lining. Furthermore, the introduction of cluster evolution analysis dynamically distinguishes different modes such as accelerated erosion, stable scour expansion, and static stable wear, generating furnace maintenance priorities and repair recommendations matching different degrees of deterioration. This avoids the blindness of maintenance scheduling based on experience, improving the targeting and economy of furnace maintenance strategies. Attached Figure Description

[0050] Figure 1 This is a flowchart illustrating the first embodiment of a method for identifying abnormal wear of the borehole lining of a ferrosilicon manganese submerged arc furnace based on the borehole orientation thermal field and stable flow stream offset, according to the present invention.

[0051] Figure 2 This is a schematic diagram of the time-series heat map of the azimuth thermal field anomaly degree of the first embodiment of the method for identifying abnormal losses of the lining of a ferrosilicon manganese submerged arc furnace based on the azimuth thermal field and the offset of stable flow streams of the present invention.

[0052] Figure 3 This is a schematic diagram of asymmetric scouring efficiency sector mapping, representing the first embodiment of a method for identifying abnormal losses in the borehole lining of a ferrosilicon manganese submerged arc furnace based on the borehole orientation thermal field and stable flow stream offset, according to the present invention.

[0053] Figure 4 This is a schematic diagram of the hysteresis convolution spatiotemporal correlation intensity map of the first embodiment of the present invention, which is a method for identifying abnormal losses of the furnace lining of a ferrosilicon manganese submerged arc furnace based on the furnace hole orientation thermal field and stable flow stream offset. Detailed Implementation

[0054] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0055] 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.

[0056] Example 1: As Figure 1 The diagram shown is a flowchart illustrating the first embodiment of a method for identifying abnormal wear of a ferrosilicon manganese submerged arc furnace lining based on the thermal field of the furnace eye orientation and the offset of a stable flow stream, according to the present invention.

[0057] In the first embodiment, the method for identifying abnormal wear of the borehole lining of a ferrosilicon manganese submerged arc furnace based on the borehole orientation thermal field and stable flow stream offset includes:

[0058] Step S10: Acquire synchronous observation data of the furnace eye consisting of thermal imaging sequences around the furnace eye and videos of the iron tapping stream. Based on the synchronous observation data of the furnace eye, perform the task of constructing the azimuth thermal field anomaly using a thermal response dynamic baseline compensation mechanism, and output azimuth thermal flow characterization data.

[0059] Synchronous observation data of the furnace eye refers to two types of observation information generated simultaneously within the same tapping furnace from the opening to the closing of the eye. These include a thermal image sequence of the furnace eye periphery and a video of the tapping flow aligned with the infrared thermal imager calibration. The thermal image sequence of the furnace eye periphery records the temperature distribution changes on the outer surface of the refractory lining around the furnace eye outlet, while the video of the tapping flow records the outline, center position, and deflection direction of the flow after the molten iron leaves the furnace eye. The dynamic baseline compensation mechanism for thermal response does not simply make a temperature threshold judgment on the thermal images. Instead, it sets an annular region of interest around the geometric center of the furnace eye outlet cross-section, divides this annular region into sector areas at equal angles along the circumference, extracts the highest and average temperatures for each sector, and performs moving median filtering according to the acquisition time. The dynamic thermal response baseline is jointly determined by the reference temperature before the opening of the current furnace, the historical empirical equilibrium temperature rise of the sector area, and the thermal response term that gradually approaches after the opening time. M represents the most recent preset number of normal furnaces, used to provide historical references for the highest temperature fluctuations during the stable tapping stage of the sector area. The resulting azimuth heat flow characterization data after this processing includes at least the azimuth thermal field anomaly sequence, annular sector division, and video of the tapping stream associated with the same furnace.

[0060] The azimuth heat flow characterization data transforms the originally dispersed pixel temperatures in continuous thermal images into an anomaly sequence arranged in circumferential sector areas of the furnace eye and progressing according to tapping time. Each sector area has an independent dynamic baseline and historical fluctuation scale. The current highest temperature only forms a high anomaly when it exceeds the reasonable thermal response range of that azimuth. The output obtained in this way can be directly used for step S20 to extract the flow video of the same furnace batch, and can also provide azimuth time sequence input for the thermal field side for step S30. Since the annular sector division is consistent with the furnace eye design center, after the subsequent flow offset azimuth is discretized to the same sector, the data of the two modes can be aligned on the same spatial index, avoiding the use of different coordinate references for the thermal field anomaly location and the flow scouring direction.

[0061] If a fixed temperature threshold or the overall average temperature trend is used for judgment, normal areas may be marked as abnormal when the overall furnace temperature rises, and early local losses may be masked when the furnace temperature is low. If only the overall average value is calculated, specific lining orientation differences such as the upper right, lower right, or left side will be smoothed out. This implementation method establishes an independent dynamic baseline for each sector area and standardizes the current deviation using the historical standard deviation of normal furnace cycles. This ensures that the thermal field anomaly degree retains its orientation positioning capability while adapting to the slow drift of heat loads in different furnace cycles. Common phenomena in on-site infrared images, such as slag buildup at the furnace mouth, flue gas obstruction, or local thermal shock at the moment of opening the furnace eye, are not easily converted into stable high anomalies under moving median filtering and dynamic baseline constraints, thus reducing false hotspot input for subsequent cross-modal verification.

[0062] During continuous tapping monitoring of a ferrosilicon manganese submerged arc furnace, slight spalling of the lining at approximately 210° azimuth to the lower right of the furnace bore was previously noted in the maintenance record. In the first two minutes after the borehole was opened in a particular furnace, the highest temperature in each sector increased synchronously with the molten iron flow. After the fourth minute, the highest temperature in the sector at approximately 210° azimuth remained consistently above its dynamic thermal response baseline, with the azimuth thermal anomaly index maintained at around 3.0, while adjacent sectors fluctuated between 1.0 and 1.3. Since this anomaly did not increase synchronously across the entire circumference but rather occurred concentrated in the same azimuth range, the azimuth thermal flow characterization data retained it as evidence of a local thermal field. The video of the tapping stream for this furnace was simultaneously stored in the same data structure. Subsequent step S20 could then extract the stable stream offset from this video and determine whether the offset direction corresponded to the scouring pattern at approximately 210° azimuth.

[0063] For example, such as Figure 2As shown, panel A constructs an azimuth thermal anomaly matrix using sector indices and sampling numbers. Darker colors indicate that certain sectors gradually experience abnormal increases in temperature after stable iron tapping, rather than a uniform temperature rise across the entire circumference. This heatmap visually shows that anomalies are not randomly scattered across all directions, but rather form continuous banded enhancements along the target sector and its neighboring sectors, indicating a clear circumferential direction in the thermal field changes. Panel B compares the azimuth profiles before and after stable iron tapping. In the first section, the anomalies in each sector are lower and more evenly distributed, while in the second section, a peak forms at approximately 75° azimuth, reflecting a gradual strengthening of the local thermal response of the lining during iron tapping. This figure corresponds to the time warping and anomaly construction process in S103 and S104, demonstrating that moving median filtering and dynamic baseline compensation do not simply generate a single final alarm value, but rather preserve the details of sector evolution over time. Taking a furnace with localized slag shedding around the furnace bore as an example, the instantaneous thermal image may show sharp high temperatures at several sampling points, but moving median filtering can suppress such short-term jumps. If the temperature rises continuously in a certain direction during the stable tapping section, and the adjacent sectors show a banded structure that gradually decreases with angle, it is more consistent with the heat transfer characteristics after the local thermal resistance of the lining decreases. For the subsequent S30 hysteresis convolution, this two-dimensional structure of time and orientation is very important because the thermal response caused by the flow scouring may appear with a lag. Only by retaining the complete matrix can it be determined whether the scouring efficiency sequence and the abnormal peak of the thermal field are in the same orientation and have a reasonable time lag, thereby avoiding taking the overall furnace temperature rise, camera exposure fluctuations, or single-point thermal noise as the basis for abnormal losses.

[0064] Step S20: Based on the azimuth heat flow characterization data, an asymmetric scour efficiency mechanism is used to perform the flow stream migration extraction task, and a heat flow coupled observation sequence is output;

[0065] After the azimuth heat flow characterization data enters step S20, the stream offset extraction task extracts the iron stream video and uses the stream instance segmentation model to identify the foreground region of the molten iron stream frame by frame. The so-called asymmetric scouring efficiency mechanism considers the distance, direction, and angle of the stream offset relative to the radial normal of the corresponding sector area, rather than recording only one offset angle. In specific processing, a detection cross-section is set at a preset detection distance from the furnace outlet. The preset detection distance can be determined according to a preset multiple L of the furnace design diameter. The geometric center of the stream foreground region is calculated on this cross-section and compared with the furnace design center axis to obtain the offset distance and offset azimuth. Subsequently, a stability index is determined by combining the time change rate of the cross-sectional area of ​​the stream foreground region and the angle change rate of the offset azimuth. Only when the stability index exceeds a preset stability threshold is it included in the valid offset sequence.

[0066] The effective migration sequence is further discretized into sector indices according to the annular sector obtained in step S10. A migration sequence with asymmetric scouring efficiency is obtained by multiplying the migration distance by the absolute value of the sine of the angle between the migration distance and the migration azimuth relative to the radial normal of the corresponding sector. This product implies that migrations of the same magnitude do not necessarily produce the same scouring effect on the lining; when the migration direction is closer to the radial force direction of a sector, the scouring efficiency of that sector is higher; when the migration direction is approximately along the circumference, the positive erosion contribution to that sector is relatively low. The final output thermal-fluid coupling observation sequence combines the azimuth thermal anomaly sequence and the migration sequence with asymmetric scouring efficiency, enabling step S30 to perform hysteresis convolution matching within the same furnace batch, the same time axis, and the same sector space.

[0067] Traditional stream flow video analysis often only extracts the average offset distance, maximum offset angle, or the main offset direction of the entire tapping section. These scalar results are insufficient to determine whether the offset is continuous within the stable tapping section, nor can they distinguish between approximately positive scouring and glancing flow along the wall. Initial splashing during the opening process, pre-blocking flow dispersion, and short-term furnace charge surface disturbances can cause significant jumps in the offset angle; without stability screening, these transient disturbances will be incorrectly included in the lining scouring assessment. This implementation first eliminates non-steady-state offsets, then converts the remaining offsets into scouring efficiencies arranged in fan-shaped zones. This transforms the stream flow mode from video contour information into a load sequence that can be compared point-by-point with thermal field orientation anomalies, preventing subsequent correlation analysis from being influenced by random fluctuations.

[0068] In a 12-minute video of molten iron being tapped, the flow stream appears as a continuous bundle for about 7 minutes, with minimal changes in cross-sectional area. The azimuth offset remains relatively stable around 75°, with a displacement distance of approximately 18 mm and azimuth angle fluctuations mostly within 6°. If the 75° area corresponds to sector 15 in the annular sector division, this sector exhibits higher asymmetric scouring efficiency due to the proximity of its radial normal to the offset direction. While adjacent sectors 14 and 16 are also affected, their scouring efficiency is significantly reduced due to the increased angle between them. This result does not simply assume that the entire flow stream shifts to the upper left; instead, it outputs a time-varying thermal-fluid coupling observation sequence with sector affiliation. If sector 15 experiences an abnormal increase in thermal field at a later time in step S10, step S30 can verify whether the two constitute a true synergy in the corresponding azimuth and lag time.

[0069] For example, such as Figure 3As shown, panel A maps the stable flow offset direction to the circumferential sectors of the borehole and uses color to represent the asymmetric scouring efficiency of each sector. The red arrow points to the main stable offset direction, indicating that the flow is not simply "offset in a certain direction," but rather generates a concentrated scouring load on the lining of the corresponding sector. Panel B further magnifies the scouring efficiency of the target sector and adjacent sectors, showing that the target sector has the highest efficiency, while the efficiency of adjacent sectors decreases rapidly as the angle between the offset azimuth and the radial normal increases. This figure illustrates the effect of "the product of the offset distance and the absolute value of the sine of the angle between the offset azimuth and the radial normal of the sector" in S204: at the same offset distance, the erosion efficiency caused by scouring perpendicularly to a sector and scouring passing through that sector are not the same. Taking the case of slight ellipticization on the inner side of the borehole outlet as an example, the flow may be offset to one side for a long time, but the effective cutting component of the offset vector for different sectors is not consistent; if only the offset distance is recorded, it is easy to regard all adjacent azimuths as equally damaged, resulting in an overly wide repair range. S204 breaks down the flow offset into 72 sectors, creating a peak in the target sector and an interpretable attenuation band in adjacent sectors. This aligns with the mechanical directionality of the flow's local scouring of the liner and facilitates sector-by-sector alignment with the azimuth thermal anomaly output by S10. Thus, the video information is no longer an isolated image measurement but is transformed into a scouring efficiency sequence that can be modeled together with the thermal imaging sequence. This provides input for S30 to determine the hysteresis relationship of "scouring occurring first, followed by thermal anomalies," and allows for the differentiation between primary repair locations and transitional inspection locations in subsequent maintenance recommendations.

[0070] Step S30: Based on the heat flow coupled observation sequence, a hysteresis convolution matching mechanism is used to perform a spatiotemporal correlation modeling task, and the orientation cooperative feature sequence is output;

[0071] The thermal-fluid coupling observation sequence is a dual-modal time series data composed of an azimuth thermal field anomaly sequence and an offset sequence with asymmetric scour efficiency. When processing this sequence using the hysteresis convolution matching mechanism, the two types of sequences are first resampled along their time axes according to a preset sampling interval, ensuring that the thermal field anomaly and scour efficiency are at the same time grid. Then, taking each sector as a unit, candidate hysteresis values ​​are attempted one by one within the range from zero hysteresis to the preset maximum hysteresis time. After shifting the scour efficiency sequence according to the candidate hysteresis values, it is matched and accumulated with the azimuth thermal field anomaly of the same sector within the stable tapping time interval, forming a spatiotemporal correlation intensity map arranged by sector and hysteresis time. The sector and hysteresis time corresponding to the global peak position in the map represent the azimuth and response delay where the flow scour and thermal field response are most consistent within that furnace cycle.

[0072] After obtaining the spatiotemporal correlation intensity map, the system extracts the global peak intensity, peak lag time, and peak sector index. Then, it determines the azimuth coordination index by combining the total scouring efficiency accumulation of this peak sector within the stable tapping section. The azimuth coordination index does not simply reflect whether the thermal anomaly is high, nor does it simply reflect whether the stream offset is large. Instead, it measures whether the stream scouring evidence and the thermal field response evidence are consistent in azimuth and whether there is a reasonable lag in time within the same heat. The output azimuth coordination feature sequence arranges the global peak position, global peak intensity, and azimuth coordination index in heat order, enabling step S40 to use the dual-modal coordination results of each heat as candidate anomaly pointing events, rather than reprocessing the original thermal images and videos.

[0073] Directly comparing thermal anomalies and flow deviations at the same moment can lead to a suppression of the true correlation due to the heat capacity of the lining material, the heat conduction path, and the imaging delay of the outer surface. Averaging over the entire tapping section can also smooth out the scouring peaks that continuously act on a certain sector within a short period. This implementation searches for the optimal match in both orientation and lag time dimensions, identifying the physical process by which flow scouring occurs first and thermal anomalies follow. Simultaneously, it uses the accumulated total scouring efficiency to constrain the peak value's scale, preventing differences in absolute correlation values ​​caused by variations in tapping time, total molten iron volume, or higher flow velocities in individual furnaces from directly becoming anomaly judgments. The resulting orientation-coordinated feature sequence is more suitable for cross-furnace comparisons.

[0074] Within the stable tapping section of a certain furnace, the scouring effect caused by the flow deviation concentrated on sector 40 starting approximately 2 minutes after tapping began, while the azimuth thermal anomaly of the same sector began to rise significantly after approximately 2.5 minutes. If compared with zero hysteresis, the thermal field had not yet increased during the initial scouring enhancement, resulting in a low matching strength. However, after using hysteresis convolution matching, when the candidate hysteresis time was approximately 30 seconds, the correlation strength of sector 40 reached the global peak for that furnace. Based on this, the system recorded the peak sector as sector 40, the peak hysteresis time as approximately 30 seconds, and obtained a high azimuth coordination index after combining it with the total scouring effect of that sector. This furnace was then sent to step S40 as evidence of a dual-modal anomaly with clear azimuth and hysteresis characteristics.

[0075] For example, such as Figure 4As shown, panel A presents a spatiotemporal correlation intensity map with sector index and lag time as axes. The brightest region appears near the target sector and corresponds to a lag of approximately 30 seconds, indicating that after the scouring efficiency of the stable flow stream is enhanced, the thermal anomaly does not appear immediately and synchronously, but rather with a delay due to the effects of the liner's heat capacity and heat conduction process. Panel B compares the lag profiles of the target sector and adjacent weakly correlated sectors. The target sector forms a peak at approximately 30 seconds, while the curves of adjacent sectors are relatively flat, indicating that this peak is not caused by global thermal noise or uniform temperature rise, but only appears when both orientation and lag conditions are met. This figure corresponds to S301 to S303, demonstrating that lag convolution matching simultaneously solves two problems: first, the thermal response has a time delay relative to the scouring effect; second, the anomaly must fall within the corresponding orientation to have physical meaning. Taking a stable flow stream in a certain furnace batch that is continuously biased towards the upper right of the furnace bore as an example, the offset has already formed in the early part of the video, but the temperature anomaly in this location in the thermal image often only increases significantly after the lining heat conduction, surface radiation enhancement, and thermal image sampling accumulation. If only synchronous correlation is performed, the peak value of the thermal field will be misaligned with the peak value of the flow stream, resulting in a low correlation coefficient; if only average correlation of the entire furnace bore is performed, the normal temperature rise in other sectors will dilute the target location signal. S30, by performing convolution matching on the same sector at multiple lag times, reveals the physically reasonable delayed response as a local peak value, enabling the system to lock the dual-modal collaborative anomaly to a specific sector and reasonable lag time, and use the position of this peak value as the key basis for the S40 anomaly pointing to the event.

[0076] Step S40: Based on the azimuth collaborative feature sequence, a spatiotemporal density clustering mechanism is used to perform an anomaly verification task and output anomaly loss azimuth cluster data;

[0077] After the azimuth coordination feature sequence enters step S40, the system first calculates the historical average and historical standard deviation of the azimuth coordination index according to the furnace order, and uses a preset threshold coefficient K to determine an adaptive threshold that updates with historical data. When the azimuth coordination index of the current furnace exceeds this threshold, it indicates that there is a bimodal consistency between the thermal field anomaly and the flow stream scouring of the furnace that exceeds the recent background level. At this time, the global peak position, azimuth coordination index, and global peak intensity of the current furnace are recorded as valid bimodal anomaly pointing events. Subsequently, the system selects N valid bimodal anomaly pointing events of the most recent preset number of furnaces, uses the peak sector angle, peak lag time, and azimuth coordination index as three-dimensional features, and calculates the spatial proximity distance between events according to the angle difference, lag time difference, and corresponding mapping coefficient.

[0078] The spatiotemporal density clustering mechanism searches for dense regions with similar orientations, lags, and high coherence within the effective event set. For candidate clusters, only when the number of sample points exceeds the preset minimum sample size MinPts, and the events within the cluster exhibit stable consistency in terms of center orientation and average lag time, is the corresponding sector area of ​​the candidate cluster identified as an anomalous loss orientation cluster. The confidence level is jointly determined by the candidate cluster's center orientation, average lag time, cluster density, and total weight. The total weight reflects the cumulative strength of bimodal evidence from multiple furnace cycles, while the cluster density reflects the concentration of this evidence in the orientation and lag space. The final output of anomalous loss orientation cluster data includes not only anomalous orientations but also confidence levels, center orientations, average lag times, cluster density, and total weight, providing a continuously trackable object for step S50 to determine the loss pattern.

[0079] Single-heater independent alarms are easily affected by sporadic factors such as slag buildup, short-term flue gas obstruction, and instantaneous changes in the taphole shape. Any anomaly in the thermal imaging or the flow stream can trigger an excessive alarm. However, simply requiring both modes to simultaneously exceed a fixed threshold in the same heat can miss the true early losses when the thermal field baseline drifts or the flow stream scouring is weak. This implementation first uses an adaptive threshold to measure the coordination index back in the recent heat condition background, and then uses multi-heater spatiotemporal density clustering to determine whether the abnormal evidence repeatedly occurs in the same location. Even if isolated high-coordination events have large values, they are difficult to form an abnormal loss location cluster due to the lack of cross-heater repetition, thus making the output results closer to the actual loss location of the lining.

[0080] When continuously monitoring 20 furnace runs, step S30 may record 7 valid bimodal anomalous pointing events. Six of these events have peak azimuths concentrated between 18° and 22°, and peak lag times concentrated between 25 and 35 seconds. The remaining event is located near 120°. After clustering, the first six events, due to their similar azimuths and lag times, and the fact that the number of sample points meets the MinPts requirement, form an anomalous wear azimuth cluster with a central azimuth of approximately 20°. This cluster has high density and a large total weight, achieving a confidence level of approximately 0.9. Although the isolated event near 120° has a high synergy index within a single furnace run, it does not form a dense distribution with other furnace runs and therefore will not be output as an anomalous wear azimuth cluster. This result indicates that continuous wear is more likely to exist in the furnace eye lining at an azimuth of approximately 20°, and this judgment is supported by cross-furnace evidence.

[0081] Step S50: Based on the abnormal loss location cluster data, the cluster evolution analysis mechanism is used to perform the loss pattern inference task and output the furnace eye maintenance and disposal data;

[0082] After the abnormal loss azimuth cluster data enters step S50, the system does not only read the abnormal azimuth at a single moment, but extracts multiple consecutive abnormal loss azimuth cluster data according to the tapping furnace sequence, comparing the center azimuth and average lag time of adjacent clusters. The change in center azimuth is divided by the corresponding tapping furnace interval to obtain the azimuth evolution rate; the change in average lag time is divided by the corresponding tapping furnace interval to obtain the lag time evolution rate. These two together form a cluster evolution vector, used to describe whether the abnormal loss evidence is moving in the circumferential direction of the furnace eye, and whether the thermal field response is gradually advancing or delaying relative to the flow stream scouring. Simultaneously, the system generates a total weight change curve based on the total weight in the consecutive clusters, used to describe whether the loss evidence is strengthening.

[0083] The loss mode inference task classifies the loss patterns based on cluster evolution vectors, total weight change curves, and cluster density changes. When the lag time evolution rate is less than a preset negative lag threshold, and the total weight change curve growth rate is greater than a preset growth threshold, it indicates that the time required for the same type of scouring to trigger a thermal response on the outer surface is shortening, and the dual-modal evidence is continuously strengthening, thus identifying it as an accelerated erosion mode. When the azimuth evolution rate is greater than a preset azimuth change threshold, the average lag time is within a preset stable range, and the cluster density is decreasing, it indicates that abnormal evidence is expanding along the circumference of the borehole, thus identifying it as a stable scouring expansion mode. When the absolute values ​​of both the azimuth evolution rate and the lag time evolution rate do not exceed a preset stable evolution threshold, and the total weight change curve growth rate does not exceed a preset gradual growth threshold, it can be identified as a static stable loss mode. The system then determines the borehole maintenance priority and repair recommendations based on the loss mode, abnormal loss azimuth clusters, and confidence level, and combines the two into borehole maintenance and disposal data.

[0084] Traditional maintenance strategies, even if they detect abnormal thermal imaging or flow patterns in a certain location, typically only provide a static alarm location, failing to indicate whether the wear at that location is accelerating, expanding circumferentially, or already in a slow, stable state. Fixed-cycle maintenance may prematurely shut down the furnace when wear is still minor, or fail to address the issue in a timely manner during the accelerated erosion phase. This implementation method connects clusters of abnormal wear locations from multiple furnaces into an evolution trajectory, allowing maintenance priorities to be determined by actual trends: accelerated erosion patterns correspond to higher priority, stable erosion expansion patterns correspond to near-term planned repairs, and static stable wear patterns can continue to be observed and incorporated into routine maintenance. The output furnace maintenance and handling data thus includes not only location but also the timing of handling and the direction of repair, facilitating more rational on-site arrangements between safety risks and shutdown costs.

[0085] In the tracking of 15 consecutive furnace runs, the cluster of abnormal wear at approximately 30° azimuth initially had a stable center azimuth, but starting from the 4th furnace run, it drifted approximately 0.3° to adjacent azimuths per furnace run, reaching a center azimuth of around 36° by the 10th furnace run. Simultaneously, the average lag time remained around 25 seconds, the cluster density gradually decreased, and the total weight increased approximately linearly. Based on this, the system determined that it conformed to a stable erosion expansion pattern, and the output furnace eye maintenance data could be prioritized for near-term planned repairs, with a recommendation to expand the scope of review and repair around the azimuth between 30° and 36°. Another cluster located around 60°, whose average lag time decreased from 28 seconds to 12 seconds over 5 furnace runs, and whose total weight increased even faster, was identified as an accelerated erosion pattern, with a higher maintenance priority than the aforementioned expanded cluster.

[0086] Example 2: Furthermore, the present invention provides a system for identifying abnormal wear of the ferrosilicon manganese submerged arc furnace lining based on the borehole orientation thermal field and stable flow stream offset. This system employs a method for identifying abnormal wear of the ferrosilicon manganese submerged arc furnace lining based on the borehole orientation thermal field and stable flow stream offset described in the above embodiments, thus solving the technical problem of identifying abnormal wear of the ferrosilicon manganese submerged arc furnace lining based on the borehole orientation thermal field and stable flow stream offset. The beneficial effects of the system for identifying abnormal wear of the ferrosilicon manganese submerged arc furnace lining based on the borehole orientation thermal field and stable flow stream offset provided by the present invention are the same as those of the method for identifying abnormal wear of the ferrosilicon manganese submerged arc furnace lining based on the borehole orientation thermal field and stable flow stream offset described in the above embodiments. Other technical features of the system for identifying abnormal wear of the ferrosilicon manganese submerged arc furnace lining based on the borehole orientation thermal field and stable flow stream offset are the same as those disclosed in the methods of the above embodiments, and will not be repeated here.

[0087] Example 3: This invention provides a device for identifying abnormal wear of the borehole lining of a ferrosilicon manganese submerged arc furnace based on the borehole orientation thermal field and stable flow stream offset. The device includes at least one processor and a memory communicatively connected to the processor. The memory stores instructions executable by the processor, which are then executed to enable the processor to perform the method for identifying abnormal wear of the borehole lining of a ferrosilicon manganese submerged arc furnace based on the borehole orientation thermal field and stable flow stream offset described in Example 1. The device for identifying abnormal wear of the borehole lining of a ferrosilicon manganese submerged arc furnace based on the borehole orientation thermal field and stable flow stream offset in this invention can include, but is not limited to, mobile terminals such as mobile phones, laptops, digital radio receivers, PDAs (Personal Digital Assistants), PADs (Portable Application Description), PMPs (Portable Media Players), and vehicle terminals (e.g., vehicle navigation terminals), as well as fixed terminals such as digital TVs and desktop computers. A device for identifying abnormal wear of the borehole lining of a ferrosilicon manganese submerged arc furnace based on the borehole orientation thermal field and stable flow stream offset is merely an example and should not limit the functionality or scope of application of the embodiments of the present invention. Such a device may include a processing unit (e.g., a central processing unit, a graphics processing unit, etc.) that can perform various appropriate actions and processes according to a program stored in a read-only memory or a program loaded from a storage device into a random access memory. The random access memory also stores various programs and data required for the operation of the device. The processing unit, the read-only memory, and the random access memory are interconnected via a bus. I / O interfaces are also connected to the bus. Typically, the following systems can be connected to the I / O interface: input devices including touchscreens, touchpads, keyboards, mice, image sensors, microphones, accelerometers, gyroscopes, etc.; output devices including liquid crystal displays (LCDs), speakers, vibrators, etc.; storage devices including magnetic tapes, hard disks, etc.; and communication devices. The communication device allows a device for identifying abnormal wear of the ferrosilicon manganese submerged arc furnace lining based on the borehole orientation thermal field and stable flow stream offset to exchange data wirelessly or via wired communication with other devices. While an abnormal wear identification device for the ferrosilicon manganese submerged arc furnace lining based on the borehole orientation thermal field and stable flow stream offset has been described, it should be understood that implementation or possession of all described systems is not required.It can be implemented alternatively or with more or fewer systems.

[0088] Example 4: This invention also provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the above-described method for identifying abnormal wear of a ferrosilicon manganese submerged arc furnace lining based on the borehole orientation thermal field and stable flow stream offset. The computer program product provided by this invention can solve the technical problem of identifying abnormal wear of a ferrosilicon manganese submerged arc furnace lining based on the borehole orientation thermal field and stable flow stream offset. Compared with the prior art, the beneficial effects of the computer program product provided by this invention are the same as those of the above-described method for identifying abnormal wear of a ferrosilicon manganese submerged arc furnace lining based on the borehole orientation thermal field and stable flow stream offset, and will not be repeated here.

[0089] In particular, according to the embodiments disclosed in this invention, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of this invention include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device, or installed from a storage device, or installed from a read-only memory. When the computer program is executed by a processing device, it performs the functions defined in the methods of the embodiments disclosed in this invention.

[0090] It should be understood that the various parts disclosed in this invention can be implemented using hardware, software, firmware, or a combination thereof. In the description of the above embodiments, specific features, structures, materials, or characteristics may be combined in any suitable manner in one or more embodiments or examples.

[0091] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the present invention and its equivalents, the present invention also intends to include these modifications and variations.

Claims

1. A method for identifying abnormal wear of the borehole lining in a ferrosilicon manganese submerged arc furnace based on the borehole orientation thermal field and stable flow stream offset, characterized in that, The methods include: Step S10: Acquire synchronous observation data of the furnace eye consisting of thermal imaging sequences around the furnace eye and videos of the iron tapping stream. Based on the synchronous observation data of the furnace eye, perform the task of constructing the azimuth thermal field anomaly using a thermal response dynamic baseline compensation mechanism, and output azimuth thermal flow characterization data. Step S20: Based on the azimuth heat flow characterization data, an asymmetric scour efficiency mechanism is used to perform the flow stream migration extraction task, and a heat flow coupled observation sequence is output; Step S30: Based on the heat flow coupled observation sequence, a hysteresis convolution matching mechanism is used to perform a spatiotemporal correlation modeling task, and the orientation cooperative feature sequence is output; Step S40: Based on the azimuth collaborative feature sequence, a spatiotemporal density clustering mechanism is used to perform an anomaly verification task and output anomaly loss azimuth cluster data; Step S50: Based on the abnormal loss location cluster data, the cluster evolution analysis mechanism is used to perform the loss pattern inference task and output furnace eye maintenance and disposal data.

2. The method for identifying abnormal wear of the borehole lining of a ferrosilicon manganese submerged arc furnace based on the borehole orientation thermal field and stable flow stream offset as described in claim 1, characterized in that, Step S10 involves acquiring synchronous observation data of the furnace borehole, consisting of a thermal imaging sequence of the furnace borehole periphery and a video of the tapping iron stream. Based on this synchronous observation data, a dynamic baseline compensation mechanism for thermal response is used to construct the azimuth thermal field anomaly, and the azimuth thermal flow characterization data is output. Specifically, this includes: Step S101: Set up an infrared thermal imager in front of the furnace eye, and set up a visible light camera or near-infrared camera that is aligned with the calibration of the infrared thermal imager. Synchronously collect thermal image sequences and iron flow videos of the furnace eye periphery area during each tapping furnace from opening to closing. Associate the thermal image sequences and the iron flow videos according to the collection timestamp to obtain the synchronous observation data of the furnace eye. Step S102: Extract the thermal image sequence from the synchronous observation data of the furnace eye, set an annular region of interest with the geometric center of the furnace eye outlet section as the origin, set the inner diameter of the annular region of interest to 0.7 to 0.9 times the design diameter of the furnace eye, set the outer diameter of the annular region of interest to 2.5 to 3.5 times the design diameter of the furnace eye, and divide the annular region of interest into 72 sector areas at equal angles along the circumference to obtain the annular sector division; Step S103: For each frame of thermal image of each tapping furnace, extract the highest temperature and average temperature of each sector area, and perform moving median filtering on the highest temperature and average temperature arranged according to the acquisition time to obtain a time-regular azimuth thermal field sequence. Step S104: For each sector, divide the difference between the current highest temperature and the dynamic thermal response baseline by the standard deviation of the highest temperature during the stable tapping stage of the most recent preset number of normal furnaces M in that sector to obtain the azimuth thermal field anomaly sequence. The dynamic thermal response baseline consists of the reference temperature before the furnace opening, the historical empirical equilibrium temperature rise of the sector, and an exponential approach term determined by the time after the furnace opening and the thermal response time constant of the sector. Extract the tapping stream video from the furnace eye synchronous observation data, and combine the azimuth thermal field anomaly sequence, the annular sector division, and the tapping stream video into the azimuth thermal flow characterization data.

3. The method for identifying abnormal wear of the borehole lining of a ferrosilicon manganese submerged arc furnace based on the borehole orientation thermal field and stable flow stream offset as described in claim 2, characterized in that, Step S20, which involves performing the stream migration extraction task using an asymmetric scour efficiency mechanism based on the azimuth heat flow characterization data and outputting a heat flow coupled observation sequence, specifically includes: Step S201: Extract the iron-producing stream video from the azimuth heat flow characterization data, and extract the stream foreground region from the iron-producing stream video using a stream instance segmentation model; Step S202: Calculate the offset of the geometric center of the flow path area relative to the design center axis of the furnace eye on the detection cross section at a preset detection distance from the furnace eye outlet, and obtain the offset distance and offset orientation; Step S203: Determine the stability index based on the cross-sectional area change rate over time of the foreground region of the stream and the angular change rate of the offset orientation. When the stability index is greater than a preset stability threshold, add the offset distance and offset orientation at the corresponding time to the effective offset sequence. Step S204: Extract the annular sector division and the azimuth thermal field anomaly sequence from the azimuth thermal flux characterization data; discretize the continuous offset azimuth in the effective offset sequence into sector indices according to the annular sector division; calculate the product of the offset distance and the absolute value of the sine of the angle between the offset azimuth and the radial normal of the corresponding sector to obtain the offset sequence with asymmetric scour effect; combine the azimuth thermal field anomaly sequence and the offset sequence with asymmetric scour effect into the thermal flux coupling observation sequence.

4. The method for identifying abnormal wear of the borehole lining of a ferrosilicon manganese submerged arc furnace based on the borehole orientation thermal field and stable flow stream offset as described in claim 3, characterized in that, In step S202, the preset detection distance is a preset multiple L of the furnace eye design diameter; the offset distance and the offset orientation are obtained by determining the geometric center of the flow foreground area on the detection cross section and calculating the offset of the geometric center relative to the furnace eye design center axis.

5. The method for identifying abnormal wear of the borehole lining of a ferrosilicon manganese submerged arc furnace based on the borehole orientation thermal field and stable flow stream offset as described in claim 4, characterized in that, Step S30, which involves performing a spatiotemporal correlation modeling task based on the heat-fluid coupling observation sequence using a hysteresis convolution matching mechanism and outputting a directional cooperative feature sequence, specifically includes: Step S301: Extract the azimuth thermal field anomaly sequence and the offset sequence with asymmetric scour effect from the thermal flux coupling observation sequence, and perform time axis alignment resampling on the azimuth thermal field anomaly sequence and the offset sequence with asymmetric scour effect according to a preset sampling interval. Step S302: Perform hysteresis convolution matching on the time-axis aligned azimuth thermal anomaly sequence and the offset sequence with asymmetric scour effect to construct a spatiotemporal correlation intensity map, which is calculated according to the following formula: Step S303: Extract the global peak position and global peak intensity of the spatiotemporal correlation intensity map. Determine the azimuth coordination index based on the total scour efficiency accumulation of the sector corresponding to the global peak intensity and the global peak position. The azimuth coordination index is calculated according to the following formula: In the formula, Indicates the first The blast furnace tapping time is delayed. and sector area The spatiotemporal correlation strength at that location; Indicates the first The fan-shaped area of ​​each tapping furnace At any moment The directional thermal field anomaly; Indicates the first The fan-shaped area of ​​each tapping furnace At any moment Asymmetric scouring efficiency; Represents a time variable. Indicated by time For integration variables; Indicates the lag time. The value range is from 0 to the preset maximum lag time. ; Indicates the first The global peak intensity of the spatiotemporal correlation intensity map of each tapping furnace; This indicates the peak lag time corresponding to the global peak position; This represents the sector index corresponding to the global peak position; Indicates the first The directional coordination index of each tapping furnace is obtained; the lower limit of integration, "stable tapping segment", represents the stable tapping time interval of the corresponding tapping furnace; the global peak position, the global peak intensity and the directional coordination index are arranged in the order of tapping furnaces to obtain the directional coordination feature sequence.

6. The method for identifying abnormal wear of the borehole lining of a ferrosilicon manganese submerged arc furnace based on the borehole orientation thermal field and stable flow stream offset as described in claim 5, characterized in that, Step S40, which involves performing anomaly verification based on the azimuth coordination feature sequence using a spatiotemporal density clustering mechanism and outputting anomalous loss azimuth cluster data, specifically includes: Step S401: Calculate the historical average value and historical standard deviation of the azimuth coordination index according to the order of tapping furnaces, and determine the adaptive threshold by multiplying the historical average value by the preset threshold coefficient K and the historical standard deviation; when the azimuth coordination index of the current tapping furnace exceeds the adaptive threshold, record the global peak position, azimuth coordination index and global peak intensity of the current tapping furnace as a valid bimodal abnormal pointing event. Step S402: Select the most recent preset number N effective bimodal anomaly pointing events from the azimuth coordination feature sequence, and use the peak sector angle, peak lag time and azimuth coordination index of each effective bimodal anomaly pointing event as three-dimensional features. Determine the spatial proximity distance based on the angle difference, lag time difference and corresponding preset mapping coefficients between two effective bimodal anomaly pointing events, and perform spatiotemporal density clustering based on the spatial proximity distance. Step S403: For candidate clusters formed by spatiotemporal density clustering, when the number of sample points of the candidate cluster is greater than the preset minimum number of samples MinPts, the sector corresponding to the candidate cluster is determined as an abnormal loss orientation cluster, and the confidence level is determined according to the center orientation, average lag time, cluster density and total weight of the candidate cluster. The abnormal loss orientation cluster, the confidence level, the center orientation, the average lag time, the cluster density and the total weight are combined into the abnormal loss orientation cluster data.

7. The method for identifying abnormal wear of the borehole lining of a ferrosilicon manganese submerged arc furnace based on the borehole orientation thermal field and stable flow stream offset as described in claim 6, characterized in that, Step S50, which involves using a cluster evolution analysis mechanism to perform a loss pattern inference task based on the abnormal loss location cluster data and outputting furnace eye maintenance and disposal data, specifically includes: Step S501: Extract multiple consecutive abnormal loss azimuth cluster data according to the order of tapping furnaces, calculate the change in center azimuth and the change in average lag time of adjacent abnormal loss azimuth clusters respectively, and divide the change in center azimuth and the change in average lag time by the corresponding tapping furnace interval to obtain the azimuth evolution rate and the lag time evolution rate, and combine the azimuth evolution rate and the lag time evolution rate into a cluster evolution vector. Step S502: Generate a total weight change curve based on the total weight in the continuous multiple abnormal loss azimuth cluster data; when the lag time evolution rate is less than a preset negative lag threshold and the growth rate of the total weight change curve is greater than a preset growth threshold, determine the loss mode as an accelerated erosion mode; when the azimuth evolution rate is greater than a preset azimuth change threshold, the average lag time is within a preset stable range, and the cluster density decreases, determine the loss mode as a stable scouring expansion mode; when the absolute values ​​of the azimuth evolution rate and the lag time evolution rate are both not greater than a preset stable evolution threshold and the growth rate of the total weight change curve is not greater than a preset smooth growth threshold, determine the loss mode as a static stable loss mode. Step S503: Determine the furnace eye maintenance priority and repair recommendations based on the loss pattern, the abnormal loss location cluster, and the confidence level, and combine the furnace eye maintenance priority and the repair recommendations into the furnace eye maintenance and disposal data.

8. A system for identifying abnormal wear of the borehole lining of a ferrosilicon manganese submerged arc furnace based on the borehole orientation thermal field and stable flow stream offset, applied to the method for identifying abnormal wear of the borehole lining of a ferrosilicon manganese submerged arc furnace based on the borehole orientation thermal field and stable flow stream offset as described in any one of claims 1 to 7, characterized in that, The abnormal wear identification system for the borehole lining of the ferrosilicon manganese submerged arc furnace includes: The azimuth heat flow characterization module is used to acquire synchronous observation data of the furnace eye consisting of thermal image sequences of the furnace eye periphery and videos of the iron tapping stream. Based on the synchronous observation data of the furnace eye, the azimuth thermal field anomaly degree construction task is performed using a thermal response dynamic baseline compensation mechanism, and the azimuth heat flow characterization data is output. The scour efficiency extraction module is used to perform the stream offset extraction task based on the azimuth heat flow characterization data using an asymmetric scour efficiency mechanism, and output the heat flow coupled observation sequence. The spatiotemporal correlation matching module is used to perform spatiotemporal correlation modeling tasks based on the heat flow coupled observation sequence using a hysteresis convolution matching mechanism, and output the orientation cooperative feature sequence. The abnormal loss confirmation module is used to perform an abnormal verification task based on the azimuth collaborative feature sequence using a spatiotemporal density clustering mechanism, and output abnormal loss azimuth cluster data. The loss pattern inference module is used to perform loss pattern inference tasks based on the abnormal loss location cluster data using a cluster evolution analysis mechanism, and output furnace eye maintenance and disposal data.

9. A device for identifying abnormal wear of the borehole lining in a ferrosilicon manganese submerged arc furnace based on the borehole orientation thermal field and stable flow stream offset, characterized in that, The device for identifying abnormal wear of the ferrosilicon manganese submerged arc furnace lining based on the borehole orientation thermal field and stable flow stream offset includes: a memory, a processor, and a program for identifying abnormal wear of the ferrosilicon manganese submerged arc furnace lining based on the borehole orientation thermal field and stable flow stream offset stored in the memory and executable on the processor. When the program for identifying abnormal wear of the ferrosilicon manganese submerged arc furnace lining based on the borehole orientation thermal field and stable flow stream offset is executed by the processor, it implements the method for identifying abnormal wear of the ferrosilicon manganese submerged arc furnace lining based on the borehole orientation thermal field and stable flow stream offset as described in any one of claims 1 to 7.

10. A computer program product, characterized in that, The computer program product includes a program for identifying abnormal wear of the lining of a ferrosilicon manganese submerged arc furnace based on the thermal field of the borehole orientation and the offset of the stable flow stream. When the program is executed by the processor, it implements a method for identifying abnormal wear of the lining of a ferrosilicon manganese submerged arc furnace based on the thermal field of the borehole orientation and the offset of the stable flow stream, as described in any one of claims 1 to 7.

Images