An intelligent inspection system and method for an industrial furnace

Abstract

This invention discloses an intelligent inspection system and method for industrial furnaces and kilns, belonging to the field of high-temperature furnace and kiln monitoring technology. It includes: a mobile carrier terminal for furnace and kiln inspection, a network transmission layer, a data processing layer, and an application output layer. The mobile carrier terminal includes a dual-view infrared thermal imaging module, a full-frame temperature acquisition unit, an embedded heterogeneous processor, a preprocessing unit, a dynamic target recognition unit, a dual-view geometric registration unit, an oxide spot recognition and correction unit, a temperature calculation unit, and an industrial-grade wireless transmission module. The network transmission layer includes an industrial server cluster for receiving, parsing, load balancing, and caching furnace and kiln inspection data. The data processing layer includes a time-series database and a data analysis server for storing historical data of the furnace and kiln, detecting safety hazards, analyzing temperature trends, and generating maintenance instructions. This system solves the problem of temperature measurement distortion in furnaces and kilns, improves inspection efficiency, and reduces temperature measurement errors to a level suitable for intelligent inspection and safety monitoring of various types of furnaces and kilns.

Description

Technical Field

[0001] This invention belongs to the field of high-temperature furnace and kiln detection technology, and particularly relates to an intelligent inspection system and inspection method for industrial furnaces and kilns. Background Technology

[0002] Industrial furnaces and kilns operate under harsh conditions of high temperature, high pressure, and high electric field for extended periods. The surfaces of heated metal surfaces and conductive components are prone to developing oxide spots that are dark in color, loose in structure, and irregular in texture due to thermal stress, electro-corrosion, and moisture oxidation. These oxide spots not only accelerate equipment aging and reduce its service life, but also alter the surface infrared emissivity and reflection characteristics, severely interfering with the accuracy of infrared temperature measurement. They can even mask potential localized overheating hazards and lead to major safety accidents such as pipe bursts, equipment breakdowns, and arc short circuits.

[0003] Current inspection methods for this type of equipment mainly rely on manual inspection using handheld infrared detectors. However, under the special conditions of high temperature and high electric field, the following problems exist: First, the operation carries extremely high safety risks, making live / loaded online inspection impossible. The high electric field environment presents serious risks of electric shock and arc burns. The high-temperature furnace also carries additional safety hazards such as high-temperature burns, poisoning from harmful gases, and confined space hazards. This constitutes a high-risk special operation with extremely high requirements for operator qualifications and protection levels. Production continuity and stability are generally maintained, and shutdowns for inspection are not typically employed; a single furnace shutdown or power outage can result in economic losses exceeding one million yuan.

[0004] Second, the inspection efficiency is extremely low, with numerous blind spots. Manual handheld inspection of the entire furnace of a single 300MW unit takes 3 to 5 days, resulting in extremely low efficiency. Due to limitations in personnel's working range and equipment structure, manual inspection cannot cover areas such as high-altitude corners and narrow passages, leaving numerous blind spots. This makes it extremely easy to miss early safety hazards such as localized overheating and oxidation corrosion. Industry data shows that the failure rate of manual inspection can be as high as 40%.

[0005] Third, the combined effects of high electric field and strong electromagnetic interference, along with oxide spots, severely compromise temperature measurement accuracy. Strong electromagnetic radiation in high electric field environments severely interferes with the signal acquisition, transmission, and processing of handheld infrared detectors, leading to image distortion, stripes, noise, and data jumps, making it impossible to obtain stable, continuous full-frame temperature data. Simultaneously, abrupt changes in emissivity caused by oxide spots on the equipment surface mean that single-view handheld detection cannot decouple the "true equipment temperature" from the "false temperature deviation caused by emissivity anomalies," resulting in high temperature measurement errors. This makes it difficult to accurately identify actual overheating hazards, easily leading to false alarms and missed hazard detections, failing to meet the accuracy requirements for equipment safety monitoring.

[0006] Fourth, the low level of data standardization makes it impossible to achieve intelligent closed-loop management. Manual handheld inspection relies on operators to manually record data and judge anomalies. The data has a low level of standardization and poor traceability, making it impossible to achieve long-term tracking and intelligent early warning of equipment temperature trends. A large amount of inspection data cannot be automatically uploaded and centrally analyzed, making it difficult to form a closed-loop management of the entire process of "detection, early warning, handling, and verification", and failing to meet the needs of intelligent operation and maintenance.

[0007] 5. During long-term operation of the electrolytic cell, a small area of ​​low-temperature deposits may appear on the surface, which may obscure the true temperature of the cell. The temperature data collected directly cannot reflect the actual state of the obscured area, which may lead to missed or misjudged high-temperature points and affect the judgment of the electrolytic cell's operating status. Summary of the Invention

[0008] To address the shortcomings of existing technologies, the present invention aims to provide an intelligent inspection system and method for industrial furnaces and kilns, thus solving the problems in the background technology.

[0009] This invention provides the following technical solution: An intelligent inspection system for industrial furnaces and kilns includes: a mobile carrier terminal for furnace and kiln inspection, a network transmission layer, a data processing layer, and an application output layer. The furnace inspection mobile carrier terminal includes a dual-view infrared thermal imaging module, a full-frame temperature acquisition unit, an embedded heterogeneous processor, a preprocessing unit, a dynamic target recognition unit, a dual-view geometric registration unit, an oxide spot recognition and correction unit, a temperature calculation unit, and an industrial-grade wireless transmission module. The network transmission layer includes an industrial server cluster, which is used for receiving, parsing, load balancing and caching of furnace inspection data; The data processing layer includes a time-series database and a data analysis server, which are used for storing historical data of the furnace and kiln, detecting safety hazards, analyzing temperature trends, and generating maintenance instructions. The application output layer includes an on-site audible and visual alarm system, a central control room management platform, and a mobile APP, which are used for alarms, data display, equipment control, and work order closed-loop processing.

[0010] Preferably, an intelligent inspection method for industrial furnaces and kilns includes the following steps: S1 is equipped with a dual-view infrared thermal imaging module on a mobile carrier for furnace and kiln inspection. As the carrier moves along a preset path, it synchronously collects infrared data streams on the heated surfaces of high-temperature furnaces and kilns. Through the device's open interface, it obtains full-frame one-dimensional temperature raw data, reconstructs it into a two-dimensional temperature matrix, and converts it into a standard Celsius temperature distribution map. S2 preprocesses the two-dimensional temperature matrix to eliminate image distortion, stripes, noise, and flickering under high dust and high electric field electromagnetic interference environments in furnaces and kilns. S3 acquires the preprocessed full-frame temperature matrix and uses a dynamic threshold algorithm to achieve adaptive separation between the target on the furnace heating surface and the furnace background. It filters out interference noise through morphological operations, extracts the contour, circumscribed rectangle, and geometric center of the effective target area, completes multi-target localization and numbering, and realizes simultaneous identification of multiple furnace heating surfaces. When there are attachments on the surface of the measured object, it accurately identifies small-scale low-temperature attachments on the surface of the measured object, eliminates large-area low-temperature background interference, and combines multi-layer annular radial temperature sampling and trend judgment to restore the true temperature characteristics of the area obscured by the attachments, thereby eliminating the impact of the attachments on the accuracy of electrolytic cell temperature acquisition.

[0011] S4 addresses the interference of temperature measurement on oxide spots on the heated metal surface of furnaces by performing geometric transformation and registration on the synchronously acquired main and auxiliary viewing temperature images to obtain an auxiliary viewing temperature matrix that is pixel-level aligned with the main viewing angle. Based on the physical characteristics that the actual temperature is independent of the shooting angle and the apparent temperature of the oxide spots is strongly correlated with the shooting angle, the S5 identifies and marks the interference area of ​​oxide spots on the heated surface of the furnace through the temperature difference of the dual perspective. S6 performs temperature correction on the marked oxide spot interference area to obtain the true temperature; for each furnace heating surface target area, it calculates key temperature indicators based on the true temperature; the inspection carrier terminal reduces noise and compresses the data, and then transmits it to the backend through an encrypted industrial wireless network; the backend server cluster completes data reception, parsing, load balancing and storage, and performs anomaly detection, identifies furnace safety hazards and generates warnings and maintenance orders.

[0012] Preferably, the preprocessing process in step S2 specifically includes: sequentially performing two-point non-uniformity correction (NUC), multi-frame temporal median filtering, neighborhood bad pixel detection and replacement, transmission CRC verification, and missing data interpolation and completion; wherein the NUC correction automatically performs a dynamic update once every preset time interval to adapt to sensor temperature drift and electromagnetic interference under the high temperature and high electric field environment of the furnace.

[0013] Preferably, in step S3, the dynamic threshold algorithm obtains the optimal segmentation threshold by calculating the inter-class variance between the background and the target, and adapts to the temperature changes in the furnace environment; the morphological operation adopts an opening operation of first erosion and then expansion to filter out scattered noise points with an area smaller than the preset threshold, so that the target area is complete and continuous.

[0014] Preferably, the geometric transformation registration in step S4 specifically involves performing radial and tangential distortion correction, three-dimensional coordinate mapping, affine transformation, and bilinear interpolation resampling on the secondary view temperature image in sequence, ultimately obtaining a secondary view temperature matrix that is pixel-level aligned with the primary view, with a registration error of less than 1 pixel; the preset angle between the optical axes of the primary and secondary views is 30°-60°.

[0015] Preferably, the oxide spot identification in step S5 specifically involves: calculating the temperature difference between the main view and the registered auxiliary view pixel by pixel; when the temperature difference is greater than a preset threshold, the area of ​​the connected region is within a preset range, or the texture of the region is irregular, it is determined to be an oxide spot interference area; at the same time, a secondary verification is performed to eliminate interference from random noise, furnace welds, and rigid beam structures.

[0016] Preferably, the temperature correction in step S6 specifically involves: using a dual-view reliable temperature weighted fusion method or fitting the mean of the effective temperature in the neighborhood non-interference method to replace and correct the false temperature values ​​in the oxide spot region, thereby ensuring a continuous and smooth temperature field.

[0017] Preferably, the dual-criteria anomaly detection in step S6 is as follows: initial judgment is completed through temperature threshold and temperature rise rate rule, and fault probability verification is completed through LSTM time series prediction model. When the rule and model are triggered simultaneously, it is determined to be a real anomaly; wherein, the highest temperature of the water-cooled wall exceeding 580℃ is determined to be a first-level overheating alarm, and the temperature rise rate of the heated surface is greater than 3℃ / min, which is determined to be a second-level warning.

[0018] Preferably, the furnace inspection mobile carrier is a circular track inspection trolley. The carrier moves along a preset path, with detection points set at preset intervals. Upon reaching its destination, it automatically locks and simultaneously triggers dual-view acquisition to achieve full coverage of the entire furnace without blind spots. The trolley is slidably positioned below the track, which is connected to support rods that are connected to the perimeter of the furnace body. A heat-conducting pad layer is provided on the inner layer of the trolley shell, evenly laid on the inner wall of the trolley shell. The heat-conducting pad layer is made of graphite, which evenly distributes the heat introduced into the trolley from the outside within the trolley shell, avoiding localized high temperatures inside the trolley and preventing localized high temperatures in the infrared thermal imaging system from causing measurement distortion, thus improving the accuracy of the temperature measurement results. The infrared imaging module inside the trolley has an ellipsoidal metal shell on its outer side. The ellipsoidal design reduces electromagnetic flux and the heat generated by eddy currents, preventing high temperatures caused by eddy currents inside the trolley and allowing the trolley to better adapt to high-temperature environments.

[0019] Preferably, a circular track trolley equipped with a dual-view thermal imaging module moves along a preset path, with a detection point set at 50mm intervals. After the trolley reaches a point, the position sensor triggers the trolley to mechanically lock, and at the same time sends a synchronous trigger signal to the dual-view thermal imaging module to ensure that the two views capture the same heated surface area at the same time and position. After completing the acquisition of a single point, the carrier automatically moves to the next point until all points in the entire furnace are detected, achieving 100% coverage without blind spots, and eliminating the need for manual entry into the high-temperature and high-electric-field dangerous areas.

[0020] Preferably, full-frame temperature data is requested through a thermal imaging device; a one-dimensional temperature array is received and reconstructed into a two-dimensional matrix according to width W and height H: T(i,j)=T1[(i-1)×W+(j-1)] and converted to degrees Celsius according to the special calibration formula for the high-temperature section of the furnace: T℃=K×T original+B where K and B are the factory calibration coefficients of the high-temperature section of the equipment, which are suitable for the temperature measurement range of furnaces from 0℃ to 1200℃. Two-point NUC calibration satisfies: T_corrected(i,j)=Gain(i,j)×T_raw(i,j)+Offset(i,j), and the calibration coefficient is dynamically updated every 30 minutes to eliminate fixed stripes and blocky screen distortion caused by sensor temperature drift; 5-frame temporal median filtering: eliminates transient noise and flickering screen distortion caused by smoke and dust; 3×3 bad pixel detection and replacement: if |T(i,j)-μ|>3σ, it is judged as a bad pixel and replaced with the neighborhood mean to repair dead pixels and burst pixels; CRC32 check + linear interpolation: fills in lost data during transmission and avoids screen distortion during transmission.

[0021] The preferred method for automatic target identification of furnace heating surfaces is as follows: Step 1: Dynamic threshold segmentation: Using the Otsu method, the inter-class variance g(T0) = ω0 × ω1 × (μ0 - μ1) is calculated. 2 The optimal segmentation threshold is set when g(T0) is maximized, enabling adaptive separation of the heated surface from the furnace background. A 3×3 structuring element is used to perform the opening operation T∘s=(T⊙s)⊕s, filtering dust noise with an area <20 pixels to ensure the target area is complete and continuous. Target localization and numbering: The Suzuki-8 neighborhood algorithm is used to extract the contour, and the minimum bounding rectangle and geometric center cx=(minX+maxX) / 2 and cy=(minY+maxY) / 2 are calculated. Multiple targets are numbered from top to bottom and from left to right.

[0022] Dual-view geometric transformation registration methods include: 1. Perform radial and tangential distortion correction on the secondary viewpoint pixels: Coordinate mapping: Project the secondary view coordinates onto the primary view plane using the intrinsic parameter matrix, rotation matrix R, and translation vector t; Fine registration using affine transformation: ; Bilinear interpolation resampling: T(u a v a =(1-α)(1-β)T(u0,v0)+(1-α)βT(u0,v0+1)+α(1-β)T(u0+1,v0)+αβT(u0+1,v0+1); finally, the auxiliary viewing temperature matrix is ​​obtained, which is aligned with the main viewing angle at the pixel level.

[0023] By simultaneously acquiring data from both perspectives and registering it at the pixel level, the temperature difference is calculated pixel by pixel: in the actual overheated pipe wall area, the temperature difference between the two perspectives is ≤3℃, independent of the viewing angle; in the area of ​​oxide spot interference, the temperature difference between the two perspectives is ≥5℃, changing significantly with the viewing angle. Combined with secondary verification using area and texture features, it can 100% distinguish between actual pipe wall overheating and false anomalies caused by oxide spots, fundamentally solving the problem of temperature measurement distortion in single-view furnaces.

[0024] Oxidation spot region temperature correction method: dual-view weighted fusion: T_fusion=ω1・T1+ω2・T2', ω1+ω2=1, giving high weight to the view with stable temperature change; neighborhood mean fitting: for regions with insufficient confidence, the mean of the non-interference effective temperature of the 3×3 neighborhood is used for repair; output interference-free full-frame true temperature matrix.

[0025] Temperature statistics algorithm: Maximum temperature: Tmax=max{T_final(i,j)|(i,j)∈R} Average temperature: Tavg=(1 / N)ΣT_final(i,j) Center temperature: Tcenter=T_final(cx,cy)N is the number of effective temperature points in the target area.

[0026] Compared with the prior art, the present invention has the following beneficial effects: This invention relates to an intelligent inspection system and method for industrial furnaces and kilns. It utilizes a mobile platform to replace manual entry into high-temperature, high-electric-field hazardous areas, enabling full-area inspections without furnace shutdown or power outages. This avoids safety risks such as electric shock, burns, and poisoning, and a single inspection can prevent economic losses exceeding one million yuan due to furnace shutdowns and power outages, ensuring continuous furnace production. The full furnace inspection time is reduced from 3-5 days to 4-6 hours, significantly improving inspection efficiency. Pre-defined paths ensure full coverage detection, completely eliminating blind spots in manual inspections and reducing the rate of missed hazard detection to zero. It completely resolves image distortion and noise issues under high electric field environments, achieving an image effectiveness rate of 99.8%. A dual-view temperature difference recognition scheme addresses oxidation spot temperature measurement interference at its physical source, achieving an oxidation spot recognition accuracy of ≥96% and a post-correction temperature measurement error of <1%, significantly improving temperature measurement accuracy. Local preprocessing and anomaly initial judgment at the terminal only upload abnormal data and key indicators, reducing data transmission volume by over 80%, making it suitable for low-bandwidth, high-interference transmission scenarios under high electric field environments. Anomaly detection reduces false alarm rate by 90%; the edge-cloud collaborative architecture enables automatic data upload, centralized analysis, intelligent early warning, and work order closure, fully adapting to the digital operation and maintenance needs of smart power plants and smart grids. Its low-power embedded design is compatible with various inspection platforms, including rail-mounted, tracked, and UAV-borne systems, and can be directly used for new unit construction and in-service unit upgrades without large-scale equipment replacement. Attached Figure Description

[0027] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained from these drawings without creative effort.

[0028] Figure 1 This is a schematic diagram of the overall layout structure of the inspection system of the present invention.

[0029] Figure 2 This is the invention Figure 1 Enlarged view of a portion of the diagram.

[0030] Figure 3 This is a schematic diagram of the inspection mobile trolley structure of the present invention.

[0031] Figure 4 This is a schematic diagram of the internal structure of the mobile inspection vehicle of the present invention.

[0032] Figure 5 This is a schematic diagram of the inspection method of the present invention.

[0033] Figure 6 The image captured is a distorted image, as described in an embodiment of the present invention.

[0034] Figure 7 These are clear images captured after processing, as shown in the embodiments of the present invention. Detailed Implementation

[0035] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0036] Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0037] Example 1: like Figure 1-7As shown, an intelligent inspection system for industrial furnaces and kilns is described. The target of the inspection is a 300MW subcritical pulverized coal furnace and kiln with a rated main steam temperature of 540℃ and a rated main steam pressure of 16.7MPa. The core inspection areas are the furnace water-cooled walls, the screen-type superheater, and the high-temperature reheater. Long-term high-temperature operation of the heating surfaces has resulted in large areas of oxidation spots. A circular track 2 is installed on the outer side of the furnace body 1, running around the outer wall of the furnace. An inspection trolley 3 is mounted on the track 2. Inspection points are set at 50mm intervals along the circular track 2, for a total of 10,800 inspection points throughout the furnace, achieving 100% coverage of the heating surfaces without blind spots. The acquisition module is a dual-view 640×512 uncooled infrared thermal imager with a temperature range of 0℃-1200℃. The optical axis angle between the main and secondary viewing angles is 45°, and the synchronous triggering accuracy is <1ms, allowing for simultaneous acquisition of dual-view data of the same heating surface area after the trolley is in position.

[0038] Processing unit: ARM+DSP+FPGA embedded heterogeneous processor, configured with ARM Cortex-A72 dual-core main processor, TIC6000DSP, Xilinx Artix-7FPGA, 8GB eMMC storage, industrial-grade 5G / Wi-Fi dual-mode wireless module, and intrinsically safe explosion-proof audible and visual alarm.

[0039] Backend system: Industrial server cluster (2 main servers + 3 node servers), InfluxDB time series database, central control room computer management platform, and mobile APP for operation and maintenance personnel.

[0040] The inspection method includes step S1: Circular trolley inspection and full-frame temperature acquisition: The inspection trolley moves along a circular track for detection, and after reaching its position, it moves 360° around the circular track; at the same time, it sends a synchronous trigger signal to the dual-view thermal imager, which simultaneously acquires infrared data streams of the same water-cooled wall area; after completing the acquisition of a single point, the locking mechanism is unlocked, and the trolley automatically moves to the next detection point, repeating the above acquisition process until all points of the entire furnace body are detected. After acquiring the infrared data stream, the thermal imager sends a data request through the official open interface of the device. The request header carries an authentication token token=GL20240601 and specifies the video channel number channel=0 to establish legitimate encrypted communication. The interface returns a whole frame of raw temperature data, including a one-dimensional temperature array, an image width W=640, a height H=512, and a temperature data bit depth of 14 bits. The system reorganizes the one-dimensional temperature array into a two-dimensional temperature matrix according to the image width and height. The reconstruction formula is: T(i,j)=T1[(i-1)×640+(j-1)], so that each matrix position corresponds one-to-one with the pixel of the thermal imaging image. According to the factory calibration coefficients K=0.08 and B=-10 for the high temperature section of the device, the raw values ​​are converted into standard degrees Celsius according to the calibration formula T℃=0.08×T raw-10, forming a full-frame temperature distribution map that can be directly used for analysis.

[0041] Step S2: Embedded multi-level preprocessing. The reconstructed two-dimensional temperature matrix undergoes multi-level preprocessing in the vehicle's embedded processor: ① Two-point non-uniformity correction NUC: Correction is performed according to the formula T_corrected(i,j)=Gain(i,j)×T_raw(i,j)+Offset(i,j), where Gain(i,j) and Offset(i,j) are the blackbody calibration coefficients from the device's factory settings. The system automatically performs dynamic correction every 30 minutes to adapt to sensor temperature drift in high-temperature furnace environments and eliminate fixed stripes and blocky screen artifacts; ② Multi-frame temporal median filtering. Wave: Establish a continuous 5-frame temperature buffer queue, take the median of 5 consecutive frames for each pixel position, and remove transient noise and screen flicker caused by furnace dust and flue gas; ③ Neighborhood bad pixel detection and replacement: For each pixel, calculate the mean μ and standard deviation σ of its 3×3 neighborhood. If |T(i,j)-μ|>3σ, it is determined to be a bad pixel. Replace the temperature value of the point with the neighborhood mean to repair dead pixels and burst pixels; ④ Transmission preprocessing: Add CRC32 check code to each frame of temperature data. If the check fails, trigger local retransmission. Use uplink and downlink linear interpolation to fill in the lost lines during transmission to avoid image screen distortion and data loss caused by transmission.

[0042] Step S3: Automatic identification of furnace heating surface targets. After obtaining the pre-processed full-frame temperature matrix, the system automatically identifies furnace heating surface targets in the image: ① Dynamic threshold segmentation: Using the Otsu method, the temperature data is divided into background class C0 (furnace environment) and target class C1 (heating surface). The pixel weights ω0 and ω1, and the mean values ​​μ0 and μ1 of the two classes are calculated according to the formula g(T0)=ω0×ω1×(μ0-μ1). 2 Calculate the inter-class variance, and take the temperature value corresponding to the maximum g(T0) as the optimal segmentation threshold T_opt=320℃; determine the area with temperature ≥320℃ as the measured object on the furnace heating surface, and the area with temperature <320℃ as the furnace environment background, to achieve adaptive separation of target and background and adapt to the temperature changes of the furnace environment; ② Noise reduction processing: use 3×3 rectangular structural elements to perform an opening operation of first erosion and then expansion to filter scattered high-temperature noise and dust interference with an area <20 pixels, eliminate edge burrs, and make the target area more complete and continuous; ③ Target positioning Location and Numbering: The Suzuki-8 neighborhood algorithm is used to extract the outer contour of each effective target area. The minX, maxX, minY, and maxY of each area are calculated to obtain the minimum bounding rectangle, which satisfies the formulas cx=(minX+maxX) / 2 and cy=(minY+maxY) / 2 to obtain the geometric center. According to the rule of top to bottom and left to right, the five effective targets of the upper section of the water-cooled wall, the middle section of the water-cooled wall, the lower section of the water-cooled wall, the screen superheater, and the high-temperature reheater are located and numbered to realize the simultaneous identification of multiple heating surfaces of the furnace.

[0043] Step S4: Dual-view geometric transformation registration. To address the interference from oxide spots on the heated surfaces of the furnace, geometric transformation registration is performed on the synchronously acquired auxiliary-view temperature images: ① Radial and tangential distortion correction: The radial distortion coefficients k1=-0.235, k2=0.126 and the tangential distortion coefficients p1=0.003, p2=-0.002, as specified by the equipment manufacturer, are read. After normalization of each pixel in the auxiliary view, distortion correction is completed, eliminating image stretching and bending caused by lens distortion; ② Three-dimensional coordinate mapping: The intrinsic parameter matrix K1 of the main view, the intrinsic parameter matrix K2 of the auxiliary view, the rotation matrix R between the two views, and the translation matrix are read. Vector t converts the auxiliary viewpoint pixel coordinates into 3D coordinates in the camera coordinate system, and after rotation and translation, projects them onto the main viewpoint pixel plane to complete coarse registration; ③ Fine registration through affine transformation: fine adjustments are made through translation, rotation, scaling, and shearing to ensure that the auxiliary viewpoint and the main viewpoint are perfectly matched in position, orientation, and size; ④ Bilinear interpolation resampling: for the transformed non-integer coordinate positions, bilinear interpolation is used to calculate the temperature value to ensure continuous and smooth temperature data; finally, the auxiliary viewpoint temperature matrix is ​​obtained and aligned with the main viewpoint at the pixel level. After registration, the root mean square error of feature points RMSE = 0.68 pixels, which is less than 1 pixel.

[0044] Step S5: Based on the physical characteristics that the actual temperature is independent of the shooting angle and the apparent temperature of the oxide spots is strongly correlated with the shooting angle, identify the oxide spots on the heated surface of the furnace: ① Calculate the main viewing angle temperature matrix T pixel by pixel. 1（ The absolute temperature difference between the registered auxiliary viewing temperature matrix T2'(i,j) and the temperature difference between i,j) and T2'(i,j) is: ΔT(i,j)=|T(i,j)-T2'(i,j)|; ② Initial judgment rule: Set a temperature difference threshold of 4℃. When ΔT(i,j)>4℃, it is marked as a suspected oxide spot area; ③ Secondary verification: Extract connected components from the suspected areas, count the area of ​​the connected components, and exclude random noise with an area <5 pixels, furnace welds with an area >50 pixels, and rigid beam structures; verify the texture irregularity by calculating the temperature gradient variance of the area. If the gradient variance is >15, it is judged as a texture irregular area; Finally, 186 areas that simultaneously meet the criteria of "ΔT>4℃, area 5-50 pixels, texture irregularity" are identified as oxide spot interference areas and marked.

[0045] Step S6: Temperature Correction for Oxidation Spot Regions. Temperature correction is performed on the marked oxide spot interference regions: ① Prioritize dual-view reliable temperature weighted fusion: For each spot region, compare the smoothness of the dual-view temperatures. Assign a high weight ω2=0.7 to the auxiliary view with stable temperature changes that conform to the overall temperature rise trend of the pipe wall, and a weight ω1=0.3 to the main view, satisfying the formula T_fusion=0.3×T1+0.7×T2' to complete the fusion correction; ② For spot regions with insufficient confidence, use the 3×3 neighborhood non-interference effective temperature mean to fit and repair, ensuring the continuous smoothness of the pipe wall temperature field; ③ Finally output the full-frame true temperature matrix T_final without oxide spot interference. For each identified target area R of the furnace heating surface, the key temperature indicators are calculated based on the corrected real temperature matrix: ① Maximum temperature: Tmax=max(T_final(i,j)), extracting the maximum temperature in the area to determine whether there is a risk of overheating in the pipe wall; ② Average temperature: Tavg=(1 / N)ΣT_final(i,j), calculating the average value of all effective temperature points in the area, reflecting the overall temperature rise level of the heating surface, where N is the number of effective temperature points in the area; ③ Center point temperature: Tcenter=T_final(cx,cy), obtaining the temperature value at the geometric center of the target, benchmarked against the traditional contact temperature measurement method; In this embodiment, the calculation results for the middle section of the left wall of the water-cooled wall are: Tmax=592℃, Tavg=547℃, Tcenter=551℃, and the number of effective temperature points N=12458. Local noise reduction at the terminal: The temperature matrix is ​​denoised using a three-layer decomposition of db4 wavelet to achieve soft threshold noise reduction, further improving data quality; Data compression: The temperature data is compressed losslessly using the LZ77 algorithm, achieving a compression ratio of 6:1; Local anomaly detection: The terminal performs initial anomaly detection locally, identifying overheating in the middle section of the left wall of the water-cooled wall. Only complete data of the abnormal area is uploaded first, while only temperature statistics are uploaded for the remaining normal heated surfaces, reducing the overall data transmission volume by 83%; Encrypted transmission: The data is encrypted using the AES-256 industrial encryption algorithm and transmitted to the backend server cluster via a 5G industrial wireless module. CRC32 checksum is retained throughout the transmission process to ensure data integrity.The backend dual-criteria anomaly detection industrial server cluster receives data uploaded from the terminal, completes protocol parsing and data verification; it distributes the data to different processing nodes through load balancing, and writes the inspection data of the entire heated surface of the furnace into the InfluxDB time-series database according to the dimensions of equipment number, inspection point, and inspection time; the data analysis server performs dual-criteria anomaly detection: rule criterion: the highest temperature of the middle section of the left wall of the water-cooled wall is 592℃>580℃, triggering a first-level overheating alarm; the temperature rise rate v=3.7℃ / min>3℃ / min, triggering a second-level warning; AI criterion: inputting 10 consecutive frames of time-series temperature data of this area into a trained LSTM time-series prediction model, the model outputs an overheating fault probability P=0.94≥0.85, which is judged as a real anomaly; comprehensive judgment: the rule criterion and AI criterion are triggered simultaneously, and finally confirmed as a first-level overheating hazard, automatically generating an anomaly alarm and maintenance work order instructions. Multi-terminal linkage alarm and work order closed loop: On-site audible and visual alarms are triggered to remind on-site inspection personnel; the large screen of the central control room management platform highlights the abnormal location, temperature data, and historical trend curves in red; the mobile APP of maintenance personnel pushes alarm information, abnormal details, and maintenance work orders, supporting one-click navigation to the fault location; after the maintenance personnel arrive on-site and confirm the thickening of the oxide scale on the pipe wall and local overheating, they promptly perform grinding treatment. After the treatment is completed, the work order status is updated in the APP, forming a complete closed loop of alarm-dispatch-handling-review, thus avoiding furnace tube rupture accidents. Implementation effect verification: In this embodiment, the oxide spot recognition accuracy rate is 96.3%, the average temperature measurement error before correction is 9.7%, and the average error after correction is 0.72%; the whole furnace inspection time is 5 hours, which is 92% more efficient than manual inspection; the target recognition accuracy rate of the furnace heating surface is 100%, successfully identifying 3 overheating hazards obscured by oxide spots, with no missed detections and no false alarms, fully meeting the accuracy and reliability requirements of intelligent furnace inspection.

[0046] Example 2: like Figure 1-5 As shown, an intelligent inspection system for electrolytic cells is described. The thickness of the oxide scale on the outer wall of the electrolytic cell can reach up to 1.2 mm in some areas, making it impossible to accurately identify the true temperature of the tube wall through manual inspection. The system employs an inspection trolley mounted on a track; a detection point is set every 100 mm along the circular track, resulting in a total of 15,000 detection points throughout the furnace, achieving 100% coverage of the heated surface without any blind spots.

[0047] Dual-view 384×288 uncooled infrared thermal imager, temperature measurement range 0℃-1200℃, main and auxiliary view optical axis angle 60°, dustproof and shockproof protection, synchronous trigger accuracy <0.5ms, suitable for harsh working conditions of CFB furnaces with high dust and strong wear. Processing unit: Low-power ARM+DSP heterogeneous processor, industrial-grade Wi-Fi 6 wireless module, built-in supplementary lighting and dust removal device, and on-site explosion-proof audible and visual alarm; Backend systems: industrial server cluster, InfluxDB time-series database, central control room management platform, and mobile APP.

[0048] Specific inspection steps S1: The inspection probe moves along a preset circular path, setting a detection point every 100mm. Upon reaching the target position, the track is automatically locked, and dual-view thermal imaging acquisition is triggered simultaneously. After completing the single-point detection, it continues to crawl until the entire electrolytic cell is fully covered. The one-dimensional temperature raw data is obtained through the open interface of the thermal imager, reconstructed into a 384×288 two-dimensional temperature matrix, and converted into a standard Celsius temperature distribution map according to the CFB furnace-specific calibration coefficient.

[0049] Step S2: Embedded preprocessing in harsh environments performs dynamic NUC calibration to eliminate sensor temperature drift, 8-frame temporal median filtering to eliminate transient noise and flicker in high-dust environments, 3×3 neighborhood bad pixel detection to replace image bad pixels caused by wear, CRC check and interpolation to complete lost transmission data, and completely eliminates screen tearing and stripe interference. (Reference) Figure 6 , Figure 7 .

[0050] Step S3: Automatic identification of targets on the furnace heating surface. The OTSU method dynamic threshold is used to achieve adaptive separation between the heating surface and the furnace background. 3×3 morphological opening operation is used to filter noise caused by dust and wall unevenness, extract the effective heating surface target area, and complete contour extraction, circumscribed rectangle calculation, geometric center positioning and multi-target numbering.

[0051] Step S4: Dual-view geometric transformation registration sequentially performs distortion correction, 3D coordinate mapping, affine transformation, and bilinear interpolation resampling on the auxiliary view temperature image to adapt to slight angular deviations during robot crawling. Finally, an auxiliary view temperature matrix is ​​obtained that is pixel-level aligned with the main view. The registration error is 0.82 pixels, which is less than 1 pixel and meets the detection requirements.

[0052] Step S5: Identification of Oxidation Spots and Wear Areas. Based on the physical characteristics that the actual temperature is independent of the viewing angle and the temperature of the oxidation spots is strongly correlated with the viewing angle, the temperature difference between the two viewing angles is calculated pixel by pixel. The temperature difference threshold of 5℃ is set. Combined with the connected region area of ​​5-50 pixels and the irregular texture features, the interference of wall welds, nails, and wear-resistant castings is eliminated through secondary verification. Finally, 217 oxidation spots and wear areas are identified and marked.

[0053] Step S6: Temperature Correction. For the marked interference areas, a dual-view reliable temperature weighted fusion method is used for correction. For uneven areas caused by wall wear, the mean of the effective neighborhood temperature is used for fitting and repair. The final output is an interference-free true temperature matrix of the electrolytic cell wall. Key temperature index calculations are based on the corrected true temperature matrix, calculating the highest temperature, average temperature, and center point temperature of each heated surface target area. The local highest temperature of the electrolytic cell wall is 612℃, exceeding the safety threshold. The data transmission terminal performs initial anomaly judgment locally, encrypting and uploading only the complete data of the abnormal area and the temperature statistics of the normal area, reducing the overall data transmission volume by 81% and adapting to the weak wireless signal environment inside the electrolytic cell. Dual-criteria anomaly detection: Cloud-based rule-based judgment triggers a highest temperature of 612℃ > 580℃, a temperature rise rate of 4.2℃ / min > 3℃ / min; the LSTM time series model outputs a fault probability of 0.96 ≥ 0.85, classifying it as a level one overheating wear hazard, and automatically generating an emergency repair work order. The alarm and work order closed-loop system is triggered by the on-site audible and visual alarms. The control room's large screen highlights the location of the potential hazard, the degree of wear, and temperature data. Emergency alarms and work orders are pushed to maintenance personnel via mobile devices. After shutting down the furnace, maintenance personnel perform wear-resistant welding on the affected area, eliminating the risk of tube rupture. Implementation effect verification: In this embodiment, the accuracy rate for identifying oxide spots and wear areas is 95.8%. The average error in temperature measurement before correction was 11.3%, and the average error after correction was 0.85%. The entire furnace inspection took 8 hours, an 85% improvement in efficiency compared to manual shutdown inspection. Four wear and overheating hazards obscured by oxide spots were successfully identified. There were no missed detections or false alarms, fully adapting to the intelligent inspection needs under harsh electrolytic cell operating conditions.

[0054] Example 3 The difference from Embodiments 1 and 2 is that when there are small-scale low-temperature deposits on the surface of the electrolytic cell that obscure the true temperature of the cell and the directly collected temperature data cannot reflect the actual state of the obscured area, the electrolytic cell temperature matrix collected by the thermal imaging device is used as a basis. First, the small-scale deposits on the surface of the electrolytic cell are accurately identified by temperature threshold and area screening. Then, multi-layer annular temperature sampling of 1 / 2 / 3 pixels is performed with the deposits as the center. Finally, the radial temperature distribution trend is used to determine whether there is a high-temperature point below the deposits that is being obscured. This corrects the temperature collection results of the deposit area and eliminates the interference of the deposits on the true temperature.

[0055] The specific process is as follows: First, prepare the thermal imaging image of the electrolytic cell and the temperature matrix that matches the corresponding pixels one by one. The image width is denoted as W and the height is denoted as H. The temperature matrix can be expanded into a one-dimensional array T[i] (index relationship i=y×W+x, where x and y are pixel coordinates). The first step is to convert the original temperature matrix data into actual temperatures, and the conversion formula satisfies: T real(x,y)=T(x,y) / 10; (The original temperature data is stored in a 10x magnification format). The converted true temperature matrix is ​​then subjected to a 3×3 neighborhood mean filter to remove single-point noise. The filtering formula is: T smooth (x,y)=(1 / 9) T real (x+dx,y+dy); This yields a noise-free, smooth temperature matrix, which serves as the basis for subsequent analysis. The second step involves filtering out small-scale attachments on the surface of the electrolytic cell that meet the characteristics, while excluding large-area low-temperature background, noise points, and other interference. The specific steps are as follows: a) Based on the normal operating temperature threshold of the electrolytic cell, pixels in the smoothed temperature matrix whose temperature is significantly lower than the normal operating temperature of the electrolytic cell are marked as candidate debris points; b) Adjacent low-temperature candidate points are merged into independent regions according to the 8-connectivity rule (adjacent in the top, bottom, left, right, and four diagonal directions). Each region is considered a debris to be judged, and the total number of pixels S in each independent region is counted (S = ...). 1. R is a region), only the region satisfying 3≤S≤50 is retained (S min =3 pixels to filter noise points, S max =50 pixels to limit a small area of ​​attachments), this type of area is the actual attachments on the surface of the electrolytic cell.

[0056] The third step involves taking each identified attachment as the object and its geometric center as the sampling origin, performing 1 / 2 / 3 pixel three-layer annular temperature sampling. The specific operation is as follows: Determine the center of the attachments: Calculate the geometric center of the connected domain of each attachment: C = (x c ,y c ); Define a three-layer annular sampling region: with C as the center, set three layers of annular borders according to pixel distance, using the following formula: First ring (r=1 pixel): R1={(x,y)∣max(∣x-xc∣,∣y-yc∣)=1} Second ring (r=2 pixels): R2={(x,y)∣max(∣x-xc∣,∣y-yc∣)=2} Third ring (r=3 pixels): R3={(x,y)∣max(∣x-xc∣,∣y-yc∣)=3}; Count the effective pixel counts N1, N2, and N3 within each annular region, and calculate the true average temperature of each ring. The temperature of each ring satisfies the formula: T1=(1 / N1)∑T real (x,y) (where (x,y)∈R1); T2=(1 / N2)∑T real(x,y) (where (x,y)∈R2); T3=(1 / N3)∑T real (x,y) (where (x,y)∈R3).

[0057] The fourth step is to determine whether the area beneath the deposit is a high-temperature point of the electrolytic cell that is being blocked, based on the average temperature distribution trend of the three-ring structure. This is used to correct the temperature acquisition results in the deposit area. The determination rule is that if any condition is met, it is determined to be a high-temperature point that is being blocked. Condition 1 (radial center increasing trend): T1>T2>T3 indicates that the deposits cover the high temperature heat source of the electrolytic cell, and the closer to the deposits, the higher the temperature, which is consistent with the typical temperature distribution of high temperature points. Therefore, the actual temperature of this area is determined to be high temperature. Condition 2 (Overall High Temperature Characteristics): If T1 > 250℃, T2 > 250℃, and T3 > 250℃, it indicates that a large area surrounding the deposit is in a high-temperature state within the electrolytic cell, and the actual temperature of this area is determined to be high. If none of these conditions are met, the area with the deposit is determined to be a normal low-temperature area within the electrolytic cell, and no correction is required.

[0058] The fifth step involves correcting the temperature of the attachment areas identified as high-temperature obstructions to the average high temperature of the surrounding annular area (such as T1 or the average temperature of the annular area); the original collected temperatures of attachment areas not identified as high-temperature obstructions are retained. Finally, the corrected global accurate temperature matrix of the electrolytic cell is output, completing the temperature acquisition.

[0059] The above methods improve the accuracy of temperature acquisition, providing accurate temperature data for subsequent early warning and judgment steps. By identifying small-area deposits (3-50 pixels) on the surface of the electrolytic cell, interference from large-area low-temperature backgrounds and single-point noise is effectively eliminated, preventing the misjudgment of non-deposited low-temperature areas as the true low temperature of the electrolytic cell. A three-layer ring sampling system (1 / 2 / 3 pixels) with the geometric center of the deposit as the origin accurately captures the temperature distribution characteristics around the deposit. Combined with radial temperature trend and global high-temperature dual judgment logic, it can effectively distinguish between "high-temperature points obscured by deposits" and "true low-temperature areas of the electrolytic cell," eliminating the bias in temperature acquisition caused by deposits at its source. No physical contact or shutdown of the electrolytic cell is required; the true temperature of areas obscured by deposits is restored without interfering with normal electrolytic cell production, meeting the needs of continuous monitoring in industrial settings.

[0060] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention. For those skilled in the art, the present invention can be modified and varied in various ways. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. An intelligent inspection system for industrial furnaces and kilns, characterized in that, include: Furnace and kiln inspection mobile carrier terminal, network transmission layer, data processing layer, and application output layer; The furnace inspection mobile carrier terminal includes a dual-view infrared thermal imaging module, a full-frame temperature acquisition unit, a processor, a preprocessing unit, a dynamic target recognition unit, a dual-view geometric registration unit, an oxide spot recognition and correction unit, a temperature calculation unit, and a wireless transmission module. The network transmission layer includes an industrial server cluster, which is used for receiving, parsing, load balancing and caching of furnace inspection data; The data processing layer includes a time-series database and a data analysis server, which are used for storing historical data of the furnace and kiln, detecting safety hazards, analyzing temperature trends, and generating maintenance instructions. The application output layer includes an on-site audible and visual alarm system, a central control room management platform, and a mobile APP, which are used for alarms, data display, equipment control, and work order closed-loop processing.

2. An intelligent inspection method for industrial furnaces and kilns, employing the inspection system as described in claim 1, characterized in that, Includes the following steps: S1: A dual-view infrared thermal imaging module equipped with a mobile carrier for furnace and kiln inspection moves along a preset path and synchronously collects infrared data streams on the heated surface of the high-temperature furnace and kiln. It obtains full-frame one-dimensional temperature raw data through the open interface of the device, reconstructs it into a two-dimensional temperature matrix, and converts it into a standard Celsius temperature distribution map. S2: Preprocessing the two-dimensional temperature matrix to eliminate image distortion, stripes, noise, and flickering under high dust and high electric field electromagnetic interference environments in furnaces and kilns; S3: Obtain the preprocessed full-frame temperature matrix, and use a dynamic threshold algorithm to achieve adaptive separation between the target on the furnace heating surface and the furnace background. Filter interference noise through morphological operations, extract the contour, circumscribed rectangle, and geometric center of the effective target area, complete multi-target localization and numbering, and realize simultaneous identification of multiple furnace heating surfaces. When there are attachments on the surface of the measured object, accurately identify small-scale low-temperature attachments on the surface of the measured object, eliminate large-area low-temperature background interference, and combine multi-layer annular radial temperature sampling and trend judgment to restore the true temperature characteristics of the area obscured by the attachments, thereby eliminating the impact of the attachments on the accuracy of electrolytic cell temperature acquisition. S4: To address the interference of temperature measurement on oxide spots on the heated metal surface of the furnace, geometric transformation and registration are performed on the synchronously acquired main and auxiliary viewing temperature images to obtain an auxiliary viewing temperature matrix that is pixel-level aligned with the main viewing angle. S5: Based on the physical characteristics of strong correlation between real temperature and shooting angle, and apparent temperature of oxide spots and shooting angle, the interference area of ​​oxide spots on the heated surface of the furnace is identified and marked by dual-view temperature difference. S6: Perform temperature correction on the marked oxide spot interference area to obtain the true temperature; For each target area of ​​the furnace heating surface, key temperature indicators are calculated based on the actual temperature; the inspection carrier terminal reduces noise and compresses the data, and then transmits it to the backend via an encrypted industrial wireless network; the backend server cluster completes data reception, parsing, load balancing and storage, and performs anomaly detection, identifies furnace safety hazards and generates warnings and maintenance orders.

3. The intelligent inspection method for industrial furnaces and kilns according to claim 1, characterized in that, The preprocessing process in step S2 is as follows: two-point non-uniformity correction (NUC), multi-frame temporal median filtering, neighborhood bad pixel detection and replacement, transmission CRC verification and lost data interpolation and completion are performed in sequence; among them, the NUC correction is automatically dynamically updated once every preset time interval to adapt to the sensor temperature drift and electromagnetic interference under the high temperature and high electric field environment of the furnace.

4. The intelligent inspection method for industrial furnaces and kilns according to claim 1, characterized in that, In step S3, the dynamic threshold algorithm obtains the optimal segmentation threshold by calculating the inter-class variance between the background and the target, and adapts to the temperature changes in the furnace environment. The morphological operation adopts an opening operation of first erosion and then expansion to filter out scattered noise points with an area smaller than the preset threshold, so that the target area is complete and continuous.

5. The intelligent inspection method for industrial furnaces and kilns according to claim 1, characterized in that, In step S4, the geometric transformation registration is specifically performed as follows: radial and tangential distortion correction, three-dimensional coordinate mapping, affine transformation, and bilinear interpolation resampling are sequentially performed on the secondary view temperature image to finally obtain the secondary view temperature matrix that is pixel-level aligned with the primary view, with a registration error of less than 1 pixel; the preset angle between the optical axes of the primary and secondary views is 30°-60°.

6. The intelligent inspection method for industrial furnaces and kilns according to claim 1, characterized in that, In step S5, the identification of oxide spots specifically involves: calculating the temperature difference between the main view and the registered auxiliary view pixel by pixel; when the temperature difference is greater than a preset threshold, the area of ​​the connected region is within a preset range, or the texture of the region is irregular, it is determined to be an oxide spot interference area; at the same time, a secondary verification is performed to eliminate interference from random noise, furnace welds, and rigid beam structures.

7. The intelligent inspection method for industrial furnaces and kilns according to claim 1, characterized in that, In step S6, temperature correction specifically involves replacing and correcting spurious temperature values ​​in the oxide spot region by using dual-view reliable temperature weighted fusion or fitting the mean of the effective temperature in the neighborhood without interference, thus ensuring a continuous and smooth temperature field.

8. The intelligent inspection method for industrial furnaces and kilns according to claim 1, characterized in that, The dual-criteria anomaly detection in step S6 is as follows: the initial judgment is completed by temperature threshold and temperature rise rate rule, the fault probability is verified by LSTM time series prediction model, and the anomaly is determined to be real when the rule and model are triggered at the same time. A maximum water-cooled wall temperature exceeding 580℃ is classified as a Level 1 overheat alarm, and a temperature rise rate of the heated surface exceeding 3℃ / min is classified as a Level 2 warning.

9. The intelligent inspection method for industrial furnaces and kilns according to claim 1, characterized in that, The furnace inspection mobile carrier is a circular track inspection trolley; the carrier moves along a preset path, and a detection point is set at a preset distance. After reaching the target position, it automatically locks and simultaneously triggers dual-view acquisition to complete the full furnace coverage without blind spots.

10. The intelligent inspection system for industrial furnaces and kilns according to claim 1, characterized in that, The mobile carrier terminal for furnace inspection is a trolley (3). The trolley (3) is slidably set under the track (2). The track (2) is connected to the support rod, which is connected to the periphery of the furnace body (1). The inner layer of the outer shell of the trolley (3) is provided with a heat-conducting pad layer. The heat-conducting pad layer is evenly laid on the inner wall of the outer shell of the trolley. The heat-conducting pad layer is made of graphite material to avoid local high temperature inside the trolley. The infrared imaging module inside the trolley is provided with an ellipsoidal metal outer side. The ellipsoidal setting reduces electromagnetic flux and reduces the heat generation of eddy current phenomenon.

Images