A method and system for detecting thermal shock damage to a furnace lining based on thermal imaging

Abstract

The present application relates to the field of furnace lining damage detection, and more particularly to a furnace lining thermal shock damage detection method and system based on thermal imaging, the method comprising: collecting a thermal image sequence during the furnace lining temperature rising process; constructing a spatial neighborhood consensus temperature curve; calculating the instantaneous deviation between the actual temperature time sequence of each pixel and its corresponding spatial neighborhood consensus temperature curve; filtering the instantaneous deviation, calculating the thermal hysteresis index of the damage degree of each pixel; aggregating the thermal hysteresis index of all pixels to calculate the global damage cumulative value of the current detection period; storing the global damage cumulative value into a historical trend sequence; updating the smoothing level and the current trend of the sequence; calculating and outputting the remaining service life of the furnace lining, so as to realize the detection and life prediction of the thermal shock damage of the furnace lining. The present application effectively solves the problem that the prior art cannot distinguish between normal thermal characteristic difference and damage thermal hysteresis.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of furnace lining damage detection, and in particular to a furnace lining thermal shock damage detection method and system based on thermal imaging. BACKGROUND

[0002] The furnace lining is a key component of high-temperature industrial furnaces, such as post-explosion welding follow-up heat treatment furnaces for titanium-steel clad plates, nickel-steel clad plates, etc. During long-term service, the furnace lining needs to withstand repeated and severe temperature changes, i.e. thermal cycles, which can easily cause thermal stress in the material and induce thermal shock damage. These damages initially appear as micro-cracks, which close during static constant temperature or cooling of the furnace, making it difficult for a thermal imager to detect them. However, during the dynamic process of heating up, these micro-cracks will instantaneously open due to thermal expansion of the furnace lining material, and the air filled in the cracks becomes a poor thermal conductor, increasing the local thermal resistance at that point. This results in poor heat transfer efficiency at the damaged point, and the surface temperature rise rate of the damaged point will continue to lag behind that of the surrounding healthy area, appearing as a dynamic abnormal feature of thermal lag or instantaneous cold spot in the thermal image sequence.

[0003] In the prior art, time-series features (such as heating rate, fluctuation peak value) are usually combined with spatial density algorithms (such as Local Outlier Factor, LOF) for detection. However, such methods have the following defects: first, feature confusion, the furnace lining itself has normal spatial heterogeneity. For example, the area near the furnace door sealing strip, the area where different refractory materials are spliced, their normal heating curves (thermal inertia) are different from those of the center area of the furnace wall. These normal boundary points also appear as heterogeneous points in the time-series feature space, confusing with the features of damaged points.

[0004] Secondly, the above feature confusion makes it difficult for spatial density algorithms such as LOF to distinguish between damage heterogeneity and normal heterogeneity, and no matter how the parameters or thresholds are adjusted, a large number of false positives will be generated at the normal boundaries, or in order to avoid false positives, the threshold is increased, resulting in missed detection of real early damage. Moreover, the LOF algorithm based on high-dimensional feature space and K-d tree has high computational complexity, which is difficult to meet the real-time processing needs of a large number of pixels in an industrial field. Therefore, there is an urgent need for a detection method that can distinguish between normal thermal characteristic differences and damage thermal lag from a mechanism. SUMMARY

[0005] In order to solve the defect that it is difficult to distinguish between damage lag and normal boundary in the prior art, the present application provides a furnace lining thermal shock damage detection method and system based on thermal imaging.

[0006] In a first aspect, the present application provides a furnace lining thermal shock damage detection method based on thermal imaging, which adopts the following technical solution:

[0007] A furnace lining thermal shock damage detection method based on thermal imaging, comprising the steps of:

[0008] The thermal image sequence in the furnace lining temperature rising process is collected to obtain the actual temperature time sequence of each pixel; and a median filter is applied to the thermal image sequence to calculate the neighborhood consensus temperature value of the pixel to construct the spatial neighborhood consensus temperature curve of the corresponding pixel as a dynamic baseline representing the local area health thermal characteristics of the pixel; the instantaneous deviation between the actual temperature time sequence of each pixel and the corresponding spatial neighborhood consensus temperature curve thereof is calculated; the instantaneous deviation is processed by one-way lag filter to retain the positive deviation value representing thermal hysteresis, and the filtered deviation value is accumulated over the whole time of the temperature rising process, and after normalization, the thermal hysteresis index representing the damage degree of each pixel is obtained;

[0009] The thermal hysteresis indexes of all pixels are aggregated to calculate the global damage accumulation value of the current detection period; and the global damage accumulation value is stored in the historical trend sequence, which records the global damage accumulation values of the previous detection periods; based on the historical trend sequence, a trend extrapolation model is applied to update the smoothing level and the current trend of the sequence; and according to the preset furnace lining failure threshold, the smoothing level and the current trend, the remaining service life of the furnace lining is calculated and outputted to realize the detection and life prediction of the thermal shock damage of the furnace lining.

[0010] By constructing the spatial neighborhood consensus temperature curve of each pixel as a dynamic baseline, the problem that the prior art cannot distinguish between normal thermal characteristic difference and damage point thermal hysteresis is fundamentally solved. By calculating the instantaneous positive deviation of the actual temperature of the pixel from the dynamic baseline and accumulating it over the whole time, the thermal hysteresis signal caused by damage can be effectively amplified from the mechanism, while the interference of normal boundary points and noise is suppressed, and the accuracy and robustness of early micro-damage detection are significantly improved. In addition, by analyzing and extrapolating the historical trend of the global damage accumulation value, not only the damage can be detected, but also the remaining service life of the furnace lining can be predicted, realizing fault diagnosis and health management.

[0011] The calculation method of the neighborhood consensus temperature value is as follows:

[0012]

[0013] wherein, the pixel at the time is the neighborhood consensus temperature value, the spatial neighborhood window centered at the pixel , the original temperature value of the pixel in the neighborhood window , the median function , and the pixel in the neighborhood window .

[0014] The neighborhood consensus temperature is calculated using the median function. This method leverages the high robustness of the median to outliers, ensuring that the constructed dynamic baseline can accurately reflect the general thermal characteristics of the local healthy area. This provides a more reliable and interference-resistant reference standard for subsequent deviation calculations, further improving the accuracy of damage identification.

[0015] Preferably, the thermal hysteresis index is calculated as follows:

[0016]

[0017] in, For pixels The thermal hysteresis index, This represents the total time step of the heating process. It is the highest target temperature for the furnace lining. For pixels exist The original temperature at that moment It is a pixel exist The neighborhood consensus temperature value at any given moment. It is a function for maximizing the value.

[0018] One-way hysteresis filtering is achieved by accumulating the positive deviation between the consensus temperature of the neighborhood and the actual temperature of the pixel. Compared with the method of simply calculating the difference, this method can accurately capture the temperature rise hysteresis phenomenon caused by the increase in thermal resistance due to damage, while effectively ignoring the temperature overshoot caused by noise or sensor fluctuations, i.e., negative deviation. This makes the calculated thermal hysteresis index more pure and stable in representing the degree of damage, thus enhancing the reliability and robustness of the detection results.

[0019] Preferably, the detection method further includes: constructing a thermal hysteresis index map based on the thermal hysteresis index; and setting a damage judgment threshold. ,when At that time, determine the pixel These are damaged pixels.

[0020] After calculating the thermal hysteresis index of each pixel, a thermal hysteresis index map was further constructed and a damage judgment threshold was set. This allows the damage status of the furnace lining to be presented intuitively in image form, facilitating rapid location and assessment of the distribution and severity of damaged areas. At the same time, the threshold judgment enables automated identification of damaged pixels, improving the practicality and efficiency of the detection.

[0021] Preferably, the global damage accumulation value is calculated as follows:

[0022] ;

[0023] in, This represents the cumulative global damage value. It is a pixel The final thermal hysteresis index.

[0024] By aggregating the final thermal hysteresis index of all pixels, a global cumulative damage value characterizing the overall health of the furnace lining is obtained. This global index provides a macroscopic, quantitative standard for assessing furnace lining health, offering a single, crucial input data for subsequent long-term health trend tracking and lifespan prediction.

[0025] Preferably, the trend extrapolation model is a quadratic exponential smoothing model.

[0026] The quadratic exponential smoothing model can better handle time series data with trends, such as damage accumulation processes, and can more accurately capture the dynamic changes in damage accumulation, thus providing a more reliable trend extrapolation basis for predicting remaining useful life.

[0027] The preferred method for updating the smoothing level and current trend of the sequence is as follows:

[0028]

[0029]

[0030] in, For the current testing cycle, This represents the cumulative global damage value for the current period. It's the end of the cycle. The smoothness level of the sequence up to now, It's the end of the cycle. The smoothing trend of the sequence up to now, and These represent the smoothing level and smoothing trend of the previous period, respectively. and These are the coefficients for the smoothing level and the smoothing trend, respectively.

[0031] Through iterative calculations, the latest detection data can be dynamically incorporated into historical trends, enabling the model to adaptively adjust its judgment on the future rate of damage deterioration. Compared with prediction methods using fixed model parameters, this method has higher prediction accuracy and timeliness.

[0032] Preferably, the remaining useful life is calculated as follows:

[0033]

[0034] in, For the remaining service life, The preset furnace lining failure threshold, It's the end of the cycle. a smoothing level of the sequence up to the time t, is a period a smoothed trend of the sequence up to the time t, is a floor function.

[0035] Based on the extrapolated damage trend and the preset failure threshold, the specific calculation method of the remaining useful life (RUL) is given. Compared with the traditional periodic maintenance or after-maintenance, this prediction-based maintenance strategy can provide clear decision basis for equipment managers, which is helpful to realize condition-based maintenance.

[0036] Preferably, the detection method further comprises: drawing a histogram of the thermal hysteresis index of all pixels, for showing the overall damage distribution of the current furnace lining.

[0037] The histogram can clearly show the separation of healthy pixels and damaged pixels from the statistical distribution, intuitively verify the effectiveness of the method in distinguishing normal and damaged signals, and be used to evaluate the overall severity and distribution characteristics of the furnace lining damage.

[0038] In a second aspect, the present application provides a thermal imaging-based furnace lining thermal shock damage detection system, which adopts the following technical scheme:

[0039] A thermal imaging-based furnace lining thermal shock damage detection system comprises a processor and a memory, and the memory stores computer program instructions which, when executed by the processor, implement the thermal imaging-based furnace lining thermal shock damage detection method described above.

[0040] The thermal imaging-based furnace lining thermal shock damage detection method described above is generated into a computer program and stored in the memory to be loaded and executed by the processor, so that the system is made according to the memory and the processor, and is convenient to use.

[0041] The present application has the following technical effects:

[0042] Firstly, a dynamic temperature baseline based on the median of the neighborhood of each pixel is constructed, which effectively solves the problem that the prior art cannot distinguish between normal thermal characteristic differences and damaged thermal hysteresis; secondly, the one-way hysteresis deviation of the actual temperature of the pixel relative to the dynamic baseline is accumulated, which realizes robust amplification and accurate quantification of early micro-damage signals; finally, the global damage is trend-extrapolated by a quadratic exponential smoothing model, which not only can accurately detect damage, but also can predict the remaining useful life of the furnace lining, and realizes the conversion from diagnosis to prediction. BRIEF DESCRIPTION OF DRAWINGS

[0043] Figure 1 is a flowchart of the thermal imaging-based furnace lining thermal shock damage detection method of the present application.

[0044] Figure 2 This is the thermal hysteresis index histogram of the present invention.

[0045] Figure 3 This is a trend chart of furnace lining health according to the present invention. Detailed Implementation

[0046] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. 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.

[0047] This invention discloses a method for detecting thermal shock damage to furnace linings based on thermal imaging, referring to... Figure 1 This includes the following steps:

[0048] S101: Acquire thermal image sequences during the furnace lining heating process, obtain the actual temperature time series of each pixel; and apply a median filter to the thermal image sequences to construct a spatial neighborhood consensus temperature curve for each pixel, which serves as a dynamic baseline characterizing the healthy thermal properties of the local area where the pixel is located.

[0049] First, during a typical heating cycle of the furnace lining, for example, from room temperature... Heat to target operating temperature , The value is 850°C, and thermal image sequences are continuously acquired using a thermal imager at a fixed frame rate. ,in For pixel coordinates, For time steps, i.e., frame index, frame rate This allows us to obtain the raw temperature of each pixel. .

[0050] To avoid using potentially damaged pixels The data serves as a benchmark for subsequent comparisons, for each pixel. Construct a desired temperature rise curve. This curve is determined by its spatial neighborhood. The other pixels within the image collectively determine the healthy heating pattern at that location. Specifically, this involves analyzing the thermal image sequence... The formula for applying the mean filter in the spatial dimension is:

[0051]

[0052] in, For pixels exist The neighborhood consensus temperature value at any given moment. For a centered spatial neighborhood window with a window size of 5x5, a neighborhood window pixels within the original temperature value, a median function, a neighborhood window pixels within.

[0053] This step uses the median instead of the mean, which is highly robust against outliers. For example, in a 5x5 window there are 25 pixels, even if a pixel itself is a damage point with a low temperature value, or there is a bad pixel in the neighborhood with a high or low temperature value, these extreme values will be placed at both ends after sorting, and the median will be taken from the middle, representing the most common, healthy thermal characteristic dynamic baseline of the local area.

[0054] In this way, by collecting thermal image sequences and using a median filter to construct a neighborhood consensus curve, a reliable and anti-interference reference benchmark is provided for subsequent differentiation of normal and abnormal thermal behavior.

[0055] S102: Calculate the instantaneous deviation between the actual temperature time series of each pixel and its corresponding spatial neighborhood consensus temperature curve; perform one-way lag filter processing on the instantaneous deviation to retain the positive deviation value representing thermal hysteresis, and accumulate the filtered deviation value over the entire time course of the temperature rise process, and after normalization, obtain the thermal hysteresis index representing the damage degree of each pixel.

[0056] Thermal shock damage causes the thermal resistance to increase during heating, so that continuously lower than , calculate the thermal hysteresis index , the expression is:

[0057]

[0058] wherein, is the final thermal hysteresis index of pixel , i.e. the damage index, is the total time step of the temperature rise process, is the neighborhood consensus temperature value of pixel at time , is the original temperature of pixel at time , is the instantaneous deviation between the expected neighborhood consensus temperature and the actual measured temperature. This is the maximum value function used for one-way hysteresis filtering. When the deviation is positive and hysteresis occurs, the deviation value is retained; when the deviation is negative and overshoot or noise occurs, it is set to 0 and not included in the accumulation. It is a full-time cumulative operation. It is the highest target temperature of the furnace lining, for example, its value is 850°C, as a fixed normalization factor.

[0059] For example:

[0060] Assumption (5 frames in total) .

[0061] Example 1: Healthy pixel (center area)

[0062] This pixel is highly consistent with its neighborhood and contains only random noise.

[0063] t=1: Deviation = 0.1. .

[0064] t=2: Deviation = -0.1. .

[0065] t=3: Deviation = 0. .

[0066] t=4: Deviation = -0.2. .

[0067] t=5: Deviation = 0.1. .

[0068] accumulation .

[0069] (Very close to 0).

[0070] Example 2: Damaged pixel (microcrack)

[0071] Due to thermal resistance, the temperature rise of this pixel lags behind that of its healthy neighborhood.

[0072] t=1: Deviation = 0.5. .

[0073] t=2: Deviation = 0.7. .

[0074] t=3: Deviation = 1.0. .

[0075] t=4: . Bias = 1.2. .

[0076] t=5: . Bias = 1.5. .

[0077] Cumulative .

[0078] (significantly larger than 0).

[0079] Example 3: Normal boundary point (near the furnace door)

[0080] Both the pixel and its neighborhood are warming up normally and slowly.

[0081] t=1: . Bias = 0.1. .

[0082] t=2: . Bias = -0.1. .

[0083] t=3: . Bias = 0. .

[0084] t=4: . Bias = -0.1. .

[0085] t=5: . Bias = 0.1. .

[0086] Cumulative .

[0087] (very close to 0).

[0088] As can be seen from this example, this step can effectively distinguish between damaged pixels (Example 2) and healthy pixels (Example 1) and normal boundary points (Example 3), the latter two of which have values close to 0, fundamentally solving the problem of false positives.

[0089] In this way, by calculating the one-way cumulative thermal hysteresis index, robust amplification of the tiny damage signal is achieved, and the damage heterogeneity and normal heterogeneity are effectively distinguished.

[0090] S103: Aggregate the thermal hysteresis index of all pixels, and calculate the global damage accumulation value of the current detection period; and store the global damage accumulation value in the historical trend sequence, which records the global damage accumulation values of the previous detection periods.

[0091] Construct a thermal hysteresis index map based on the results of S102 . Set a fixed damage determination threshold , for example , which is obtained by statistical analysis of healthy furnace lining data. When , it is determined as a damaged pixel. Then, calculate the global damage accumulation value of the entire furnace lining , the expression is: ; The value represents the total sum of the cumulative temperature loss of the entire furnace lining surface, which is a reliable indicator of the overall health of the furnace lining.

[0092] Finally, store the global damage accumulation value obtained in this detection in the historical trend database to obtain the historical trend sequence , . Among them, represents the global damage accumulation value obtained in the period.

[0093] In this way, by aggregating the global damage accumulation value and constructing the historical sequence, a data basis is provided for macroscopic evaluation and future trend prediction of the health of the furnace lining.

[0094] S104: Based on the historical trend sequence, apply a trend extrapolation model to update the smoothing level and current trend of the sequence; and according to the pre-set furnace lining failure threshold, combine the smoothing level and the current trend to calculate and output the remaining service life of the furnace lining, to realize the detection and life prediction of the thermal shock damage of the furnace lining.

[0095] According to industry standards, define a failure threshold for the total damage accumulation value . When exceeds , the furnace lining must be immediately shut down for maintenance, for example .

[0096] Since the historical trend sequence is usually a time sequence that shows a clear upward trend (damage accumulation) as the detection period increases, a quadratic exponential smoothing (Holts Method) is used to extrapolate the sequence. The iterative calculation formula of the quadratic exponential smoothing (Holts Method) is:

[0097]

[0098]

[0099] wherein, is the current detection cycle, is the global damage accumulation value for the current cycle. is the smoothed level of the sequence up to cycle , is the smoothed trend of the sequence up to cycle . and are the smoothed level and the smoothed trend of the previous cycle, respectively, and are the coefficients of the smoothed level and the smoothed trend, respectively, for example .

[0100] Finally, based on the level and the trend at the last time , the damage values for the next cycles are predicted, and the remaining useful life RUL is calculated by solving the value of at time :

[0101]

[0102] wherein is the floor function, and the number of cycles that still need to accumulate damage to reach the failure threshold is calculated by this formula. When the calculated smoothed trend is less than or equal to a very small positive threshold, for example Bc≤0.001, it indicates that there is no worsening trend in damage, and in this case the remaining useful life RUL can be considered as a specific maximum value representing a safe state, in order to avoid calculation errors and ensure the robustness of the model.

[0103] The calculation process of RUL is illustrated below through a simplified calculation example:

[0104] Assumptions: failure threshold . Smoothed coefficients .

[0105] Historical trend sequence: .

[0106] Initialization (based on data from cycles 1 and 2):

[0107]

[0108]

[0109] Iterative calculation:

[0110] Assuming after multiple iterations, at the... When the period is defined, the calculated smoothing value and trend are as follows:

[0111]

[0112]

[0113] The current number is Period: Calculated ;renew and :

[0114]

[0115]

[0116]

[0117]

[0118] The current estimated cumulative damage level is 199,750, and the damage deterioration rate is 12,625 per cycle.

[0119] RUL calculation:

[0120]

[0121]

[0122] The remaining service life is calculated to be 103 testing cycles.

[0123] Reference Figure 2 As shown, in step S103, all pixels are drawn. The histogram. Due to the robust design of the S102, when the Y-axis (number of pixels) of this graph is displayed on a logarithmic coordinate system, it clearly shows: There is an extremely high gray spike (from a large number of healthy pixels, including normal boundary points); in The presence of red distribution (damaged pixels) in the area, along with a clear isolation zone between the two, demonstrates that this solution can separate normal and damaged signals.

[0124] Reference Figure 3 As shown, in step S104, the historical trend sequence It is plotted as a line chart. The X-axis represents the detection period, and the Y-axis represents... The figure shows a blue broken line steadily rising with the period (for example, 1 to 10), which intuitively proves that the detection method of S101-S103 can stably capture the accumulation of damage, and the step S104 is just to extrapolate the RUL by using the trend of the blue line. By applying the trend extrapolation model, the method further provides the prediction of the remaining useful life on the basis of accurately detecting the damage, which provides a theoretical basis for equipment maintenance.

[0125] The embodiment of the present application also discloses a furnace lining thermal shock damage detection system based on thermal imaging, comprising a processor and a memory, the memory stores computer program instructions, and when the computer program instructions are executed by the processor, a furnace lining thermal shock damage detection method based on thermal imaging is realized.

[0126] The above system also includes a communication bus and a communication interface and other components well known to those skilled in the art, the setting and function of which are known in the art, and thus will not be repeated here.

[0127] The above are preferred embodiments of the present application, not to limit the protection scope of the present application, therefore: any equivalent changes made according to the structure, shape, principle of the present application should be covered within the protection scope of the present application.

Claims

1. A method for detecting thermal shock damage to a furnace lining based on thermography, characterized by, The method comprises the steps of: collecting a thermal image sequence during the furnace lining temperature rising process to obtain an actual temperature time sequence of each pixel; applying a median filter to the thermal image sequence to calculate a neighborhood consensus temperature value of the pixel to construct a spatial neighborhood consensus temperature curve of the corresponding pixel as a dynamic baseline representing the local area health thermal characteristics of the pixel; and calculating the instantaneous deviation between the actual temperature time sequence of each pixel and the corresponding spatial neighborhood consensus temperature curve; performing one-way lag filter processing on the instantaneous deviation to retain the positive deviation value representing thermal hysteresis, and accumulating the filtered deviation value over the entire time course of the temperature rising process to obtain a thermal hysteresis index representing the damage degree of each pixel after normalization; aggregating the thermal hysteresis indices of all pixels to calculate a global damage accumulation value of the current detection period; storing the global damage accumulation value into a historical trend sequence, which records the global damage accumulation values of previous detection periods; based on the historical trend sequence, applying a trend extrapolation model to update the smoothing level and smoothing trend of the sequence; and according to a preset furnace lining failure threshold, combining the smoothing level and smoothing trend to calculate and output the remaining service life of the furnace lining to realize the detection and life prediction of the thermal shock damage of the furnace lining.

2. A method of detecting thermal shock damage to a furnace lining based on thermography according to claim 1, characterized in that, The calculation method of the neighborhood consensus temperature value is: wherein, is a pixel In the neighborhood consensus temperature value at time t, is the original temperature value of a pixel is a spatial neighborhood window centered at is a pixel in the neighborhood window is the median function is a pixel in the neighborhood window in the neighborhood window 3. A method of detecting thermal shock damage to a furnace lining based on thermography according to claim 2, characterized in that, The calculation method of the thermal hysteresis index is: in, For pixels The thermal hysteresis index, This represents the total time step of the heating process. It is the highest target temperature for the furnace lining. For pixels exist The original temperature at that moment It is a pixel exist The neighborhood consensus temperature value at any given moment. It is a function for maximizing the value.

4. A method of detecting thermal shock damage to a furnace lining based on thermography according to claim 3, characterized in that, The detection method further comprises: constructing a thermal hysteresis index map based on the thermal hysteresis index; and setting a damage determination threshold When , the pixel is determined as a damage pixel.

5. The method of claim 1, wherein, The calculation method of the global damage accumulation value is: ； wherein, is a global damage accumulation value, is a thermal hysteresis index of the pixel .

6. The method of claim 1, wherein, The trend extrapolation model is a quadratic exponential smoothing model.

7. The method of claim 1, wherein the method further comprises: The method for updating the smoothing level and smoothing trend of the sequence is: wherein, is the current detection period, is the global damage cumulative value for the current period, is the smoothed level of the sequence up to period is the smoothed trend of the sequence up to period is the smoothed level of the sequence up to period is the smoothed trend of the sequence up to period and are the smoothed level and the smoothed trend of the previous period, respectively, and are the coefficients of the smoothed level and the smoothed trend, respectively.

8. The method of claim 1, wherein the method further comprises: The calculation method of the remaining service life is: wherein, is the remaining useful life, is a preset threshold of lining failure, is the smoothed level of the sequence up to cycle is the smoothed trend of the sequence up to cycle is the smoothed trend of the sequence up to cycle is the smoothed trend of the sequence up to cycle is a floor function.

9. The method of claim 1, wherein the method further comprises: The detection method further comprises: drawing a histogram of the thermal hysteresis indices of all pixels to show the overall damage distribution of the current furnace lining.

10. A thermal imaging based furnace lining thermal shock damage detection system, characterized in that, comprise: a processor and a memory, the memory storing computer program instructions which, when executed by the processor, implement a thermal imaging-based furnace lining thermal shock damage detection method according to any one of claims 1-9.

Images