TW201118317A - Vision-based method for combustion process monitoring, diagnosis, and computer-readable medium using the same - Google Patents

Abstract

A vision-based method for combustion process monitoring and diagnosis is disclosed, in which an original image is obtained, and then is converted to a converted image using at least one color space. A feature curve extraction is performed to the converted image and at least one curve section of the feature curve is selected, wherein the curve section corresponds to a specific image region of the converted image. Whether all or a portion of regions of the original image are abnormal have been monitored according to the feature curve or the selected curve section to monitor and diagnose the combustion process

Description

201118317 VI. Description of the Invention: [Technical Field] The present invention relates to an image-based process monitoring and diagnosis method, and particularly relates to a flame image-based combustion process monitoring and anomaly phenomenon and a cause diagnosis method. [Prior Art] The boiler system is the source of process production power in the current chemical plant, Dianwei or traditional manufacturing industry. However, it has been affected by the fluctuation of international oil prices and environmental awareness in recent years, and the increasing emphasis on industrial safety. The development of more efficient, more environmentally friendly emission standards and safer combustion monitoring systems has become an important issue in boiler equipment development and combustion process monitoring. The traditional combustion process monitoring method is for visual monitoring and diagnosis through an off-line industrial camera or peephole. This method relies on human experience and is not immediacy. While some systems use automated flame detectors to determine fire/no fire, these methods do not provide knowledge of the state of combustion and the quality of the combustion. In addition, some systems use multi-point thermocouples to sense the temperature on the wall of the furnace to determine the combustion state of the flame, but the thermocouple sensor is easily covered by unburned toner, and the measured temperature is not the combustion temperature around the flame. Even the sensor is prone to aging due to the influence of high temperature environment, which makes the data distorted. In order to obtain more information about the flame, some sensor manufacturers have developed direct measurement of flame temperature using ultraviolet, visible spectrum and infrared measurement techniques. Although they can learn more information than traditional temperature measurement methods, they are more optical. Higher price and maintenance is not easy, not 201118317 is suitable for general small and medium-sized factory investment. In recent years, due to the maturity of CCD sensing component technology, the cost of many image capturing devices such as cameras has become lower and lower. The use of CCD cameras to capture the flame image in the furnace can be used to instantly monitor the combustion situation in the furnace and solve the traditional measurement only. It can solve problems such as single point measurement or high cost of sensing equipment. However, because the amount of image information is very large, a single image is hundreds of thousands of pixels, and many are tens of millions of elements. Therefore, how to extract a few important and useful image feature information is very important for image pre-processing steps. issue. In addition, the stability of the combustion system is related to safety, fuel cost and environmental protection issues. Therefore, how to effectively monitor the stability of the combustion process and how to quickly detect abnormalities and diagnose the cause is also an urgent problem. SUMMARY OF THE INVENTION Based on the above objects, the present invention provides an image-based combustion process monitoring and diagnosis method for capturing visible light flame images in a furnace in a high temperature environment by a multi-dimensional image capturing device, and via a computer or the like. • The calculation unit and its internal algorithm program achieve the goal of combustion system monitoring and diagnosis. Embodiments of the present invention provide an image-based combustion process monitoring method, including the following steps. First, an original image of a flame image in a furnace is obtained, and the original image is converted into a converted image by using at least one color space. Then, a feature curve is captured from the converted image and at least one curve segment of the feature curve is selected, wherein the curve segment corresponds to a specific image region of one of the converted images. After that, the original monitoring combustion process is monitored based on the characteristic curve or the selected 201118317 curve section. Whether all or part of the area of the Qin image is abnormal [Embodiment] The description of the invention provides different embodiments to explain the technical features of the embodiment. Wherein, the arrangement of each component in the example of 眚^^J^1 is used by 〇明明, and the invention is not limited by 1". In the embodiments, the reference numerals are used to simplify the description, and are not meant to be between different embodiments: Embodiments of the present invention disclose an image-based combustion process monitoring and diagnosis method. The image-based combustion process monitoring and diagnosis method of the embodiment of the present invention utilizes a multi-dimensional image capturing device (for example, a CCD camera) to capture a characteristic curve obtained by capturing visible light flame images in a cavity at a temperature of two temperatures. Perform multi-dimensional image visual monitoring and diagnose abnormal patterns. Please refer to Figure 1. Figure 1 is a flow chart showing an image-based combustion process monitoring method in accordance with an embodiment of the present invention. First, in step S110, a multi-dimensional image capturing device (e.g., a CCD camera) is used to obtain an original image of a flame image contained in a hearth. After the image capturing device captures the image information, in step S120, the original image is converted into a converted image by using at least one color space for appropriate color space conversion. For example, the color space used in the image-based combustion program monitoring and diagnosis method of the embodiment of the invention 201118317 includes: gray scale, principal component, RGB, HSB (Hue/Saturation/

Brightness ), HSL ( Hue / Saturation / Lightness ), HSV ( Hue / Saturation / Value ) 'YIQ ( Luminace / Inphase / Quadrature ) ' YUV ( Luminace / Chrominance ) , YCbCr ( Luminace / blue

And red Chrominance) or temperature field, but it is not intended to limit the invention. In general, among the many color spaces, the gray space is the most frequently selected space because it has the smallest memory space. In order to simplify the description, in the following embodiments, the gray-scale color space is taken as an example. Therefore, the foregoing color space conversion system converts the original image into a converted image by using a gray-scale color space, that is, the converted image is a gray scale. image. Take Bumu, such as 芡骒S130, and draw the characteristic curve black of the image for 锝 to screen out the important curve segments in the characteristic curve as the basis for subsequent monitoring and diagnosis. The (4) dragon curve refers to the curve function of the screaming film, and the image can be converted into a feature (4) by an algorithm. For example, in the implementation of the financial, the Jixian image contains dirty 3D image information, and the image of the degree = 492x658 is taken as an example. The image includes a slice of pixels. Because the amount of data is too large, it is necessary to extract the conversion from the representative (10) fresh pixels. In the invention method, the original image is spatially converted. The η instance can be divided into the gray-scale space. The Guangyuan original # image is converted to the eight-dimensional gray-scale space based on the brightness value, and then the cumulative distribution function of the degree value (Cu_a" is used to reduce the number of points and the number of the fields. After knowing the characteristics of each dimension, 201118317 generates feature vectors based on all the image features obtained, and then uses these feature vectors to form a feature curve. In addition, different gray-scale feature curve segments correspond to different image regions of the original image. Referring to Fig. 2, Fig. 2 is a schematic diagram showing an image characteristic curve according to an embodiment of the present invention. As shown in Fig. 2, the characteristic curve 20 corresponds to a converted image of a raw image containing a flame image in a furnace. The image characteristic curve 20 drawn by CDF after the original image of a furnace flame is subjected to one-dimensional gray-scale space conversion, wherein the horizontal axis represents the brightness value of 0 to 255, and the vertical axis The dimension pixel surface is represented as a scaled value of 0 to 1. In the present embodiment, the characteristic curve 20 is divided into the characteristic curve sections 21-25, wherein the area where the gray scale intensity is small is the part where the original image brightness is dark, for example, The gray scale intensity value <38 indicates the furnace wall portion (the right furnace wall region 21 and the left furnace wall region 22), and the region with a large gray scale intensity is a portion where the original image brightness is bright, for example, the gray scale intensity is -222, Indicates the inner core of the flame (the inner region of the flame 25). The medium gray intensity indicates the flame (the inner ring region 24 of the flame) and its outer portion (the outer ring region 23 of the flame), so that different characteristic curves can be selected from the characteristic curves of the image. The curved section, in turn, interprets the condition of the flame. The important curve segments are selected from the characteristic curve, and the appropriate threshold values can be found by manual, automatic or semi-automatic analysis methods to distinguish the desired image regions, so that the characteristic curves correspond to In the present embodiment, the grayscale image is used for description. The so-called manual analysis method refers to 201118317. The grayscale value of the grayscale image is sequentially increased from 0 to 2. Generating an image pattern corresponding to the grayscale value of each, and then thoroughly == ratio is changed to form a pattern, measured way to find can be distinguished: Long grayscale serving as an important region of the value - the real seed,

Like a graph, by observing the change in the anti-white area of the image_, the gray-scale threshold value of the important n-domain of the sub-image is found. The so-called automatic analysis method refers to the use of the algorithm to define the similarity of the gate interpolation with the large characteristic, and then selects the gray-scale gate value of the corresponding image region from the similarity of the gate weights. For example, if the user wants to distinguish N image areas, the automated method uses N-1 gray thresholds as a reference. Please refer to Figure 3. Fig. 3 is a flow chart showing a method of automatically analyzing a segment of a characteristic curve according to an embodiment of the present invention for finding a complex grayscale threshold value of a complex image region corresponding to a converted grayscale image. In this embodiment, in order to avoid the problem of poor image partition caused by similar gray scale values, the distance between any two gray scale thresholds should not be less than a predetermined value H (for example, 1〇). As shown in FIG. 3, the automatic segment analysis method according to the embodiment of the present invention first calculates the threshold value similarity {τ" for each grayscale value i of the grayscale image (step 310). 0 to 255, and assume that the characteristic curve is a sequence {Ci} 'the first-order difference sequence is {Di}, the second-order difference sequence is {twist}, and the symbol sequence of the second-order difference is {Ei}. If Ei>0, let Ai =卜若Ei<〇,令

Ai=-1, Pi represents the percentile of Di, and the threshold value similarity {Ti} is defined as follows: 201118317 /-1 /+///2 is =-. k = i - H / 2 k = i + \ Next, all the calculated threshold similarities {Ti} are sorted from large to small, and a sorted sequence {Si} is obtained (step S320). Then, according to the number of image regions to be distinguished, a corresponding grayscale threshold value is selected from the sorting sequence {Si} to provide a corresponding grayscale threshold value (step S330). For example, if you want to distinguish N image regions, you can select the previous N-1 grayscale values from {Si} as the threshold value. If the distance between any two threshold values is less than the predetermined value, then After shaving the smaller value, select other values from {Si} in order until all N-1 thresholds satisfy the requirement that the distance of any two thresholds is greater than the established value H. Fig. 4 is a view showing the result of automatic analysis of the characteristic curve according to the embodiment of the present invention. It can be seen from the results in Fig. 4 that the grayscale values of 10, 37, 80, and 222 have relatively high similarities, indicating that the grayscale image is in the grayscale value = 10, 37, 80, 222. When the user wants to distinguish five image areas, the automatic analysis method outputs four gray level thresholds, 10, 37, 80, and 222. The so-called semi-automatic analysis method first uses the automatic method to find the similarity of the threshold value, and selects some initial threshold values. If the initial threshold value cannot effectively distinguish the N image areas, the invalid threshold value can be shaved first, and the manual threshold is referred to. As a result of the analysis, the appropriate threshold value is added, and finally N-1 threshold values are found to distinguish N important image regions. After using the aforementioned manual, from 201118317 'moving or semi-automatic analysis method to find the appropriate threshold value, the threshold value and image area name can be further marked on the characteristic curve for subsequent monitoring. Returning to FIG. 1 , after extracting the characteristic curve of the image and filtering out the important curve segments in the characteristic curve, as in step S140, all the characteristic curves or subsequent combustion using the specific characteristic curve segments may be utilized. Process image monitoring and diagnosis. If the user wants to monitor and diagnose all the image areas, the entire characteristic curve can be used for monitoring. If the user only wants to monitor and diagnose certain image areas, such as the fire knocking area or the furnace wall area, a specific characteristic curve section can be screened (such as the flame inner core area 25 or the right furnace wall shown in Fig. 2). Zone 21 and left furnace wall zone 22), and only this section is used for monitoring and diagnosis. Please refer to Figure 2, if the user only wants to monitor the area of the inner ring of the flame (including the inner core), just filter the characteristic curve section with the grayscale value of 80~255, if you want to monitor the whole image (including the entire flame zone and φ furnace wall area), the entire combustion process state monitoring and abnormality diagnosis can be carried out by including the characteristic curve of gray scale value 0~255. For the convenience of explanation, the following monitoring and diagnostic method descriptions are based on the global monitoring and diagnosis of the entire characteristic curve. In the monitoring phase, it is first necessary to collect normal images for a period of time and convert them into characteristic curves, and then calculate a curve control interval corresponding to the characteristic curves according to the characteristic curves, for example, calculate each longitudinal data of the characteristic curves. The average value produces a normal image of the average feature 11 201118317 curve, followed by a +3 or -3 standard deviation, and a curve control interval is calculated. Please refer to Figure 6A and Figure 6B. Fig. 6A is a diagram showing a characteristic curve distribution of a normal image according to an embodiment of the present invention. Fig. 6B is a diagram showing a characteristic curve having a control section in accordance with an embodiment of the present invention. From Fig. 6A, the distribution of the characteristic curves of these normal images in some specific regions can be found. As shown in Fig. 6B, 51 and 53 indicate the control boundary corresponding to the normal characteristic curve (between 51 and 53) of Fig. 6A, and the control boundaries 51 and 53 form a control section having a center line 54 ( That is, the average characteristic curve of the normal _ image. In the subsequent image monitoring, subsequent image monitoring can be performed according to the control boundary. If the characteristic curve of the subsequent image exceeds the control interval, it can be judged that the image is suspected to be abnormal. In one embodiment, the control interval may be represented in a conspicuous color such as red to facilitate interpretation. Although the aforementioned curve monitoring can provide more abnormal information, such as: abnormal image area, abnormal condition (for example, the upper boundary of the control area φ table flame becomes smaller, and the lower boundary of the control area represents the flame becomes larger), however, the curve After all, the monitoring is more complicated than the single point monitoring. Therefore, the embodiment of the present invention further provides a single point monitoring mode, which first calculates the characteristic index of the single point corresponding to the characteristic curve, and uses the indicator to monitor the image. In the embodiment of the present invention, the following steps are used to convert the curve into a point to calculate a characteristic index of a single point corresponding to the characteristic curve. First, using the method as described above, collecting a plurality of normal images, and converting the collected normal image 12 201118317 into a complex normal characteristic curve, and averaging the plurality of normal characteristic curves to obtain an average characteristic curve of the 'one plus ten shirt image' The average characteristic curve of this normal image is the center line 54 which can be the control interval shown in Fig. 6B. Then, the similarity between the characteristic curve and the normal image average characteristic curve is calculated for the subsequent image to be monitored as a characteristic index of the corresponding characteristic curve. The similarity of the aforementioned characteristic curves can be obtained by calculating the distances of the two curves, e.g., the Euclidean distance. It should be noted that this distance is not limited to Euclidean distance.

Other distances, such as: Squared Euclidean distance, City-block (Manhattan) distance, Chebychev distance or Mahalanobis distance. Referring to Figure 7, there is shown a schematic diagram of the similarity of the characteristic curves according to the embodiment of the present invention. In this implementation, it is assumed that the characteristic curve of the image to be monitored is G = ("1-2,...%), and the average characteristic curve of normal image transmission is h(W2,.", vJ ^ Λ The similarity can be obtained by calculating the Euclidean distance of the two curves. The formula is as follows: To calculate the distance between the characteristic line of the image to be monitored and the center line between the five systems, using the above formula, When the distance is too high, it can be judged that the image is abnormal. The control boundary is more 'in the above embodiment, the control area is the positive characteristic curve of the positive image and the +3 or _3 line is the value of the standard deviation of 3 times. Fig. 8 shows _. 75 8 brother 8 is not in accordance with the embodiment of the present invention Γ 5: 1 13 201118317 characteristic index f _. In the actual closed towel monitoring image, a total of continuous shooting just seconds, will take 1 per shot Waiting, the Japanese, the line 1340 represents the control limit, the curve is the connection of the "one-time _ material", like the connection of similarity, the mourning and the eve of the natal society, the curve of the curve in the bar η η Special 4, _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ This can be regarded as normal, and the guiding characteristic index falls on the daily limit beyond the control limit. At this time, a small abnormality has occurred, and the flame burning situation is the same as B in Fig. 8, and when the characteristic index falls within the =1330, it greatly exceeds Control · 134〇, indicating a large abnormality, the flame burning situation is like c and d in Figure 8. In Figure 8, it shows that the Lai indicator (4) gradually goes up, indicating that the fire record is getting more and more deviated. The trend of the normal range. It is worth noting that in an embodiment, although the characteristic index in Fig. 8 exceeds the control limit, the warning is abnormal. However, since the combustion flame has the dynamic characteristics of flicker, "continuous Μ can be added. There are N points (for example, there are 7 points in consecutive 1G points). The monitoring principle must be issued when the control limit is exceeded to reduce the misjudgment caused by the fire. Among them, M and N can be used by engineers according to actual use. For the image monitoring flowchart described above, please refer to FIG. 5. FIG. 5 is a flow chart showing image monitoring according to another embodiment of the present invention, such as normal image as 201118317 Positive = dimension image (step S51 〇), the normal image of the normal image is generated by the normal image of the normal image, and the positive image of the image to be monitored is turned over. The similarity of the characteristic curve, the relative control (four) levy index (step S54G), whether the characteristic index has exceeded the pipe; Γ二:^ [when the line or indicator exceeds the regulatory limit, the indication === for subsequent Abnormal diagnosis (step S58〇)〇m was born in the stool

In the image monitoring phase, the original jade will be continuously monitored. "Step S150." If the diagnosis is that there is no abnormality (Step S150, return to Step S14G to continue the subsequent image monitoring. If the step S150 is YES, then step S16〇-S17〇 is performed. After the abnormality of the dish is controlled, 'the abnormal phenomenon diagnosis must be performed (step sl6〇) j often causes the diagnosis (step S17〇) to know the phenomenon of the different f. And the type and the possible cause. The image-based combustion king monitoring and recording method in the embodiment of the present invention is in the diagnostic part of the different f phenomenon, and is beyond the image area of the characteristic curve of the control boundary = and the image of the reverse image to know The corresponding different hanging area' is shown in Fig. 10. The first and second drawing shows a schematic diagram showing the normal and abnormal result display according to the embodiment of the present invention. The thin line in the figure f and the figure 10B represent the monitored normal and abnormal, respectively. The characteristic curve and the 10Cth and ith〇d diagrams represent the normal and abnormal original images corresponding to the first 〇A picture and the first 〇B picture respectively. Obviously, the image of the 帛刚图 201118317 monitoring characteristic curve falls under the control Area In addition to the difference, it can be found that an abnormality occurs, and the observed characteristic curve falls outside the control interval. Through the classification of the characteristic curve segment, the region and phenomenon in which the abnormality occurs can be preliminarily judged. Compare the difference between the two with the abnormal original image to diagnose the abnormal phenomenon. In the part of the abnormal cause diagnosis, in this embodiment, it is assumed that several specific typical abnormal images have been defined in advance in a historical abnormal database. The characteristic curve corresponding to the specific typical abnormal images mentioned above is generated. Therefore, the concept of the similarity comparison of the characteristic curves can be used to compare the characteristic curves of the image to be diagnosed with the characteristic curves of the typical abnormal images in sequence, according to the comparison. As a result, the abnormal type is judged, and the typical abnormality with the highest similarity is regarded as the priority suspected abnormal type. Among them, the relative characteristic index of any image refers to the distance between the characteristic curve and the characteristic curve of the normal average image, the distance The shorter, the greater the similarity. For example, please refer to Figure 11. Section 11 A characteristic curve comparison diagram according to an embodiment of the present invention is shown, which respectively includes a characteristic curve Fault1-Fault3 of three typical abnormal images and a characteristic curve 1110 of a pending image. As can be seen from the figure, the characteristic curve 1110 of the image to be judged is Abnormal category 3 (curve Fault3) is the closest, with the highest similarity, exception category 2 (curve Fault2) is the second, and the exception category 1 (curve fault1) has the largest difference, with the lowest similarity. Therefore, the image to be judged is given priority. It is suspected that it is of the type of abnormal category 3. Similarly, the distance can be used to measure the similarity of the curve, and the distance between the 16 201118317 characteristic curve of the image to be diagnosed and the characteristic curve of the typical abnormal image is calculated. FIG. 12 shows the implementation according to the present invention. A schematic diagram of the distance between the characteristic curve of the image to be diagnosed and the characteristic curve of the typical abnormal image. As shown in Fig. 12, the characteristic curve of the image to be judged has the shortest Euclidean distance from the abnormal category 3 (Fault 3), and therefore has the highest similarity. Therefore, the typical abnormal category 3 can be regarded as the type or cause of the suspected priority. Then, through the experience of the engineering staff, find the real reason. For the image diagnosis flowchart described above, please refer to Figure 9. Fig. 9 is a view showing an image diagnosis flow chart according to another embodiment of the present invention. As shown in Fig. 9, when an abnormal image is detected (900), an abnormality diagnosis (902) and an abnormal cause diagnosis (908) are performed. In the case of abnormality diagnosis, the image region of the characteristic curve exceeding the regulatory limit is used to discriminate the region where the abnormality occurs (904). Or, the anti-white normal is compared with the abnormal image to know the corresponding abnormal region (906). The above two (904, 906) can be selectively executed to initially diagnose anomalies. Of course, in an embodiment, both (904, 906) can be performed to diagnose anomalies. • For the diagnosis of abnormal causes, first enter the characteristic curve of the image to be judged (910), and then retrieve the typical abnormal image of the specific known abnormal category and cause from the historical abnormal database (916), and then capture the typical abnormal image. Corresponding to the characteristic curve and filtering the curve segment (920), after which the characteristic curve of the typical abnormal image and the characteristic curve of the image to be judged are compared with the characteristic curve similarity (912), and the typical abnormality with the highest similarity is selected as The type or reason for which the priority is suspected (914). Therefore, engineers can quickly find the real cause based on the type or cause of the selected suspected priority and its experience. In addition, if the engineer finds that the abnormality is a new cause of abnormality after diagnosis, the related information such as abnormal images and causes can be added to the database after the problem r 17 201118317 is resolved, so as to facilitate subsequent diagnosis. An embodiment is exemplified below to further illustrate how the image-based combustion process monitoring and diagnosis method of the embodiment of the present invention can be used for combustion process monitoring and diagnosis, but is not intended to limit the present invention. In the present embodiment, how to diagnose an abnormal phenomenon is shown by two examples. Suppose the first example finds a characteristic index value anomaly, and its characteristic curve can find that the gray scale value 0~10 is out of bounds. From the previous characteristic curve analysis, it can be known that the gray scale value 0~10 represents the right furnace wall, so It is judged that this is an abnormality of the right furnace wall. At this time, the abnormal region of the abnormal image can be highlighted, and compared with the normal image, the abnormal phenomenon can be initially diagnosed. In the second example, an abnormality of the characteristic index is found, that is, the similarity between the characteristic curve and the average characteristic curve of the normal image exceeds the control interval, and the characteristic curve can be found that the grayscale value is about 30~120 beyond the limit, and the previous characteristic curve The analysis can know that this section belongs to the category of the periphery of the flame, so it can be judged that this is an abnormality related to the periphery of the flame, and can be confirmed by means of anti-white image and the like. In order to further understand the types of abnormalities and possible causes, further diagnosis and classification are necessary. In this practical case, we define three typical categories, the first is a typical normal image, taken from the normal average image, and the second is the typical anomaly category 1, which means that the air flow is too small, but the concentration of exhaust gas does not exceed the environmental protection. Standard question. The third type is the typical anomaly category 2, which means that the air flow is too small, it will smoke, and there is a risk of generating CO 18 201118317. After defining three typical categories, you can calculate their individual feature curves, and calculate their characteristic indicators for all images and plot them into trend graphs, as shown in Figure 13. Figure 13 shows a trend graph of similarity indicators for three typical types in accordance with an embodiment of the present invention. In this embodiment, one image of the image to be diagnosed is continuously captured every second for a total of 180 seconds, and after each image is converted into a characteristic curve, the similarity is compared with the characteristic curves of the above three typical types, from the trend. It can be seen that in the φ image of each photo (every 1 second), three distance-based similarity indicators can be obtained. The category with the smallest distance represents the maximum similarity between the image to be diagnosed and its existence, so the image to be diagnosed Will be judged as this category. For example, when the image to be diagnosed is at the 150th month (ie, 150 seconds), the line 1440 as shown in FIG. 13 intersects with three typical types at three points, of which the abnormal category 2 (as shown by 1430). The distance is the shortest, the distance of the abnormal category 1 (as shown by the 1420) is the second, and the distance from the normal image (such as the 1410 shown) is the longest, and the image to be diagnosed is classified as the abnormal category 2. # In addition, in order to simplify the classification of anomalies and make the anomaly categories predictable, all the abnormal category curves can be subtracted from the normal curve, and the corresponding anomaly index curves are obtained respectively. These corresponding anomaly index curves are drawn to obtain the corrected similarity. For the indicator trend chart, please refer to Figure 14. As shown in Figure 14, this modified similarity indicator trend graph can be regarded as a control chart with a large characteristic (the larger the index value is, the better), and the lower control limit is 0 (as shown in Figure 14) Post-similarity index 1550), and judge whether it is abnormal by the following principle: f 19 201118317 L If all abnormality index curves are greater than 〇, then (as shown in the interval 151〇). ~ Like the normal indicator curve "attached _, the fuzzy zone must be different (four) alert (as shown in the interval (10))., at this time 3. Right some abnormal index curve is significantly smaller than G, then the minimum preparation = judgment The abnormal category. The abnormal curve of the interval shown in the figure _ has the most ' 疋 abnormal category is the abnormal class, and the abnormality in the interval (10) ^ therefore judges F to read 2 minimum values, so the abnormal type is determined as the abnormal category ^ ie / 吏 all indicators anomaly curves are greater than ... if there are also ~

Potential: For example: Compliance with "Continuous 7 o'clock ^ line slow will occur. In other words: out, because there may be some abnormal process monitoring and diagnosis at this time 乂 Example based on the image C method: According to the image-based combustion quality monitoring Representing the characteristic curve of the image feature, and then the equal-area=two-symbol curve is divided into several representatives. 4: Use * (4) Lk 'The user can decide to use the U image, like monitoring 'or taking a specific characteristic curve section' For the whole domain, control. In addition, the shadow method of the base area of the present invention proposes two monitoring methods: characteristic curve monitoring t == 20 201118317 It is judged whether the combustion process is normal, and if there is an abnormality, the area where the image is abnormal is displayed in a reversed manner. The case also proposes a measure of the similarity of abnormal types to find the case most similar to historical anomalies, and assist engineers in diagnosing the cause of the abnormality, the action plan, and even the process impact assessment. Embodiments of the present invention further disclose a storage medium for a computer program for performing an image-based combustion process monitoring method. Figure 15 is a schematic diagram showing a computer readable storage medium in accordance with an embodiment of the present invention. The computer readable storage medium 1700 of the embodiment of the present invention stores a computer program 1600. The electrical φ brain program 1600 is used to load into a computer system and cause the computer system to perform the steps of the image-based combustion process monitoring method as described above. The computer program 1600 mainly includes a program logic 1610 for obtaining an original image of a flame image in a furnace, a program logic 1620 for converting the original image into a converted image by using at least one color space, capturing a characteristic curve from the converted image and filtering The program logic 1630 of at least one curve segment in the characteristic curve, using the characteristic curve or the selected curve segment, performing program monitoring logic 1640, using the difference between the abnormal image and the normal image characteristic curve to perform an abnormal phenomenon The program logic 1650 of the diagnosis and the program logic 1660 for the diagnosis of the abnormal cause are performed by using the characteristic curve of the abnormal image and the characteristic similarity and distance of the specific image in the database. The method of the invention, or a particular version or portion thereof, may exist in the form of a code. The code can be included in a physical medium such as a floppy disk, a CD, a hard disk, or any other machine readable (eg computer readable) storage medium in which the code is loaded and executed by a machine such as a computer. At this time, the machine becomes a device for participating in the present invention. The code can also be transmitted via some transmission medium, such as wire or cable, fiber optic, or any transmission type, where the machine becomes available when the code is received, loaded, and executed by a machine such as a computer. To participate in the device of the present invention. When implemented in a general purpose processing unit, the code combination processing unit provides a unique means of operation similar to the application specific logic. While the present invention has been described above by way of a preferred embodiment, it is not intended to limit the invention, and the present invention may be modified and modified without departing from the spirit and scope of the invention. The scope of protection is subject to the definition of the scope of the patent application.

22 201118317 [Simple Description of the Drawings] Fig. 1 is a flow chart showing an image-based combustion process monitoring method according to an embodiment of the present invention. Figure 2 is a diagram showing an image characteristic curve in accordance with an embodiment of the present invention. Figure 3 is a flow chart showing a method for automatically analyzing a segment of a characteristic curve according to an embodiment of the present invention. Fig. 4 is a view showing the result of automatic solution of the characteristic curve according to the embodiment of the present invention. Figure 5 is a flow chart showing the image monitoring according to another embodiment of the present invention. Fig. 6A is a diagram showing the distribution of characteristic curves of a normal image according to an embodiment of the present invention. Fig. 6B is a diagram showing a characteristic curve having a control section in accordance with an embodiment of the present invention. Fig. 7 is a schematic view showing the similarity of a characteristic curve according to an embodiment of the present invention. Figure 8 is a diagram showing a feature index control map in accordance with an embodiment of the present invention. Figure 9 is a flow chart showing the image diagnosis according to another embodiment of the present invention. Figures 10A-10D are schematic diagrams showing the display of normal and abnormal results in accordance with an embodiment of the present invention. Figure 11 shows a comparison of characteristic curves in accordance with an embodiment of the present invention 23 201118317. Fig. 12 is a view showing the distance between the characteristic curve of the image to be diagnosed and the characteristic curve of the typical abnormal image according to an embodiment of the present invention. Figure 13 is a graph showing the trend of similarity indicators for three typical types in accordance with an embodiment of the present invention. Figure 14 is a graph showing a trend of similarity indicators in accordance with a modified embodiment of the present invention.

Figure 15 is a diagram showing a computer readable storage medium in accordance with an embodiment of the present invention. [Main component symbol description] S110-S170~ execution step; 20~ characteristic curve; 21-25~ characteristic curve section; 51, 53~ control boundary line; 54~ center line; S310-S330~ execution step; S510-S580~ Execution steps; A, B, C, D ~ flame combustion situation; u, v ~ characteristic curve; 900, 902, 904, ..., 920 ~ process steps; 1110 ~ characteristic curve of the image to be judged;

Faultl-Fault3~ characteristic curve of typical abnormal image; 24 201118317 1310, 1320, 1330~ interval; 1340~ control upper limit; 1410-1430~ similarity index; 1510-1540~ interval; 1550~corrected similarity index; 1600~ Computer program; 1610..1660 ~ program logic; 1700 ~ computer can read storage media.

25

Claims (1)

201118317 Seven patent application scope: 1. Image-based combustion process monitoring method, including: using a multi-dimensional image capture to obtain an original image of a flame image in a cavity; Converting the original image into a conversion image, extracting a characteristic curve from the converted image, and filtering out the characteristic curve, the curve corresponding to the conversion image according to the characteristic curve or the selected image of the curved segment image or Whether part of the area is abnormal to monitor the job = original 2. The combustion system described in the application of the fourth patent (4), wherein the method of converting the original image includes: a million image. The grayscale color space is used to convert the original image into a grayscale shadow package. For example, the method of monitoring the process described in the first item of the patent (4), when the original image is abnormal, the characteristic curve and the frequency are used. A comparison of characteristic curve differences of hard-numbered images, execution-abnormality diagnosis and diagnosis of anomalies. 'Ding Yi, 1 (4), the combustion process monitoring method, all or part of the initial image θ θ Μ amp amp amp amp 控 控 控 控 控 控 控 控 控 控 控 控 控 控 控 控 控 控 控 控 控 控 控 控 控 控 控 控 控 控 控 控 控 控 控 控 26 26 201118317 Si:: The normal image is converted into a complex normal characteristic curve; whether the different y characteristic curve and the curve control interval are monitored, and when the original image initial curve is monitored beyond the curve control interval, the original (4) (4) 2 items are determined. The process monitoring method includes: stepping out at least one curve segment of the characteristic curve to further convert: 2: 2: automatic or half automatic resolution method to find the corresponding complex gray scale like the cut image series Threshold value, and the segment uses the gray scale 嶋 ' to filter out the curve in the characteristic curve. ==; The combustion process monitoring method described in the item, - gray scale value pair binary value:: increment by 255 ' And generating a combustion process monitoring method for each of the image patterns 'finding the corresponding image area of the basin 2::::Li snake circumference item 6. The method for the eight-dimensional image of the image pattern further includes: = such Like the graphics, the reversed area by observation of such images or graphics, such as those to be found on the gray scale value of the video door sowing area. Wherein the combustion process monitoring method described in the item ===; r λ I 27 201118317 defines a complex threshold value similarity with a large characteristic by using an algorithm; and automatically selects the correspondingness among the threshold similarities The grayscale threshold values of the image area. 9. The method of monitoring a combustion process as described in claim 8, wherein the method for defining a similarity of a complex threshold value having a large characteristic includes: calculating a threshold similarity for each grayscale value of the grayscale image . 10. The method of monitoring a combustion process as described in claim 9, wherein the method for selecting the gray thresholds corresponding to the image regions is selected as follows: the calculated similarity of the thresholds is as large as Smallly arranged to obtain a sorted column; and the grayscale thresholds are sequentially selected from the sorted column. 11. For the combustion process monitoring method described in claim 8, wherein the distance between any two gray scale thresholds is not less than a predetermined value. 12. The combustion process monitoring method according to claim 5, wherein the semi-automatic analysis method comprises: calling an algorithm to define a complex threshold value similarity with a large characteristic; from the similarity of the threshold values Selecting a plurality of initial threshold values; and manually adjusting the initial threshold values to obtain the grayscale threshold values corresponding to the image regions. 13. The combustion process monitoring method according to claim 1, wherein the step of monitoring whether all or part of the original image is abnormal according to the characteristic curve or the selected curve segment comprises: 28 201118317 collecting a plurality of normal images; converting the normal images collected into a plurality of normal characteristic curves; averaging the normal characteristic curves to obtain an average characteristic curve of a normal image; calculating the characteristic curve corresponding to the original image and the normal image The similarity of the average characteristic curve is used as a characteristic index corresponding to the characteristic curve; and according to the characteristic index, whether the original image is abnormal or not is determined. B. The combustion process monitoring method of claim 13, wherein the similarity is equal to a distance between the characteristic curve corresponding to the original image and an average characteristic curve of the normal image. 15. The method of monitoring a combustion process as described in claim 3, wherein the step of performing the abnormality diagnosis and the abnormal cause diagnosis by using the characteristic curve and the characteristic curve similarity of the specific images in the database comprises: : defining a plurality of typical anomalous images; • generating a complex anomaly characteristic curve corresponding to the typical anomalous images; and comparing the similarity between the anomalous characteristic curves and the characteristic curve of the original image, and determining an abnormality according to the comparison result Types of. 16. The combustion process monitoring method according to claim 15, wherein the ratio is similar to the characteristic curve of the original image and the step of determining the abnormal type according to the comparison result. The method includes: calculating the abnormal characteristic curve of each of the typical abnormal images and the occurrence of the characteristic curve of the occurrence of the 2011 2011 1717 * and the class of the distances to determine the original shadow distance and the smallest occurrence of the abnormality Method, ==== 监控 监控 : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 依据 依据 依据 依据 依据 依据 依据Judging the original shadow method, the combustion process monitoring party, ?, is normal or abnormal; = including the "! lower limit" to determine the limit, the part curve of the d constant index curve is significantly smaller than the control ^ will have a small value of the table _ 4t is the abnormal category of the determination. The abnormal category corresponding to the line is set to = the second part of the combustion process monitoring method as described in item 18 of the patent application scope:::, the abnormal index curve and the lower limit of the regulation, and the judgment is made. The original two, the normal or abnormal steps include: #曲t Γ 异 异 指标 指标 指标 指标 均 均 指标 且 且 且 且 且 且 且 且 且 且 且 且 且 且 且 且 且 且 异 异 异 异 异 异 异 异 下 下 下 下 下 下 下The combustion process monitoring method described in the third paragraph of the patent scope, 30 201118317, wherein the similarity of the characteristic curve and the data curve is used, and the image region of the characteristic curve of the abnormality diagnosis image is determined The combustion process monitoring method described in item 3 of the patent scope 1 of the production, the curve similarity, the step of performing the abnormality diagnosis step, and the comparison of the different (four) images with the normal image to know the correspondence 22· Please refer to the characteristic curve or the selected curve area in the combustion process supervision mentioned in the scope of the patent item: the step of whether the initial image part or part of the area is abnormal, and the original ghost set is included. a normal multi-dimensional image; the normal image that the image is to be collected is obtained through the conversion space - the normal conversion Lu line; the shoddy from the normal converted image - positive The image average feature curve calculates the similarity of the feature curve or the selected average feature curve to obtain a relatively normal image index; a feature of the Λ characteristic curve establishes a control limit of the feature index; ;:== the original image Carry out a continuous limit of w. Whether the characteristic indicator of the original shirt has exceeded the regulatory boundary 31