201239100 六、發明說明： 【發明所屬之技術領域] 本揭露是有關於一種高爐風口監控方法，特別是有關 於一種應用攝影系统來監測高爐風口的監控方法。 【先前技術】 高爐冶煉係一種將鐵礦石還原成生鐵的連續生產過 程。鐵礦石、焦炭和熔劑等固體原料根據爐況來從爐頂裝 料裝置分批送入高爐’並使爐喉料面保持一定的高度。焦 炭和礦石在爐内形成交替分層架構。礦石料在下降過程中 逐步被還原、熔化成鐵和渣，聚集在爐床中，並分別從鐵 口和渣口放出。 高爐風徑區焦炭與粉煤燃燒反應供應高爐煉鐵所需的 熱與還原氣體’因此高爐風徑區的反應穩定度直接影響高 爐產鐵的效率。為了監控高爐風徑區的反應情況，目前的 作法大多是在高爐鼓風口上設置窺視孔，以供高爐操作人 員觀察高爐風徑區的反應情況並做出適當的處理。 請參照第1圖，其係繪示習知鼓風嘴件組10的結構示 意圖。鼓風嘴件組10包含鼓風管u、粉煤喷鎗12、窺視 孔13及侧管14。鼓風管11係插設於高爐之侧壁15之風 口 15a上。粉煤喷鎗π係連設於鼓風管η旁，以將粉煤 輸送至鼓風管11之前端。窺視孔13係位於鼓風管11之後 端，用以提供操作員觀察風徑區16的情況。侧管14係與 鼓風管11連通，以供熱風進入鼓風管η中。 目前對於高爐風徑區的監控係以人工巡爐為主，例如i 4 201239100 • 每8個小時派人從窺視孔13監控高爐風徑區的反應情況。 然而’對於瞬息萬變的爐況而言，巡爐的間隔時間仍然太 長。但是，若再增加巡視頻率，相對地會增加人事成本， 同時對於緊急狀況的預測或防止亦極為有限。 因此，業界開發了多種的高爐監控方法來監控高爐風 徑區的狀況。 在中華民國專利號第1246537號案中，採用了光度計 來偵測粉煤燃燒的情況。然而，光度計的取像解析度非常 低’因此僅能就燃燒區域求一平均亮度，而無法對其它高 爐異常狀況進行監控。201239100 VI. Description of the invention: [Technical field to which the invention pertains] The present disclosure relates to a method for monitoring a blast furnace tuyere, and more particularly to a monitoring method for monitoring a blast furnace tuyere by using an imaging system. [Prior Art] Blast Furnace Smelting is a continuous production process for reducing iron ore to pig iron. The solid raw materials such as iron ore, coke and flux are fed into the blast furnace in batches from the top charging device according to the furnace conditions, and the throat surface is maintained at a certain height. The coke and ore form an alternating layered structure within the furnace. The ore material is gradually reduced, melted into iron and slag during the descending process, collected in the hearth, and discharged from the iron mouth and the slag port respectively. The coke and pulverized coal combustion reaction in the blast furnace wind zone provides the heat and reducing gas required for blast furnace ironmaking. Therefore, the reaction stability of the blast furnace wind-diameter zone directly affects the efficiency of blast furnace iron production. In order to monitor the reaction situation of the blast furnace wind tunnel area, most of the current methods are to set a peephole on the blast furnace blast opening for the blast furnace operator to observe the reaction situation of the blast furnace wind tunnel area and make appropriate treatment. Referring to Fig. 1, there is shown a schematic view of the structure of a conventional blast nozzle assembly 10. The blast nozzle assembly 10 includes a blast tube u, a pulverized coal lance 12, a peephole 13 and a side tube 14. The blast tube 11 is inserted into the tuyere 15a of the side wall 15 of the blast furnace. The pulverized coal gun π is connected to the blast tube η to deliver the pulverized coal to the front end of the blast tube 11. The peephole 13 is located at the rear end of the blast tube 11 to provide an operator with an observation of the wind tunnel region 16. The side tube 14 is in communication with the blast tube 11 for hot air to enter the blast tube n. At present, the monitoring of the blast furnace wind tunnel area is mainly based on manual patrol, for example, i 4 201239100 • Every 8 hours, people are sent from the peephole 13 to monitor the reaction of the blast furnace wind tunnel. However, for the ever-changing furnace conditions, the interval between the furnaces is still too long. However, if the video surveillance rate is increased, the personnel cost will be relatively increased, and the prediction or prevention of emergency situations is extremely limited. Therefore, the industry has developed a variety of blast furnace monitoring methods to monitor the condition of the blast furnace wind zone. In the case of the Republic of China Patent No. 1246537, a photometer was used to detect the combustion of pulverized coal. However, the photometric resolution of the photometer is very low. Therefore, only an average brightness can be obtained for the combustion area, and other abnormal conditions of the blast furnace cannot be monitored.
S 在美國專利號第5223908號案中，採用了光感測器來 擷取高爐風徑區的燃燒狀況，再利用分析穿透(Transmitted) 與反射(Reflection)或散射（Scattering)關係，以計算風口相 關參數。然而，光感測器僅能擷取亮度的相關資訊，而無 法用來對其它的高爐異常狀況進行監控。 在美國專利號第7209871號案中，主要利用實驗方法 與維度分析（dimensional analysis)來找出氣體流速、壓力等 關係。然而，這種方式只能建立製程參數，對異常狀況的 監測並無太大的幫助。 在美國專利號第7209871號案中，主要利用實驗方法 與維度分析(dimensional analysis)來找出氣體流速、壓力等 關係。然而，這種方式只能建立製程參數，對異常狀況的 監測並無太大的幫助。 . 在美國專利號第4842253號案中，主要利用探棒從風 . 口伸入來觀測高爐燃燒狀況。然而，由於高爐具有高溫高 5 201239100 壓之特性，因此這種方法有實施的危險性。 在美國專利號第5481247號案中，採用由直流電源、 電位計及光感測器所構成之高爐風口感測系統來監測高 爐。藉由高爐風口感測系統的電壓輸出，可以進行警報輸 出、風口阻塞、喷煤停止或感測系統損害之異況診斷。然 而，此案所能使用的資訊較少，無法發展較複雜的分析工 作。 鑒於以上分析，在此提供一種新的高爐風口監控方 法，以自動地監控高爐的爐況。 【發明内容】 本發明之一方面是在提供一種高爐風口監控方法，以 透過電腦系統與設置於風口的攝影機來判斷高爐風徑區中 是否有異常狀況產生。 根據本發明之一實施例，在此監控方法中，首先利用 攝影機來透過高爐之風口擷取出風口影像，其中此風口影 像係對應至高爐之風徑區。此風口影像係由複數條像素列 所構成，而每一像素列係由複數個像素所構成。接著，對 風口影像進行影像二值化處理，以獲得風口影像之二值化 影像，其中此二值化影像包含前景區域和背景區域。然後， 進行凸包計算步驟，以根據二值化影像來計算凸包區域。 接著，進行異常偵測步驟，以根據凸包區域來判斷是否有 異常狀況發生。In U.S. Patent No. 5,223,908, a light sensor is used to extract the combustion condition of the blast furnace wind tunnel, and then the relationship between Transmitted and Reflection or Scattering is used to calculate Air outlet related parameters. However, the light sensor can only capture information about the brightness and cannot be used to monitor other blast furnace anomalies. In U.S. Patent No. 7,208,871, experimental methods and dimensional analysis are mainly used to find out the relationship between gas flow rate and pressure. However, this method can only establish process parameters and does not help the monitoring of abnormal conditions. In U.S. Patent No. 7,208,871, experimental methods and dimensional analysis are mainly used to find out the relationship between gas flow rate and pressure. However, this method can only establish process parameters and does not help the monitoring of abnormal conditions. In U.S. Patent No. 4,842,253, the probe is mainly used to observe the combustion condition of the blast furnace from the wind. However, this method has a risk of implementation because of the high temperature and high pressure of the 2012-05100 blast furnace. In U.S. Patent No. 5,481,247, a blast furnace tuyere sensing system consisting of a DC power source, a potentiometer, and a light sensor is used to monitor the blast furnace. With the voltage output of the blast furnace tucus sensing system, it is possible to diagnose the alarm output, the tuyere blockage, the coal injection stop or the damage of the sensing system. However, the case can be used with less information and cannot develop more complex analytical work. In view of the above analysis, a new blast furnace tuyere monitoring method is provided to automatically monitor the furnace condition of the blast furnace. SUMMARY OF THE INVENTION One aspect of the present invention provides a blast furnace tuyere monitoring method for judging whether an abnormal condition occurs in a blast furnace wind tunnel region through a computer system and a camera disposed at a tuyere. According to an embodiment of the present invention, in the monitoring method, first, a camera is used to take out a tuyere image through a tuyere of a blast furnace, wherein the tuyere image corresponds to a wind tunnel area of the blast furnace. The tuyere image is composed of a plurality of pixel columns, and each pixel column is composed of a plurality of pixels. Then, the image of the tuyere image is binarized to obtain a binarized image of the tuyere image, wherein the binarized image includes a foreground area and a background area. Then, a convex hull calculation step is performed to calculate the convex hull region from the binarized image. Next, an abnormality detecting step is performed to determine whether an abnormal condition has occurred based on the convex hull area.
S 【實施方式】 6 201239100 β月參照第2圖，其係纟會示根據本發明實施例之爐況監 控β又備。爐況監視e又備包含攝影機21 〇以及電腦系統220。 攝影機210係设置於該窺視孔13外，用以從窺視孔13擁 取尚爐風口 15a之影像。電腦系統220係用以從攝影機21 〇 接收風口影像，以進行一高爐風口監控方法來判斷是否有 異常狀況產生。 凊參照第3a圖，其係繪示根據本發明實施例之高爐風 口監控方法300的流程示意圖。在高爐風口監控方法3〇〇 中’首先，進行取像步驟310，以利用攝影機來擷取風口 影像，其中此風口影像可對應顯示出高爐風徑區的實際狀 況。然後，進行二值化處理步驟320，以將風口影像二值 化。在二值化處理步驟320中，風口影像的像素係根據其 亮度來排列成如第3b圖所示之直方圖，以統計出各亮度所 對應的像素個數，其中縱軸為像素的個數，而橫軸為像素 冗度。接著’利用Ostu方法或最大熵法則（Maximum Entropy) 來算閥值’如此即可將風口影像二值化。 接著’進行凸包（Convex Hull)計算步驟330，以根據二 值化後的風口影像來計算凸包區域CHA，如第3c圖所示。 凸包計算步驟330係根據影像中的白色區域（圖中的點狀剖 線區域)之邊界點，如pA、pB、pc等來計算出包圍白色區 域的凸包線CHL’而被凸包線CHL所包圍的影像區域即為 凸包區域CHA(亦可稱為R〇i(Regi〇I1 Of Interest)圓）。在本 實施例中’凸包計算步驟330係採用快包法(Quick Hull)來 進行凸包區域CHA的計算，但本發明之實施例並不受限於 此。在本發明之其他實施例中，亦可使用增量法 7 i 201239100 (Incremental)、賈維斯步進法(Jarvis，March)等演算法來計 算凸包區域CHA。凸包計算步驟330係用以估測風口在影 像中的位置，如第3c圖所示，凸包線CHL·可代表風口在 影像中的位置。 然後，進行異常狀況偵測步驟340，以利用R0I來偵 測異常狀況。在本發明之實施例中，異常狀況偵測步驟會 根據風口影像來偵測出以下幾個情況攝 否模糊_峨否歪斜;(3)風徑區是否有=象象， (4)粉煤喷搶是否正在噴煤；（5)是否有大塊落料在風徑區 中。以下’將分別就上述的異常情況來說明本發明之實施 例如何_異常狀況。值得注意的是，本發明實施例不僅 可同時進行上述5個異常狀況的_，亦可僅進行-個或 兩個異常狀況的偵測。 況偵測步驟的流程示意圖。 判斷步驟400，其係用以你 判斷步驟400中，首先進子 包頂點，如PA、PB、PC等所 明參照第4圖’其係繪示根據本發明實施例之異常狀 此異常狀況偵測步驟包含模糊S [Embodiment] 6 201239100 β month refers to Fig. 2, which shows that the furnace condition monitoring according to the embodiment of the present invention is further prepared. The condition monitoring e includes a camera 21 and a computer system 220. The camera 210 is disposed outside the peephole 13 for capturing an image of the tuyere 15a from the peephole 13. The computer system 220 is configured to receive the tuyere image from the camera 21 to perform a blast furnace tuyere monitoring method to determine whether an abnormal condition has occurred. Referring to Figure 3a, there is shown a flow diagram of a blast furnace tuyere monitoring method 300 in accordance with an embodiment of the present invention. In the blast furnace tuyere monitoring method 3 ’ first, an image capturing step 310 is performed to capture the tuyere image by using a camera, wherein the tuyere image can correspond to the actual condition of the blast furnace wind tunnel area. Then, a binarization processing step 320 is performed to binarize the tuyere image. In the binarization processing step 320, the pixels of the tuyere image are arranged according to the brightness thereof as a histogram as shown in FIG. 3b, to count the number of pixels corresponding to each brightness, wherein the vertical axis is the number of pixels. And the horizontal axis is pixel redundancy. Then, using the Ostu method or Maximum Entropy to calculate the threshold value, the tuyere image can be binarized. Next, a Convex Hull calculation step 330 is performed to calculate the convex hull area CHA from the binarized tuyere image, as shown in Fig. 3c. The convex hull calculation step 330 calculates the convex envelope line CHL' surrounding the white area and is convexly wrapped according to the boundary points of the white area (the dotted line area in the figure) in the image, such as pA, pB, pc, and the like. The image area enclosed by CHL is the convex hull area CHA (also referred to as R〇i (Regi〇I1 Of Interest) circle). In the present embodiment, the convex hull calculation step 330 uses the Quick Hull method to perform the calculation of the convex hull area CHA, but the embodiment of the present invention is not limited thereto. In other embodiments of the present invention, an algorithm such as the incremental method 7 i 201239100 (Incremental) or Jarvis step method (Jarvis, March) may be used to calculate the convex hull region CHA. The convex hull calculation step 330 is used to estimate the position of the tuyere in the image. As shown in Fig. 3c, the convex envelope line CHL· can represent the position of the tuyere in the image. An abnormal condition detection step 340 is then performed to utilize the ROI to detect an abnormal condition. In the embodiment of the present invention, the abnormal condition detecting step detects whether the following conditions are blurred or not according to the image of the tuyere; (3) whether there is a shadow image in the wind tunnel area, (4) pulverized coal Whether the spray is blasting coal; (5) whether there is a large block in the wind tunnel area. The following description will respectively explain the implementation of the present invention, such as the _ abnormal condition. It should be noted that the embodiment of the present invention can perform not only the _ of the above five abnormal conditions but also the detection of one or two abnormal conditions. A schematic diagram of the process of the detection step. The determining step 400 is used in the determining step 400, first entering the sub-packet vertices, such as PA, PB, PC, etc., referring to FIG. 4, which shows the abnormal condition of the abnormality according to the embodiment of the present invention. Measurement step contains blur
广平均對焦值F定為: 士 |>(ά) 8 201239100 其中η為凸包頂點的個數。 以凸丄要儘量避免中央嘴煤部位，因此先 =度值，再::所== 映i實界在=得梯度值無， :ί對=ί罩處理後的影像進:^::二) 凸包可以真正含蓋二四周擴散以包圍凸包，讓 判斷=3,，接著進行判斷步驟，以 ，於預設之對焦閥值時，則進行判定步驟右:〇 = 疋風口影像為模糊狀態。在本實關巾預設 46灰階/像素’意即當平均對焦值F小於％時，^餅 220即判斷風口影像模糊。 、” 請參照第5圖，其係緣示根據本發明實施例之異 況偵測步驟的流程示意圖，此異常狀況偵測步 a 判斷步驟5G()，則貞測風徑區是否有浮造產生。值^、主= 的是，在本實施射，高爐風口監控方法3⑽ 未作業時(例如高爐定修停爐時)的風口影像⑽下稱為: 風口影像)來進行’以提供標準的⑽圓，而浮斷^ 5〇〇係針對高爐作業時的風口影像來進行偵& 業是否出現異常。 乂制冋爐作 在此異常狀況_步财，首先進行像素 步驟训’料減π騎巾聽水平料Μ像^梯度The wide average focus value F is set to: 士|>(ά) 8 201239100 where η is the number of convex vertices. In order to avoid the central mouth coal part, the first = degree value, then::=================================================================================== The convex hull can be covered with a cover for two weeks to surround the convex hull, so that the judgment = 3, and then the judgment step is performed, so that when the preset focus threshold is used, the determination step is performed right: 〇 = the image of the hurricane is Fuzzy state. In the actual cover towel preset 46 gray scale / pixel 'that is, when the average focus value F is less than %, ^ cake 220 to determine the tuyere image blur. Please refer to FIG. 5 , which is a schematic diagram showing the flow of the abnormal condition detecting step according to the embodiment of the present invention. The abnormal condition detecting step a determines the step 5G ( ), and then whether the wind path area is generated or not. The value ^, the main = is, in this implementation, the blast furnace tuyere monitoring method 3 (10) is not working (for example, when the blast furnace is fixed and shut down), the tuyere image (10) is called: tuyere image) to perform 'to provide the standard (10) circle And the floating ^ 5〇〇 system is used to detect whether there is an abnormality in the blasting and blasting operation of the blast furnace. The 冋 冋 作 作 在 异常 异常 _ _ _ _ , , , , 像素 像素 像素 像素 像素 像素 像素 像素 像素 像素 像素 像素 像素 像素 像素 像素Listening to the horizontal image
S 9 201239100 平句值例如，將一條像素行中所有傻+斜& 行加總後平均，如此可 應的灰階值進 來再對連續像素列平均灰階進行^列的平均灰階。接下 度值》對於整個梯度運算，以獲得平均梯 所代表之矩陸 f>像為M(M為多個像辛值 所代表之矩陣），影像之寬度為w 列轉京值 為h個像素大小，則像素梯产 、，、’影像之尚度 ]像素梯度句值Dh計算方法如下··S 9 201239100 The flat sentence value, for example, sums all the silly + oblique & lines in a pixel row and averages them so that the grayscale value of the corresponding pixel ranks is then averaged grayscale of the average grayscale of the consecutive pixel columns. The value of the next degree is calculated for the entire gradient to obtain the moment represented by the average ladder. The image is M (M is a matrix represented by multiple symplectic values), and the width of the image is w. The pixel size, then the pixel ladder, and the 'image image' pixel gradient sentence value Dh is calculated as follows··
Dh=▽士卿山.，1]τ 其中’中括號内的1共w個。 像素梯度平均值係代表影像 St::條水平像素列之像素梯 的梯度:Γ ，其可表示出每-水平像素列所對應 值行基線決定步驟520，以根據像素列梯度參考 來決^基線。在本實施例中，預設梯度 灰階/像素，而超過63的梯度參考值ς係對 f二度為D的像素列’因一此像素列來做為浮渣基 附件二所示。然後，進行判斷步驟530,以判斷基 否大_設之距離閱值。在本實施例中，基線 為風口位置下緣(即標準咖圓之凸包線CHL下緣） =的距離，而預設距離閱值為2〇個像素。當基線高度 ;20時’電腦系統220便進行判定步驟54〇，以判定 有洋渣產生，並立即通知高爐操作人員。 由上述說明可知’本實施例之浮渔判斷步驟则係利 用水平技衫梯度向量來估測浮《查液位，以判斷是否有浮渣 201239100 產生。另外’上述之浮渣判斷步驟谓係針對水平的 液位來進行侧。若浮錄位不為水平時， 二 方法來偵測浮渣液位。 知用八他 請參照第6 ® ’其係、㈣轉本發明實施例之異常狀 況偵測步驟的流程示意圖’此異常狀況躺步驟包含浮潰 判斷步驟_，則貞觀徑區是否有浮渣產生。浮渣判斷 步驟600係類似於浮潰判斷步驟，，但不同之處在於浮 產，斷步驟6GG係用以偵測非水平之浮渣液位。值得注意 的是，在本實施例中，高爐風口監控方法300係針對高爐 未作業時的歷史風口影像來進行，以提供標準的R〇I圓， 而浮渣判斷步驟600係針對高爐作業時的風口影像來進 行，以偵測高爐作業是否出現異常。 在浮渣判斷步驟600中，首先進行長度計算步驟61〇， 以計算風口影像中標準ROI圓内的凸包區域的凸包下緣線 段長度。接著，進行長度判斷步驟620，以判斷凸包下緣 線段的長度是否大於預設之長度閥值。然後，當凸包下緣 線段的長度大於預設長度閥值時，進行角度判斷步驟630， 以判斷凸包下緣線段之傾斜角是否小於預設之角度閥值。 在本實施例中，凸包下緣線段之傾斜角係指凸包下緣線段 與水平線段之夾角。接著，當凸包下緣線段之傾斜角小於 預設之角度閥值時’進行判定步驟640，以判定有傾斜浮 渣產生。 對於高爐作業時的風口影像而言，若風徑區有非水平 之浮渣液位，會使凸包區域出現傾斜的下緣區段’如附件 三所示。因此，浮渣判斷步驟600係計算出凸包下緣線段 201239100 的長度L以及凸包下緣線段的傾斜角度^，並利用預設的 長度閥值和角度閥值來確定此凸包下緣線段是否為傾斜的 浮渣液面。藉由適當地設定長度閥值和角度閥值，即可正 確地判斷岐否有料產生。在本實施射，長度闕值為 63個像素’而角度閥值為25度，但本發明之實施例並不 受限於此。 另外，值得注意的是，本發明實施例之長度判斷步驟 620與角度判斷步驟63〇的進行順序並不受限於本實施 例。在本發明之其他實施例中，可先進行角度判斷步驟 630，接著再進行長度判斷步驟62〇，或者同時進行角度判 斷步驟630和長度判斷步驟62〇〇當凸包下緣線段的長度 大於預設之長度閥值時’以及凸包下緣線段之傾斜角小於 預設之角度閥值時，即判斷有傾斜浮渣產生。 請同時參照第7a圖和第7b圖，第7a圖係繪示根據本 發明實施例之異常狀況偵測步驟的流程示意圖，第7b圖係 繪示根據本發明實施例之大塊落料時的風口影像。在本實 施例中’異常狀況偵測步驟包含大塊落料判斷步驟700, 以偵測風徑區是否有大塊落料。值得注意的是，在本實施 例中’高爐風口監控方法300係針對高爐未作業時的歷史 風口影像來進行，以提供標準的R0I圓，而大塊落料判斷 步驟700係針對高爐作業時的風口影像來進行，以偵測高 爐作業是否出現異常。 在大塊落料判斷步驟700中，首先進行面積計算步驟 710，以計算標準ROI圓内之子區域SCHA的面積，其中 子區域SCHA為風口位置(r〇i圓）中的白色區塊。在第7b 12 201239100 圖中，僅繪示有一個子區域SCHA，然而在本發明之其他 實施例中，子區域可能有兩個以上，此係因為當有大塊落 料產生時，原本風口位置中的白色區域可能會被大塊落料 切割為數個小區域。另外，子區域SCHA的面積計算方法 為本領域之公知常識，故不在此贅述。 在步驟710之後，接著進行判斷步驟720，以判斷所 有子區域SCHA的總面積是否小於預設之面積閥值。若所 有子區域SCHA的總面積小於預設之面積閥值時，則進行 判定步驟730，以判定有大塊落料掉入風徑區。在本實 施例中，面積閥值為2000像素平方，但本發明之實施例 並不受限於此。 請同時參照第8a圖和第8b圖，第8a圖係繪示根據本 發明實施例之異常狀況偵測步驟的流程示意圖，第8b圖係 繪示歷史喷槍區塊估測步驟的流程示意圖。在本實施例 中，異常狀況偵測步驟包含喷煤判斷步驟800,以判斷粉 煤喷槍是否停喷。 在喷煤判斷步驟800中，首先進行歷史喷槍區塊提供 步驟810，以提供歷史喷搶區塊與其面積值。歷史喷槍區 塊係指在喷搶尚未噴煤時，喷槍在風口影像中對應的影像 區塊。歷史喷槍區塊的提供可於高爐未作業階段下進行， 意即本實施例之歷史喷搶區塊提供步驟可合併步驟310至 步驟330的結果，來於高爐未作業的階段中進行，以提供 標準的ROI圓以及標準的喷搶區塊。為了方便說明起見， 以下先就歷史喷槍區塊的提供來進行說明。 在歷史喷搶區塊提供步驟810中，首先進行背景區塊 13 201239100 決定步驟812,以從二值化後的歷史風口影像（即在步驟330 中所獲得的風口影像）中決定出目標背景區塊。請參照第8c 圖和第8d圖，其係繪示喷槍在未喷煤時的風口影像。在第 8c圖中，可明顯看出噴搶係對應至兩個歷史影像區塊（突出 的剖線區塊部分），為了從影像中決定出此兩個區塊，在背 景區塊決定步驟812中，將歷史凸包區域（即凸包區域CHA) 内的歷史背景區塊定義出來（白色區塊部分），如第8d圖所 示，再接著進行面積計算步驟814，以計算這些歷史背景 區塊的面積。由第8d圖可看出，位在ROI圓内的歷史背 景區塊（白色區塊），除了噴搶所對應的區塊外，其他的區 塊都很小，因此在面積計算步驟814後，接著進行判斷步 驟816，以個別判斷每一區塊的面積是否大於預設的面積 閥值。若區塊的面積大於預設的面積閥值，則判定此區塊 為歷史喷槍區塊，即喷搶在風口影像中對應的影像區塊。 在本實施例中，面積閥值為93x93像素平方。 從上述說明中可知喷搶區塊提供步驟810可根據風口 影像來估測出喷搶的位置，而當喷槍區塊決定後，即可將 此喷搶區塊應用於喷搶是否停喷的判斷流程中。 請回到第8a圖，在歷史喷槍區塊提供步驟810之後， 接著進入喷煤判斷階段，以根據高爐作業時的風口影像來 判斷喷槍是否停喷。在喷煤判斷階段中，首先進行二值化 處理步驟820，以將風口影像二值化。二值化處理步驟820 係類似於二值化處理步驟320，故不在此贅述。接著，進 行凸包計算步驟830，以根據二值化後的風口影像來計算 凸包區域CHR，如第8e圖所示，其中第8e圖係繪示喷搶 201239100 喷煤時的二值化影像。凸包計算步驟830係類似於凸包計 算步驟330，故其計算原理不在此贅述。然後，進行背景 區塊決定步驟840,以決定出凸包區域CHR内的目標背景 區塊B1和B2。背景區塊決定步驟840係類似於背景區塊 決定步驟812，皆用以決定出凸包區域内的背目標景區塊， 因此背景區塊決定步驟840的決定流程不在此贅述。然 後，進行判斷步驟850，以判斷目標背景區塊B1和B2的 面積是否大於相應的歷史喷搶區塊面積，若背景區塊的面 積未大於歷史喷搶區塊面積時，則進行判定步驟860，以 判定喷搶停喷。 由第8e圖可看出，當影像中的左侧喷槍在喷煤時，凸 包區域CHR内的黑色區塊（圖中的斜剖線區塊）面積B1會 變得較大，因此將背景區塊B1與歷史喷搶區塊的面積進行 比較，若背景區塊B1的面積大於相應的歷史喷槍區塊的面 積，則表示左側喷槍可能正在喷煤。類似地，若是影像中 的右側喷槍正在喷煤，凸包區域CHR内的黑色區塊面積 B2也會變得較大，因此若背景區塊B2的面積大於相應的 歷史喷搶區塊的面積，則表示右側喷槍可能正在喷煤。 由上述說明可知，本實施例係利用面積比較的方式來 判斷粉煤喷槍是否停喷，然而在本發明之其他實施例中亦 可使用其他方式來判斷粉煤喷槍是否停喷。 請參照第9a圖，其係繪示根據本發明實施例之異常狀 況偵測步驟的流程示意圖，此異常狀況偵測步驟包含喷煤 判斷步驟900，以判斷粉煤喷搶是否停喷。在喷煤判斷步 驟900中，首先進行二值化處理步驟820以及凸包計算步 15 201239100 驟830，以得到如第8e圖所示之風口影像。接著，進行内 切圓面積計算步驟940,以計算凸包區域CHR内的最大内 切圓CA的面積，如第9b圖所示。然後，進行判斷步驟950, 以判斷最大内切圓的面積是否大於預設圓面積閥值。若最 大内切圓的面積大於預設圓面積閥值，則進行判定步驟 960，以判定粉煤喷搶停喷。 由上述說明可知，本實施例之喷煤判斷步驟900係利 用最大内切圓來判斷喷搶是否停喷。由第8e圖可看出，當 噴槍在喷煤時CHR内的最大内切圓面積會變得相當小，因 此藉由適當地設計圓面積閥值，即可利用最大内切圓來判 斷粉煤喷槍是否停喷。在本實施例中，圓面積閥值為800 像素平方，意即當CHR内的最大内切圓面積大於800時判 定喷槍停喷。 請參照第10a圖，其係繪示根據本發明實施例之異常 狀況偵測步驟的流程示意圖，此異常狀況偵測步驟包含攝 影機狀態判斷步驟1000，以根據高爐作業時的風口影像來 判斷攝影機是否處於歪斜狀態。在本實施例中，高爐風口 監控方法300係針對高爐未作業時的歷史風口影像來進 行，以提供標準的ROI圓，而攝影機狀態判斷步驟1〇〇〇 係針對高爐作業時的風口影像來進行，以判斷攝影機在高 爐作業時是否出現異常。 在攝影機狀態判斷步驟1000中，首先進行二值化處理 步驟820以及凸包計算步驟830，以計算風口影像的凸包 區域。接著，進行判斷步驟1100，以根據風口影像的凸包 區域、歷史風口影像的標準ROI圓以及距離閥值來判斷攝 201239100 影機是否歪斜。請參照 歪斜狀況。當攝5機/ 圖’其係繪示攝影機可能的 會⑽會往;像;方住移上動方=日卜口影像的凸包區域 時風口影像的凸包區域/ 地，當攝影機往右方歪斜 操取到風〇的_會往左方移動。當攝影機無法完全 緣截掉。gj此 *風像的凸包㈣便會被影像邊 像凸包區域的中ΪΓΓ係將標準R〇1圓的中心與風口影 部份，然後㈣二」如此可找出風°頭出影像外圍的 D2、D3或D Λ犬出部份與影像邊界的距離，例如IV 斜距離閥值4, 設歪斜距_值。若超過預設歪 雖缺k ㈣譲，叫线職歪斜。 ^、、、本發明已以數個實_揭露如上，然 二本發明所屬技術領域中任何具撕知識 不脫離本發明之精神和範圍内，當可 所::者::本發明之保護範圍當視後附之申請專利範圍 【圖式簡單說明】 ，讓本發明之上述和其他目的、特徵、和優點能更明 二易懂，上文特舉數個較佳實施例，並配合所附圖式，作 洋細說明如下： 第1圖係繪示習知鼓風嘴件組的結構示意圖。 第2圖係繪示根據本發明一實施例之爐況監控設備。 第3a圖係繪示根據本發明一實施例之高爐風口監控 方法的流程示意圖。 第3b圖係繪示根據本發明一實施例之風口影像的亮 17 201239100 度直方圖。 第3c圖係繪示根據本發明一實施例之風口影像的凸 包區域。 第4圖係繪示根據本發明一實施例之模糊判斷步驟的 流程示意圖。 第5圖係繪示根據本發明一實施例之浮渣判斷步驟的 流程不意圖。 第6圖係繪示根據本發明一實施例之浮渣判斷步驟的 流程示意圖。 第7a圖係繪示根據本發明一實施例之大塊落料判斷 步驟的流程示意圖。 第7b圖係繪示根據本發明一實施例之大塊落料時的 風口影像。 第8a圖係繪示根據本發明一實施例之喷煤判斷步驟 的流程不意圖。 第8b圖係繪示根據本發明一實施例之歷史噴搶區塊 估測步驟的流程示意圖。 第8c圖係繪示根據本發明一實施例之喷搶未喷煤時 的風口影像。 第8d圖係繪示根據本發明一實施例之喷槍未喷煤時 的風口影像。 第8e圖係繪示根據本發明一實施例之喷槍喷煤時的 二值化影像。 第9a圖係繪示根據本發明一實施例之喷煤判斷步驟 的流程不意圖。 第9b圖係繪示根據本發明一實施例之凸包區域内的5 201239100 最大内切圓的示意圖。 第10a圖係繪示根據本發明一實施例之攝影機狀態判 斷步驟的流程示意圖。 第10b圖係繪示根據本發明一實施例之攝影機的可能 歪斜狀況。 【主要元件符號說明】 10 :鼓風嘴件組 12 :粉煤喷鎗 14 :側管 15a :風口 210 :攝影機 300 :高爐風口監控方法 320 :二值化處理步驟 340 :異常狀況偵測步驟 400 :模糊判斷步驟 420 :平均值計算步驟 440 :判定步驟 500 :浮渣判斷步驟 520 :基線決定步驟 540 :判定步驟 600 :浮渣判斷步驟 620 :長度判斷步驟 640 :判定步驟 700 :大塊落料判斷步驟 11 :鼓風管 13 :窺視孔 15 :側壁 16 :風徑區 220 :電腦系統 310 :取像步驟 330 :凸包計算步驟 410 :梯度值計算步驟 430 :判斷步驟 510 :梯度計算步驟 530 :判斷步驟 610 :長度計算步驟 630 :角度判斷步驟 710 :面積計算步驟 201239100 720 判斷步驟 800 喷煤判斷步驟 812 背景區塊決定步驟 816 判斷步驟 830 凸包計算步驟 850 判斷步驟 900 喷煤判斷步驟 950 判斷步驟 1000 :攝影機狀態判斷步驟 1200:判定步驟 B1 :背景區塊 CA :内接圓 CHA :凸包區域 CHL :凸包線 PA、PB、PC :凸包頂點 730 :判定步驟 810 :歷史喷槍區塊提供步驟 814 :面積計算步驟 820 :二值化處理步驟 840 :背景區塊決定步驟 940 :内切圓面積計算步驟 960 :判定步驟 1100 :判斷步驟 B2 :背景區塊 CHR :凸包區域 D!、D2、D3、D4 :距離 SCHA :子區域Dh=▽士卿山.,1]τ where 1 of the brackets in the brackets. The pixel gradient average represents the gradient of the pixel ladder of the image St:: horizontal pixel column: Γ , which can represent the value of each horizontal pixel column baseline determination step 520 to determine the baseline according to the pixel column gradient reference . In this embodiment, the gradient gray scale/pixel is preset, and the gradient reference value of more than 63 is the pixel column of the second dimension of D. The pixel column of the second degree is shown as the scum base. Then, a decision step 530 is performed to determine whether the base is greater than the distance read value. In this embodiment, the baseline is the distance from the lower edge of the tuyere position (ie, the lower edge of the convex line of the standard coffee circle CHL), and the preset distance is 2 pixels. When the baseline height; 20 o'clock, the computer system 220 performs a decision step 54 to determine the generation of foreign slag and immediately notify the blast furnace operator. It can be seen from the above description that the floating fishing judgment step of the present embodiment uses the horizontal skill gradient vector to estimate the floating liquid level to determine whether or not scum 201239100 is generated. Further, the above-described dross determining step is performed on the side with respect to the horizontal liquid level. If the floating position is not horizontal, the second method is to detect the scum level. Please refer to the flow chart of the abnormality detection step of the sixth embodiment of the invention, which is referred to as the "6", and the fourth step of the invention, and the step of detecting the abnormal condition is included in the step of arranging the abnormal condition. . The scum determination step 600 is similar to the pulverization determination step, but the difference is in the floating production, and the breaking step 6GG is used to detect the non-level scum level. It should be noted that in the present embodiment, the blast furnace tuyere monitoring method 300 is performed for the historical tuyere image when the blast furnace is not working to provide a standard R〇I circle, and the dross determining step 600 is for the blast furnace operation. The tuyere image is used to detect whether the blast furnace operation is abnormal. In the dross determining step 600, a length calculating step 61 is first performed to calculate the length of the lower edge of the convex hull of the convex hull area in the standard ROI circle in the tuyere image. Next, a length judging step 620 is performed to determine whether the length of the lower edge line of the convex hull is greater than a preset length threshold. Then, when the length of the lower edge of the convex hull is greater than the preset length threshold, the angle determining step 630 is performed to determine whether the inclination angle of the lower edge of the convex hull is less than a preset angle threshold. In this embodiment, the inclination angle of the lower edge line segment of the convex hull refers to the angle between the lower edge line segment of the convex hull and the horizontal line segment. Next, when the inclination angle of the lower edge line of the convex hull is less than the preset angle threshold, a decision step 640 is performed to determine that the scum is generated. For the tuyere image during blast furnace operation, if there is a non-level scum level in the wind tunnel area, the inclined lower edge section will appear in the convex hull area as shown in Annex III. Therefore, the dross determining step 600 calculates the length L of the lower edge line segment 201239100 of the convex hull and the inclination angle ^ of the lower edge line segment of the convex hull, and determines the lower edge line segment of the convex hull by using a preset length threshold and an angle threshold. Whether it is a sloping liquid level. By appropriately setting the length threshold and the angle threshold, it is possible to correctly determine whether or not the material is generated. In the present embodiment, the length 阙 is 63 pixels' and the angle threshold is 25 degrees, but the embodiment of the present invention is not limited thereto. Further, it is to be noted that the order of the length judging step 620 and the angle judging step 63 of the embodiment of the present invention is not limited to the embodiment. In other embodiments of the present invention, the angle determining step 630 may be performed first, followed by the length determining step 62, or the angle determining step 630 and the length determining step 62, respectively, when the length of the lower edge of the convex hull is greater than the pre- When the length threshold is set and the inclination angle of the lower edge of the convex hull is less than the preset angle threshold, it is judged that the scum is generated. Please refer to FIG. 7a and FIG. 7b simultaneously. FIG. 7a is a schematic flow chart showing an abnormal condition detecting step according to an embodiment of the present invention, and FIG. 7b is a schematic diagram showing a bulk blanking according to an embodiment of the present invention. Air outlet image. In the present embodiment, the abnormal condition detecting step includes a bulk blanking determining step 700 to detect whether there is a large blanking in the wind tunnel area. It should be noted that in the present embodiment, the 'blast furnace tuyere monitoring method 300 is performed for the historical tuyere image when the blast furnace is not working to provide a standard R0I circle, and the bulk blanking judging step 700 is for the blast furnace operation. The tuyere image is used to detect whether the blast furnace operation is abnormal. In the bulk blanking determination step 700, an area calculation step 710 is first performed to calculate the area of the sub-area SCHA within the standard ROI circle, where the sub-area SCHA is the white block in the tuyere position (r〇i circle). In the figure 7b 12 201239100, only one sub-area SCHA is shown, however, in other embodiments of the present invention, there may be more than two sub-areas, because the original tuyere position is generated when a large blank is generated. The white areas in the area may be cut into small areas by large blanks. Further, the method of calculating the area of the sub-area SCHA is common knowledge in the art, and therefore will not be described here. After step 710, a decision step 720 is then performed to determine if the total area of all sub-areas SCHA is less than a predetermined area threshold. If the total area of all sub-areas SCHA is less than the preset area threshold, then a decision step 730 is performed to determine that a large blank has fallen into the wind path area. In the present embodiment, the area threshold is 2000 pixels square, but the embodiment of the present invention is not limited thereto. Please refer to FIG. 8a and FIG. 8b simultaneously. FIG. 8a is a schematic flow chart showing an abnormal condition detecting step according to an embodiment of the present invention, and FIG. 8b is a flow chart showing a historical spray gun block estimating step. In the present embodiment, the abnormal condition detecting step includes a coal injection determining step 800 to determine whether the powder gun is stopped. In the coal injection determination step 800, a history spray gun block providing step 810 is first performed to provide a historical spray block and its area value. The historical spray gun section refers to the corresponding image block of the spray gun in the image of the tuyere when the coal has not been sprayed. The provision of the historical spray gun block can be performed in the blast furnace non-operational stage, that is, the historical spray block providing step of the embodiment can combine the results of steps 310 to 330 to perform in the stage of the blast furnace not working, A standard ROI circle is provided as well as a standard spray block. For the sake of convenience, the following description of the history of the spray gun block is provided. In the historical squirting block providing step 810, the background block 13 201239100 decision step 812 is first performed to determine the target background area from the binarized historical tuyere image (ie, the tuyere image obtained in step 330). Piece. Please refer to the 8c and 8d drawings, which show the image of the tuyere when the gun is not fired. In Fig. 8c, it can be clearly seen that the blasting system corresponds to two historical image blocks (the highlighted section line block portion), and in order to determine the two blocks from the image, the background block decision step 812 is made. In the middle, the historical background block in the historical convex hull area (ie, the convex hull area CHA) is defined (white block part), as shown in Fig. 8d, and then the area calculating step 814 is performed to calculate these historical background areas. The area of the block. It can be seen from Fig. 8d that the historical background block (white block) located in the ROI circle, except for the block corresponding to the squirting, the other blocks are small, so after the area calculation step 814, A decision step 816 is then performed to individually determine if the area of each block is greater than a predetermined area threshold. If the area of the block is greater than the preset area threshold, it is determined that the block is a historical spray gun block, that is, the corresponding image block in the tuyere image is sprayed. In this embodiment, the area threshold is 93 x 93 pixels squared. It can be seen from the above description that the spray block providing step 810 can estimate the position of the spray according to the tuyere image, and when the spray block is determined, the spray block can be applied to the spray or stop spray. In the judgment process. Returning to Figure 8a, after step 810 of the historical spray gun block is provided, the coal injection judging phase is then entered to determine whether the spray gun is stopped based on the tuyere image of the blast furnace operation. In the coal injection judging phase, a binarization processing step 820 is first performed to binarize the tuyere image. The binarization processing step 820 is similar to the binarization processing step 320 and will not be described here. Next, a convex hull calculation step 830 is performed to calculate the convex hull region CHR according to the binarized tuyere image, as shown in FIG. 8e, wherein the 8th image shows the binarized image when the blasting 201239100 coal injection . The convex hull calculation step 830 is similar to the convex hull calculation step 330, so the calculation principle is not described here. Then, a background block decision step 840 is performed to determine the target background blocks B1 and B2 within the convex hull region CHR. The background block decision step 840 is similar to the background block decision step 812, and is used to determine the back target scene block in the convex hull area. Therefore, the decision process of the background block decision step 840 is not described herein. Then, a determining step 850 is performed to determine whether the area of the target background blocks B1 and B2 is greater than the area of the corresponding historical shot block. If the area of the background block is not greater than the area of the historical shot block, then a decision step 860 is performed. In order to determine the spray stop and spray. It can be seen from Fig. 8e that when the left side spray gun in the image is coal injection, the area B1 of the black block (the oblique line block in the figure) in the convex hull area CHR will become larger, so The background block B1 is compared with the area of the historical spray block. If the area of the background block B1 is larger than the area of the corresponding historical spray block, it means that the left side spray gun may be spraying coal. Similarly, if the right side of the image is being shot, the black area B2 in the convex hull area CHR will also become larger, so if the area of the background block B2 is larger than the area of the corresponding historical spurt block , indicating that the right side of the spray gun may be spraying coal. As can be seen from the above description, this embodiment uses the area comparison method to determine whether the pulverized coal lance is stopped. However, in other embodiments of the present invention, other methods can be used to determine whether the pulverized coal lance is stopped. Please refer to FIG. 9a, which is a flow chart showing an abnormal condition detecting step according to an embodiment of the present invention. The abnormal condition detecting step includes a coal injection determining step 900 to determine whether the coal shot is stopped or not. In the coal injection determination step 900, a binarization processing step 820 and a convex hull calculation step 15 201239100 step 830 are first performed to obtain a tuyere image as shown in Fig. 8e. Next, an inscribed circle area calculation step 940 is performed to calculate the area of the maximum inscribed circle CA in the convex hull region CHR as shown in Fig. 9b. Then, a determination step 950 is performed to determine whether the area of the largest inscribed circle is greater than a preset circular area threshold. If the area of the largest inscribed circle is greater than the preset circular area threshold, then a decision step 960 is performed to determine the pulverized coal shot stop. As apparent from the above description, the coal injection determining step 900 of the present embodiment uses the maximum inscribed circle to determine whether or not the spray is stopped. It can be seen from Fig. 8e that the maximum inscribed circle area in the CHR becomes quite small when the spray gun is injected, so by appropriately designing the circular area threshold, the maximum inscribed circle can be used to judge the pulverized coal. Whether the spray gun stops spraying. In the present embodiment, the circular area threshold is 800 pixels square, which means that the spray gun is stopped when the maximum inscribed circle area in the CHR is greater than 800. Please refer to FIG. 10a, which is a flow chart showing an abnormal condition detecting step according to an embodiment of the present invention. The abnormal condition detecting step includes a camera state determining step 1000 to determine whether the camera is based on the tuyere image during the blast furnace operation. It is in a skewed state. In the present embodiment, the blast furnace tuyere monitoring method 300 is performed for the historical tuyere image when the blast furnace is not in operation to provide a standard ROI circle, and the camera state judging step 1 is performed for the tuyere image during the blast furnace operation. To determine if the camera is abnormal during blast furnace operation. In the camera state determination step 1000, a binarization process step 820 and a convex hull calculation step 830 are first performed to calculate a convex hull region of the tuyere image. Next, a determination step 1100 is performed to determine whether the 201239100 camera is skewed based on the convex hull area of the tuyere image, the standard ROI circle of the historical tuyere image, and the distance threshold. Please refer to the skew condition. When the camera 5 / picture 'the picture shows the camera may be (10) will go; like; square shift to the convex side = the convex area of the image of the mouth of the mouth when the image of the convex hull area / ground, when the camera goes to the right The square 歪 obliquely fetched to the wind _ will move to the left. When the camera is not completely cut off. Gj This convex image of the wind image (4) will be the center of the standard R〇1 circle and the shadow of the wind tunnel by the middle of the image convex area, and then (4) 2” The distance between the D2, D3, or D dog part and the image boundary, such as the IV slanting distance threshold of 4, is set to the slanting distance _ value. If the default is exceeded, although k (four) is missing, the line is called skewed. The present invention has been disclosed in the above, and it is to be understood that the scope of the present invention is not limited by the spirit and scope of the present invention. The above and other objects, features, and advantages of the present invention will become more apparent from the aspects of the appended claims. The drawings are described as follows: Figure 1 is a schematic view showing the structure of a conventional blaster nozzle set. Fig. 2 is a diagram showing a condition monitoring device according to an embodiment of the present invention. Fig. 3a is a flow chart showing a method of monitoring a blast furnace tuyere according to an embodiment of the present invention. Figure 3b is a diagram showing the brightness of the tuyere image according to an embodiment of the present invention. Figure 3c is a diagram showing a convex hull region of a tuyere image in accordance with an embodiment of the present invention. Figure 4 is a flow chart showing the steps of the blur determination according to an embodiment of the present invention. Fig. 5 is a flow chart showing the flow of the dross determining step according to an embodiment of the present invention. Figure 6 is a flow chart showing the steps of determining the dross according to an embodiment of the present invention. Fig. 7a is a flow chart showing the step of judging the blanks according to an embodiment of the present invention. Figure 7b is a view of the tuyere when the bulk is blanked according to an embodiment of the present invention. Fig. 8a is a schematic view showing the flow of the coal injection judging step according to an embodiment of the present invention. Figure 8b is a flow chart showing the estimation step of the historical spray block according to an embodiment of the present invention. Fig. 8c is a view showing the tuyere image when the unsprayed coal is sprayed according to an embodiment of the present invention. Fig. 8d is a view showing a tuyere image when the spray gun is not fired according to an embodiment of the present invention. Figure 8e is a diagram showing a binarized image of a spray gun when coal is injected according to an embodiment of the present invention. Fig. 9a is a schematic view showing the flow of the coal injection judging step according to an embodiment of the present invention. Figure 9b is a schematic diagram showing the maximum inscribed circle of 5 201239100 in the convex hull region according to an embodiment of the present invention. Figure 10a is a flow chart showing the steps of the camera state determination according to an embodiment of the present invention. Figure 10b is a diagram showing the possible skew of the camera in accordance with an embodiment of the present invention. [Main component symbol description] 10: blast nozzle group 12: pulverized coal lance 14: side tube 15a: tuyere 210: camera 300: blast furnace tuyere monitoring method 320: binarization processing step 340: abnormal condition detecting step 400 : Fuzzy determination step 420: average value calculation step 440: determination step 500: scum determination step 520: baseline determination step 540: determination step 600: scum determination step 620: length determination step 640: determination step 700: bulk blanking Judgment step 11: blast tube 13: peephole 15: side wall 16: wind path area 220: computer system 310: image taking step 330: convex hull calculation step 410: gradient value calculation step 430: decision step 510: gradient calculation step 530 Step 610: Length calculation step 630: Angle determination step 710: Area calculation step 201239100 720 Decision step 800 Coal injection determination step 812 Background block decision step 816 Decision step 830 Convex calculation step 850 Judgment step 900 Coal injection determination step 950 Judgment step 1000: camera state determination step 1200: decision step B1: background block CA: inscribed circle CHA: convex hull area CHL: convex hull line PA, PB, PC: convex hull vertex 730: decision step 810: history spray gun block providing step 814: area calculation step 820: binarization processing step 840: background block decision step 940: inscribed circle area calculation step 960: decision Step 1100: Judgment step B2: background block CHR: convex hull area D!, D2, D3, D4: distance SCHA: sub-area