TW202314411A - Method and system for processing alarm event in factory - Google Patents

Method and system for processing alarm event in factory Download PDF

Info

Publication number
TW202314411A
TW202314411A TW110134726A TW110134726A TW202314411A TW 202314411 A TW202314411 A TW 202314411A TW 110134726 A TW110134726 A TW 110134726A TW 110134726 A TW110134726 A TW 110134726A TW 202314411 A TW202314411 A TW 202314411A
Authority
TW
Taiwan
Prior art keywords
sensing data
outlier
sensing
solutions
probability
Prior art date
Application number
TW110134726A
Other languages
Chinese (zh)
Other versions
TWI784718B (en
Inventor
魏山雄
徐國容
李若蘭
李玟毅
Original Assignee
和碩聯合科技股份有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 和碩聯合科技股份有限公司 filed Critical 和碩聯合科技股份有限公司
Priority to TW110134726A priority Critical patent/TWI784718B/en
Application granted granted Critical
Publication of TWI784718B publication Critical patent/TWI784718B/en
Publication of TW202314411A publication Critical patent/TW202314411A/en

Links

Images

Landscapes

  • Emergency Alarm Devices (AREA)
  • Alarm Systems (AREA)

Abstract

A method and a system for processing an alarm event in a factory are provided. The system provides a calculation platform and at least one sensing device within a factory. The system performs the method, in which the calculation platform receives sensing data from the at least one sensor device and clusters the sensing data. When outlier data is obtained, the system records multiple solutions with respect to the outlier data and determines that the outlier data is about a normal event or an abnormal event. The related solutions are then directed to a probability allocation mode for obtaining a ranking of probabilities of the solutions corresponding to the outlier data. When a new sensing data is received, the ranking probabilities can be used to determine which solution is performed if the sensing data is determined to be the outlier data of an abnormal event. At last, a speaker is provided for outputting a voice relating to a description of a corresponding scenario and processing steps.

Description

廠區告警事件處理方法與系統Plant area alarm event processing method and system

說明書提出一種廠區內處理異常事件的方法,特別是一種通過數據學習以預測異常事件的廠區告警事件處理方法與系統。The specification proposes a method for handling abnormal events in the factory area, especially a method and system for processing alarm events in the factory area to predict abnormal events through data learning.

根據現行工廠安全管理的措施,大多仰賴人員從監視器監看或到現場巡邏查看,並可配合廠區設置的各種感測裝置與監控設備,如煙霧警報器、溫濕度感測器、機台監控設備等,再通過電子看板監視。當廠區產生異常事件,如火災、煙霧,可通過人員的經驗判斷,並可加上監控設備產生的警報訊號,判斷後執行後續措施。According to the current factory safety management measures, most of them rely on personnel to monitor from the monitor or go to the site to patrol and check, and can cooperate with various sensing devices and monitoring equipment installed in the factory area, such as smoke alarms, temperature and humidity sensors, and machine monitoring. Equipment, etc., and then monitor through the electronic kanban. When an abnormal event occurs in the factory area, such as fire and smoke, it can be judged by the experience of the personnel, and the alarm signal generated by the monitoring equipment can be added, and follow-up measures can be implemented after the judgment.

然而,但當遇到人員疏忽或是認知的不同,又或者非過去所已知的異常時,就無法即時的進行提醒,最主要原因是過去都是將資訊依照已知的條件、規則或門閥值進行彙總,以儀錶板的方式提供管理人員進行監看與查詢,才進行下一步措施。However, when encountering personnel negligence or differences in cognition, or abnormalities that were not known in the past, it is impossible to give immediate reminders. The main reason is that in the past, the information was based on known conditions, rules or gates The values are summarized and provided to managers in the form of a dashboard for monitoring and query before taking the next step.

本揭露所要解決的是先前技術中無法達到即時告警並同時提供有效方案的問題。What the present disclosure aims to solve is the problem that the prior art cannot achieve instant warning and provide an effective solution at the same time.

為了解決上述問題,本揭露提供一種廠區告警事件處理方法,運作於一運算平台,包括:接收由至少一感測裝置產生的多個第一感測數據;集群該些第一感測數據以得到一離群感測數據;記錄針對離群感測數據的多個解決方案,並基於該些解決方案判定離群感測數據為一正常事件或一異常事件;將該些解決方案導入機率分配模式,得到離群感測數據所對應的該些解決方案的一機率排序;接收第二感測數據並判定第二感測數據歸類於異常事件的離群感測數據後,基於機率排序執行該些解決方案的至少其一。In order to solve the above problems, the present disclosure provides a plant alarm event processing method, which operates on a computing platform, including: receiving a plurality of first sensing data generated by at least one sensing device; clustering the first sensing data to obtain An outlier sensing data; recording a plurality of solutions for the outlier sensing data, and judging the outlier sensing data as a normal event or an abnormal event based on the solutions; importing these solutions into a probability distribution mode , to obtain a probability sorting of the solutions corresponding to the outlier sensing data; after receiving the second sensing data and determining that the second sensing data is classified as the outlier sensing data of an abnormal event, the probability sorting is performed based on the probability sorting at least one of these solutions.

為了解決上述問題,本揭露提供另外一種廠區告警事件處理系統,包括一運算平台以及至少一感測裝置,運算平台運行廠區告警事件處理方法,在方法中,運算平台接收由至少一感測裝置產生的多個第一感測數據,分析以擷取感測數據的特徵,經集群所得到的第一感測數據號得到離群感測數據時,記錄針對離群感測數據的多個解決方案,並基於所述解決方案判定離群感測數據為一正常事件或一異常事件,接著將解決方案導入一機率分配模式,得到離群感測數據所對應的解決方案的一機率排序,如此,當接收第二感測數據並判定第二感測數據歸類於異常事件的離群感測數據後,基於機率排序執行些解決方案的至少其一。In order to solve the above problems, the present disclosure provides another plant alarm event processing system, including a computing platform and at least one sensing device, and the computing platform runs a plant alarm event processing method. In the method, the computing platform receives information generated by at least one sensing device. A plurality of first sensing data, analyze to extract the characteristics of the sensing data, and record multiple solutions for the outlier sensing data when the outlier sensing data is obtained by clustering the first sensing data number , and based on the solution, it is determined that the outlier sensing data is a normal event or an abnormal event, and then the solution is introduced into a probability allocation mode to obtain a probability ranking of the solutions corresponding to the outlier sensing data, thus, After receiving the second sensing data and determining that the second sensing data is classified as the outlier sensing data of the abnormal event, performing at least one of the solutions based on probability sorting.

基於上述,藉由本揭露的廠區告警事件處理方法及系統,因針對異常事件的離群感測數據紀錄多個解決方案,故待接收到新的感測數據(即第二感測數據)且新的感測數據被判定為異常事件的離群感測數據後,就可即時提供解決方案的至少其一,以解決因人員疲勞或是經驗認知上的不同而忽略該事件需要立即採取對應對做的失誤。Based on the above, with the factory alarm event processing method and system disclosed in this disclosure, multiple solutions are recorded for the outlier sensing data of abnormal events, so new sensing data (ie, second sensing data) is received and new After the sensing data is judged as the outlier sensing data of an abnormal event, at least one of the solutions can be immediately provided to solve the problem of ignoring the event due to personnel fatigue or differences in experience and cognition and need to take immediate countermeasures mistakes.

為使能更進一步瞭解本發明的特徵及技術內容,請參閱以下有關本發明的詳細說明與圖式,然而所提供的圖式僅用於提供參考與說明,並非用來對本發明加以限制。In order to further understand the features and technical content of the present invention, please refer to the following detailed description and drawings related to the present invention. However, the provided drawings are only for reference and description, and are not intended to limit the present invention.

以下是通過特定的具體實施例來說明本發明的實施方式,本領域技術人員可由本說明書所公開的內容瞭解本發明的優點與效果。本發明可通過其他不同的具體實施例加以施行或應用,本說明書中的各項細節也可基於不同觀點與應用,在不悖離本發明的構思下進行各種修改與變更。另外,本發明的附圖僅為簡單示意說明,並非依實際尺寸的描繪,事先聲明。以下的實施方式將進一步詳細說明本發明的相關技術內容,但所公開的內容並非用以限制本發明的保護範圍。The implementation of the present invention is described below through specific specific examples, and those skilled in the art can understand the advantages and effects of the present invention from the content disclosed in this specification. The present invention can be implemented or applied through other different specific embodiments, and various modifications and changes can be made to the details in this specification based on different viewpoints and applications without departing from the concept of the present invention. In addition, the drawings of the present invention are only for simple illustration, and are not drawn according to the actual size, which is stated in advance. The following embodiments will further describe the relevant technical content of the present invention in detail, but the disclosed content is not intended to limit the protection scope of the present invention.

應當可以理解的是,雖然本文中可能會使用到“第一”、“第二”、“第三”等術語來描述各種元件或者信號,但這些元件或者信號不應受這些術語的限制。這些術語主要是用以區分一元件與另一元件,或者一信號與另一信號。另外,本文中所使用的術語“或”,應視實際情況可能包括相關聯的列出項目中的任一個或者多個的組合。It should be understood that although terms such as "first", "second" and "third" may be used herein to describe various elements or signals, these elements or signals should not be limited by these terms. These terms are mainly used to distinguish one element from another element, or one signal from another signal. In addition, the term "or" used herein may include any one or a combination of more of the associated listed items depending on the actual situation.

本揭露提出一種廠區告警事件處理方法與系統,方法適用於工廠或相關場域中針對異常事件告警的需求。所述廠區告警事件處理系統除各硬體設備外,還引用軟體手段運行一機器學習方法。根據定義好的條件、規則以及門閥值等參數學習廠區內各式感測設備與監控設備產生的數據。利用循環且自動優化的機器學習演算法建立一基於人工智能技術實現的廠區告警模型,使得在系統中運行的廠區告警事件處理方法中,只需將廠區內各式感測裝置(如影像與各種環境監控設備)即時產生的數據接入根據歷史數據建立所述的廠區告警模型,即自動區分出正常現象或有異常的信息,藉此預測異常事件,並產生告警信息。This disclosure proposes a method and system for processing alarm events in a factory area, and the method is suitable for the needs of abnormal event alarms in factories or related fields. In addition to each hardware device, the plant area alarm event processing system also uses software means to run a machine learning method. According to the defined conditions, rules, thresholds and other parameters, learn the data generated by various sensing devices and monitoring devices in the factory area. A cyclic and automatically optimized machine learning algorithm is used to establish a plant area alarm model based on artificial intelligence technology, so that in the plant area alarm event processing method running in the system, it is only necessary to integrate various sensing devices in the plant area (such as images and various (environmental monitoring equipment) real-time data access to establish the factory alarm model based on historical data, that is, to automatically distinguish normal phenomena or abnormal information, thereby predicting abnormal events and generating alarm information.

進一步地,根據廠區告警事件處理方法實施例,透過機器學習方法分類所接收的廠區內各設備產生的數據,即便相關事件最後人為介入標記為正常事件,也可循環加入訓練,作為下次判斷依據。所述方法的目的之一是能夠通過所建立的廠區告警模型在異常事件發生前,使系統的特定裝置自動產生告警信息,以文字、語音、燈號或影像等方式提供相關人員面臨的情境及對應處理步驟。讓所述人員可以立即知道接下來要做出何種動作,解決因人員疲勞或是經驗認知上的不同而忽略該事件需要立即採取對應失誤的措施。特別的是,系統所產生的告警信息為針對即將發生的異常事件組合語音檔案形成一音頻數據112,再通過揚聲器以語音方式(實際實施並不排除可以其他形式告警)播出對應的情境描述語音及情境解決步驟語音,這些通過機器學習得出的後續處理措施可能是過去解決相同異常事件的措施。Further, according to the embodiment of the factory area alarm event processing method, the received data generated by each device in the factory area is classified through the machine learning method. Even if the related event is marked as a normal event by human intervention in the end, it can also be added to the training cycle as the basis for the next judgment . One of the purposes of the method is to enable a specific device of the system to automatically generate alarm information before an abnormal event occurs through the established plant alarm model, and provide the situation and information faced by the relevant personnel in the form of text, voice, light signal or image. Corresponding processing steps. The personnel can immediately know what action is to be taken next, and to solve the problem of ignoring the event due to fatigue or differences in experience and cognition of the personnel needs to take immediate measures corresponding to the error. In particular, the alarm information generated by the system is to form an audio data 112 by combining voice files for the upcoming abnormal event, and then broadcast the corresponding situation description voice in voice mode through the speaker (actual implementation does not exclude other forms of alarm) And situational resolution steps voice, these follow-up actions derived through machine learning may be the measures to solve the same abnormal event in the past.

所述廠區告警事件處理系統可參考圖1顯示的架構實施例示意圖,廠區告警事件處理系統提出一運算平台10,運算平台10以電腦設備配合軟體實現各種功能模組以運作廠區告警事件處理方法。當廠區內感測裝置111感測到事件110而產生感測數據,如即時影像、聲音、信號或環境數據(如溫溼度、煙霧、進出入事件等)時,感測數據被傳送到運算平台10。其中運行的廠區告警事件處理方法可參考圖2顯示的實施例流程圖,以及圖5顯示的方法實施範例。The factory area alarm event processing system can refer to the schematic diagram of the architecture embodiment shown in FIG. 1. The factory area alarm event processing system proposes a computing platform 10. The computing platform 10 uses computer equipment and software to realize various functional modules to operate the factory area alarm event processing method. When the sensing device 111 in the factory senses the event 110 and generates sensing data, such as real-time images, sounds, signals or environmental data (such as temperature and humidity, smoke, entry and exit events, etc.), the sensing data is transmitted to the computing platform 10. For the method for processing alarm events in the factory area, reference may be made to the flow chart of the embodiment shown in FIG. 2 and the implementation example of the method shown in FIG. 5 .

在一實施例中,所述感測裝置111數量依照廠區需求而定,可為一個或是設於不同位置的多個感測裝置111,用於取得廠區內一或多個特定監視區域的感測數據,讓運算平台10可即時取得大量串流載入的感測數據,在此實施例表示為多個第一感測數據(步驟S201)。感測裝置111可為攝影機、音訊接收器或特定感測器等,而這些感測數據可為對應某監視區域中發生的任一事件110的即時影像、聲音、信號或環境數據,可以作為機器學習的樣本。In one embodiment, the number of sensing devices 111 is determined according to the needs of the factory area. It can be one or multiple sensing devices 111 located at different locations to obtain the sensing of one or more specific monitoring areas in the factory area. measurement data, so that the computing platform 10 can obtain a large amount of streaming-loaded sensing data in real time, which is represented as a plurality of first sensing data in this embodiment (step S201 ). The sensing device 111 can be a video camera, an audio receiver, or a specific sensor, etc., and these sensing data can be real-time images, sounds, signals or environmental data corresponding to any event 110 that occurs in a certain monitoring area, and can be used as a machine Learning samples.

接著以集群(clustering)方法對第一感測數據進行分群,藉此得出離群感測數據(步驟S203)。根據實施例,感測數據例如為廠區內攝影機拍攝在對應設置的監視區域中工作人員操作機器產生的即時影像,以音訊接收器擷取聲響,或是特定感測器取得機台或環境的信息。當運算平台10接收到感測數據,運算平台10中軟體程序可運行一非監督式學習(unsupervised learning),藉由分群來得到離群感測數據(outlier)。根據實施例之一,可以利用一種基於密度空間聚類(Density-Based Spatial Clustering of Applications with Noise, DBSCAN)而得離群數據,可參考圖8所示實施範例。Next, the first sensing data is grouped by a clustering method, thereby obtaining outlier sensing data (step S203 ). According to the embodiment, the sensing data is, for example, the real-time images produced by the workers operating the machines in the corresponding monitoring area captured by the cameras in the factory area, the sound is captured by the audio receiver, or the information of the machine or the environment is obtained by specific sensors. . When the computing platform 10 receives the sensing data, the software program in the computing platform 10 can run an unsupervised learning (unsupervised learning) to obtain outlier sensing data (outlier) by clustering. According to one of the embodiments, outlier data obtained by using Density-Based Spatial Clustering of Applications with Noise (DBSCAN) can be used, and reference can be made to the implementation example shown in FIG. 8 .

進一步地,當系統從第一感測數據得到離群資料後,在步驟S205中,廠區告警事件處理系統可以通過管理人員或機器針對一段時間內產生的感測數據所區隔出的離群資料進行標記,對離群數據設定解決方案,其中同一個離群感測數據可能有一或多個解決方案(步驟S205),也會判定此離群資料為正常事件或異常事件(步驟S207)。運算平台10即記錄管理人員或機器針對離群資料設定的解決方案,並儲存至資料庫105。藉此,經標記的異常數據仍回饋到運算平台10中機器學習樣本建立與學習的程序中,以自動循環學習。Further, after the system obtains the outlier data from the first sensing data, in step S205, the plant alarm event processing system can use the management personnel or machines to isolate the outlier data from the sensing data generated within a period of time Marking and setting solutions to the outlier data, wherein the same outlier sensing data may have one or more solutions (step S205 ), and the outlier data is also determined to be a normal event or an abnormal event (step S207 ). The computing platform 10 is to record the solutions set by management personnel or machines for outlier data, and store them in the database 105 . In this way, the marked abnormal data is still fed back to the program of machine learning sample creation and learning in the computing platform 10 for automatic cyclic learning.

之後,將判定過的異常事件的相關離群感測數據與解決方案導入機率分配模式,本實施例即通過運算平台10對各離群感測數據的解決方案進行機率排序(步驟S209)。Afterwards, the outlier sensing data and solutions related to the determined abnormal events are imported into the probability allocation mode, and in this embodiment, the solutions of each outlier sensing data are sorted by probability through the computing platform 10 (step S209 ).

請參考圖3,圖3是離群事件之解決方案之機率矩陣的實施例圖。圖3顯示根據各筆離群感測數據的解決方案的機率值形成的機率排序301。機率分配模式係將各解決方案賦予一權重而得一機率值,進而排序得出的多個機率值而得所述的機率排序。舉例來說,機率矩排序301是依照各事件發生情境設定權重303,再依照機率值由高至低排序得到機率排序301。權重303的設定是依據實際應用而定,例如,當發生異常事件的特定情境下會有多種解決方案,而系統設定權重303時,可針對比較常用的解決方案設定較高權重,較少使用的解決方案即設定較低權重。如此,在根據機率值排序提供解決方案時,權重303將成為決策的重要參數。同時,運算平台10可基於解決方案判定離群感測數據為正常事件305或異常事件307。Please refer to FIG. 3 . FIG. 3 is an embodiment diagram of a probability matrix of a solution to an outlier event. FIG. 3 shows a probability ranking 301 formed according to the probability values of solutions for each outlier sensing data. The probability distribution mode assigns a weight to each solution to obtain a probability value, and then sorts the obtained probability values to obtain the probability ranking. For example, the probability moment sorting 301 is to set the weight 303 according to each event occurrence scenario, and then sort according to the probability value from high to low to obtain the probability sorting 301 . The setting of the weight 303 is based on the actual application. For example, when an abnormal event occurs in a specific situation, there will be multiple solutions, and when the system sets the weight 303, a higher weight can be set for the more commonly used solutions, and less used ones The solution is to set a lower weight. As such, the weight 303 will become an important parameter for decision-making when ordering solutions according to probability values. Meanwhile, the computing platform 10 can determine the outlier sensing data as a normal event 305 or an abnormal event 307 based on the solution.

關於權重303的決定情境的一實施範例,不同管理人員對同一個離群感測數據可能有相同的做法,假設針對同一個離群感測數據, 第一、第二與第三管理人員都是提供相同A解決方案,運算平台就會對A解決方案記錄權重為:1+1+1,另有第四工作人員提供B解決方案,所以這個離群感測數據就會有A解決方案的機率值為3,而B解決方案的機率值為1。據此,得到此離群感測數據之解決方案的機率排序。Regarding an implementation example of the decision context of the weight 303, different managers may have the same approach to the same outlier sensing data, assuming that for the same outlier sensing data, the first, second and third managers are all Provide the same solution A, the computing platform will record the weight of solution A as: 1+1+1, and another fourth staff member provides solution B, so this outlier sensing data will have the probability of solution A value of 3, while solution B has a probability value of 1. Accordingly, the probabilities of solutions for the outlier sensing data are obtained.

在一實施範例中,解決方案可包括對所述離群感測數據處理的影像串流、對應影像串流的情境描述語音及情境解決步驟語音。其中情境描述語音是指事件發生之後接續可能發生什麼情境,而情境解決步驟語音表示發生事情後的處理步驟。另外,可利用揚聲器播放情境描述語音及情境解決步驟語音。In an example embodiment, the solution may include an image stream processed for the outlier sensing data, a situation description voice corresponding to the image stream, and a situation solution step voice. The voice of the situation description refers to what situation may occur next after the event occurs, and the voice of the situation resolution step indicates the processing steps after the event occurs. In addition, the speaker can be used to play the situation description voice and the situation solution step voice.

接著,廠區告警事件處理系統繼續運作,繼續接收由感測裝置111所產生新的感測數據,在此實施例稱第二感測數據(步驟S211),並判定第二感測數據歸類於異常事件的該離群感測數據後,可基於根據上述方法計算得出的機率排序執行多個解決方案的至少其一(步驟S213)。詳細來說,運算平台10接收到新的第二感測數據後,且經集群也被判定為離群感測數據。接下來就要判定第二感測數據被歸類於哪一異常事件的離群感測數據。舉例來說,運算平台10可計算新的感測數據(即第二感測數據)的特徵值與每一離群感測數據的特徵值之間的距離。基於距離小於一閥值而判定新的感測數據歸類於離群感測數據,其中計算距離的方式可採用用於表示多維空間中兩個點之間的歐式距離(Euclidean distance)的計算方法。Next, the factory area alarm event processing system continues to operate, and continues to receive new sensing data generated by the sensing device 111, which is referred to as the second sensing data in this embodiment (step S211), and determines that the second sensing data belongs to After the outlier sensing data of the abnormal event, at least one of multiple solutions may be executed based on the probability ranking calculated according to the above method (step S213 ). In detail, after the computing platform 10 receives the new second sensing data, it is also determined to be outlier sensing data through clustering. The next step is to determine the outlier sensing data of which abnormal event the second sensing data is classified into. For example, the computing platform 10 can calculate the distance between the feature value of the new sensing data (ie, the second sensing data) and the feature value of each outlier sensing data. Based on the distance being less than a threshold, it is determined that the new sensing data is classified as outlier sensing data, wherein the distance calculation method can be used to represent the Euclidean distance between two points in a multi-dimensional space (Euclidean distance) calculation method .

當新的感測數據被歸類後,對應被歸類的離群感測數據的多個解決方案的至少其一可被執行,例如可通過移動設備113的揚聲器115或固定於廠區內各式的非移動設備114的揚聲器116發出解決方案。執行解決方案亦可稱告警信息,即以語音方式播放解決方案。例如通過揚聲器115、116告知廠區內監視區域中的現場人員117已經發生或即將發生的異常事件,還可進一步播出對應的情境描述及對應處理步驟,讓現場人員117可以立即知道接下來要做出何種動作。在此一提的是,上述廠區告警事件處理方法實施例以語音方式形成告警信息,另仍不排除可以文字、影像或燈號等方式通過非移動設備114告知現場人員117已經發生或即將發生的異常事件,亦包括可將各式信息播送至現場人員117所持的移動設備113。When the new sensory data is classified, at least one of a plurality of solutions corresponding to the classified outlier sensory data can be implemented, for example, through the speaker 115 of the mobile device 113 or various devices fixed in the factory area. The speaker 116 of the non-mobile device 114 emits the solution. The execution solution can also be called an alarm message, that is, the solution is played in a voice mode. For example, the speakers 115 and 116 are used to inform the on-site personnel 117 in the monitoring area of the factory that an abnormal event has occurred or is about to occur, and the corresponding situation description and corresponding processing steps can be further broadcasted, so that the on-site personnel 117 can immediately know what to do next what kind of action. What should be mentioned here is that the above-mentioned embodiment of the plant alarm event processing method forms alarm information in the form of voice, and it is still not ruled out that the on-site personnel 117 can be notified through the non-mobile device 114 that the incident has occurred or is about to occur. Abnormal events also include broadcasting various information to the mobile device 113 held by the on-site personnel 117 .

在一實施例中,離群感測數據的相關數據如影像數據、現場聲響、環境數據與/或機台運作數據等,亦能一併輸入機器學習演算法進行學習,同樣地提供解決方案的廠區告警模型。舉例來說,經觸發某事件後,運算平台10將相關感測數據儲存於資料庫中。針對離群事件,待現場人員處理完成或排除觸發事件後,可將副本資料進行儲存。也就是說,當在上述集群步驟(如步驟S203)中所得到的離群感測數據組為多個時,可針對每一離群感測數據組及對應的解決方案建立一查詢索引。在此一提的是,系統可提供相關人員檢視副本資料檢索,並提供編輯介面供修正或增補資訊。In one embodiment, data related to the outlier sensing data, such as image data, on-site sound, environmental data and/or machine operation data, etc., can also be input into the machine learning algorithm for learning, and the solution is also provided. Plant alarm model. For example, after a certain event is triggered, the computing platform 10 stores the relevant sensing data in the database. For outlier events, the copy data can be stored after the on-site personnel complete the processing or eliminate the trigger event. That is to say, when there are multiple outlier sensing data sets obtained in the above clustering step (such as step S203 ), a query index can be established for each outlier sensing data set and the corresponding solution. What is mentioned here is that the system can provide relevant personnel to view the copy data retrieval, and provide an editing interface for correction or supplementary information.

如此,運算平台10即針對每筆正常、異常事件的離群感測數據、所建立的查詢索引以及相關感測數據一併儲存至資料庫,並能在標記後自動循環學習,對於新出現的場景並不需要投入額外的開發資源。In this way, the computing platform 10 stores the outlier sensing data of each normal and abnormal event, the established query index and related sensing data in the database, and can automatically learn in a loop after marking. Scenes do not require additional development resources to be invested.

在上述實施例流程中,系統通過機器學習演算法學習即時取得的感測數據,即第一感測數據與第二感測數據,如圖2步驟S205,廠區告警事件處理系統離群資料的標記並設定解決方案後,建立判斷與預測異常事件的廠區告警模型,同時也不斷地學習即時產生的感測數據以更新廠區告警模型。In the process of the above-mentioned embodiment, the system learns the sensing data obtained immediately through machine learning algorithms, that is, the first sensing data and the second sensing data, as shown in step S205 in Figure 2, the marking of outlier data in the plant alarm event processing system After the solution is set, a plant alarm model for judging and predicting abnormal events is established, and at the same time, the sensor data generated in real time is continuously learned to update the plant alarm model.

根據以上描述的流程實施例,廠區告警事件處理系統通過軟體程序自動學習標記結果,並擷取隨時間戳記記錄的場景特徵,未來若出現相同特徵時候,就可以達到事件發前的預警通知的目的。According to the process embodiment described above, the plant alarm event processing system automatically learns the marking results through software programs, and extracts the scene features recorded with time stamps. If the same features appear in the future, the purpose of early warning notification before the event occurs can be achieved. .

基於以上實施例所描述的方法流程,以下描述以影像數據為實施方式的廠區告警事件處理方法實施例。Based on the method flow described in the above embodiments, the following describes an embodiment of a method for processing an alarm event in a plant using image data as an embodiment.

圖4所示的方法實施例流程運用影像特徵建立廠區內廠區告警事件處理系統中的廠區告警模型,所述感測裝置可為設於廠區各處的監視攝影機。監視攝影機用於拍攝所設的監視區域中工作人員操作機器的即時影像,形成影像數據。在沒有人為介入的情況下,讓機器能夠自動將傳入的串流影像進行場景分類,並隨著時間序可以自動識別新出現的場景,可以使得系統可以根據感測數據中的特徵預測到異常事件即將發生,並提供對應的情境描述以及情境對應的處理步驟。根據圖5顯示廠區告警事件處理方法的實施例流程圖,可配合圖1所示之系統實施例示意圖,以及圖5至圖8所示之實施例圖。The flow of the method embodiment shown in FIG. 4 uses image features to establish a plant alarm model in the plant alarm event processing system in the plant area. The sensing device can be a surveillance camera installed in various places in the plant area. Surveillance cameras are used to capture real-time images of workers operating machines in the set surveillance area to form image data. Without human intervention, the machine can automatically classify the incoming streaming images into scenes, and automatically identify new scenes over time, so that the system can predict abnormalities based on the features in the sensing data The event is about to occur, and the corresponding situation description and the processing steps corresponding to the situation are provided. According to FIG. 5 , the flow chart of an embodiment of the method for processing an alarm event in the factory area is shown, which can cooperate with the schematic diagram of the system embodiment shown in FIG. 1 and the embodiment diagrams shown in FIGS. 5 to 8 .

以影像監控而產生影像數據為例,感測裝置用於拍攝廠區中特定監視區域的影像,形成上述實施例中的第一感測數據與第二感測數據,可以是所述實施例中至少一感測裝置所取得對應設置的監視區域中工作人員操作機器的即時影像、聲響或信號,感測數據以串流方式載入運算平台。特別的是,這些影像、聲響或信號數據可以作為機器學習的樣本,較佳可以取得高畫質檔案。根據實施例,在運行廠區告警事件處理方法之前,通過運算平台中的軟體程序啟始一控制介面,用於設定監視區域,例如可將廠區區分多個區域(步驟S401)。可參考圖5所示系統中提供給管理人員監視廠區的圖形使用者介面示意圖,此圖中顯示一以軟體手段實現的設定監視區域的介面52,管理人員選擇其中廠區布局圖501選項,右方的設定畫面503中顯示將一個廠區分為多個區域的畫面,標示出每個區域的位置、相關資訊以及其中設備,如圖中代表的一監視區域505,每個監視區域可依照其場域特性設有一或多個各種感測裝置。Taking the image data generated by image monitoring as an example, the sensing device is used to take images of a specific monitoring area in the factory area to form the first sensing data and the second sensing data in the above embodiment, which can be at least A sensing device obtains the real-time images, sounds or signals of the workers operating the machines in the corresponding monitoring area, and the sensing data is loaded into the computing platform in a streaming manner. In particular, these image, sound or signal data can be used as samples for machine learning, preferably high-quality files can be obtained. According to an embodiment, before running the factory area alarm event processing method, a control interface is started through a software program in the computing platform for setting monitoring areas, for example, the plant area can be divided into multiple areas (step S401 ). Refer to the schematic diagram of the graphic user interface provided to the management personnel to monitor the factory area in the system shown in Figure 5. This figure shows an interface 52 for setting the monitoring area realized by software means. The management personnel selects the option 501 of the factory layout diagram, and the The setting screen 503 of the display shows a screen that divides a factory area into multiple areas, marking the location of each area, related information and equipment in it, such as a monitoring area 505 represented in the figure, and each monitoring area can be configured according to its field Features are provided with one or more various sensing devices.

根據所選定的監視區域505,設定其中感測裝置(步驟S403)。可參考圖6所示實施例,其中顯示一個提供管理人員設定各個監視區域中感測裝置的表格,此例為設定感測裝置的介面60,當中列出各個感測裝置的資訊,如所屬區域、機台位址、通訊埠等,可供管理人員查詢與修改設定。According to the selected monitoring area 505 , a sensing device therein is set (step S403 ). Reference can be made to the embodiment shown in FIG. 6 , which shows a table for managers to set the sensing devices in each monitoring area. This example is an interface 60 for setting sensing devices, which lists the information of each sensing device, such as the area to which it belongs. , machine address, communication port, etc., for management personnel to query and modify settings.

在此實施例中,感測裝置如攝影機,用於拍攝廠區的特定監視區域。完成上述設定後即開始運作,即時產生大量而高畫質的影像數據,並由系統中的運算平台即時取得大量串流載入的影像數據(步驟S405)。接著是進行數據分析,並取得影像特徵值(步驟S407)。其中可運用軟體方法取得影像中的影像特徵,例如在影像處理方法中,將影像分割為多區,從影像畫素值得出各區最大值、最小值與平均值,用以偵測影像中的邊緣與形狀,可比對資料庫中影像特徵的數據,執行影像匹配;在另一方式中,可以採用一種卷積神經網路(convolutional neural network,CNN)建立影像特徵。In this embodiment, a sensing device such as a video camera is used to photograph a specific surveillance area of the factory. After the above settings are completed, it will start to operate, and generate a large amount of high-quality image data in real time, and the computing platform in the system can obtain a large amount of stream-loaded image data in real time (step S405 ). Next is to perform data analysis and obtain image feature values (step S407 ). Among them, software methods can be used to obtain the image features in the image. For example, in the image processing method, the image is divided into multiple regions, and the maximum value, minimum value and average value of each region are obtained from the pixel value of the image to detect the image in the image. Edge and shape can be compared with image feature data in the database to perform image matching; in another way, a convolutional neural network (CNN) can be used to establish image features.

利用卷積神經網路建立影像特徵的實施例示意圖可參考圖7A至7C。圖7A顯示一影像幀中的每個畫素值,圖7B顯示將影像分區後取得當中最大畫素值,另亦可針對最小或平均畫素值來處理,接著圖7C顯示進行卷積運算(convolutional operation )中池化演算(pooling)後保留重要資訊後的特徵圖,即可得到所選擇的廠區內特定區域的影像特徵。A schematic diagram of an embodiment of constructing an image feature using a convolutional neural network may refer to FIGS. 7A to 7C . Figure 7A shows the value of each pixel in an image frame, Figure 7B shows the maximum pixel value obtained after the image is partitioned, and can also be processed for the minimum or average pixel value, and then Figure 7C shows the convolution operation ( In the convolutional operation), the feature map after the important information is retained after the pooling operation (pooling), can obtain the image features of the specific area in the selected factory area.

根據一實施例,經選定監視區域時,系統利用軟體程序在記憶體或資料庫中建立一個暫存空間,依照時間順序儲存所接收的影像數據,並在運作時可載入至系統記憶體中,影像可以關鍵幀(key frame)為主。在一實施例中,暫存的記憶體空間可以儲存最少幀數(例如32幀)的影像為參考,之後開始擷取其中特徵。在類神經網路的機器學習方法中,特徵擷取使用不同數量的輸入層、卷積層、池化層及不同初始參數推疊組合而成,以此得出影像特徵,再將特徵結果存放於記憶體中。以上步驟為持續根據即時接收的影像數據運行,新輸入影像可替換掉原暫存空間中的最久遠影像。接著,可針對記憶體中暫存的影像特徵演算特徵距離,進行前後次的殘差計算,作為分類的依據(步驟S409),此步驟可以通過機器學習的技術自動擷取影像特徵值,以根據影像之間的特徵差異進行場景分類。According to one embodiment, when the monitoring area is selected, the system uses a software program to create a temporary storage space in the memory or database, stores the received image data in chronological order, and can be loaded into the system memory during operation , the image can be based on key frames. In one embodiment, the temporary memory space can store images with a minimum number of frames (for example, 32 frames) as a reference, and then start to extract features therein. In the neural network-like machine learning method, feature extraction uses different numbers of input layers, convolutional layers, pooling layers and different initial parameters to stack and combine to obtain image features, and then store the feature results in in memory. The above steps are continuously run based on the image data received in real time, and the new input image can replace the oldest image in the original temporary storage space. Next, the feature distance can be calculated based on the image features temporarily stored in the memory, and the previous and subsequent residual calculations can be performed as the basis for classification (step S409). feature differences between images for scene classification.

特別的是,所述感測數據為影像數據時,演算平台可以取得隨著時間取得連續幀影像,並擷取其中各幀的影像特徵,如果所監視的現場沒有太大的變化,則如圖8顯示一種基於密度空間聚類(DBSCAN)的方法,顯示經過分類形成的群組數量會趨於穩定,如圖8中的(a)至(c)圖例,演變到(d)至(f)圖,產生變化,表示監視區域中的密度特徵改變,演算平台中的軟體程序將可自動識別出這個新的群組(離群),即可據此發出信息。在此進行分群演算法如圖形匹配(pattern match)、k-平均演算法(k-means clustering)或特定演算法,其主要概念是應用演算法取得影像特徵,並學習其中群聚現象,並據此執行分類以取得不同的群組(image classification)。In particular, when the sensing data is image data, the calculation platform can obtain continuous frames of images over time, and extract the image features of each frame. If the monitored scene does not change much, the 8 shows a method based on density spatial clustering (DBSCAN), which shows that the number of groups formed by classification will tend to be stable, as shown in the legends (a) to (c) in Figure 8, and evolve to (d) to (f) Changes in the graph indicate changes in density characteristics in the monitoring area, and the software program in the calculation platform will automatically recognize this new group (outlier), and then send a message accordingly. Here, clustering algorithms such as pattern match, k-means clustering, or specific algorithms are performed. The main concept is to apply algorithms to obtain image features, and to learn the clustering phenomenon, and according to This performs classification to obtain different groups (image classification).

當通過一段時間影像特徵變化判斷有離群事件時,系統端的管理人員將對離群事件進行標記(步驟S411)。離群事件可經標記屬於正常群組或異常群組,同樣加入機器學習樣本建立與學習的程序,與前述標記的正常群組或異常群組都可回饋到運算平台中機器學習樣本建立與學習的程序以自動循環學習,最後形成廠區告警模型,能根據異常事件的機率排序使得系統可對相關廠區人員進行示警(步驟S413)。When it is judged that there is an outlier event through the change of image features over a period of time, the management personnel at the system end will mark the outlier event (step S411 ). The outlier events can be marked as belonging to the normal group or the abnormal group, and the program of machine learning sample creation and learning is also added, and the above-mentioned marked normal group or abnormal group can be fed back to the computing platform for machine learning sample creation and learning The program learns in an automatic cycle, and finally forms a plant alarm model, which can be sorted according to the probability of abnormal events so that the system can warn relevant plant personnel (step S413).

於一實施範例中,廠區告警事件處理系統利用攝影機拍攝廠區內產線上的機台與工作人員,攝影機持續拍攝工作人員操作機台的動作,並可同時擷取現場的聲響與機台運作產生的信息。系統中的運算平台可以即時取得各式感測裝置產生的數據,通過以上實施例描述的廠區告警事件處理方法,通過廠區告警模型判斷是否會發生異常事件,並執行相關告警措施,例如,將即將發生的事件情境轉換成音頻數據112,再通過揚聲器自動說出情境描述及對應處理步驟,讓現場的工作人員可以立即知道接下來要做出何種動作。In an implementation example, the factory area alarm event processing system uses cameras to shoot machines and staff on the production line in the factory area. The camera continues to shoot the actions of the staff operating the machines, and can simultaneously capture the sound of the scene and the sound generated by the operation of the machines. information. The computing platform in the system can obtain the data generated by various sensing devices in real time, through the factory area alarm event processing method described in the above embodiments, and through the plant area alarm model to judge whether an abnormal event will occur, and execute relevant alarm measures, for example, the upcoming The event situation that occurs is converted into audio data 112, and then the situation description and the corresponding processing steps are automatically spoken out through the speaker, so that the on-site staff can immediately know what action to take next.

根據所提出方法的目的之一是,在一工廠的場域中,針對異常事件發生之前即發出告警信息,即為情境描述及對應處理步驟。若廠區中工作人員執行某項動作時,系統擷取相關感測數據,即時擷取其中特徵並分析後,判斷出一處理措施,系統在尚未發生異常事件前即以語音方式告知工作人員的後續處理措施。One of the purposes of the proposed method is to send out warning information before abnormal events occur in the field of a factory, that is, the description of the situation and the corresponding processing steps. If the staff in the factory area performs a certain action, the system captures the relevant sensing data, captures the characteristics in real time and analyzes it, and judges a treatment measure. The system will inform the staff of the follow-up by voice before any abnormal event occurs. Handling measures.

綜上所述,根據上述廠區告警事件處理方法與系統的實施例,利用機器學習的方法得出的廠區告警模型來預測並說出即將發生的異常事件,執行對應處理步驟,特別可以語音方式播出告警信息,讓相關人員可以立即知道接下來要做出何種動作,並反饋後續執行產生的數據,實現即時自我循環的機器學習流程,解決過往因傳輸速度慢而需要事前訓練的問題。To sum up, according to the above-mentioned embodiments of the factory area alarm event processing method and system, the plant area alarm model obtained by the machine learning method is used to predict and tell the upcoming abnormal events, and execute the corresponding processing steps. The alarm information is issued, so that relevant personnel can immediately know what action to take next, and feedback the data generated by the subsequent execution, realize the machine learning process of instant self-circulation, and solve the problem of prior training due to the slow transmission speed in the past.

以上所公開的內容僅為本發明的優選可行實施例,並非因此侷限本發明的申請專利範圍,所以凡是運用本發明說明書及圖式內容所做的等效技術變化,均包含於本發明的申請專利範圍內。The content disclosed above is only a preferred feasible embodiment of the present invention, and does not therefore limit the scope of the patent application of the present invention. Therefore, all equivalent technical changes made by using the description and drawings of the present invention are included in the application of the present invention. within the scope of the patent.

110:事件 111:感測裝置 10:運算平台 105:資料庫 112:音頻數據 113:移動設備 114:非移動設備 115、116:揚聲器 117:現場人員 301:機率排序 303:權重 305:正常事件 307:異常事件 52:設定監視區域的介面 501:廠區布局圖 503:設定畫面 505:監視區域 60:設定感測裝置的介面 步驟S201~S213:廠區告警事件處理方法的實施例流程之一 步驟S401~S413:廠區告警事件處理方法的實施例流程之二 110: Event 111: Sensing device 10:Computing platform 105: Database 112: audio data 113:Mobile equipment 114:Non-mobile device 115, 116: speaker 117: On-site personnel 301: Probability sorting 303: weight 305: normal event 307: Abnormal event 52: Interface for setting monitoring area 501: Plant layout 503: Setting screen 505: Surveillance area 60: Setting the interface of the sensing device Steps S201-S213: one of the embodiment flow of the plant area alarm event processing method Steps S401-S413: the second embodiment process of the factory area alarm event processing method

圖1顯示廠區告警事件處理系統的架構實施例示意圖;Fig. 1 shows the schematic diagram of the framework embodiment of factory area alarm event processing system;

圖2顯示廠區告警事件處理方法的實施例之一流程圖;Fig. 2 shows one flow chart of the embodiment of plant area alarm event processing method;

圖3顯示離群感測數據之解決方案之機率矩陣的實施例圖;Figure 3 shows an example diagram of a probability matrix for a solution to outlier sensing data;

圖4顯示廠區告警事件處理方法的實施例之二流程圖;Fig. 4 shows the second flow chart of the embodiment of the plant area alarm event processing method;

圖5顯示設定監視區域的圖形使用者介面實施例示意圖;Figure 5 shows a schematic diagram of an embodiment of a graphical user interface for setting a monitoring area;

圖6顯示設定感測裝置的圖形使用者介面實施例示意圖;6 shows a schematic diagram of an embodiment of a graphical user interface for setting a sensing device;

圖7A至圖7C顯示影像處理過程的實施例圖;以及7A to 7C are diagrams showing an embodiment of the image processing process; and

圖8顯示隨時間變化的特徵值的分群分佈實施例示意圖。FIG. 8 shows a schematic diagram of an embodiment of the group distribution of feature values over time.

110:事件 110: Event

111:感測裝置 111: Sensing device

10:運算平台 10:Computing platform

105:資料庫 105: Database

112:音頻數據 112: audio data

113:移動設備 113:Mobile equipment

114:非移動設備 114:Non-mobile device

115、116:揚聲器 115, 116: speaker

117:現場人員 117: On-site personnel

Claims (16)

一種廠區告警事件處理方法,運作於一運算平台,包括: 接收由至少一感測裝置產生的多個第一感測數據; 集群該些第一感測數據以得到一離群感測數據; 記錄針對該離群感測數據的多個解決方案,並基於該些解決方案判定該離群感測數據為一正常事件或一異常事件; 將該些解決方案導入一機率分配模式,得到該離群感測數據所對應的該些解決方案的一機率排序;以及 接收一第二感測數據並判定該第二感測數據歸類於該異常事件的該離群感測數據後,基於該機率排序執行該些解決方案的至少其一。 A method for processing plant alarm events, operating on a computing platform, comprising: receiving a plurality of first sensing data generated by at least one sensing device; clustering the first sensing data to obtain an outlier sensing data; recording a plurality of solutions for the outlier sensing data, and determining the outlier sensing data as a normal event or an abnormal event based on the solutions; importing the solutions into a probability allocation model to obtain a probability ranking of the solutions corresponding to the outlier sensing data; and After receiving a second sensing data and determining that the second sensing data belongs to the outlier sensing data of the abnormal event, at least one of the solutions is executed based on the probability ranking. 如請求項1所述的廠區告警事件處理方法,其中該運算平台接收設於一廠區內一或多個監視區域中的該至少一感測裝置所產生的該些第一感測數據及該第二感測數據。The factory area alarm event processing method as described in claim 1, wherein the computing platform receives the first sensing data and the second sensing device generated by the at least one sensing device installed in one or more monitoring areas in a factory area 2. Sensing data. 如請求項2所述的廠區告警事件處理方法,其中該些第一感測數據及該第二感測數據分別包括該至少一感測裝置所取得對應設置的監視區域中工作人員操作機器的即時影像、聲響或信號。The factory area alarm event processing method as described in claim 2, wherein the first sensing data and the second sensing data respectively include the real-time information of the staff operating the machine in the corresponding monitoring area obtained by the at least one sensing device image, sound or signal. 如請求項1所述的廠區告警事件處理方法,其中該解決方案包括對該離群感測數據處理的一影像串流及對應該影像串流的一情境描述語音及一情境解決步驟語音。The factory area alarm event processing method as described in claim 1, wherein the solution includes an image stream for processing the outlier sensing data, a situation description voice and a situation solution step voice corresponding to the image stream. 如請求項4所述的廠區告警事件處理方法,其中執行該些解決方案的其中之一是利用一揚聲器播放該情境描述語音及該情境解決步驟語音。The method for processing plant alarm events as described in claim 4, wherein implementing one of the solutions is to use a loudspeaker to play the situation description voice and the situation resolution step voice. 如請求項1所述的廠區告警事件處理方法,其中該機率分配模式係將各該解決方案賦予一權重而得一機率值,進而排序該些機率值而得該機率排序。The method for processing plant alarm events as described in Claim 1, wherein the probability distribution mode is to assign a weight to each solution to obtain a probability value, and then sort the probability values to obtain the probability ranking. 如請求項6所述的廠區告警事件處理方法,其中在該集群步驟中所得到的該離群感測數據組為多個時,針對每一該離群感測數據組及對應的該解決方案建立一查詢索引。The factory area alarm event processing method as described in claim item 6, wherein when there are multiple outlier sensing data groups obtained in the clustering step, for each outlier sensing data group and the corresponding solution Create a query index. 如請求項1所述的廠區告警事件處理方法,更包括計算該第二感測數據的特徵與該離群感測數據的特徵之間的一距離,基於該距離小於一閥值,判定該第二感測數據歸類於該離群感測數據。The factory area alarm event processing method as described in Claim 1 further includes calculating a distance between the feature of the second sensing data and the feature of the outlier sensing data, and judging the first based on the distance being smaller than a threshold Two sensing data are classified as the outlier sensing data. 一種廠區告警事件處理系統,包括: 一運算平台;以及 至少一感測裝置; 其中該運算平台運行一廠區告警事件處理方法,該方法包括: 接收由該至少一感測裝置產生的多個第一感測數據; 集群該些第一感測數據以得到一離群感測數據; 記錄針對該離群感測數據的多個解決方案,並基於該些解決方案判定該離群感測數據為一正常事件或一異常事件; 將該些解決方案導入一機率分配模式,得到該離群感測數據所對應的該些解決方案的一機率排序;以及 接收一第二感測數據並判定該第二感測數據歸類於該異常事件的該離群感測數據後,基於該機率排序執行該些解決方案的至少其一。 A plant alarm event processing system, comprising: a computing platform; and at least one sensing device; Wherein the computing platform runs a plant area alarm event processing method, the method comprising: receiving a plurality of first sensing data generated by the at least one sensing device; clustering the first sensing data to obtain an outlier sensing data; recording a plurality of solutions for the outlier sensing data, and determining the outlier sensing data as a normal event or an abnormal event based on the solutions; importing the solutions into a probability allocation model to obtain a probability ranking of the solutions corresponding to the outlier sensing data; and After receiving a second sensing data and determining that the second sensing data belongs to the outlier sensing data of the abnormal event, at least one of the solutions is executed based on the probability ranking. 如請求項9所述的廠區告警事件處理系統,其中該運算平台接收設於一廠區內一或多個監視區域中的該至少一感測裝置所產生的該些第一感測數據及該第二感測數據。The factory area alarm event processing system as described in claim 9, wherein the computing platform receives the first sensing data and the first sensing data generated by the at least one sensing device installed in one or more monitoring areas in a factory area 2. Sensing data. 如請求項10所述的廠區告警事件處理系統,其中該些第一感測數據及該第二感測數據分別包括該至少一感測裝置所取得對應設置的監視區域中工作人員操作機器的即時影像、聲響或信號。The plant area alarm event processing system as described in claim item 10, wherein the first sensing data and the second sensing data respectively include the real-time information of the staff operating the machine in the corresponding monitoring area obtained by the at least one sensing device. image, sound or signal. 如請求項9所述的廠區告警事件處理系統,其中該解決方案包括對該離群感測數據處理的一影像串流及對應該影像串流的一情境描述語音及一情境解決步驟語音。The factory area alarm event processing system as described in claim 9, wherein the solution includes an image stream for processing the outlier sensing data, a situation description voice and a situation solution step voice corresponding to the image stream. 如請求項12所述的廠區告警事件處理系統,其中執行該些解決方案的其中之一是利用一揚聲器播放該情境描述語音及該情境解決步驟語音。In the factory area alarm event processing system as described in claim 12, one of the solutions is implemented by using a speaker to play the situation description voice and the situation resolution step voice. 如請求項9所述的廠區告警事件處理系統,其中該機率分配模式係將各該解決方案賦予一權重而得一機率值,進而排序該些機率值而得該機率排序。The factory area alarm event processing system as described in Claim 9, wherein the probability allocation mode is to assign a weight to each solution to obtain a probability value, and then sort the probability values to obtain the probability ranking. 如請求項14所述的廠區告警事件處理系統,其中在該集群步驟中所得到的該離群感測數據組為多個時,針對每一該離群感測數據組及對應的該解決方案建立一查詢索引。The factory area alarm event processing system as described in claim 14, wherein when there are multiple outlier sensing data groups obtained in the clustering step, for each outlier sensing data group and the corresponding solution Create a query index. 如請求項9所述的廠區告警事件處理系統,更包括計算該第二感測數據的特徵與該離群感測數據的特徵之間的一距離,基於該距離小於一閥值,判定該第二感測數據歸類於該離群感測數據。The plant alarm event processing system as described in Claim 9 further includes calculating a distance between the feature of the second sensing data and the feature of the outlier sensing data, and judging the first based on the distance being smaller than a threshold Two sensing data are classified as the outlier sensing data.
TW110134726A 2021-09-17 2021-09-17 Method and system for processing alarm event in factory TWI784718B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
TW110134726A TWI784718B (en) 2021-09-17 2021-09-17 Method and system for processing alarm event in factory

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
TW110134726A TWI784718B (en) 2021-09-17 2021-09-17 Method and system for processing alarm event in factory

Publications (2)

Publication Number Publication Date
TWI784718B TWI784718B (en) 2022-11-21
TW202314411A true TW202314411A (en) 2023-04-01

Family

ID=85794617

Family Applications (1)

Application Number Title Priority Date Filing Date
TW110134726A TWI784718B (en) 2021-09-17 2021-09-17 Method and system for processing alarm event in factory

Country Status (1)

Country Link
TW (1) TWI784718B (en)

Family Cites Families (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
TW201717684A (en) * 2015-11-12 2017-05-16 博世科智能股份有限公司 Monitoring system for area
TWI639978B (en) * 2017-07-19 2018-11-01 和碩聯合科技股份有限公司 Video surveillance system and video surveillance method
US10802488B1 (en) * 2017-12-29 2020-10-13 Apex Artificial Intelligence Industries, Inc. Apparatus and method for monitoring and controlling of a neural network using another neural network implemented on one or more solid-state chips
CA3099659A1 (en) * 2018-05-07 2019-11-14 Strong Force Iot Portfolio 2016, Llc Methods and systems for data collection, learning, and streaming of machine signals for analytics and maintenance using the industrial internet of things
EP3764184A4 (en) * 2019-03-26 2021-04-21 Toshiba Mitsubishi-Electric Industrial Systems Corporation Abnormality determination assistance device

Also Published As

Publication number Publication date
TWI784718B (en) 2022-11-21

Similar Documents

Publication Publication Date Title
JP6555547B2 (en) Video monitoring system, video processing apparatus, video processing method, and video processing program
Ravanbakhsh et al. Plug-and-play cnn for crowd motion analysis: An application in abnormal event detection
KR101925907B1 (en) Apparatus and method for studying pattern of moving objects using adversarial deep generative model
US7667596B2 (en) Method and system for scoring surveillance system footage
US9141184B2 (en) Person detection system
CN110428522A (en) A kind of intelligent safety and defence system of wisdom new city
US20180261071A1 (en) Surveillance method and system based on human behavior recognition
US20150262068A1 (en) Event detection apparatus and event detection method
KR102149832B1 (en) Automated Violence Detecting System based on Deep Learning
US11288954B2 (en) Tracking and alerting traffic management system using IoT for smart city
CN110659397A (en) Behavior detection method and device, electronic equipment and storage medium
KR102511287B1 (en) Image-based pose estimation and action detection method and appratus
CN114359976B (en) Intelligent security method and device based on person identification
CN107122743A (en) Security-protecting and monitoring method, device and electronic equipment
KR20210062256A (en) Method, program and system to judge abnormal behavior based on behavior sequence
CN106781167B (en) Method and device for monitoring motion state of object
JP6909657B2 (en) Video recognition system
TWI784718B (en) Method and system for processing alarm event in factory
WO2023279716A1 (en) Device linkage method and apparatus, and device, storage medium, program product and computer program
KR20230103890A (en) Video surveillance system based on multi-modal video captioning and method of the same
CN113012006A (en) Intelligent investigation and research method, system, computer equipment and storage medium
CN111291597A (en) Image-based crowd situation analysis method, device, equipment and system
US11809675B2 (en) User interface navigation method for event-related video
CN116129490A (en) Monitoring device and monitoring method for complex environment behavior recognition
Lau et al. A real time aggressive human behaviour detection system in cage environment across multiple cameras