201219756 六、發明說明： 【發明所屬之技術領域】 本提案係種針對動力設備運作的制與診斷，特別 係指一種動力設備異常檢測裴置及其檢測方法。 【先前技術】 -般來說’動力②備在轉發生前，常常會出現性能的衰 退與耗能的增加，但這些現象並不會立即影響到設備的運轉， 因此時常不會被使用者所發現，此結絲了會增加故障發生機 率與縮短設備壽命外’動力設備效能降低與耗能的增加，對於 產業競爭力與環境保護都有著負面的影響。 動力設備(例如馬達)的診斷模式衫是湘—感測器來感 測動力設備運作的狀態，再透過有線或無線傳輸方式將喊二 到的貪料傳送到後端系統做進一步的分析。但此種模式需要透 過大量的資料傳輸頻寬，主要原因在於為了確保分析診斷的有 政f生及穩疋性，盡可能地將所有感測器所接收到運作狀態資料 完整的發送至後端系統中。 因此，如何能夠透過-種方法或手段，除了保有一樣精確 的分析診斷能力，先將感測的運作狀態資料進行處理，以減少 f要傳送錢端祕的麵量，進喊財效降低資料傳輸頻 寬大小、提昇資料傳輸穩定性、縮短診斷更新時間與降低建置 成本等功效，長久以來一直是相關廠商努力的目標。 【發明内容】 201219756 鑒於以上的問題’本提案提供—鶴力設備異f檢測装置 及其檢測方法。藉由將感測的運倾態資先進行處理，以 達到有效降低:#料傳輸織別、、提昇資料傳輸穩定性、縮短 診斷更新時間與降低建置成本。 根據本提案所揭露之動力設備異常檢測裝置，係包括一感 測·、—處理模組、—最佳化處理模組及—分齡斷模組。201219756 VI. Description of the invention: [Technical field to which the invention pertains] This proposal relates to the system and diagnosis of power equipment operation, and in particular to a power equipment anomaly detection device and a detection method thereof. [Prior Art] - Generally speaking, before the power generation, the performance degradation and energy consumption often occur, but these phenomena do not immediately affect the operation of the equipment, so they are often not found by users. This knot will increase the probability of failure and shorten the life of the equipment. 'The reduction of power equipment efficiency and energy consumption have a negative impact on industrial competitiveness and environmental protection. The diagnostic mode shirt of the power equipment (such as the motor) is a Hunan-sensor to sense the state of operation of the power equipment, and then transmits the greedy material to the back-end system for further analysis through wired or wireless transmission. However, this mode requires a large amount of data transmission bandwidth. The main reason is that in order to ensure the rationality and stability of the analysis and diagnosis, the operational status data received by all the sensors is sent to the back end as completely as possible. In the system. Therefore, how can we pass the method or means, in addition to maintaining the same accurate analytical diagnostic ability, first process the sensed operational status data to reduce the amount of money to be transmitted, and to reduce financial data. The effects of bandwidth size, improved data transmission stability, shortened diagnostic update time, and reduced construction costs have long been the goal of related vendors. SUMMARY OF THE INVENTION 201219756 In view of the above problems, the present proposal provides a Heli equipment iso-f detection device and a detection method thereof. By processing the sensed transport state first, it can effectively reduce: #material transmission weaving, improve data transmission stability, shorten diagnostic update time and reduce construction cost. The power equipment abnormality detecting device disclosed in the present proposal includes a sensing, a processing module, an optimization processing module, and an ageing module.
感測模組係用以感測一動力設備以取得多個運轉訊號。處理模 組係連接__餘，以接收該些運轉訊號，並依序自各該 運轉訊號取得多個特徵值。 取佳化處理模組係連接該處理模組，以接收該些特徵值， 類該些特徵值來建立多_素群組。其中，最佳化處理模 ☆可利用m素分析方法，將特徵值侧聯性分類出多細素群 2各朗鱗轉彳—餘·鱗組讀賤徵值，該些 夂"特徵值之數量係少於軸韻值之數量。 分類診斷模_連接該最佳化處理模_以接收該些因 此預二規=依據—預設規則與該翻素群組發送—狀態訊號。 常項iH—分類啦，分類列表包括-正常項目及一異 為表示運轉二二為動力设備運轉時的正常情況，異常項目則 平衡情況、^對:“:’異常項目可包括但不限於不 輛彎曲情況、^滑情況、共振情況、轴承損壞情況、 諧波倍崎狀贿如。目立斜鱗況、電料平衡情況、 201219756 因此，藉由上述之動力設備異常檢測裝置，檢測裝置可μ 置於一動力設備上，透過最佳化處理模組利用因素分析方、去將 動力設備上所感測之運轉訊號簡化，直接透過分類診斷模組來 進行動力設備運作情況的判斷，無須將所感測的運轉訊號發送 至後端系統來進行直接且即時的處理，以達到縮短診斷更新時 間與降低建置成本。再者，即便未來仍需要後端系、絲進行處 理(例如：透過一遠端伺服器匯整多個動力設備的運作狀態）地 經因素分析方法所歸納出的變異特徵值的數量低於自各該運 轉訊號所取得之繼值的數4，故可_降低⑽傳輪^大 小和提昇資料傳輸穩定性的功效。 根據本提案所揭露之動力賴異常檢測方法，透過债測動 力設備運_資訊來進行異f參數的檢_輯。動力設備異 常檢測方法，首先_—峨叙方法自該動力設備取得多個 運轉訊號，並自運轉訊號中擷取多個特徵值。接著，再將特徵 值依關聯性進行分類以建立多個因素群組，而各該因素群組具 有一變異特徵值。最後再將所取得變異特徵值大於1之因素群 組’利用_經網路和經驗法·到此動力設備的運作狀離， 並依一預設規_嶋力設備之運作狀態是否異常。 ”中w撕㈣路和經驗法則所觸的運作狀態不一致 據_因素群組修正類神經網路之模型，直到兩者判斷 出來的運作狀態結果一致。 運d虎可騎力設備之振動職、溫度減、磁通訊 201219756 號、電流訊號或電壓訊號。處理模組係將所感測之運轉訊號透 過一時域轉換處理或-多尺度網(Multiscale Entropy, mse)運 算以取得特徵值’特徵值可為振動峨之倍頻峰值或特徵頻率 . *。時域轉換處理可制—離散傅立葉轉換處理(D肅eteThe sensing module is configured to sense a power device to obtain a plurality of operating signals. The processing module is connected to the __ to receive the operation signals, and sequentially obtains a plurality of characteristic values from the operation signals. The processing module is connected to the processing module to receive the feature values, and the feature values are used to establish a multi-prime group. Among them, the optimization processing model ☆ can use the m-analytic method to classify the eigenvalues side-by-side into the multi-fine group 2, and the scaly scales are the 夂 quot quot quot quot quot quot quot The number is less than the number of axis values. The classification diagnostic mode _ connects the optimization processing mode _ to receive the pre-two rules = according to the preset rule and the layer-sending group-state signal. The normal item iH-category, the classification list includes - normal items and one different means that the operation is the normal situation when the power equipment is running, and the abnormal items are balanced, ^: ": 'Exception items may include but are not limited to No bending, slipping, resonance, bearing damage, harmonics, such as brittles, etc. slanting scales, balance of electric materials, 201219756 Therefore, the above-mentioned power equipment abnormality detecting device, detecting device The μ can be placed on a power device, and the optimization analysis module can be used to simplify the operation signal sensed on the power device by using the factor analysis party, and directly judge the operation state of the power device through the classification diagnosis module, without The sensed operation signal is sent to the back-end system for direct and immediate processing to shorten the diagnostic update time and reduce the construction cost. Moreover, even if the back-end system and wire are needed for processing in the future (for example, through a long distance) The end server collects the operating states of multiple power devices. The number of eigenvalues summarized by the ground factor analysis method is lower than that of each The number 4 of the success value obtained by the signal can reduce the size of the transmission wheel and improve the stability of data transmission. According to the power detection method disclosed in this proposal, the information is transmitted through the debt measurement power equipment. The detection of the different f-parameters. The power equipment anomaly detection method firstly obtains a plurality of operation signals from the power device, and extracts a plurality of characteristic values from the operation signal. Then, the feature values are correlated. Sexuality is classified to establish a plurality of factor groups, and each of the factor groups has a variability characteristic value. Finally, the group of factors whose obtained eigenvalues are greater than 1 'utilizes _ via the network and the empirical method to this power The operation of the equipment is different, and according to a preset regulation, the operation state of the equipment is abnormal. "The operation state of the tearing (four) road and the rule of thumb is inconsistent according to the _ factor group correction type neural network model, Until the two judged the operational status results are consistent. Vibration, temperature reduction, magnetic communication 201219756, current signal or voltage signal. The processing module transmits the sensed operation signal through a time domain conversion process or a multiscale Entropy (mse) operation to obtain a feature value. The feature value can be a frequency doubling peak or a characteristic frequency of the vibration .. Time domain conversion processing can be made - discrete Fourier transform processing (D ete ete
Founer Transform，DFT)、-快速傅立葉轉換處理（_ F〇urier Transfonn，FFT)、一離散餘弦轉換處理（以似咖c〇sine Tmnsf〇rmati〇n，DCT)、一離散哈特利轉換處理㈣咖出咖 • TranSf〇rm，DHT)、一小波轉換處理(Wavelet Transform, WT)或 一功率頻率處理(Power Spectrum)。 用以觸動力設備之運作狀態是否異常之預設規則可為 -分類列表，分類列表包括—正常項目及—異常項目。正常項 目為動力設備運轉時的正常情況，異常項目則絲示運轉時異 常的情況’例如，異常項目可包括但不限於不平衡情況、不對 4況、顺情況、共振情況、轴承損壞情況、軸胃曲情況、 _ 鬆歸況、她砰衡情況、電料平衡航、舰倍頻情況 及短路情況。 於此，當透過類神經網路取得動力設備之運作狀態後，可 根據上述之預設規則判斷此動力設備可能是哪—部份發生異 吊it況。類神經網路則可採用一倒傳遞類神經網路(6&成 Propagation Network, BPN) ^ -1 f ,^(H〇pfield NeuralFouner Transform, DFT), fast Fourier transform processing (_F〇urier Transfonn, FFT), a discrete cosine transform processing (like coffee c〇sine Tmnsf〇rmati〇n, DCT), a discrete Hartley conversion process (4) Coffee Maker • TranSf〇rm, DHT), Wavelet Transform (WT) or Power Spectrum. The preset rule for whether the operating state of the power-operated device is abnormal may be - a classification list, the classification list includes - normal items and - abnormal items. Normal items are normal conditions when the power equipment is running, and abnormal items indicate abnormal conditions during operation. For example, abnormal items may include, but are not limited to, unbalanced conditions, incorrect conditions, conditions, resonance, bearing damage, and shafts. Gastric conditions, _ pine return, her balance, electric balance, octave and short circuit. Here, after obtaining the operating state of the power device through the neural network, it can be determined according to the above-mentioned preset rules, which part of the power device may be. The neural network can use a reverse-transfer-like neural network (6& into Propagation Network, BPN) ^ -1 f , ^(H〇pfield Neural
NetWOTk，麵）、一徑向基底類神、經網路(Radial Basis Function Network RBFK)、-模糊類神經網路(Fuzzy跑㈤ 201219756 麵)或-_鏈路類神經網路(Functi〇nal㈣Ne— Netw〇rk flnn)。經驗法縣i徵觸 '—轉門檻、—軌跡圖、一 包絡線、一諧波分析或其組合。 因此’透過本提案之動力設備異常檢測方法，可透過感測 動力叹備之運轉喊’並透過因素分析方法簡化自運轉訊號 所，得之特徵值的數量和大小’可_於:細肖耗較小的微處 理盗令直接進订運异處理來判斷動力設備的運作狀態，無須 將所感測_運轉職傳送至後料統，後㈣統僅需要接收 判斷結果，可達到降低#料傳輸頻寬大小、提昇資料傳輸穩定 性、縮短診斷更新顧與降低建置成本的功t 有關本提案的特徵、實作與功效，兹配合圖式作實施例詳 細說明如下。 【實施方式】 請參考『第1圖』所示，『第1圖』係本提案之動力設備 異常檢職㈣示意圖。動力設備異輪職置⑽可為一後 *-^^^.I«£(Personal Digital Assent, 屢卿料處理裝置，動力設備__置_、設置於一 動力設備200並包括一感測模組11〇、一處理模組⑽、一最 佳化處理模組13〇及一分類診斷模組14〇。 感測模組110係用以感測動力設備100以取得多個運轉訊 號’以馬達為例，感測模組110除透過振動程度感測動力設備 100之振動職資料外，亦取得溫度、磁通、紐、轉速、電 201219756 壓等馬達運作時的運轉訊號。 處理換組120根據感測模組11〇所感測到的運轉訊號，利 用-訊號處理方法自運轉訊號中擷取多個特徵值。以振動訊號 絲’處理触m _蝴社_姚财絲運轉訊號 自诚轉軸賴’並魏鮮⑽魏減絲頻，再依 序自頻域之振動訊賴取出Q 5倍頻、1倍齡12倍頻，所掏 取之倍頻訊断為對應振動訊號之特徵值。NetWOTk, a radial base, a Radial Basis Function Network RBFK, a fuzzy neural network (Fuzzy Run (5) 201219756) or a - link neural network (Functi〇nal (4) Ne- Netw〇rk flnn). Experience law county i touch '- turn-turn threshold, - trajectory map, an envelope, a harmonic analysis or a combination thereof. Therefore, 'through the power equipment anomaly detection method of this proposal, you can simplify the self-running signal through the method of sensing the power sigh's operation, and the number and size of the characteristic values obtained by the factor analysis can be: The smaller micro-processing piracy order directly determines the operation status of the power equipment, and does not need to transmit the sensed_operation job to the back-end system. After the (4) system only needs to receive the judgment result, the reduction of the material transmission frequency can be achieved. The advantages of wide size, improved data transmission stability, shortened diagnostic update and reduced construction cost. The characteristics, implementation and efficacy of this proposal are described in detail below with reference to the drawings. [Embodiment] Please refer to the "Fig. 1", and "Fig. 1" is a schematic diagram of the power equipment abnormality inspection (4) of this proposal. The power equipment different wheel position (10) can be a post-*-^^^.I«£ (Personal Digital Assent, a power processing device, a power device __set_, set in a power device 200 and includes a sensing module The group 11〇, a processing module (10), an optimization processing module 13〇, and a classification diagnostic module 14〇. The sensing module 110 is configured to sense the power device 100 to obtain a plurality of operating signals 'to the motor For example, the sensing module 110 not only transmits the vibration information of the power device 100 through the degree of vibration, but also obtains the operation signals of the motor, the magnetic flux, the magnetic flux, the new speed, the rotational speed, and the electric power such as the 201219756 pressure. The sensing signal sensed by the sensing module 11 is used to extract a plurality of characteristic values from the running signal by using the signal processing method. The vibration signal wire is used to process the touch m _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ Wei Xian (10) Wei reduced the frequency of the wire, and then took the Q 5 frequency multiplier and the 1 time old 12 times frequency according to the vibration signal in the frequency domain. The multiplied signal is the characteristic value of the corresponding vibration signal.
/敢佳化處理模組130係連接該處理模組12〇，以接收該些 特徵值JIE分類該些特徵值來建立多個因素群組。其中，最佳 处极、、旦130可利用.因素分析方法，將特徵值依關繼分類 $多個因素群組。各該因素群組具有—代表_素群組之變異 特徵值’該些變異特徵值之數量係少於該些特徵值之數量。。 』分類診斷模組140係連接該最佳化處理模組13〇用以 。亥一口素群組’亚依據一預設規則與該些因素群組發送一狀能 訊號，以判斷動力設備的運作情況。此預設規則可為一分划 包括一正常項目及一異常項目。正常項目為:力 常情況，異常項目縣麵·時異常的情 、/'、項目可包括但不限於不平衡情況、不對心 潤利況、共振情況、轴承損壞情況、轴彎曲情況I 相位：衡情況、電位不平衡情況、讀波倍頻情況及短 備里2考『弟2A圖』所示，『第2A圖』係本提案之動於 裝置-實施例的示意圖。動力設備異常檢卿置更又 201219756 包括-j示裝置15G，該警示裝置i5G係用以接收該狀態訊 號’並當該狀態訊縣該異科，料通知使用者。警示裂置 I/O可為但不限於一振動模組、—發光模組、一顯示模組、一 聲響柄組或其組合，以透過振動警示、燈光警示'訊息警示或 聲音警不等方絲通知使用者動力設備的運作發生異常。 清參考『第2B圖』所示，『第2B圖』係本提案之動力設 備異吊檢概置另—實施例的示意圖。動力設備異常檢測裳置 更包括傳輸模組160，傳輸模組160係連接該分類診斷模組 140，用以接收該狀態訊號，並透過有線或無線的傳輸方式將 該狀態訊號發送至警示裝置150。 請參考『第2C圖』所示，『第2c圖』係本提案之動力設 備異系檢測裝置又一實施例的示意圖。動力設備異常檢測裝置 更包括一記憶模組17〇,記憶模組no用以儲存該動力設備2〇〇 之該些運轉訊號’當使用者有需要讀取動力設備之運轉訊號來 進行進一步的分析，再透過存取記憶模組17〇來取得所需的運 轉訊號。δ己憶模組170可供設置記憶卡以儲存感測之運轉訊 號’ s己憶卡可為一小型快閃(Compact Flash, CF)記憶卡、一微 型硬碟(Micro Drive, MD)記憶卡、一安全數位(Secure Digital， SD)記憶卡、一微型安全數位(Micro SD)記憶卡、一多媒體(Multi Media Card, MMC)記憶卡、一長條(Memory Stick，MS)記憶卡 或一微型長條(Micro MS)記憶卡。 因此’藉由上述之動力設備異常檢測裝置，檢測裝置可設 201219756 置於一動力設備上，透過最佳化處理模組利用因素分析方法將 自動力β又備上所感測之運轉訊號簡化，直接透過分類診斷模組 來進行動力設備運作情況的判斷，無須將所感測的運轉訊號發 ' 达至後端系統來進行直接且即時的處理，以達到縮短診斷更新 ' ㈣與降低建置成本。再者’即便未來仍需要後端系統來進行 處理(例如：透過一遠端伺服器匯整多個動力設備的運作狀 態)，經因素分析方法所歸納出的變異特徵值的數量低於自各該 • 運轉訊麵取得之雜_數量，故可達晴赌料傳輸頻寬 大小和提昇資料傳輸穩定性的功效。 里火請參考『第3圖』所示，『第3圖』係本提案之動力設備 異常檢測方法的步驟流程圖。動力設備異常檢測方法係應用於 一動力設備，包括： 步驟S300:利用一訊號處理方法自該動力設備取得多個運 轉訊號； 擊步驟S310:自該些運轉訊號取得對應於各該運轉訊贫之 個特徵值； & 么^驟S320 .將該些特徵值進行分組，以建立多個因素群 組，各該因素群組具有一變異特徵值； 動力設備運轉之一 步驟S340 :相 裝置運轉之—第二 ^驟S33G .根據該翻素群組，_ —類神經網路判斷該 ^―第一運作狀態； :根據該些特徵值，利用一經驗法則判斷該動力 $二運作狀態；The daemonization processing module 130 is connected to the processing module 12 to receive the feature values JIE to classify the feature values to establish a plurality of factor groups. Among them, the best position, the Dan 130 can use the factor analysis method to classify the eigenvalues into more than one factor group. Each of the factor groups has a representative value of the representative _ prime group. The number of the eigenvalues is less than the number of the eigenvalues. . The classification diagnostic module 140 is connected to the optimization processing module 13 for use. The Haiyi Group consists of a preset rule and a group of factors to send a signal to determine the operation of the power equipment. This preset rule can be a division including a normal item and an abnormal item. The normal project is: the situation of the normal situation, the abnormal situation of the county, the time of the abnormal project, /', the project may include but is not limited to the unbalanced situation, the unbalanced condition, the resonance condition, the bearing damage condition, the shaft bending condition I phase: The balance case, the potential imbalance, the read wave multiplier, and the short test are shown in the "2A diagram", and the "2A diagram" is a schematic diagram of the device-implementation of the present proposal. The power equipment abnormality checker is further set to 201219756 to include the -j indicating device 15G, the warning device i5G is used to receive the status signal 'and the state informs the user of the state, and informs the user. The warning split I/O can be, but is not limited to, a vibration module, a light module, a display module, a sounding handle group or a combination thereof, through a vibration warning, a light warning, a message warning or an audible alarm. The square wire informs the user that the operation of the power equipment is abnormal. Refer to "2B" for the sake of clarity, and "2B" is a schematic diagram of the power device of this proposal. The power device abnormality detecting device further includes a transmission module 160. The transmission module 160 is connected to the classification diagnostic module 140 for receiving the status signal, and transmitting the status signal to the warning device 150 through a wired or wireless transmission manner. . Please refer to the "2Cth diagram", and the "2c diagram" is a schematic diagram of still another embodiment of the power equipment different detection apparatus of the present proposal. The power device abnormality detecting device further includes a memory module 17 for storing the operation signals of the power device 2 when the user needs to read the operation signal of the power device for further analysis. And accessing the memory module 17 to obtain the required operation signal. The δ recall module 170 can be used to set a memory card to store the sensed operation signal. The memory card can be a Compact Flash (CF) memory card or a Micro Drive (MD) memory card. , a Secure Digital (SD) memory card, a Micro SD memory card, a Multi Media Card (MMC) memory card, a Memory Stick (MS) memory card or a mini Long (Micro MS) memory card. Therefore, by the above-mentioned power equipment abnormality detecting device, the detecting device can be set to be placed on a power device in 201219756, and the optimization signal is used to simplify the operation signal of the automatic force β by using the factor analysis method. Through the classification diagnostic module to judge the operation of the power equipment, it is not necessary to send the sensed operation signal to the back-end system for direct and immediate processing, so as to shorten the diagnostic update' (4) and reduce the construction cost. Furthermore, 'even if the back-end system is still needed for processing in the future (for example, the operation state of multiple power devices is collected through a remote server), the number of eigenvalues summarized by the factor analysis method is lower than that. • The number of _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ Please refer to "Figure 3" for the fire, and "3" is the flow chart of the steps for the power equipment abnormality detection method of this proposal. The power device abnormality detecting method is applied to a power device, and includes: Step S300: acquiring a plurality of operation signals from the power device by using a signal processing method; and performing step S310: obtaining, from the operation signals, corresponding to each of the operation signals And the feature values are grouped to establish a plurality of factor groups, each of the factor groups having a variation characteristic value; one of the power device operations step S340: the phase device is operated - a second step S33G. According to the group of vertices, the _-type neural network determines the first operational state; based on the eigenvalues, using a rule of thumb to determine the power $2 operating state;
S 11 201219756 步驟S350.比較該第一運作狀態與該第二運作狀態是否相 同； 步驟S36G:當該第—運作狀態與該第二運作狀態不相同 時，根據該些因素群組修正該類神經網路，直到該第一運作狀 態與該第二運作狀態相同； 步驟S37〇:當該第一運作狀態與該第二運作狀態相同時， 根據一預設規則判斷該第一運作狀態是否異常； 步驟S380 :若靖該第一運作狀態為異f，則發送一異常 訊號；以及 步驟S390 .若判斷言亥第一運作狀態為正常，則發送一正常 訊號。 請參考『第4八圖』所示，『第4八圖』係第3圖中步驟咖 之a靶例μ私圖。步驟S31〇所述之自該些運轉訊號取得對 應於各該運轉峨之多個值，運觀號包括振動訊號、溫 度訊號、磁通訊號、電流訊號或電壓訊號。步驟S310包括： Y驟S311 動力設備’以取得該些運轉訊號； ，驟 利用日t域轉換處理，將該運轉訊號之一時域 資料轉換為一頻域資料；以及 步驟SM3 :自物域資料娜多個特徵值。 其中’時域轉換處理可為一離散傅立葉轉換處理、一快速 傅立葉轉換處理、-離散餘_換歧、—離散哈_轉換處 理、-小波轉換處理或—辨頻率處理。 12 201219756 以馬達的振動訊號來說，當振動訊號經過快速傅立葉轉換 處理’基頻(諧波)可透過以下公式計算： 第-基頻位置=((1*轉速*運轉訊號之#料長度)/6〇)頻譜 - 擷取頻率； - 苐二基頻位置=((2*轉速*運轉訊號之資料長度)/6〇)頻譜 擷取頻率；以此類推。 舉例··假設感測一每分鐘1800轉的馬達，來取得一 16千 • 位元組的運轉訊號，取頻率為12千赫茲的頻域，則第一基頻 的位置可為40。 據此，當步驟S311取得一代表馬達振動訊號之運轉訊號 時，透過步驟S312將運轉訊號自時域資料轉換為頻域資料， 再依序自頻域之運轉訊號擷取出〇.5倍頻、丨倍頻至12倍頻位 置的值，所擷取之倍頻訊號即為對應振動訊號之特徵值，這些 特徵值的數量為24個，並根據倍頻的大小分別定義為〇 5χ、 φ 1χ、15χ、2χ、2 5χ、3χ、3.5χ、4χ、4,5χ、5χ、5.5χ、6χ、6.5χ、 7χ、7.5χ、8χ、8.5χ、9χ、9.5χ、ΙΟχ、1〇.5χ、11χ、11.5χ 及 12χ。 請參考『第4Β圖』所示，『第4Β圖』係第3圖中步驟S310 之另一實施例流程圖。相較第4Α圖所示之步驟S3i〇的實施 例，第4Β圖所示之步驟S31〇實施例係透過多尺度熵(Multiscale Entropy, MSE)運异來取得運轉訊號之特徵值，此實施例包括： 步驟S314 :感測該動力設備，以取得該些運轉訊號；以及 步驟S315:將去除雜訊後之該些運轉訊號透過一多尺度熵 13 201219756 (Multiscale Entropy，MSE)運算’以取得對應該運轉訊號之該些 特徵值。 步驟S314和S315之間更可包括一步驟S316,步驟S316 : 利用小波轉換對該些運轉訊號進行雜訊處理。主要原因在於經 過感測所取得之運轉訊號可能具有雜訊，透過小波轉換處理可 以達到抑制雜訊的功效。 » * · Τ·ν«χ |I0j』 3圖中步驟s32〇之步驟流程B。待取得該麵應於運轉訊號之 多個_值(步驟綱），動力設備異常檢測方料透過时分 析方法簡化這些繼值，如步驟S32G職之將触特徵值進 ^驟^::晴餐細滩科-變異特徵 該些因步=1姻蝴刪_錄群，以建立 -變異特齡⑽ _值，以] 步驟S323 :保留該些變異特徵值大於 藉由上述之步驟S321至步驟助=牛徵值 201219756 因素群組名稱 變數 變異特徵值 解釋變異(％) 因素一 1.5 X 4x 4.5 X 5x 5.5 x 6 x 6.5 x 7x 8.5 x 10X 7.431 30.962 因素二 3 x 1 x 2.5 x 9x 8x 2 x 7.5 x 3.5 x 3.257 13.573 因素三 lOx 10.5x 1.258 5.244 15 201219756S11 201219756 Step S350. Compare whether the first operational state and the second operational state are the same; Step S36G: when the first operational state is different from the second operational state, correct the neural group according to the group of factors a network, until the first operational state is the same as the second operational state; Step S37: when the first operational state is the same as the second operational state, determining whether the first operational state is abnormal according to a preset rule; Step S380: If the first operational state is different from f, an abnormal signal is sent; and step S390. If it is determined that the first operational state is normal, a normal signal is sent. Please refer to "4th 8th", "4th 8th" is the target private image of the step a in the third figure. In step S31, a plurality of values corresponding to each of the operation signals are obtained from the operation signals, and the observation number includes a vibration signal, a temperature signal, a magnetic communication number, a current signal or a voltage signal. Step S310 includes: Y step S311, the power device 'to obtain the operation signals; and using the day t domain conversion process to convert the time domain data of one of the operation signals into a frequency domain data; and step SM3: from the object domain data Multiple feature values. The 'time domain conversion process may be a discrete Fourier transform process, a fast Fourier transform process, a discrete residual_division, a discrete Ha_transformation process, a wavelet transform process, or a discrimination frequency process. 12 201219756 In the case of motor vibration signals, when the vibration signal is subjected to fast Fourier transform processing, the fundamental frequency (harmonic) can be calculated by the following formula: - fundamental frequency position = ((1 * speed * running signal # material length) /6〇) Spectrum - Capture frequency; - 苐2 fundamental frequency position = ((2 * speed * data length of the operation signal) / 6 〇) spectrum acquisition frequency; and so on. For example, if a motor with a frequency of 1800 rpm is sensed to obtain a running signal of 16 kilobytes, taking the frequency domain of 12 kHz, the first fundamental frequency can be 40. According to this, when step S311 obtains an operation signal representing the motor vibration signal, the operation signal is converted from the time domain data to the frequency domain data through step S312, and then the frequency signal of the frequency domain is sequentially extracted and extracted.丨 multiplier to the value of the 12-octave position, the multiplied signal is the characteristic value of the corresponding vibration signal. The number of these characteristic values is 24, and is defined as 〇5χ, φ 1χ according to the size of the multiplication frequency. , 15χ, 2χ, 2 5χ, 3χ, 3.5χ, 4χ, 4,5χ, 5χ, 5.5χ, 6χ, 6.5χ, 7χ, 7.5χ, 8χ, 8.5χ, 9χ, 9.5χ, ΙΟχ, 1〇.5χ , 11χ, 11.5χ and 12χ. Please refer to FIG. 4, and FIG. 4 is a flow chart of another embodiment of step S310 in FIG. Compared with the embodiment of step S3i of FIG. 4, the step S31 shown in FIG. 4 is obtained by multi-scale entropy (MSE) to obtain the characteristic value of the operation signal. The method includes: step S314: sensing the power device to obtain the operation signals; and step S315: transmitting the operation signals after removing the noise by a multi-scale entropy 13 201219756 (Multiscale Entropy, MSE) operation to obtain a pair These characteristic values of the signal should be run. Step S316 and S315 may further include a step S316. Step S316: performing noise processing on the operation signals by using wavelet transform. The main reason is that the operation signal obtained through sensing may have noise, and the effect of suppressing noise can be achieved by wavelet conversion processing. » * · Τ·ν«χ |I0j』 Step B of step s32 in the figure. To obtain the _ value (step outline) of the operation signal, the analysis method of the power equipment abnormality detection simplifies these relay values, for example, the step S32G will touch the characteristic value into the ^^:: The Shoal Division-variation feature is determined by the step = 1 marriage deletion _ record group to establish - variability age (10) _ value, to] step S323: retaining the variogram characteristic values greater than the above step S321 to step assist = cattle levy 201219756 factor group name variable variation eigenvalue interpretation variation (%) factor one 1.5 X 4x 4.5 X 5x 5.5 x 6 x 6.5 x 7x 8.5 x 10X 7.431 30.962 factor two 3 x 1 x 2.5 x 9x 8x 2 x 7.5 x 3.5 x 3.257 13.573 factor three lOx 10.5x 1.258 5.244 15 201219756
8χ、8.5x、9x、9·5χ、1〇χ、10 ^8χ, 8.5x, 9x, 9·5χ, 1〇χ, 10^
ux川义、11又、115叉及12\)算出樣 本共變數矩陣S，共變數矩陣表示如下： s = var(〇.5X) cov(lZ,〇.5Z) c〇v(〇.5Z,lx) ... c〇v(〇.5X,12X)' var(lZ) : 〇v(12^〇.5x) c〇v(12Z,1X) ... var(12x) 其中var表示變異數，c〇v表示共變異數。 接著，再從共變數矩陣中算出24個變異特徵值A，入， 分別為方程式 det(5-A/) = var(0.5 斗;tc〇v(l^,〇.5x) cov(〇.5X,lX) var(l^)-/l c〇v(l2X,〇.5x) c〇v(12X,1^) cov(〇.5X,12X) ：=0 vai{l2X)~ λUx Chuanyi, 11 again, 115 fork and 12\) calculate the sample covariate matrix S, the common variable matrix is expressed as follows: s = var(〇.5X) cov(lZ,〇.5Z) c〇v(〇.5Z, Lx) ... c〇v(〇.5X,12X)' var(lZ) : 〇v(12^〇.5x) c〇v(12Z,1X) ... var(12x) where var represents the variance , c〇v represents the total number of variations. Then, calculate 24 eigenvalues A from the common variable matrix, and enter, respectively, the equation det(5-A/) = var(0.5 bucket; tc〇v(l^,〇.5x) cov(〇.5X ,lX) var(l^)-/lc〇v(l2X,〇.5x) c〇v(12X,1^) cov(〇.5X,12X) :=0 vai{l2X)~ λ
的解。所以即可解出 A - 7·4：>1，12 = 3.257，= 1.258，& =丨邊，；l5 = 1.124Α24 = 0.029，再依 照計算結果’將變異特徵值大於丨作為選取因素群組個數之原 則，以表一為例，共計挑出五個因素群組作為類神經網路之輸 入變數。 16 201219756 而透過因素分析來對特徵值分群(步驟兮切），由於每一個 么頻特徵值在五_麵组之下都有其貞荷值，透過選擇在其 某一因素縣下最大之貞荷值，絲示雜值所屬之因素群 • 、组。以1X的倍頻特徵值為例，在因素-之負荷為_0.260，在因 . 素二之負荷為0.899，在因素三之負荷為_〇 〇38，在因素四之負 荷為0.092，在因素五之負荷為_〇〇15，其中又以以在因素二 下之負荷值是五個因素群組當中最大的，亦即1χ隸屬於因素 • 二的因素群組中。因此，透過相同的方式，可把其餘的23個 特徵值依照最大負荷值分別歸類在五個因素群組當中，如表一 所列’故因素.一下包含了 L5X、4Χ、4 5χ、5χ、5 5χ、6χ、 6.5Χ、7Χ、8·5Χ、12Χ ;因素二包括 ΐχ、2Χ、2.5χ、3χ、3 5χ、 7.5Χ、8Χ、9Χ ;因素三包含ι〇χ、10.5Χ ;因素四包括11χ、 11.5Χ;因素五包含 〇.5Χ、9.5Χ。 如步驟S330所述，分別將利用五個因素組合帶入類神經 • 網路來取得第一運作狀態。關於類神經網路之建立此為本領域 之技藝人士知悉，於此不加以累述。類神經網路則可採用—倒 傳遞類神經網路(Back Propagation Network，ΒΡΝ)、一霍普菲爾 網路(Hopfield Neural Network, HNN)、一徑向基底類神經網路 (Radial Basis Function Network，RBFN)、一模糊類神經網路 (Fuzzy Neural Network，FNN)或一函數鏈路類神經網路 (Functional Link Neural Network，FLNN)。經驗法則為 _ 特徵頻 譜、一臨界門檻、一軌跡圖、一包絡線、一諧波分析或其組合。Solution. So you can solve for A - 7·4:>1,12 = 3.257,= 1.258,&=丨,;l5 = 1.124Α24 = 0.029, and then according to the calculation result 'the eigenvalue is greater than 丨 as the selection factor The principle of the number of groups, taking Table 1 as an example, a total of five factor groups are selected as input variables of the neural network. 16 201219756 And by factor analysis to group eigenvalues (steps are cut), since each eigenvalue has its 贞 value under the _ 面 group, by selecting the largest 县 in one of its factors The value of the load, the number of factors that the miscellaneous value belongs to, and the group. Taking the 1X multiplier characteristic value as an example, the factor-load is _0.260, the load of factor 2 is 0.899, the load of factor 3 is _〇〇38, and the load of factor 4 is 0.092. The load of five is _〇〇15, and the load value under factor two is the largest among the five factor groups, that is, the group of factors belonging to factor two. Therefore, in the same way, the remaining 23 eigenvalues can be classified into five factor groups according to the maximum load value, as shown in Table 1. The factors include L5X, 4Χ, 4 5χ, 5χ. 5 5χ, 6χ, 6.5Χ, 7Χ, 8·5Χ, 12Χ; factors 2 include ΐχ, 2Χ, 2.5χ, 3χ, 3 5χ, 7.5Χ, 8Χ, 9Χ; factor 3 includes ι〇χ, 10.5Χ; factors Four includes 11χ, 11.5Χ; factor five includes 〇5Χ, 9.5Χ. As described in step S330, a combination of five factors is taken into the neural network to obtain the first operational state. The establishment of a neural network is known to those skilled in the art and will not be described herein. The neural network can use the Back Propagation Network (ΒΡΝ), a Hopfield Neural Network (HNN), and a Radial Basis Function Network. , RBFN), a fuzzy neural network (FNN) or a functional Link Neural Network (FLNN). The rule of thumb is _ characteristic spectrum, a critical threshold, a trajectory map, an envelope, a harmonic analysis, or a combination thereof.
S 17 201219756 同理，步驟S340所述，將特徵值利用經驗法則判斷動力 裝置之第二運作狀態。經驗法則乃是依據基礎理論所推導之規 則，以振動訊號為例，經驗法則為機械振動基礎所推導之振動 特性規則，最常見較_射各難職值(脚微頻譜) 搭配臨界π触組合加輯算，亦可使職關、包絡線等方 法進-步將變異賴錢行解_成分分析，常見的諧波分析 亦可協助取得邊頻資料。 於此，步驟S340所採用之經驗法則係利畴徵值與門檀 設定的方法，當特徵值超過_設定_可判斷動力設備的^ 作發生異常。 舉例來說，假設IX之倍頻特徵值與2χ之倍頻特徵值的 峰值以5每秒絲(millimeteiy_nd, mm/s)為門檻設定值，双 之特徵值的♦酬以2每秒絲為門麟紐，如『第$圖』 所不’『―第5圖』為第3圖中步驟⑽利用經驗法則判斷動: 裝置之第—運作狀(％之—實施例的邏輯流程圖。若π之伴頻 特徵值小於5則判斷動力設備正常，反之，在ιχ之倍觸徵 值大於5，且2Χ和3Χ之倍頻特徵值同時小於5和2，則判斷 此動力設備切作㈣發生斜_航。 、 假叹透過類神經網路所取得之第—運作狀態和經驗法 所取得之第二運作祕_時(倾s3价預設朗可為—八 類列表，分_!表包括—正常項目及-異常項目。正常項 動力設傷運轉時的正常情況，異常項目則為表示運轉時異常的 18 201219756 情死，例如，異常項目可包括但不限於不平衡情況、不對心情 =、潤滑情況、共振情況、軸承_情況、軸f曲情況、鬆動 =、她斜衡航、電位斜衡㈣、触麵情況及短 路情況。 步驟S380則根據此預設規則判斷第一運作狀態是否為里 常項目中所紀錄之運作情況，藉以判斷動力設備是否發生異常 的狀態。反之’料-運作狀態為正常項目中所紀錄之運作情 况，則判斷此動力設備之運作狀態正常。 一因此，透過本提案之動力設備異常檢測方法，可透過感測 一動力.設備之運轉訊號’並透素分析方法簡化自運轉訊號 所取得之特徵值隨量和大小，可咖於資源消耗較小的微處 =器中，直接進行運算處理來判斷動力設備的運作狀態，無須 將所感測_運轉訊麟送地_，後端_需要接收 降刪傳崎大小、提昇細輪穩定 性、縮短讀騎日_與降健置絲的功效。 雖然賴案之魏例滅如场述，非肋限定本提 案’任何熟習相關技藝者，在不脫離提案之精神和範圍内，舉 凡依申請範騎述之形狀、構造、特徵及精神當可齡許之變 =::專利保護_視本說明書所― 【圖式簡單說明] 第1圖係本提案之動力設備異常檢·置的示意圖。 £ 19 201219756 第2A圖係本提案之動力設備異常檢測裝置一實施例的示意 圖。 第2B圖係本提案之動力設備異常檢測裝置另一實施例的示意 圖。 第2C圖係本提案之動力設備異常檢測裝置又一實施例的示意 圖。 第3圖係本提案之動力設備異常檢測方法的步驟流程圖。 第4A圖係第3圖中步驟S310之一實施例流程圖。 第4B圖係第3圖中步驟S310之另-實施例流程圖。 第4C圖係第3圖中步驟S320之步驟流程圖。 第5圖係第3 U中步驟S34〇之—實施例的流程圖。 【主要元件符號說明】 動力設備異常檢測裝置 感測模組 處理模組 最佳化處理模組 分類診斷模組 警示裝置 傳輸模組 記憶模組 動力設備 100 110 120 130 140 150 160 170 20 200S 17 201219756 Similarly, as described in step S340, the feature value is used to determine the second operational state of the power unit using empirical rules. The rule of thumb is based on the rules derived from the basic theory. The vibration signal is taken as an example. The rule of thumb is the vibration characteristic rule derived from the basis of mechanical vibration. The most common is the y-shot cataract value (foot micro-spectrum) with the critical π-touch combination. Adding calculations, it is also possible to make the duties and envelopes and other methods to further analyze the _ component analysis, and the common harmonic analysis can also assist in obtaining the sideband data. Here, the rule of thumb adopted in step S340 is a method of setting the domain value and the gate setting. When the feature value exceeds _ setting_, it can be judged that the power device is abnormal. For example, suppose the octave eigenvalue of IX and the peak value of the octave eigenvalue of 2 以 are set at a threshold of 5 milliseconds (millimeteiy_nd, mm/s), and the eigenvalue of the double eigenvalue is 2 seconds. Menlinu, such as "the first map" is not "" - Figure 5 is the step (10) in Figure 3 using the rule of thumb to determine the movement: the first operation - the operation of the device (% - the logic flow chart of the embodiment. If the frequency characteristic value of π is less than 5, it is judged that the power equipment is normal. On the contrary, if the multi-touch value of ιχ is greater than 5, and the multiplication characteristic values of 2Χ and 3Χ are less than 5 and 2 at the same time, it is judged that the power equipment is cut (4)斜_航。, sigh through the neural-network-derived first-operating state and the second operational secret obtained by the empirical method _ (pour s3 price preset lang can be - eight categories list, _! table includes - Normal items and - Abnormal items. Normal conditions during normal operation of the power setting operation, and abnormal items are abnormalities indicating the abnormality of the operation. For example, the abnormal items may include, but are not limited to, imbalance, no mood =, Lubrication, resonance, bearing _ condition, shaft f-curve, loose =, she slopes the balance, the potential balance (four), the contact situation and the short circuit condition. Step S380 according to the preset rule to determine whether the first operational status is recorded in the internal project, to determine whether the power equipment occurs Abnormal state. Conversely, the 'material-operation status is the operation recorded in the normal project, then the operation status of the power equipment is judged to be normal. Therefore, through the power equipment abnormality detection method of this proposal, the power can be sensed. The operation signal of the device and the method of analyzing the simplification of the eigenvalues obtained by the self-running signal can be used to determine the operational state of the power device by directly performing arithmetic processing in the micro-storage device with small resource consumption. There is no need to send the _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ The statement, non-ribbed limited the proposal 'any familiar with the relevant art, in the spirit and scope of the proposal, the shape, structure, characteristics and The spirit of the age can be changed =:: Patent protection _ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ A schematic diagram of an embodiment of a power plant abnormality detecting device. Fig. 2B is a schematic view showing another embodiment of the power plant abnormality detecting device of the present proposal. Fig. 2C is a schematic view showing still another embodiment of the power plant abnormality detecting device of the present proposal. 3 is a flow chart of the steps of the power equipment abnormality detecting method of the present proposal. Fig. 4A is a flow chart of an embodiment of step S310 in Fig. 3. Fig. 4B is a flow chart of another embodiment of step S310 in Fig. 3. Figure 4C is a flow chart of the steps of step S320 in Figure 3. Figure 5 is a flow chart of an embodiment of step S34 in the third U. [Main component symbol description] Power equipment anomaly detection device Sensing module Processing module Optimization processing module Classification diagnostic module Warning device Transmission module Memory module Power equipment 100 110 120 130 140 150 160 170 20 200