TW201118361A - Method of monitoring and inspecting equipment - Google Patents
Method of monitoring and inspecting equipment Download PDFInfo
- Publication number
- TW201118361A TW201118361A TW98140039A TW98140039A TW201118361A TW 201118361 A TW201118361 A TW 201118361A TW 98140039 A TW98140039 A TW 98140039A TW 98140039 A TW98140039 A TW 98140039A TW 201118361 A TW201118361 A TW 201118361A
- Authority
- TW
- Taiwan
- Prior art keywords
- vibration
- reference line
- signal
- range
- eigenvalues
- Prior art date
Links
Landscapes
- Testing Of Devices, Machine Parts, Or Other Structures Thereof (AREA)
- Measurement Of Mechanical Vibrations Or Ultrasonic Waves (AREA)
Abstract
Description
201118361 六、發明說明： 【發明所屬之技術領域】 本發明係關於-種設備監診方法，詳言之，係關於一種 在設備定轉速下之監診方法。 【先前技術】 設備之故障監診是為了有效執行設備維護與管理而發 展。在習知故障監診技術開發過程中，分析流程的著眼點 是-種從上而下的方式。其中，以振動為監珍對象之分析 流程大致為振動訊號擷取、數位訊號處理、訊號特徵提取 及決策分析。 在該習知故障監診流程中，其面臨到許多選擇的問題。 首先，根據監診設備之操作狀況，要選擇量測加速度、速 度或是位移振幅訊號？訊號選擇部分會決定訊號對於故障 類型之靈敏度；其次，在數位訊號處理過程中，要選擇均 方根值（RMS)、頻譜（spectrum)、小波轉換（職以以 transform)、階次分析（〇rder traeking analysis)、或是碎形 (fractal)之維度（dimensi〇n)表示，這部分工作會影響故 障特徵選擇方式；再來，訊號特徵提取部分，如何依據選 用之分析方法’簡化故障特徵成某些特徵值；最後，在決 策分析部分’如何選擇適合之決策分析方法，做為設備狀 況識別使用’例如；類神經網路（neural netw〇rk )、專家 系統（expert system )等，但無論是選擇哪一條分析路 徑’其目標終究要以定性及定量方式，將量測數據轉換成 設備管理上之訊息（設備正常或異常）。 142310.doc 201118361 ::二:據上述習知故障監診分析流程，在推展至故障 制時，_並不完整，且難以識別設備狀 振動之物理條件._^心以、略兩個因素：―為產生 ^, ，—為/刀析設備或測點數量。基本上，機 定棘、#二動程ί ’是作用力與動態剛性之間響應關係。以 <備二！而$ ’振動之所以能作為診斷參考，係在假設 a又備作用力不 的則棱下，當動態剛性改變時（如：軸承 知壞、機座鬆動 '軸糸尤亚I、 軸系不千衡），會完全反應至設備振動 的原則’進而可利用其進行監診及判斷。 但影響動態剛性還有盆 他因素，例如：設備安裝方式' 女裝機座穩定性、安裝 響性，因此在實際狀況中Π 樣的也具有關鍵影 旦、 、 可發現兩部一樣的設備，在相 同篁測位置’使用的機座 她欲从 的形式，在歷經多年使用及反覆 、准 > 後，在相同操作條件 θ 干夂正吊情況下，產生之振動程度 :不-樣的。由此可知，若以上述習知故障監診分析流 ，無淪是採用任何決策模式’即使可做到設備定性分 ::仁疋里問題上，便難以決定’一旦設備之檢測測點或 類型眾多，問題將無法獲得解決。 广卜在設備振動定量問題無法解決的情形下，在上述 習知故障監診分析流程的邏輯下，導入人工智慧（如：類 神經網路、專家系統）在決策分析部分，纟目的在於解決 故障類型分類方面問題。 然’在實際應用上受到以下因素影響其應用至其他設備 之實用性及複製，故降低應用成效。 142310.doc 201118361 最佳正規化（normalized)問題：類神經網路之輸入值及 期望值，為配合計算過程中非線性函數，均需正規化至 [〇，1]或是[·1，1]，i規化的大小將影響辨識系統之靈敏 度’因此存在最佳正規化問題。 即使是相同設備 '相同振動量測位置，均在正常運轉 下因機座、安裝、位置不同，其正常狀況下之# (如振動量）並不-致。因此，即使是在針對某—類型設備201118361 VI. Description of the invention: [Technical field to which the invention pertains] The present invention relates to a method for inspecting a device, and more particularly, to a method for inspecting a device at a fixed rotational speed. [Prior Art] The fault diagnosis of equipment is developed to effectively perform equipment maintenance and management. In the development of conventional fault diagnosis technology, the focus of the analysis process is a top-down approach. Among them, the analysis process of vibration as the treasure object is roughly the vibration signal acquisition, digital signal processing, signal feature extraction and decision analysis. In the conventional fault monitoring process, it faces many choices. First, according to the operating conditions of the monitoring equipment, should we choose to measure the acceleration, speed or displacement amplitude signal? The signal selection part determines the sensitivity of the signal to the fault type. Secondly, in the digital signal processing process, the root mean square (RMS), spectrum, wavelet transform (transformation), and order analysis are selected. Rder traeking analysis), or fractal dimension (dimensi〇n) indicates that this part of the work will affect the fault feature selection method; then, the signal feature extraction part, how to simplify the fault feature according to the selected analysis method Some feature values; finally, in the decision analysis section, 'how to choose the appropriate decision analysis method, use as the device status identification', for example; neural network (neural netw〇rk), expert system (expert system), etc. It is the choice of which analysis path. The goal is to convert the measurement data into a device management message (normal or abnormal device) in a qualitative and quantitative manner. 142310.doc 201118361 ::2: According to the above-mentioned conventional fault diagnosis and analysis analysis process, when it is pushed to the fault system, _ is not complete, and it is difficult to identify the physical condition of the device-like vibration. _^心以, slightly two factors: ―To generate ^, , - is / knife the number of devices or points. Basically, the fixed pitch, #二动程 ί ' is the response relationship between force and dynamic stiffness. With < prepared two! And the reason why $ 'vibration can be used as a diagnostic reference is that under the assumption that a is not the force of the force, when the dynamic stiffness changes (such as: the bearing is bad, the frame is loose), the axis is not in the I, the shaft is not Qian Heng) will fully reflect the principle of equipment vibration 'and can be used for monitoring and judgment. However, there are also factors affecting the dynamic rigidity, such as: equipment installation method, the stability of the women's seat, and the installation of the sound, so in the actual situation, there are also key shadows, and two devices can be found. The same speculative position 'the frame used by the base she wants to use, after many years of use and repeated, quasi->, under the same operating conditions θ dry sling, the degree of vibration generated: not-like. It can be seen that if the above-mentioned conventional fault diagnosis and analysis flow is used, it is in any decision-making mode. Even if the equipment can be qualitatively classified: the problem of Ren Renli, it is difficult to determine 'once the detection point or type of equipment Numerous, the problem will not be solved. In the case that the equipment vibration quantification problem cannot be solved, in the logic of the above-mentioned conventional fault diagnosis and analysis analysis process, artificial intelligence (such as: neural network, expert system) is introduced in the decision analysis part, and the purpose is to solve the fault. Type classification issues. However, in practical applications, the following factors affect the practicality and replication of its application to other devices, thus reducing application effectiveness. 142310.doc 201118361 The best normalized problem: the input and expected values of the neural network are normalized to [〇, 1] or [·1, 1] in order to match the nonlinear function in the calculation process. The size of the i-regulation will affect the sensitivity of the identification system. Therefore, there is an optimal normalization problem. Even the same equipment 'the same vibration measurement position, under normal operation, due to the difference in the base, installation and position, the # (such as the amount of vibration) under normal conditions is not. So even for a certain type of device
進行類神經網路發展模式，也沒有完全的複製性，需要依 其設備特性不同而調整。 如何進行類神經網路學 在故障樣本數量不足情況下 習？ 辨識過程中如何表示趨勢分析演變過程？若類神經網路 根據輸入值（經數位訊號處理後之特徵值）大小及其相關頻 率，在良好正規化條件下可區〆分設備是否（同時或單一）發 生不平衡、偏心、或轴承損壞等故障情形，但利用此方式 • 纟故障趨勢分析中會遇到很大困難，即在相同的特徵頻率 下，特徵值依故障趨勢的增加而增加，一旦增加至某一特 定值，可錢為聚類中心的改變，又發出另日―新故障類 $ ’但其應被視為同一故障類型，而非新的故障產生，其 之間差異只在於程度上大小，因此在這樣的學習過程中， 存在過度學習及樣本***問題。 如上所述，當上述習知故障監診分析技術在實際應用中 的問題無法克服時，遑論如何利用此技術協助設備管理工 作。因此，有必要提供-創新且富有進步性之設備監珍方 142310.doc 201118361 法，以解決上述問題。 【發明内容】 本發月提供#設備監診方法，用以進行一設備之異常 檢測該&備可進行—固定轉速之運轉，該設備監診方法 包括以下步驟：_取該設備在該固定轉速下之-振動訊 號二該振動訊號具有複數個頻寬範圍；(b)根據該振動訊號 計算相應該等頻絲圍之特徵值，其中該等特徵值符合高 斯分佈；（C)根據相應頻寬範圍及特徵值計算至少一基準 線；及⑷根據該基準線、相應頻寬㈣及特徵值判斷該設 備為正常或異常。 在本發明之设備監診方法中，若振動訊號超過安全範圍 基準線，即可診斷及識別出異常及故障類型，並且更可進 一步具有自動監診及學習能力。另外，本發明之設備監診 方法不需要考慮設備整個使用壽命中可能遇到的所有故障 類型，大幅降低故障監診系統發展的規模及困難故可提 升設備狀況之可視化及增加設備之妥善率。 【實施方式】 參考圖1，其顯示本發明之設備監診方法流程圖◊本發 明之s又備監珍方法用以進行一設備之異常檢測，該設備係 可進行一固定轉速之運轉。首先，參考步驟su，擷取該 設備在該固定轉速下之一振動訊號，其中該振動訊號之頻 率具有複數個頻寬範圍。 在本實施例中，本發明之方法係連續擷取該設備在該固 定轉速下之一振動加速度訊號。其中，在步驟S11中可利 142310.doc • 6 - 201118361 用一振動感測器’在固定高通濾波器、低通濾波器戴止頻 率叹疋下掏取該振動加速度訊號。較佳地，該高通濾波器 截止頻率係設定在〇. 5 Hz，該低通濾、波器之截止頻率係 。又疋在30〇〇 Hz，每隔1小時量測該振動加速度訊號訊號， 且持續；!：測3 0至50天。在本實施例中，係將該振動加速度The neural network development model is not completely reproducible and needs to be adjusted according to its device characteristics. How to perform neural network-like learning In the case of insufficient number of fault samples? How does the process of trend analysis evolve during the identification process? If the neural network is based on the input value (the characteristic value processed by the digital signal) and its associated frequency, under good normalization conditions, whether the device can be unbalanced, eccentric, or bearing damaged (simultaneously or single) Wait for the fault situation, but use this method. 纟 There is a great difficulty in the analysis of the fault trend. At the same characteristic frequency, the eigenvalue increases according to the increase of the fault trend. Once it increases to a certain value, it can be The change of the cluster center, another day-new fault class $' is issued, but it should be regarded as the same fault type, instead of the new fault, the difference is only the extent, so in this learning process There are problems with over-learning and sample explosion. As described above, when the above-mentioned problem of the conventional fault diagnosis and analysis technology cannot be overcome, the public opinion can use this technology to assist the equipment management work. Therefore, it is necessary to provide an innovative and progressive equipment to solve the above problems. SUMMARY OF THE INVENTION The present invention provides a device monitoring method for performing an abnormality detection of a device. The device can perform a fixed-speed operation. The device monitoring method includes the following steps: _ taking the device at the fixed At the rotational speed, the vibration signal has a plurality of frequency ranges; (b) calculating a characteristic value corresponding to the equal frequency wire according to the vibration signal, wherein the characteristic values are Gaussian-distributed; (C) according to the corresponding frequency The wide range and the eigenvalues calculate at least one reference line; and (4) determine whether the device is normal or abnormal according to the reference line, the corresponding bandwidth (four), and the eigenvalue. In the device monitoring method of the present invention, if the vibration signal exceeds the safety range reference line, the abnormality and the type of the failure can be diagnosed and recognized, and the automatic monitoring and learning ability can be further improved. In addition, the device monitoring method of the present invention does not need to consider all types of faults that may be encountered during the entire service life of the device, and greatly reduces the scale and difficulty of the development of the fault monitoring system, thereby improving the visibility of the device condition and increasing the proper rate of the device. [Embodiment] Referring to Fig. 1, there is shown a flow chart of a method for monitoring a device according to the present invention. The method of the present invention is also provided for performing an abnormality detection of a device which can perform a fixed rotation speed operation. First, referring to step su, a vibration signal of the device at the fixed rotation speed is captured, wherein the frequency of the vibration signal has a plurality of bandwidth ranges. In this embodiment, the method of the present invention continuously captures a vibration acceleration signal of the device at the fixed rotational speed. Wherein, in step S11, 142310.doc • 6 - 201118361 uses a vibration sensor to capture the vibration acceleration signal under a fixed high-pass filter and a low-pass filter. Preferably, the high-pass filter cutoff frequency is set at 〇. 5 Hz, and the low-pass filter and the cutoff frequency of the waver are. At 30 Hz, the vibration acceleration signal signal is measured every hour and continues; !: 30 to 50 days. In this embodiment, the vibration acceleration is
L 訊號區分為 0-625 Hz、625-1250 Hz 及 1250-2500 Hz 之頻寬 摩巳圍’然而在其他應用中，係可將該振動加速度訊號區分 為更多頻寬範圍。 參考步驟S 12 ’根據該振動訊號計算相應該等頻寬範圍 之特徵值’其中該等特徵值係符合高斯分佈。在本實施例 中’係根據該振動加速度訊號計算相應該等頻寬範圍之速 度均方根值（RMS)及加速度波高率（Crest Factor，Cf)，在 此’該速度均方根值及該加速度波高率即為該等特徵值， 且符合高斯分佈係為該等特徵值之陡峭值（Kurt〇sis)在 3±0.5之間。 φ 在本實施例中，步驟S12包括：步驟S 121，積分該振動 加速度訊號為一振動速度訊號；步驟S122，將該振動加速 度訊號及該振動速度訊號進行離散小波轉換（Discrete wavelet transform);及步驟S123，根據該振動加速度訊號 及該振動速度訊號計算該加速度波高率及該速度均方根 值。 在本實施例中’振動訊號之分析是考量包括加速度訊號 及速度訊號二種類型訊號。其中，利用加速規擷取加速度 訊號，並透過積分使加速度訊號轉換成速度訊號。 r s: 1 142310.doc 201118361 另外’基於機械學習之智能設備診斷技術在數位訊號處 理上，一般是利用小波轉換作為前處理工具，以擷取不同 頻段時間訊號，其目的在於藉由不同故障類型，對應產生 之頻率範圍及其波形，作為故障原因識別參考。 當擷取該振動加速度訊號後，進行離散小波轉換，其可 表示為下式（1): al = x(k) αί = 1 du n dk = Σ ai"XSn-2k (1) n (k = 0,1,2,..............N — 其中， :時間數據 W:取樣點數 力(/0，公〇).共軛鏡象濾波器（QMF)2H(jw^〇G(jw)的衝 擊反應函數 :分解的層數 再運用 Mallat 算法（參考 Mallat，S.(1998) 「A WaveletThe L signal is divided into 0-625 Hz, 625-1250 Hz, and 1250-2500 Hz bandwidth. However, in other applications, the vibration acceleration signal can be divided into more bandwidth ranges. Referring to step S12', feature values corresponding to the bandwidth ranges are calculated based on the vibration signal, wherein the feature values conform to a Gaussian distribution. In this embodiment, the speed root mean square (RMS) and the acceleration wave rate (Cf) corresponding to the bandwidth ranges are calculated according to the vibration acceleration signal, where the speed rms value and the The acceleration wave height rate is the eigenvalue, and the Gaussian distribution system is such that the steepness value (Kurt〇sis) of the eigenvalues is between 3±0.5. In this embodiment, step S12 includes: step S121, integrating the vibration acceleration signal into a vibration speed signal; and step S122, performing discrete wavelet transform on the vibration acceleration signal and the vibration speed signal; Step S123, calculating the acceleration wave height rate and the speed root mean square value according to the vibration acceleration signal and the vibration speed signal. In the present embodiment, the analysis of the vibration signal considers two types of signals including an acceleration signal and a speed signal. Among them, the accelerometer is used to extract the acceleration signal, and the integral is used to convert the acceleration signal into a speed signal. Rs: 1 142310.doc 201118361 In addition, the intelligent device diagnostic technology based on mechanical learning generally uses wavelet transform as a pre-processing tool to capture time signals of different frequency bands. The purpose is to use different fault types. Corresponding to the generated frequency range and its waveform, as a fault cause identification reference. After extracting the vibration acceleration signal, a discrete wavelet transform is performed, which can be expressed as the following equation (1): al = x(k) αί = 1 du n dk = Σ ai"XSn-2k (1) n (k = 0,1,2,..............N — where, : time data W: number of sampling points (/0, metric). conjugate mirror filter (QMF) 2H(jw^〇G(jw) shock response function: the number of layers to be decomposed using the Mallat algorithm (refer to Mallat, S. (1998) "A Wavelet
Tour of Signal Processing，Academic Press」，San Dieg〇 CA，US A)，可將訊號一層層進行分解，每一層分解的結果 是將上次分解得到的低頻訊號再分解成低頻和高頻二部 分，每一次分解後的數據量減半，因此分析後得到的低頻 成分和高頻成分的時域解析度比分解前訊號減低一半。斜Tour of Signal Processing, Academic Press", San Dieg〇CA, US A), can decompose the signal layer by layer, and the result of each layer decomposition is to decompose the low frequency signal obtained by the last decomposition into low frequency and high frequency. The amount of data after each decomposition is halved, so the time domain resolution of the low frequency component and the high frequency component obtained after the analysis is reduced by half compared with the signal before the decomposition. oblique
Mallat分解法後，訊號可用重構算法進行重構，重構算法 如下式（2)所示： 142310.doc (2) 201118361 a =Σ + 广 h_2 ^ = 〇，1，2,3,......... ，Ν - \ ) 。重構算法為分解算法的逆過程，經每一層重構之後，钒 ，.數據里增加—倍，提高訊號時域解析度，因此 據可分解為下式(3)之形$ : 9After the Mallat decomposition method, the signal can be reconstructed by the reconstruction algorithm. The reconstruction algorithm is shown in the following formula (2): 142310.doc (2) 201118361 a =Σ + 广h_2 ^ = 〇,1,2,3,.. ....... , Ν - \ ). The reconstruction algorithm is the inverse process of the decomposition algorithm. After each layer is reconstructed, the vanadium, .data is increased by - times, and the signal time domain resolution is improved. Therefore, the data can be decomposed into the following equation (3): $: 9
x(k) = a + y D « = ,·-, w (3)x(k) = a + y D « = ,·-, w (3)
其中’ A為粗略子空D為細節子空間（Detail (Level) 〇 間（Approximation subspace)訊號， subspace)訊號，L為需分解之層數 利用離散小波轉換能將訊號正交分解至不同頻寬之時域 表不中，故有助於瞭解原始訊號中各組成訊號、發生頻寬 及波形’以進一步提取其波形特徵，作為故障識別之夂 考。 ' 一般而言，關於設備故障之特徵參數大致上可區分為二 類5L .此里參數及統計參數。其中，能量參數是以能量為 單位顯示故障特徵（如軸偏心、不平衡），注重其能量改變 以表示設備狀況傾向；統計參數均屬於無因次參數，注重 其統計特性（波形）改變以表示設備狀態改變◎正常訊號多 屬於高斯分佈，其陡峭值（kurt〇sis)其值約為3，若發生衝 擊波時（軸承、齒輪損壞），其陡峭值統計參數會大於3，故 障監測使用數位訊號處理方法與對應使用量化之特徵參數 選擇關係。其中，利用離散小波轉換作為數位訊號處理工 具，既可以用於能量參數使用，亦可用於統計參數使用， 且量化結果具有直觀物理意義，適合用於振動之特徵表 142310.doc „ ^ 201118361 示。Where 'A is the rough subspace D (Detail (Level) (Approximation subspace) signal, subspace) signal, L is the number of layers to be decomposed. Discrete wavelet transform can be used to decompose the signal orthogonally to different bandwidths. The time domain is not in the table, so it is helpful to understand the constituent signals, the occurrence bandwidth and the waveform of the original signal to further extract the waveform characteristics as a reference for fault identification. In general, the characteristic parameters of equipment faults can be roughly divided into two types of 5L. These parameters and statistical parameters. Among them, the energy parameter is to display the fault characteristics (such as shaft eccentricity and unbalance) in energy, and pay attention to its energy change to indicate the tendency of equipment condition; the statistical parameters are all dimensionless parameters, paying attention to the change of its statistical characteristics (waveform) to represent Equipment status change ◎ Normal signal is mostly Gaussian distribution, its steep value (kurt〇sis) has a value of about 3. If a shock wave occurs (bearing, gear damage), its steep value statistical parameter will be greater than 3, and the fault monitoring uses digital signal. The processing method selects a relationship with the characteristic parameter corresponding to the use of quantization. Among them, discrete wavelet transform is used as a digital signal processing tool, which can be used for energy parameters or statistical parameters, and the quantitative results have intuitive physical meaning, which is suitable for vibration characteristics table 142310.doc „ ^ 201118361.
參考步驟S13，根據相應頻寬範圍及特徵值（該振動加速 度訊號及該振動速度訊號）計算至少一基準線。在本實施 例中，其係根據該振動加速度訊號及該振動速度訊號，利 用標準差過程計算-第-基準線。其中，該標準差過程包 括.步驟S131，若該等特徵值之統計分析均符合高斯分 佈，分別計算各特徵值分佈之標準差及平均值；及步= S132，以該等特徵值之統計樣本之平均值為圓心座標，以 該等特徵值之統計樣本之六個標準差為二弦，以該二弦之 一斜邊為半徑畫圓形成該第一基準線。 該第一基準線係作為識別該設備振動正常與否之參考。 該第一基準線的組成，由前述步驟S12之敘述可知，訊號 特徵（該等特徵值）是以離散小波轉換為基礎，藉由長時間 蒐集設備在正常狀況下振動訊號，依據離散小波轉換之設 定下’计算相應各頻寬範圍之速度訊號之均方根值（能量 參數）及加速度訊號之波高率（統計參數）。速度均方根值 (RMS)及加速度波高率（Crest Factor，Cf)可表示為下式（4) 及（5): 1 N-\ (4) xms =(~^hy12 ^ /=0 .. —"^max xcf - , XRMS (5) 其中，= 為時間序列。 速度訊號之均方根值是表示振動程度常用參考數據，而 -10- 142310.doc 201118361 加速度訊號之波高率則是反應訊號中衝擊現象。在本實施 例中’透過統計分析各特徵值長期累積數值，是否符合高 斯分佈’再根據六個標準差概念，計算其安全與警戒之界 線（D亥第一基準線）。其中’若統計上無法收斂時，則表示 特徵值趙勢一直改變，可視為設備正處於異常狀況，且需 再進一步觀察。 其中’在步驟S132之後可另包括一第二基準線建立步 驟，該第二基準線建立步驟包括以下步驟：根據該等特徵 值之統計分析計算各特徵值分佈之標準差及平均值；及以 該等特徵值之統計樣本之平均值為圓心座標，以該等特徵 值之統計樣本之十二個標準差為二弦，以該二弦之一斜邊 為半控畫圓形成一第二基準線。 另外，在該第二基準線建立步驟之後可另包括一第三基 準線建立步驟’該第三基準線建立步驟包括以下步驟：根 據該等特徵值之統計分析計算各特徵值分佈之標準差及平 均值’及以該等特徵值之統計樣本之平均值為圓心座標， 以4等特徵值之統計樣本之十八個標準差為二弦以該二 弦之一斜邊為半徑晝圓形成一第三基準線。 參考圖2，其係顯示本發明在設備正常情況下於不同頻 寬範圍内建立基準線之示意圖。其中，圖2(約顯示根據設 備之原始振動訊號，建立正常、警告、危險三條基準線之 不意圖；圖 2(b)-2(d)分別顯示於 125〇_25〇〇 Hz、625-1250 Hz、0-625 Hz之頻寬範圍内，根據相應迷度訊號之均方根 值及加速度訊號之波高率所建立正常、警告、危險三條基 142310.doc -11- 201118361 準線（由内而外）之示意圖，其中基準線L1、L2、L3即分別 表不上述之第一基準線、第二基準線及第三基準線。 再配合參考圖1之步驟S14及圖2，根據該第一基準線、 相應頻寬範圍及特徵值判斷該設備為正常或異常。其中， 在該固定轉速下持續擷取之該振動訊號中，若相應每一頻 寬範圍之特徵值（請參考上述步驟S12中關於特徵值之計算 方式）相對位置在該第一基準線1^範圍之内，判斷該設備 ^ 為正常’若相應該等頻寬範圍之至少其中之一特徵值相對 位置在該第一基準線L1範圍之外，判斷該設備為異常，其 中，若相應該等頻寬範圍之至少其中之一特徵值相對位置 在該第二基準線L2與該第三基準線L3之間，判斷該設備之 異常為警告狀態，若相應該等頻寬範圍之至少其中之一特 徵值相對位置在該第三基準線L3之外，判斷該設備之異常 為危險狀態。 凊再配合參考圖2(a)-2(d)及步驟S14，其可清楚看出在 φ 母一頻寬範圍内之振動訊號統計資料皆位於正常基準線 (第一基準線L1)範圍内，其表示設備狀態為正常。 在本實施例中，在步驟S14中另包括一故障編碼步驟， 、該故障編碼步驟包括以下步驟：若該等頻寬範圍中相應之 特徵值相對位置在該第一基準線L1範圍之内，定義為一第 一編碼（例如：數值0)，若該等頻寬範圍中相應之特徵值相 對位置在該第一基準線“範圍之外，定義為一第二編碼 (例如.數值1);根據該第一編碼、該第二編碼及其相應頻 寬範圍判斷該設備之異常類型。 ·-· 1423I0.doc •12· 201118361 較佳地，本發明之方法可利用類神經網路（neural network)或專家系統（eXpert system)判斷該設備之異常類 型。其中，該類神經網路可選用自適應共振類神經網路 (Adaptive Resonance Theory Neural Network, ART)。另 外’步驟S14可更包括一記憶學習步驟，以該類神經網路 或專家系統記憶及儲存經編碼後之異常類型。 當完成故障編碼後，可使用自適應共振理論類神經網路 (參考 Adaptive Resonance Theory Neural Network, ART)(Carpenter G. A., Grossberg S.(1998) 「The ART 〇fReferring to step S13, at least one reference line is calculated according to the corresponding bandwidth range and the characteristic value (the vibration acceleration signal and the vibration speed signal). In this embodiment, based on the vibration acceleration signal and the vibration speed signal, the standard deviation process is used to calculate the -first reference line. The standard deviation process includes a step S131. If the statistical analysis of the eigenvalues all conform to the Gaussian distribution, respectively calculate the standard deviation and the average value of each eigenvalue distribution; and step = S132, the statistical sample of the eigenvalues The average value is a centroid coordinate, and the six standard deviations of the statistical samples of the eigenvalues are two chords, and the first reference line is formed by drawing a circle with one of the two chords as a radius. The first reference line serves as a reference for identifying whether the vibration of the device is normal or not. The composition of the first reference line is as described in the foregoing step S12. The signal characteristics (the eigenvalues) are based on the discrete wavelet transform, and the vibration signal is filtered under normal conditions by the long-time collecting device, and the discrete wavelet transform is performed. Set the 'root mean square value (energy parameter) of the speed signal corresponding to each bandwidth range and the wave height rate (statistical parameter) of the acceleration signal. The velocity root mean square (RMS) and acceleration wave rate (Cf) can be expressed as the following equations (4) and (5): 1 N-\ (4) xms = (~^hy12 ^ /=0 .. —"^max xcf - , XRMS (5) where = is the time series. The rms value of the speed signal is the commonly used reference data for the degree of vibration, and the wave height of the acceleration signal is -10- 142310.doc 201118361 The impact phenomenon in the signal. In this embodiment, 'through the statistical analysis of the long-term cumulative value of each feature value, whether it conforms to the Gaussian distribution' and then calculate the boundary between safety and alert based on the six standard deviation concepts (D Hai first baseline). If the statistical value cannot be converged, it means that the characteristic value Zhao has been changed, which can be regarded as the abnormal condition of the device, and needs to be further observed. Wherein, after step S132, a second baseline establishing step may be further included. The second baseline establishing step includes the steps of: calculating a standard deviation and an average value of each eigenvalue distribution according to statistical analysis of the eigenvalues; and calculating a mean value of the statistical samples of the eigenvalues as a centroid coordinate, with the features Value The twelve standard deviations of the statistical sample are two chords, and one of the two chords is a semi-controlled circle to form a second reference line. In addition, a third reference may be further included after the second baseline establishing step. The line establishing step 'the third baseline establishing step comprises the steps of: calculating a standard deviation and an average value of each eigenvalue distribution according to statistical analysis of the eigenvalues' and an average value of the statistical samples of the eigenvalues as a central coordinate The eighteen standard deviations of the statistical samples of the eigenvalues of four are two chords, and one of the two chords is a radius and a circle is formed to form a third reference line. Referring to FIG. 2, the present invention shows the normal condition of the device. A schematic diagram of establishing a reference line in different bandwidths. Figure 2 (approximate intention to establish normal, warning, and dangerous baselines based on the original vibration signal of the device; Figure 2(b)-2(d) Displayed in the bandwidth range of 125〇_25〇〇Hz, 625-1250 Hz, and 0-625 Hz, respectively, according to the rms value of the corresponding fascia signal and the wave height rate of the acceleration signal, establish three normal, warning, and dangerous Base 142310.doc -11- 2 01118361 A schematic diagram of a guideline (from the inside out), wherein the reference lines L1, L2, and L3 respectively represent the first reference line, the second reference line, and the third reference line, respectively. Referring again to step S14 of FIG. 2, judging whether the device is normal or abnormal according to the first reference line, the corresponding bandwidth range, and the characteristic value, wherein, in the vibration signal continuously captured at the fixed rotation speed, if each of the bandwidth ranges is characterized The value (refer to the calculation method of the feature value in the above step S12) is within the range of the first reference line 1^, and it is determined that the device ^ is normal 'if at least one of the characteristic values of the range of the bandwidths The relative position is outside the range of the first reference line L1, and the device is determined to be abnormal. If at least one of the feature values of the corresponding bandwidth ranges is relative to the second reference line L2 and the third reference line Between L3, it is determined that the abnormality of the device is a warning state, and if at least one of the characteristic value relative positions of the bandwidth ranges is outside the third reference line L3, it is determined that the abnormality of the device is a dangerous state. Referring to FIG. 2(a)-2(d) and step S14, it can be clearly seen that the vibration signal statistics in the range of φ mother-band width are all within the range of the normal reference line (first reference line L1). , which indicates that the device status is normal. In this embodiment, a fault coding step is further included in step S14, and the fault coding step includes the following steps: if the corresponding feature value relative position in the bandwidth ranges is within the first reference line L1, Defined as a first code (eg, a value of 0), if the corresponding feature value relative position in the range of the bandwidth is outside the range of the first reference line, defined as a second code (eg, value 1); Determining an abnormal type of the device according to the first code, the second code, and a corresponding bandwidth range thereof. - - 1423I0.doc • 12 · 201118361 Preferably, the method of the present invention can utilize a neural network (neural network) Or the expert system (eXpert system) determines the abnormal type of the device, wherein the neural network may use an Adaptive Resonance Theory Neural Network (ART). In addition, the 'Step S14 may further include a memory. Learning steps to remember and store the encoded anomaly type with such a neural network or expert system. When the fault coding is completed, an adaptive resonance theory-like neural network can be used (Ref. A Daptive Resonance Theory Neural Network, ART) (Carpenter G. A., Grossberg S. (1998) "The ART 〇f
Adaptive Pattern Recognition by a Self-Orginizing NeuralAdaptive Pattern Recognition by a Self-Orginizing Neural
Network」⑽ 21，pp_ 77-88·)，進行故障編碼 識別的工作。ART是模擬人的認知過程和大腦的特點，以 期在複雜、非平穩及有干擾的環境中，對各種事物進行分 類及識別，並且對於所學習的訊息的累積和儲存既具有剛 性又具有彈性，即，一方面牢固地保存已學習的訊息，另 一方面又能學習大量新的訊息，能避免先前學習模式的修 改’同時’記憶容量可隨學習樣本增加而增加。 類神經網路之工作原理為：當類神經網路接受來自環境 的輸入（例如：振動加速度訊號），網路即檢查新的輸入模 式與所儲存模式之間匹配程度，根據一預設的門檻值計算 這些輸入模式相識度，若相似度高則選擇最相似模式類型 作為該模型代表；反之網路則設立一新的模式類型並加以 儲存’以作為後續輸入模式匹配過程之參考。 參考圖3及4，其係顯示以本發明設備監診方法分別診斷 1423l〇.d〇, 201118361 二部不同定轉速風車 U早馬達之結果示意圖。其中，圖3(a)及 圖4(a)顯示風車馬達 咬 < 原始振動訊號之統計資料；圖3(b) 及圖4(b)顯示風車馬遠· 振動訊號位於1250-2500 Hz之頻 寬範圍内之統計資料.园 1貝针，圖3(c)及圖4(c)顯示風車馬達之振 動訊號位於625-1250 -Network" (10) 21, pp_ 77-88·), work on fault coding identification. ART is a model of the human cognitive process and the brain, in order to classify and identify various things in a complex, non-stationary and disturbing environment, and is rigid and flexible for the accumulation and storage of the learned information. That is, on the one hand, the learned message is firmly preserved, and on the other hand, a large number of new messages can be learned, and the modification of the previous learning mode can be avoided, and the memory capacity can be increased as the learning sample increases. The neural network works by: when the neural network accepts input from the environment (for example, vibration acceleration signals), the network checks the degree of matching between the new input mode and the stored mode, according to a preset threshold. The values calculate the eigenvalues of these input modes. If the similarity is high, the most similar mode type is selected as the model representative; otherwise, the network establishes a new mode type and stores it as a reference for the subsequent input pattern matching process. Referring to Figures 3 and 4, it is a schematic diagram showing the results of the diagnosis of the 1423l〇.d〇, 201118361 two different fixed speed windmill U early motors by the device monitoring method of the present invention. 3(a) and 4(a) show the statistics of the windmill motor bite < original vibration signal; Fig. 3(b) and Fig. 4(b) show the frequency of the windmill horse far vibration signal at 1250-2500 Hz Wide range of statistics. Park 1 pin, Figure 3 (c) and Figure 4 (c) show that the windmill motor vibration signal is located at 625-1250 -
Hz之頻寬範圍内之統計資料；圖3(d) 及圖4(d)顯示風車馬读 千勹逹之振動訊號位於0-625 Hz之頻寬範 圍内之統計資料。Statistics over the bandwidth of Hz; Figure 3(d) and Figure 4(d) show statistics for the windmill horse reading vibration signal in the bandwidth range of 0-625 Hz.
使用小波函數之離散小波轉換，根據振動訊號之三個不 同子空間（頻寬_)，計算各子空間之速度㈣之均方根 值及加速度訊號之波尚率，再利用統計分析建立原始振動 訊號及各子工間之基準線’#，曲線l ι為正常基準線，曲 線L2及L3為異常（警告及危險）基準線（第二基準線及第三 基準線)。其中’在本實例中’ 1250-2500 Hz之頻寬範圍内 之異常狀態為轴承問題及/或潤滑問題；625_l25〇 Hz之頻 寬範圍内之異常狀態為齒輪問題；〇_625 Hz之頻寬範圍内 之異常狀態為不平衡問題及/或鬆動問題。 •參考圖3(a)，明顯地，風車馬達之原始振動訊號之統計 資料超出曲線L1之範圍，但在曲線以範圍之内，即表示有 異常且為警告狀態。進一步分析異常之類型：參考圖 3(b)，振動訊號位於1250-2500 Hz之頻寬範圍内之統計資 料超出曲線L1之範圍且在曲線L2範圍之内，即表示在 1250-2500 Hz之頻寬範圍内有異常狀態（軸承問題及/或潤 滑問題），並將該狀況編碼為丨（第一編碼）；參考圖3(c), 振動訊號位於625-1250 Hz之頻寬範圍内之統計資料，超 142310.doc -14- 201118361 出曲線L1之範圍且在曲線^範圍之内，即表示在625_i2wUsing the discrete wavelet transform of the wavelet function, according to the three different subspaces (bandwidth _) of the vibration signal, calculate the rms value of each subspace (4) and the wave rate of the acceleration signal, and then use the statistical analysis to establish the original vibration. The signal and the reference line '# of each sub-workplace, the curve l is the normal baseline, and the curves L2 and L3 are the abnormal (warning and danger) baseline (the second baseline and the third baseline). The abnormal state in the bandwidth range of '1250-2500 Hz in this example is the bearing problem and/or the lubrication problem; the abnormal state in the bandwidth range of 625_l25〇Hz is the gear problem; the bandwidth of 〇_625 Hz The abnormal state within the range is an imbalance problem and/or a loose problem. • Referring to Fig. 3(a), it is apparent that the statistics of the original vibration signal of the windmill motor are outside the range of the curve L1, but within the range of the curve, it indicates that there is an abnormality and a warning state. Further analysis of the type of anomaly: Referring to Figure 3 (b), the statistical data of the vibration signal in the frequency range of 1250-2500 Hz is beyond the range of the curve L1 and within the range of the curve L2, that is, the frequency at 1250-2500 Hz There is an abnormal state (bearing problem and/or lubrication problem) in a wide range, and the condition is coded as 丨 (first code); referring to Figure 3 (c), the vibration signal is in the range of bandwidth of 625-1250 Hz Information, super 142310.doc -14- 201118361 out of the range of curve L1 and within the range of curve ^, that is expressed in 625_i2w
Hz之頻寬範圍内有異常狀態（齒輪問題），並將該狀況編碼 為1(第二編碼）；參考圖3⑷，振動訊號位於〇·625 Hz之頻 寬範圍内之統計資料未超出曲線u之範圍，即表示在心 625 Hz之頻寬範圍内無異常狀態，並將該狀況編碼為〇(第 三編碼）。在此，風車馬達之異常狀態之三位編碼可表示 為[1 1 0] 〇 參考圖4(a) ’風車馬達之原始振動訊號之統計資料超出 曲線L1之範圍，但在曲線[2範圍之内，即表示有異常且為 警告狀態。進一步分析異常之類型：參考圖4(b)，振動訊 號位於1250-2500 Hz之頻寬範圍内之統計資料未超出曲線 L1之範圍，即表示在125〇_25〇〇 Hzi頻寬範圍内無異常狀 態，並將該狀況編碼為0(第一編碼）；參考圖4(c)，振動訊 號位於625.125G Hz之頻寬範圍内之統計資料未超出曲線 L1之範圍，即表示在625_125〇 Hz之頻寬範圍内無異常狀 態，並將該狀況編碼為0(第二編碼）；參考圖4(幻，振動訊 號位於0-625 Hz之頻寬範圍内之統計資料，超出曲線以之 範圍且在曲線L2範圍之内，即表示在〇_625 Hz之頻寬範圍 内有異常狀態（平衡問題及/或鬆動問題），並將該狀況編碼 為1(第三編碼在此，風車馬達之異常狀態之三位編碼可 表不為[0 0 1]。 其中，異常狀態經編碼後，即由自適應共振理論類神經 網路進灯故障編碼識別的工作，自適應共振理論類神經網 路並累積、儲存及學習該等異常狀態之編碼及其代表意 142310.doc , 『、:' 201118361 義，以保存已學習的訊息， 以作為後續振動訊號之診斷及 識別使用。There is an abnormal state (gear problem) in the frequency range of Hz, and the condition is coded as 1 (second code); referring to Figure 3 (4), the statistical data of the vibration signal in the bandwidth range of 〇·625 Hz does not exceed the curve u The range means that there is no abnormal state in the bandwidth of the heart 625 Hz, and the condition is coded as 〇 (third code). Here, the three-digit code of the abnormal state of the windmill motor can be expressed as [1 1 0] 〇 Refer to FIG. 4(a). The statistics of the original vibration signal of the windmill motor are outside the range of the curve L1, but in the range of the curve [2] Inside, it means that there is an exception and it is a warning state. Further analysis of the type of anomaly: Referring to FIG. 4(b), the statistical data of the vibration signal in the frequency range of 1250-2500 Hz does not exceed the range of the curve L1, that is, within the range of 125〇_25〇〇Hzi bandwidth. Abnormal state, and encode the condition as 0 (first code); referring to Figure 4 (c), the statistics of the vibration signal in the bandwidth of 625.125 G Hz does not exceed the range of the curve L1, that is, at 625_125 Hz There is no abnormal state in the bandwidth range, and the condition is encoded as 0 (second code); refer to Figure 4 (magic, the vibration signal is in the range of 0-625 Hz bandwidth, beyond the curve and Within the range of the curve L2, it means that there is an abnormal state (balance problem and/or looseness problem) within the bandwidth of 〇_625 Hz, and the condition is coded as 1 (the third code is here, the abnormality of the windmill motor) The three-bit code of the state can be expressed as [0 0 1]. Among them, the abnormal state is encoded, that is, the adaptive resonance theory-like neural network enters the lamp fault code recognition work, and the adaptive resonance theory-like neural network is Accumulate, store and learn the difference Coding and representatives of the Italian 142310.doc state, ",: '201 118 361 righteousness, to save the message has been learning to use as a diagnostic and follow-up to identify the vibration signal.
能力。另外，本發 在本發明之設備監診方法中，若振動訊號超過安全範圍 準線，則予以自動編碼，i冰λ ώ 型之3乡斷及識別，同時保有自動監診及學習 本發明之設備監診方法不需要考慮設備整個 使用壽命中可能遇到的所有故障類型，大幅降低故障監診 系統發展的規模及困難，故可提升設備狀況之可視化及增 加設備之妥善率。 上述實施例僅為說明本發明之原理及其功效，並非限制 本發明。因此習於此技術之人士對上述實施例進行修改及 變化仍不脫本發明之精神。本發明之權利範圍應如後述之 申請專利範圍所列。 【圖式簡單說明】 圖1顯示本發明之設備監診方法流程圖； 圖2，其係顯示本發明在設備正常情況下於不同頻寬範 圍内建立基準線之示意圖；及 圖3及4係顯示以本發明設備監診方法分別診斷二部不同 定轉速風車馬達之結果示意圖。 142310.doc • 16·ability. In addition, in the device monitoring method of the present invention, if the vibration signal exceeds the safety range guideline, the automatic coding is performed, and the i-bing λ ώ type is disconnected and identified, and the automatic supervision and learning of the present invention are maintained. The equipment supervision method does not need to consider all types of faults that may be encountered during the entire service life of the equipment, and greatly reduces the scale and difficulty of the development of the fault diagnosis system, thereby improving the visibility of the equipment condition and increasing the proper rate of the equipment. The above embodiments are merely illustrative of the principles and effects of the invention and are not intended to limit the invention. Therefore, those skilled in the art can make modifications and changes to the above embodiments without departing from the spirit of the invention. The scope of the invention should be as set forth in the appended claims. BRIEF DESCRIPTION OF THE DRAWINGS FIG. 1 is a flow chart showing a method for inspecting a device according to the present invention; FIG. 2 is a schematic view showing the present invention for establishing a reference line in different bandwidths under normal conditions of the device; and FIGS. 3 and 4 A schematic diagram showing the results of diagnosing two different fixed speed windmill motors by the device monitoring method of the present invention. 142310.doc • 16·
Claims (1)
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| TW98140039A TWI398629B (en) | 2009-11-25 | 2009-11-25 | Equipment supervision method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| TW98140039A TWI398629B (en) | 2009-11-25 | 2009-11-25 | Equipment supervision method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| TW201118361A true TW201118361A (en) | 2011-06-01 |
| TWI398629B TWI398629B (en) | 2013-06-11 |
Family
ID=44935616
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| TW98140039A TWI398629B (en) | 2009-11-25 | 2009-11-25 | Equipment supervision method |
Country Status (1)
| Country | Link |
|---|---|
| TW (1) | TWI398629B (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| TWI476387B (en) * | 2012-07-26 | 2015-03-11 | Univ Southern Taiwan Sci & Tec | Automatic monitoring system and its method for the initial defect of a mechanical system |
| TWI777681B (en) * | 2021-07-22 | 2022-09-11 | 宇辰系統科技股份有限公司 | Vibration monitoring system for electrical machine |
Family Cites Families (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US6904371B2 (en) * | 1997-10-17 | 2005-06-07 | Test Devices, Inc. | Method and apparatus for measuring rotor unbalance |
| TW419839B (en) * | 1999-08-09 | 2001-01-21 | Nat Science Council | Device for measuring vibration and method thereof |
| TW505793B (en) * | 1999-10-07 | 2002-10-11 | Shr-Guang Lee | Autonomous phase-gain rotation/linear piezoelectric optimal sensing device |
| US7254475B1 (en) * | 2006-02-07 | 2007-08-07 | Infineon Technologies Ag | Detection systems and methods |
-
2009
- 2009-11-25 TW TW98140039A patent/TWI398629B/en active
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| TWI476387B (en) * | 2012-07-26 | 2015-03-11 | Univ Southern Taiwan Sci & Tec | Automatic monitoring system and its method for the initial defect of a mechanical system |
| TWI777681B (en) * | 2021-07-22 | 2022-09-11 | 宇辰系統科技股份有限公司 | Vibration monitoring system for electrical machine |
Also Published As
| Publication number | Publication date |
|---|---|
| TWI398629B (en) | 2013-06-11 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US10520397B2 (en) | Methods and apparatuses for defect diagnosis in a mechanical system | |
| Yan et al. | An efficient approach to machine health diagnosis based on harmonic wavelet packet transform | |
| Lei et al. | New clustering algorithm-based fault diagnosis using compensation distance evaluation technique | |
| KR102261899B1 (en) | Apparatus and method for diagnosing fault of drone | |
| CN110398647B (en) | Transformer state monitoring method | |
| CN102155988A (en) | Equipment monitoring and diagnosing method | |
| CN113086799B (en) | Elevator fault detection and early warning method based on image recognition | |
| CN112182912B (en) | Manufacturing equipment spindle bearing health assessment method based on probability description and spectrum analysis | |
| CN109214522B (en) | Equipment performance degradation evaluation method based on multiple attributes | |
| Abid et al. | Bearing fault detection of induction motor using SWPT and DAG support vector machines | |
| WO2022147684A1 (en) | Method and apparatus for identifying abnormalities in mechanical apparatus or mechanical component | |
| CN114705432B (en) | Method and system for evaluating health state of explosion-proof motor bearing | |
| CN113391207A (en) | Motor fault detection method, medium and system | |
| CN114118219A (en) | Data-driven real-time abnormal detection method for health state of long-term power-on equipment | |
| CN114486263B (en) | Noise reduction demodulation method for vibration signal of rolling bearing of rotary machine | |
| Zhang et al. | Fault diagnosis based on optimized node entropy using lifting wavelet packet transform and genetic algorithms | |
| TW201118361A (en) | Method of monitoring and inspecting equipment | |
| CN117782198B (en) | Highway electromechanical equipment operation monitoring method and system based on cloud edge architecture | |
| CN112666430B (en) | Intelligent fault detection method and system for voiceprint of transformer | |
| CN112132069A (en) | Rolling bearing weak fault intelligent diagnosis method based on deep learning | |
| CN117370879A (en) | Real-time online fault diagnosis method and system for wind turbine gearbox | |
| CN110749443A (en) | Rolling bearing fault diagnosis method and system based on high-order origin moment | |
| CN114580472A (en) | Large-scale equipment fault prediction method with repeated cause and effect and attention in industrial internet | |
| CN114993671A (en) | Vibration fault diagnosis method and system based on Q factor wavelet transform | |
| CN113505898A (en) | Equipment predictive maintenance method based on AI technology |