TWI807985B - Fault detection method of wind turbine gearbox - Google Patents

Fault detection method of wind turbine gearbox Download PDF

Info

Publication number
TWI807985B
TWI807985B TW111133237A TW111133237A TWI807985B TW I807985 B TWI807985 B TW I807985B TW 111133237 A TW111133237 A TW 111133237A TW 111133237 A TW111133237 A TW 111133237A TW I807985 B TWI807985 B TW I807985B
Authority
TW
Taiwan
Prior art keywords
component
wind turbine
gearbox
feature image
detection method
Prior art date
Application number
TW111133237A
Other languages
Chinese (zh)
Other versions
TW202411532A (en
Inventor
呂學德
王孟輝
吳家峻
Original Assignee
國立勤益科技大學
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 國立勤益科技大學 filed Critical 國立勤益科技大學
Priority to TW111133237A priority Critical patent/TWI807985B/en
Application granted granted Critical
Publication of TWI807985B publication Critical patent/TWI807985B/en
Publication of TW202411532A publication Critical patent/TW202411532A/en

Links

Images

Landscapes

  • Wind Motors (AREA)
  • Measurement Of Mechanical Vibrations Or Ultrasonic Waves (AREA)

Abstract

一種風力機齒輪箱之故障檢測方法，係為先檢測一風力發電機組之一齒輪箱的運轉的振動狀態，再將該振動狀態傳送給一電腦運算系統，該電腦運算系統再將該振動狀態透過一經驗模態分解法來找出一故障訊號的特徵資訊，並透過一對稱點座標法將該特徵訊號呈現出一對稱的二維特徵圖像，再透過一更快的區域卷積神經網路將該二維特徵圖像進行特徵提取，使其進行訓練與分類，當所有資料的分類均已完成並達到收斂，即透過一風力機齒輪箱故障瑕疵模型來辦識出該齒輪箱的故障類型。 A fault detection method for a wind turbine gearbox, which is to first detect the vibration state of a gearbox of a wind power generating set, and then transmit the vibration state to a computer operation system, and the computer operation system uses an empirical mode decomposition method to find out the characteristic information of a fault signal, and uses a symmetrical point coordinate method to present the characteristic signal into a symmetrical two-dimensional feature image, and then uses a faster regional convolutional neural network to perform feature extraction on the two-dimensional feature image, so that it can be trained and classified. Convergence, that is, identify the fault type of the gearbox through a fault model of the wind turbine gearbox.

Description

風力機齒輪箱之故障檢測方法 Fault detection method of wind turbine gearbox

本發明係關於一種風力機齒輪箱之故障檢測方法，特別是一種能籍由齒輪箱所引發的三軸方向的振動變化，使其該齒輪箱的故障初期就能發現故障類型的一種風力機齒輪箱之故障檢測方法。 The present invention relates to a fault detection method for a gearbox of a wind turbine, in particular to a fault detection method for a gearbox of a wind turbine that can detect the type of fault at the initial stage of the gearbox fault by virtue of the vibration changes in the three-axis directions caused by the gearbox.

全球對於風力發電的需求日漸增長，已成為日常生活中不可或缺的再生能源，然而風力發電機因長期暴露在戶外，而且長時間的運轉，使得機件磨損故障與零件老化發生的機率相對增加，因此從國內外相關的風電事故及歷年風機事故統計指出，迄今多起風機災損紀錄中，有高達三分之一的停機主因係源於齒輪箱的故障。 The global demand for wind power generation is growing day by day, and it has become an indispensable renewable energy source in daily life. However, due to long-term exposure to the outdoors and long-term operation of wind power generators, the probability of mechanical wear and tear failures and parts aging has increased relatively. Therefore, according to related wind power accidents at home and abroad and the statistics of wind turbine accidents over the years, as many as one third of the wind turbine disaster records so far are due to gearbox failures.

因此如何在齒輪箱運轉時，即能從該齒輪箱的振動訊號中發現故障初期的微弱的故障特徵，乃為學界與業界亟欲解決的問題之一。 Therefore, how to detect the weak fault characteristics at the initial stage of the fault from the vibration signal of the gearbox when the gearbox is running is one of the problems that the academic circles and the industry want to solve urgently.

本發明目的在於提供一種風力機齒輪箱之故障檢測方法，藉以針對齒輪箱所引發的三軸方向的振動變化，來診斷出該齒輪箱之故障初期的故障類型。 The purpose of the present invention is to provide a fault detection method for a gearbox of a wind turbine, so as to diagnose the initial fault type of the gearbox according to the vibration changes in the three-axis directions caused by the gearbox.

為達成上述目的，本發明實施例所述之風力機齒輪箱之故障檢測方法，至少包括如下步驟：第一步：接收設置於該齒輪箱內或表面的一三軸振動感測器，用以感測該齒輪箱轉動時的三軸方向的振動狀態，並產生出多個振動訊號；第二步驟：將該 多個振動訊號透過一可程式邏輯控制器傳送給一電腦運算系統；第三步驟：該電腦運算系統先透過一經驗模態分解法(Empirical Mode Decomposition，簡稱EMD)，將該多個振動訊號的各頻率分解成固有模態函數(Intrinsic Mode Function，簡稱IMF)的形式，作為故障訊號的特徵資訊；第四步驟：該電腦運算系統再透過一對稱點座標法(Symmetrized Dot Pattern，簡稱SDP)，將該已分解後的故障訊號的波形的振幅或頻率之間的差異轉換為極座標平面之定位點的位置差異與曲率變化，用以呈現出一對稱的二維特徵圖像；第五步驟：該電腦運算系統最後透過一更快的區域卷積神經網路(Faster Region-based Convolutional Neural Network，簡稱Faster R-CNN)，將該對稱的二維特徵圖像進行特徵提取以產生一第一特徵圖像，並將該第一特徵圖像透過多個錨框(Anchor Box)進行多個候選區域的生成，再將該多個候選區域進行分類，並將該分類後的候選區域與該第一特徵圖像相結合以產生一第二特徵圖像，再將該第二特徵圖像進行訓練與分類，當該分類均已完成並達到收斂，即透過一風力機齒輪箱故障瑕疵模型來辦識出該齒輪箱的故障類型。 In order to achieve the above-mentioned purpose, the fault detection method of the wind turbine gearbox described in the embodiment of the present invention includes at least the following steps: the first step: receiving a three-axis vibration sensor installed in or on the surface of the gearbox to sense the vibration state of the three-axis direction when the gearbox rotates, and generating a plurality of vibration signals; A plurality of vibration signals are transmitted to a computer computing system through a programmable logic controller; the third step: the computer computing system decomposes each frequency of the multiple vibration signals into the form of an intrinsic mode function (Intrinsic Mode Function, referred to as IMF) through an Empirical Mode Decomposition method (EMD for short), which is used as the characteristic information of the fault signal; the fourth step: the computer computing system then uses a Symmetrized Dot Pattern (Symmetrized Dot Pattern, referred to as SDP), the difference between the amplitude or frequency of the decomposed fault signal waveform is converted into the position difference and curvature change of the positioning point on the polar coordinate plane, so as to present a symmetrical two-dimensional feature image; the fifth step: the computer computing system finally uses a faster region convolutional neural network (Faster Region-based Convolutional Neural Network, referred to as Faster R-CNN) to perform feature extraction on the symmetrical two-dimensional feature image to generate a first feature image, and pass the first feature image through multiple anchors Anchor Box generates a plurality of candidate areas, and then classifies the plurality of candidate areas, and combines the classified candidate areas with the first feature image to generate a second feature image, and then trains and classifies the second feature image, and when the classification is completed and converges, the fault type of the gearbox is identified through a fault model of a wind turbine gearbox.

以上關於本發明所揭露內容的說明及以下實施方式的說明係用以示範與解釋本發明的精神與原理，並且提供本發明的專利申請範圍更進一步的解釋。 The above descriptions about the disclosure of the present invention and the following descriptions of the embodiments are used to demonstrate and explain the spirit and principle of the present invention, and to provide further explanation of the patent application scope of the present invention.

1:風力機齒輪箱之故障檢測系統 1: Fault detection system for wind turbine gearbox

11:風力發電機組 11:Wind turbine

111:齒輪箱 111: Gearbox

12:三軸振動感測器 12: Three-axis vibration sensor

13:可程式邏輯控制器 13: Programmable logic controller

14:電腦運算系統 14:Computer computing system

141:經驗模態分解法 141: Empirical Mode Decomposition Method

142:對稱點座標法 142: Symmetric point coordinate method

143:更快的區域卷積神經網路 143:Faster Region Convolutional Neural Networks

S10~S14:步驟 S10~S14: steps

S20~S24:步驟 S20~S24: steps

S30~S35:步驟 S30~S35: steps

S x :X軸訊號 S x : X axis signal

S y :Y軸訊號 S y : Y axis signal

S z :Z軸訊號 S z : Z axis signal

圖1為本發明的風力機齒輪箱之故障檢測系統的方塊示意圖。 FIG. 1 is a schematic block diagram of a fault detection system for a wind turbine gearbox of the present invention.

圖2為本發明的風力機齒輪箱之故障檢測方法的步驟流程圖。 FIG. 2 is a flow chart of the steps of the fault detection method for the wind turbine gearbox of the present invention.

圖3為本發明的經驗模態分解法的步驟流程圖。 Fig. 3 is a flow chart of the steps of the empirical mode decomposition method of the present invention.

圖4為本發明的更快的區域卷積神經網路的步驟流程圖。 FIG. 4 is a flowchart of the steps of the faster regional convolutional neural network of the present invention.

圖5為本發明的風力機齒輪箱之正常齒輪的示意圖。 Fig. 5 is a schematic diagram of a normal gear of the wind turbine gearbox of the present invention.

圖5A為本發明的正常齒輪之對稱的二維特徵圖。 Fig. 5A is a symmetrical two-dimensional characteristic diagram of a normal gear of the present invention.

圖6為本發明的風力機齒輪箱之齒輪崩齒的示意圖。 Fig. 6 is a schematic diagram of tooth chipping of the gear box of the wind turbine gearbox of the present invention.

圖6A為本發明的齒輪崩齒之對稱的二維特徵圖。 Fig. 6A is a symmetrical two-dimensional characteristic diagram of the tooth chipping of the gear of the present invention.

圖7為本發明的風力機齒輪箱之齒輪生鏽的示意圖。 Fig. 7 is a schematic diagram of rusted gears of the wind turbine gearbox of the present invention.

圖7A為本發明的齒輪生鏽之對稱的二維特徵圖。 Fig. 7A is a symmetrical two-dimensional characteristic diagram of the gear rusting of the present invention.

為使本發明實施例的目的、技術方案和優點更加清楚，下面將結合本發明實施例中的附圖，對本發明實施例中的技術方案進行清楚、完整地說明，顯然，所描述的實施例是本發明一部分實施例，而不是全部的實施例。基於本發明中的實施例，本領域普通技術人員在沒有做出創造性勞動前提下所獲得的所有其他實施例，都屬本發明保護的範圍。 In order to make the purpose, technical solutions and advantages of the embodiments of the present invention more clear, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by persons of ordinary skill in the art without making creative efforts fall within the protection scope of the present invention.

請參閱圖1，圖1為本發明的風力機齒輪之故障檢測系統的整體方塊圖。本發明實施例所述之風力機齒輪之故障檢測系統1，至少包括一風力發電機組11、一齒輪箱111、一三軸振動感測器12、一可程式邏輯控制器13及一電腦運算系統14。其中該齒輪箱111設置於該風力發電機組11內，並且該三軸振動感測器12與該齒輪箱111感應連接，該三軸振動感測器12與該可程式邏輯控制器13電性連接，該可程式邏輯控制器13與該電腦運算系統14網路連接。 Please refer to FIG. 1 . FIG. 1 is an overall block diagram of the wind turbine gear fault detection system of the present invention. The wind turbine gear fault detection system 1 described in the embodiment of the present invention at least includes a wind turbine 11 , a gear box 111 , a three-axis vibration sensor 12 , a programmable logic controller 13 and a computer computing system 14 . Wherein the gear box 111 is arranged in the wind power generating set 11, and the three-axis vibration sensor 12 is inductively connected with the gear box 111, the three-axis vibration sensor 12 is electrically connected with the programmable logic controller 13, and the programmable logic controller 13 is connected with the computer computing system 14 network.

以上，需要特別說明的是：本發明實施例之風力機齒輪之故障檢測系統1係透過該三軸振動感測器12來感測該風力發電機組11內的齒輪箱111轉動時的三軸方向的振動狀態，並產生出具有三軸方向(例如X軸、Y軸及Z軸)的多個振動訊號(例如 一X軸訊號Sx、一Y軸訊號Sy及一Z軸訊號Sz)，再透過該可程式邏輯控制器13來擷取該多個振動訊號，並將該多個振動訊號傳送給該電腦運算系統14進行分析，該電腦運算系統14會先透過一經驗模態分解法141將該多個振動訊號的各頻率分解成固有模態函數的形式，作為故障訊號的特徵資訊，然後再透過一對稱點座標法142將該已分解後的故障訊號的波形的振幅或頻率之間的差異轉換為極座標平面之定位點的位置差異與曲率變化，用以呈現出一對稱的二維特徵圖像，最後透過一更快的區域卷積神經網路143將該對稱的二維特徵圖像進行特徵提取以產生一第一特徵圖像，並將該第一特徵圖像透過多個錨框進行多個候選區域的生成，再將該多個候選區域進行分類，並將該分類後的候選區域與該第一特徵圖像相結合以產生一第二特徵圖像，再將該第二特徵圖像進行訓練與分類，當該第二特徵圖像分類完成並達到收斂，即透過一風力機齒輪箱故障瑕疵模型144來辦識出該齒輪箱的故障原因。 Above, it needs to be specifically explained that the wind turbine gear fault detection system 1 of the embodiment of the present invention uses the three-axis vibration sensor 12 to sense the vibration state of the three-axis direction when the gearbox 111 in the wind power generation unit 11 rotates, and produces a plurality of vibration signals (such as X-axis, Y-axis and Z-axis) with three-axis directions (such as X-axis, Y-axis and Z-axis) 1 X-axis signal Sx, a Y-axis signal Sthe yand a Z-axis signal Sz), and then use the programmable logic controller 13 to capture the multiple vibration signals, and send the multiple vibration signals to the computer computing system 14 for analysis. The computer computing system 14 will first decompose the frequencies of the multiple vibration signals into the form of natural mode functions through an empirical mode decomposition method 141, as the characteristic information of the fault signal, and then use a symmetrical point coordinate method 142. The difference between the amplitude or frequency of the waveform of the decomposed fault signal is converted into a polar coordinate plane. position difference and curvature change to present a symmetrical two-dimensional feature image, and finally perform feature extraction on the symmetrical two-dimensional feature image through a faster regional convolutional neural network 143 to generate a first feature image, and use the first feature image to generate multiple candidate regions through multiple anchor frames, then classify the multiple candidate regions, and combine the classified candidate regions with the first feature image to generate a second feature image, and then train and classify the second feature image. The gearbox fault defect model 144 is used to identify the fault cause of the gearbox.

請參閱圖2，圖2為本發明的風力機齒輪之故障檢測方法的步驟流程圖。本發明實施例所述之風力機齒輪之故障檢測方法，係用於檢測該風力發電機組11的一齒輪箱111的運轉狀態，其中該故障檢測方法包括如下步驟： Please refer to FIG. 2 . FIG. 2 is a flow chart of the steps of the wind turbine gear fault detection method of the present invention. The wind turbine gear fault detection method described in the embodiment of the present invention is used to detect the operating state of a gear box 111 of the wind power generating set 11, wherein the fault detection method includes the following steps:

第一步驟S10：接收設置於該齒輪箱內或表面的一三軸振動感測器，用以感測該齒輪箱轉動時的三軸方向的振動狀態，並產生出多個振動訊號； The first step S10: receiving a three-axis vibration sensor installed in or on the surface of the gear box to sense the vibration state of the three-axis direction when the gear box rotates, and generate a plurality of vibration signals;

第二步驟S11：將該多個振動訊號透過一可程式邏輯控制器傳送給一電腦運算系統； The second step S11: sending the plurality of vibration signals to a computer computing system through a programmable logic controller;

第三步驟S12：該電腦運算系統先透過一經驗模態分解法，將該多個振動訊號的各頻率分解成固有模態函數的形式，作為故障訊號的特徵資訊； The third step S12: the computer computing system decomposes each frequency of the plurality of vibration signals into the form of intrinsic mode functions through an empirical mode decomposition method, and uses it as characteristic information of the fault signal;

第四步驟S13：該電腦運算系統再透過一對稱點座標法，將該已分解後的故障訊號的波形的振幅或頻率之間的差異轉換為極座標平面之定位點的位置差異與曲率變化，用以呈現出一對稱的二維特徵圖像； The fourth step S13: the computer operation system converts the difference between the amplitude or frequency of the waveform of the decomposed fault signal into the position difference and curvature change of the positioning point on the polar coordinate plane through a symmetrical point coordinate method, so as to present a symmetrical two-dimensional characteristic image;

第五步驟S14：該電腦運算系統最後透過一更快的區域卷積神經網路，將該對稱的二維特徵圖像進行特徵提取以產生一第一特徵圖像，並將該第一特徵圖像透過多個錨框進行多個候選區域的生成，再將該多個候選區域進行分類，並將該分類後的候選區域與該第一特徵圖像相結合以產生一第二特徵圖像，再將該第二特徵圖像進行訓練與分類，當該分類均已完成並達到收斂，即透過一風力機齒輪箱故障瑕疵模型來辦識出該齒輪箱的故障類型。 The fifth step S14: the computer computing system finally uses a faster regional convolutional neural network to perform feature extraction on the symmetrical two-dimensional feature image to generate a first feature image, and uses the first feature image to generate multiple candidate regions through multiple anchor frames, and then classifies the multiple candidate regions, and combines the classified candidate regions with the first feature image to generate a second feature image, and then trains and classifies the second feature image. Type of failure of the gearbox.

請參閱圖3，圖3為本發明的經驗模態分解法141的步驟流程圖。本發明實施例的電腦運算系統14所採用的經驗模態分解法141的步驟，至少包括： Please refer to FIG. 3 . FIG. 3 is a flow chart of the steps of the empirical mode decomposition method 141 of the present invention. The steps of the empirical mode decomposition method 141 adopted by the computer computing system 14 of the embodiment of the present invention at least include:

第一步驟S20：先找出原始訊號的所有極大值和極小值，再將該所有極大值串連成上包絡線u 0(t)，將該所有極小值串連成下包絡線v 0(t)。 The first step S20: find out all the maximum and minimum values of the original signal, and then concatenate all the maximum values to form the upper envelope u 0 ( t ), and concatenate all the minimum values to form the lower envelope v 0 ( t ).

第二步驟S21：求出上下包絡線之平均，得到均值包絡線m 0(t)。其中該均值包絡線的方程式為： Second step S21: Calculate the average of the upper and lower envelopes to obtain the mean envelope m 0 ( t ). where the equation of the mean envelope is:

Figure 111133237-A0101-12-0005-1
Figure 111133237-A0101-12-0005-1

其中該u 0(t)為上包絡線；v 0(t)為下包絡線。 Wherein the u 0 ( t ) is the upper envelope; v 0 ( t ) is the lower envelope.

第三步驟S22：將原始訊號與均值包絡線相減，得到一第一分量h 1(t)。 Third step S22: Subtract the original signal from the mean envelope to obtain a first component h 1 ( t ).

第四步驟S23：檢查該第一分量是否符合一固有模態函數(Intrinsic Mode Function，簡稱IMF)的條件，如果不符合，則回到第一步驟S20，並將該第一分量當作原始訊號，並進行第二次的篩選，並重復篩選k次計算，直到該第一分量符合該固有模態函數的條件，即得到一第一固有模態函數分量。其中該進行第二次的篩選，並重復篩選k次計算的方程式為：

Figure 111133237-A0305-02-0007-1
其中該m1(t)為第二次的均值包絡線；該u1(t)為第二次的上包絡線；該v1(t)為第二次的下包絡線；h2(t)為第二次分量；hi(t)為第一次分量；mk-1(t)為k-1次的均值包絡線；該uk-1(t)為k-1次的上包絡線；該vk-1(t)為k-1次的下包絡線；hk(t)為k次分量；hk-1(t)為k-1次分量。 The fourth step S23: check whether the first component meets the condition of an intrinsic mode function (IMF), if not, return to the first step S20, and use the first component as the original signal, and perform a second screening, and repeat the calculation for k times until the first component meets the condition of the intrinsic mode function, that is, a first intrinsic mode function component is obtained. The equation for performing the second screening and repeating the calculation for k times is:
Figure 111133237-A0305-02-0007-1
 Wherein the m 1 (t) is the second average envelope; the u 1 (t) is the second upper envelope; the v 1 (t) is the second lower envelope; h 2 (t) is the second component; h i (t) is the first component; m k -1 ( t) is the k-1 average envelope; The lower envelope; h k (t) is the kth component; h k-1 (t) is the k-1 component.

第五步驟S24：原始訊號減去該第一固有模態函數分量用以得到一第一剩餘量，再將該第一剩餘量當作新的資料，重新執行第一步驟至第五步驟，以得到新的一第二剩餘量，如此重覆n次，當第n個剩餘量已成為一單調函數或其值小於預先給定的值時，則分解過程完成。其中該取得第一剩餘量r 1(t)的方程式為：

Figure 111133237-A0305-02-0007-2
The fifth step S24: subtracting the first intrinsic mode function component from the original signal to obtain a first residual quantity, and then taking the first residual quantity as new data, re-executing the first step to the fifth step to obtain a new second residual quantity, repeating this n times, when the nth residual quantity has become a monotonic function or its value is less than a predetermined value, the decomposition process is completed. Wherein the equation for obtaining the first remaining amount r 1 ( t ) is:
Figure 111133237-A0305-02-0007-2

其中該h k (t)為k次分量，設為y 1(t)，因此該y 1(t)就成為第一分量；r 1(t)為對應的第一剩餘分量；x(t)為原始信號。 The h k ( t ) is the k -order component, which is set to y 1 ( t ), so the y 1 ( t ) becomes the first component; r 1 ( t ) is the corresponding first residual component; x ( t ) is the original signal.

其中該取得第二剩餘量r 2(t)與n個剩餘量r n (t)的方程式為： Wherein the equation for obtaining the second residual amount r 2 ( t ) and n residual amounts r n ( t ) is:

Figure 111133237-A0101-12-0007-4
Figure 111133237-A0101-12-0007-4

其中該y 2(t)為第二分量；y n (t)為第n次分量；r 1(t)為對應的第一剩餘分量；r n-1(t)為對應的第n個剩餘分量。 Wherein the y 2 ( t ) is the second component; y n ( t ) is the nth component; r 1 ( t ) is the corresponding first residual component; r n -1 ( t ) is the corresponding nth residual component.

本發明實施例的電腦運算系統14所採用的對稱點座標法142的極座標平面的定位點具有一極座標的半徑γ(i)、一極座標的初始線之順時針旋轉角度α cw (i)及一極座標的初始線之逆時針旋轉角度α ccw (i)。其中該極座標的半徑γ(i)的方程式為： The positioning point of the polar coordinate plane of the symmetrical point coordinate method 142 adopted by the computer operation system 14 of the embodiment of the present invention has a polar coordinate radius γ ( i ), a clockwise rotation angle α cw ( i ) of the initial line of the polar coordinates, and a counterclockwise rotation angle α ccw ( i ) of the initial line of the polar coordinates. Wherein the equation of the radius γ ( i ) of the polar coordinate is:

Figure 111133237-A0101-12-0007-5
Figure 111133237-A0101-12-0007-5

其中該x min 為振動數值的最小振幅值；x max 為振動數值的最大振幅值；x i 為假設時間i的振動數值。 Wherein the x min is the minimum amplitude value of the vibration value; x max is the maximum amplitude value of the vibration value; x i is the vibration value at the assumed time i.

其中該極座標的初始線之順時針旋轉角度α cw (i)的方程式為： The equation of the clockwise rotation angle α cw ( i ) of the initial line of the polar coordinates is:

Figure 111133237-A0101-12-0007-6
Figure 111133237-A0101-12-0007-6

其中該x min 為振動數值的最小振幅值；x max 為振動數值的最大振幅值；x i 為假設時間i的振動數值；ψ為初始旋轉角度；△T是時間間隔(範圍值1~10之間)；k是旋轉角的放大倍數。 Among them, x min is the minimum amplitude value of the vibration value; x max is the maximum amplitude value of the vibration value; x i is the vibration value of the hypothetical time i ; ψ is the initial rotation angle; △ T is the time interval (between 1 and 10) ;

其中該極座標的初始線之逆時針旋轉角度α ccw (i)的方程式為： The equation of the counterclockwise rotation angle α ccw ( i ) of the initial line of the polar coordinates is:

Figure 111133237-A0101-12-0007-7
Figure 111133237-A0101-12-0007-7

其中該x min 為振動數值的最小振幅值；x max 為振動數值的最大振幅值；x i 為假設時間i的振動數值；ψ為初始旋轉角度；△T是時間間隔(範圍值1~10之間)；k是旋轉角的放大倍數。 Among them, x min is the minimum amplitude value of the vibration value; x max is the maximum amplitude value of the vibration value; x i is the vibration value of the hypothetical time i ; ψ is the initial rotation angle; △ T is the time interval (between 1 and 10) ;

請參閱圖4，圖4為本發明的更快的區域卷積神經網路143的步驟流程圖。本發明實施例的電腦運算系統14所採用的更快的區域卷積神經網路143的步驟，至少包括： Please refer to FIG. 4 , which is a flowchart of the steps of the faster region convolutional neural network 143 of the present invention. The steps of the faster regional convolutional neural network 143 adopted by the computer computing system 14 of the embodiment of the present invention at least include:

第一步驟S30：將對稱的二維特徵圖像輸入至共享卷積層(Shared Convolutional Layers)以進行特徵提取以產生出第一特徵圖像； The first step S30: Inputting the symmetrical two-dimensional feature image to the Shared Convolutional Layers (Shared Convolutional Layers) for feature extraction to generate a first feature image;

第二步驟S31：將該第一特徵圖像輸入一區域候選網路，並透過多個錨框(Anchor Box)進行多個候選區域的生成； The second step S31: input the first feature image into a region candidate network, and generate a plurality of candidate regions through a plurality of anchor boxes (Anchor Box);

第三步驟S32：將生成的候選區域進行聯合交集(intersection over union,IoU)，並依據該聯合交集後的客觀分數來將該候選區域進行分類；其中當該聯合交集>0.7，則該錨框為正標籤；該聯合交集<0.3，則該錨框為負標籤；當該聯合交集<0.3<0.7，則排除該錨框。其中該聯合交集的一多任務損失(Multitask loss，稱稱L)的方程式為： The third step S32: Perform joint intersection over union (IoU) on the generated candidate areas, and classify the candidate areas according to the objective score after the joint intersection; when the joint intersection > 0.7, the anchor box is a positive label; the joint intersection < 0.3, the anchor box is a negative label; when the joint intersection < 0.3 < 0.7, the anchor box is excluded. The equation of a multitask loss (Multitask loss, called L) of the joint intersection is:

Figure 111133237-A0101-12-0008-9
Figure 111133237-A0101-12-0008-9

其中該p為預測分類；u為真實分類；t u 為預測平移縮放參數；v為真實平移縮放參數；L cls 為分類損失；L reg 為回歸損失；

Figure 111133237-A0101-12-0008-35
為真實標籤；λ為平衡參數。 Among them, p is the predicted classification; u is the real classification; t u is the predicted translation scaling parameter; v is the real translation scaling parameter; L cls is the classification loss; L reg is the regression loss;
Figure 111133237-A0101-12-0008-35
is the real label; λ is the balance parameter.

其中該多任務損失的一最小化目標函數的方程式為： Wherein the equation of a minimization objective function of the multi-task loss is:

Figure 111133237-A0101-12-0008-8
Figure 111133237-A0101-12-0008-8

其中該N cls 為小批次的大小；L cls 為分類損失；L reg 為回歸損失；p i 為預測概率；

Figure 111133237-A0101-12-0008-36
為真實標籤；i為錨框的索引編號；λ為平衡參數； N reg 為錨框位置的數量；t i 為預測邊框的座標向量；
Figure 111133237-A0101-12-0009-30
為實際邊框的座標向量。 Among them, the N cls is the size of the small batch; L cls is the classification loss; L reg is the regression loss; p i is the predicted probability;
Figure 111133237-A0101-12-0008-36
is the real label; i is the index number of the anchor frame; λ is the balance parameter; N reg is the number of anchor frame positions; t i is the coordinate vector of the predicted frame;
Figure 111133237-A0101-12-0009-30
is the coordinate vector of the actual bounding box.

第四步驟S33：將該分類後的候選區域與該第一特徵圖像相結合以產生一第二特徵圖像，並將該第二特徵圖像輸入至感興趣區域(Region of Interest，簡稱ROI)的池化層中進行映射； The fourth step S33: combining the classified candidate region with the first feature image to generate a second feature image, and inputting the second feature image into a pooling layer of a region of interest (Region of Interest, ROI for short) for mapping;

第五步驟S34：再將映射後的該第二特徵圖像輸入至一全連接層(Fully Connected Layer)進行訓練，再將訓練後的結果輸出至一回歸層(Box-regression)與分類層(Box-classification)，將該候選區域所篩選出來的視窗進行匹配與分類；其中該回歸層具有四個預測偏移量及四個真實偏移量，其中該四個預測偏移量t x ,t y ,t w ,t h 的方程式為： The fifth step S34: input the mapped second feature image to a fully connected layer (Fully Connected Layer) for training, and then output the training result to a regression layer (Box-regression) and a classification layer (Box-classification), and perform matching and classification on the windows selected from the candidate area; wherein the regression layer has four predicted offsets and four real offsets, wherein the equations of the four predicted offsets t x , ty , t w , t h are:

Figure 111133237-A0101-12-0009-10
Figure 111133237-A0101-12-0009-10

其中該x,y為預測邊框中心點的座標，w為預測邊框的寬度，h為預測邊框的高度；該x a ,y a 為錨框中心點的座標，w a 為錨框的寬度，h a 為錨框的高度。其中該回饋層中的四個真實偏移量

Figure 111133237-A0101-12-0009-31
,
Figure 111133237-A0101-12-0009-32
,
Figure 111133237-A0101-12-0009-33
,
Figure 111133237-A0101-12-0009-34
的方程式為： The x, y are the coordinates of the center point of the predicted frame, w is the width of the predicted frame, h is the height of the predicted frame; the x a , y a are the coordinates of the center point of the anchor frame, w a is the width of the anchor frame, h a is the height of the anchor frame. where the four true offsets in the feedback layer
Figure 111133237-A0101-12-0009-31
,
Figure 111133237-A0101-12-0009-32
,
Figure 111133237-A0101-12-0009-33
,
Figure 111133237-A0101-12-0009-34
The equation is:

Figure 111133237-A0101-12-0009-11
Figure 111133237-A0101-12-0009-11

其中該x * ,y *為實際邊框中心點的座標，w *為實際邊框的寬度，h *為實際邊框的高度；該x a ,y a 為錨框中心點的座標，w a 為錨框的寬度，h a 為錨框的高度。 The x * , y * are the coordinates of the center point of the actual border, w * is the width of the actual border, h * is the height of the actual border; the x a , y a are the coordinates of the center point of the anchor frame, w a is the width of the anchor frame, and h a is the height of the anchor frame.

第六步驟S35：當該映射後的第二特徵圖的分類已完成並達到收斂，則透過一風力機齒輪箱故障瑕疵模型來辦識出該齒輪箱 的故障類型；反之，當該映射後的第二特徵圖的分類未完成也未達到收斂，則返回到第三步驟。 The sixth step S35: when the classification of the mapped second feature map has been completed and converged, the gearbox is identified through a fault model of a wind turbine gearbox On the contrary, when the classification of the second feature map after the mapping is not completed and convergence is not reached, return to the third step.

本發明實施例的風力機齒輪箱故障瑕疵模型144至少具有3種齒輪箱的模型： The fault defect model 144 of the wind turbine gearbox in the embodiment of the present invention has at least three models of gearboxes:

第一種為齒輪箱111內的齒輪1111為正常時，為如圖5所示，其呈現出的對稱的二維特徵圖為如圖5A所示。 The first one is when the gear 1111 in the gear box 111 is normal, as shown in FIG. 5 , and its symmetrical two-dimensional characteristic diagram is shown in FIG. 5A .

第二種為齒輪箱111內的齒輪1111出現崩齒1111a，為如圖6所示，其呈現出的對稱二維特徵圖為如圖6A所示。 The second type is that the gear 1111 in the gear box 111 has tooth chipping 1111a, as shown in FIG. 6 , and its symmetrical two-dimensional characteristic diagram is shown in FIG. 6A .

第三種為齒輪箱111內的齒輪1111已生鏽1111b，為如圖7所示，其呈現出的對稱二維特徵圖為如圖7A所示。 The third type is that the gear 1111 in the gear box 111 has rusted 1111b, as shown in FIG. 7 , and its symmetrical two-dimensional characteristic diagram is shown in FIG. 7A .

前述本實施例所揭示的圖5、圖6及圖7的齒輪1111僅為用於方便說明，並不以此為限。在其它實施例中，也可以使用例如小齒輪軸、原動軸齒輪等結構。 The aforementioned gear 1111 disclosed in FIG. 5 , FIG. 6 and FIG. 7 in this embodiment is only for convenience of description and is not limited thereto. In other embodiments, structures such as pinion shafts, drive shaft gears, etc. may also be used.

雖然本發明以前述之諸項實施例揭露如上，然其並非用以限定本發明，任何熟習相像技藝者，在不脫離本發明之精神和範圍內，當可作些許之更動與潤飾，因此本發明之專利保護範圍須視本說明書所附之申請專利範圍所界定者為準。 Although the present invention is disclosed above with the above-mentioned various embodiments, it is not intended to limit the present invention. Any person familiar with similar skills may make some changes and modifications without departing from the spirit and scope of the present invention. Therefore, the scope of patent protection of the present invention must be defined by the patent scope attached to this specification.

S10~S14:步驟 S10~S14: steps

Claims (10)

一種風力機齒輪箱之故障檢測方法，用於檢測一風力發電機組的一齒輪箱的運作狀態，其中該故障檢測方法包括如下步驟：第一步驟：接收設置於該齒輪箱內或表面的一三軸振動感測器，用以感測該齒輪箱轉動時的三軸方向的振動狀態，並產生出多個振動訊號；第二步驟：將該多個振動訊號透過一可程式邏輯控制器傳送給一電腦運算系統；第三步驟：該電腦運算系統先透過一經驗模態分解法，將該多個振動訊號的各頻率分解成固有模態函數的形式，作為一故障訊號的特徵資訊；第四步驟：該電腦運算系統再透過一對稱點座標法，將已分解後之該故障訊號的特徵資訊波形的振幅或頻率之間的差異轉換為極座標平面之定位點的位置差異與曲率變化，用以呈現出一對稱的二維特徵圖像；第五步驟：該電腦運算系統最後透過一更快的區域卷積神經網路，將該對稱的二維特徵圖像進行特徵提取以產生一第一特徵圖像，並將該第一特徵圖像透過多個錨框進行多個候選區域的生成，再將該多個候選區域進行分類，並將該分類後的候選區域與該第一特徵圖像相結合以產生一第二特徵圖像，再將該第二特徵圖像進行訓練與分類，當該分類均已完成並達到收斂，即透過一風力機齒輪箱故障瑕疵模型來辦識出該齒輪箱的故障類型。 A fault detection method for a wind turbine gearbox, which is used to detect the operating state of a gearbox of a wind power generating set, wherein the fault detection method includes the following steps: the first step: receiving a three-axis vibration sensor installed in or on the surface of the gearbox to sense the vibration state of the three-axis direction when the gearbox rotates, and generating multiple vibration signals; the second step: sending the multiple vibration signals to a computer computing system through a programmable logic controller; Each frequency of the vibration signal is decomposed into the form of an intrinsic mode function, which is used as the characteristic information of a fault signal; the fourth step: the computer operation system converts the difference between the amplitude or frequency of the decomposed characteristic information waveform of the fault signal into the position difference and curvature change of the positioning point on the polar coordinate plane through a symmetrical point coordinate method, so as to present a symmetrical two-dimensional feature image; the fifth step: the computer operation system finally uses a faster regional convolutional neural network to perform feature extraction on the symmetrical two-dimensional feature image to generate a first feature The first feature image is used to generate multiple candidate areas through multiple anchor frames, and then the multiple candidate areas are classified, and the classified candidate area is combined with the first feature image to generate a second feature image, and then the second feature image is trained and classified. When the classification is completed and converges, the fault type of the gearbox is identified through a wind turbine gearbox fault defect model. 如請求項1所述的風力機齒輪箱之故障檢測方法，其中所述之該經驗模態分解法可具有如下步驟： 第一步驟：先找出原始訊號的所有極大值和極小值，再將該所有極大值串連成上包絡線，將該所有極小值串連成下包絡線；第二步驟：求出上下包絡線之平均，得到均值包絡線；第三步驟：將原始訊號與均值包絡線相減，得到一第一分量；第四步驟：檢查該第一分量是否符合一固有模態函數的條件，如果不符合，則回到第一步驟，並將該第一分量當作原始訊號，並進行第二次的篩選，並重覆篩選k次計算，直到該第一分量符合該固有模態函數的條件，即得到一第一固有模態函數分量；第五步驟；原始訊號減去該第一固有模態函數分量用以得到一第一剩餘量，再將該第一剩餘量當作新的資料，重新執行第一步驟至第五步驟，以得到新的一第二剩餘量，如此重覆n次後，當第n個剩餘量已成為一單調函數或其值小於預先給定的值時，則分解過程完成。 The fault detection method of the wind turbine gearbox as described in claim 1, wherein said empirical mode decomposition method may have the following steps: The first step: find all the maximum and minimum values of the original signal, and then concatenate all the maximum values into the upper envelope, and connect all the minimum values into the lower envelope; the second step: find the average of the upper and lower envelopes to obtain the mean envelope; the third step: subtract the original signal from the mean envelope to obtain a first component; the fourth step: check whether the first component meets the conditions of an intrinsic mode function, if not, return to the first step, and use the first component as the original signal, and perform the second step times of screening, and repeated screeningkcalculation until the first component meets the condition of the intrinsic mode function, that is, a first intrinsic mode function component is obtained; the fifth step: subtracting the first intrinsic mode function component from the original signal to obtain a first residual quantity, and then taking the first residual quantity as new data, re-executing the first step to the fifth step to obtain a new second residual quantity, after repeating n times, when the nth residual quantity has become a monotonic function or its value is less than a predetermined value, the decomposition process is completed. 如請求項2所述的風力機齒輪箱之故障檢測方法，其中該均值包絡線的方程式為：
Figure 111133237-A0305-02-0014-3
 其中該u 0(t)為上包絡線；v 0(t)為下包絡線。 The fault detection method of the wind turbine gearbox as described in claim 2, wherein the equation of the mean value envelope is:
Figure 111133237-A0305-02-0014-3
 Wherein the u 0 ( t ) is the upper envelope; v 0 ( t ) is the lower envelope. 如請求項2所述的風力機齒輪箱之故障檢測方法，其中該進行第二次的篩選，並重復篩選k次計算的方程式為：
Figure 111133237-A0305-02-0014-4
 其中該m 1(t)為第二次的均值包絡線；該u 1(t)為第二次的上包絡線；該v 1(t)為第二次的下包絡線；h 2(t)為第二次分量；h 1(t)為第一次分量；m k-1(t)為k-1次的均值包 絡線；該u k-1(t)為k-1次的上包絡線；該v k-1(t)為k-1次的下包絡線；h k (t)為k次分量；h k-1(t)為k-1次分量。 The fault detection method of the wind turbine gearbox as described in claim 2, wherein the second screening is carried out, and the equation for repeating the calculation of k times of screening is:
Figure 111133237-A0305-02-0014-4
 Wherein the m 1 ( t ) is the second average envelope; the u 1 ( t ) is the second upper envelope; the v 1 ( t ) is the second lower envelope; h 2 ( t ) is the second component; h 1 ( t ) is the first component; m k -1 ( t ) is the k -1 average envelope; the u k -1 ( t ) is the k -1 upper envelope; the v k -1 ( t ) is the lower envelope of degree k -1; h k ( t ) is the component of degree k; h k -1 ( t ) is the component of degree k -1. 如請求項2所述的風力機齒輪箱之故障檢測方法，其中取得該第一剩餘量的方程式為：
Figure 111133237-A0305-02-0015-5
 其中該h k (t)為k次分量，設為y 1(t)，因此該y 1(t)就成為第一分量；r 1(t)為對應的第一剩餘分量；x(t)為原始信號。 The fault detection method of the wind turbine gearbox as described in claim 2, wherein the equation for obtaining the first remaining amount is:
Figure 111133237-A0305-02-0015-5
 The h k ( t ) is the k -order component, which is set to y 1 ( t ), so the y 1 ( t ) becomes the first component; r 1 ( t ) is the corresponding first residual component; x ( t ) is the original signal. 如請求項2所述的風力機齒輪箱之故障檢測方法，其中取得該第二剩餘量與n個剩餘量的方程式為：
Figure 111133237-A0305-02-0015-6
 其中該y 2(t)為第二分量；y n (t)為第n次分量；r 1(t)為對應的第一剩餘分量；r n-1(t)為對應的第n個剩餘分量。 The fault detection method for a wind turbine gearbox as described in Claim 2, wherein the equations for obtaining the second remaining quantity and the n remaining quantities are:
Figure 111133237-A0305-02-0015-6
 Wherein the y 2 ( t ) is the second component; y n ( t ) is the nth component; r 1 ( t ) is the corresponding first residual component; r n -1 ( t ) is the corresponding nth residual component. 如請求項1所述的風力機齒輪箱之故障檢測方法，其中該更快的區域卷積神經網路具有如下步驟：第一步驟：將對稱的二維特徵圖像輸入至共享卷積層以進行特徵提取以產生出第一特徵圖像；第二步驟：將該第一特徵圖像輸入一區域候選網路，並透過多個錨框進行多個候選區域的生成；第三步驟：將生成的候選區域進行聯合交集，並依據該聯合交集後的客觀分數來將該候選區域進行分類；第四步驟：將該分類後的候選區域與該第一特徵圖像相結合以產生一第二特徵圖像，並將該第二特徵圖像輸入至感興趣區域的池化層中進行映射； 第五步驟；再將映射後的該第二特徵圖像輸入至一全連接層進行訓練，再將訓練後的結果輸出至一回歸層與一分類層，將該候選區域所篩選出來的視窗進行匹配與分類；第六步驟：當該映射後的第二特徵圖的分類已完成並達到收斂，則透過一風力機齒輪箱故障瑕疵模型來辦識出該齒輪箱的故障原因；反之，當該映射後的第二特徵圖的分類未完成也未達到收斂，則返回到第三步驟。 The fault detection method for a wind turbine gearbox as described in Claim 1, wherein the faster regional convolutional neural network has the following steps: first step: inputting a symmetrical two-dimensional feature image into a shared convolutional layer for feature extraction to generate a first feature image; second step: inputting this first feature image into a region candidate network, and generating multiple candidate regions through multiple anchor frames; third step: performing joint intersection of the generated candidate regions, and classifying the candidate region according to the objective score after the joint intersection; fourth step: combining the classified candidate region with The first feature images are combined to generate a second feature image, and the second feature image is input to the pooling layer of the region of interest for mapping; The fifth step; then input the second feature image of the mapping to the one -full connection layer for training, and then output the training results to the one -return layer with a classification layer to match the window screened by the candidate area; Six Step: When the classification of the second feature diagram after the mapping has been completed and convergence, it can be learned through a wind machine gear chart failure model to understand The cause of the failure of the gear box; on the contrary, when the classification of the second feature diagram after the mapping is not completed or convergence, it returns to the third step. 如請求項7所述的風力機齒輪箱之故障檢測方法，其中該聯合交集的一多任務損失的方程式為：
Figure 111133237-A0305-02-0016-8
 其中該p為預測分類；u為真實分類；t u 為預測平移縮放參數；v為真實平移縮放參數；L cls 為分類損失；L reg 為回歸損失；
Figure 111133237-A0305-02-0016-18
為真實標籤；λ為平衡參數。 The fault detection method of the wind turbine gearbox as described in claim item 7, wherein the equation of a multi-task loss of the joint intersection is:
Figure 111133237-A0305-02-0016-8
 Among them, p is the predicted classification; u is the real classification; t u is the predicted translation scaling parameter; v is the real translation scaling parameter; L cls is the classification loss; L reg is the regression loss;
Figure 111133237-A0305-02-0016-18
is the real label; λ is the balance parameter. 如請求項8所述的風力機齒輪箱之故障檢測方法，其中該多任務損失的一最小化目標函數的方程式為：
Figure 111133237-A0305-02-0016-7
 其中該Ncls為小批次的大小；Lcls為分類損失；Lreg為回歸損失；pi為預測概率；
Figure 111133237-A0305-02-0016-20
為真實標籤；i為錨框的索引編號；λ為平衡參數；Nreg為錨框位置的數量；ti為預測邊框的座標向量；
Figure 111133237-A0305-02-0016-19
為實際邊框的座標向量。 The fault detection method of a wind turbine gearbox as described in Claim 8, wherein the equation of a minimum objective function of the multi-task loss is:
Figure 111133237-A0305-02-0016-7
 Among them, the N cls is the size of the small batch; L cls is the classification loss; L reg is the regression loss; p i is the predicted probability;
Figure 111133237-A0305-02-0016-20
is the real label; i is the index number of the anchor frame; λ is the balance parameter; N reg is the number of anchor frame positions; t i is the coordinate vector of the predicted frame;
Figure 111133237-A0305-02-0016-19
is the coordinate vector of the actual bounding box. 如請求項7所述的風力機齒輪箱之故障檢測方法，其中該回歸層具有四個預測偏移量及四個真實偏移量，其中該四個預測偏移量的方程式為：
Figure 111133237-A0305-02-0016-9
，
Figure 111133237-A0305-02-0016-10
，
Figure 111133237-A0305-02-0016-11
，
Figure 111133237-A0305-02-0016-12
；其中該四個真實偏移量的方程式為；
Figure 111133237-A0305-02-0016-13
，
Figure 111133237-A0305-02-0016-14
，
Figure 111133237-A0305-02-0016-15
，
Figure 111133237-A0305-02-0016-17
； 其中該x,y為預測邊框中心點的座標，w為預測邊框的寬度，h為預測邊框的高度；其中該x * ,y *為實際邊框中心點的座標，w *為實際邊框的寬度，h *為實際邊框的高度；其中該x a ,y a 為錨框中心點的座標，w a 為錨框的寬度，h a 為錨框的高度。 The fault detection method of the wind turbine gearbox as described in claim 7, wherein the regression layer has four predicted offsets and four real offsets, wherein the equations of the four predicted offsets are:
Figure 111133237-A0305-02-0016-9
,
Figure 111133237-A0305-02-0016-10
,
Figure 111133237-A0305-02-0016-11
,
Figure 111133237-A0305-02-0016-12
; The equations of the four real offsets are;
Figure 111133237-A0305-02-0016-13
,
Figure 111133237-A0305-02-0016-14
,
Figure 111133237-A0305-02-0016-15
,
Figure 111133237-A0305-02-0016-17
Wherein the x, y are the coordinates of the center point of the predicted frame, w is the width of the predicted frame, and h is the height of the predicted frame; wherein the x * , y * are the coordinates of the center point of the actual frame, w * is the width of the actual frame, h * is the height of the actual frame; wherein the x a , y a are the coordinates of the center point of the anchor frame, w a is the width of the anchor frame, and h a is the height of the anchor frame.
TW111133237A 2022-09-01 2022-09-01 Fault detection method of wind turbine gearbox TWI807985B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
TW111133237A TWI807985B (en) 2022-09-01 2022-09-01 Fault detection method of wind turbine gearbox

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
TW111133237A TWI807985B (en) 2022-09-01 2022-09-01 Fault detection method of wind turbine gearbox

Publications (2)

Publication Number Publication Date
TWI807985B true TWI807985B (en) 2023-07-01
TW202411532A TW202411532A (en) 2024-03-16

Family

ID=88149125

Family Applications (1)

Application Number Title Priority Date Filing Date
TW111133237A TWI807985B (en) 2022-09-01 2022-09-01 Fault detection method of wind turbine gearbox

Country Status (1)

Country Link
TW (1) TWI807985B (en)

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
TW201120310A (en) * 2009-12-09 2011-06-16 Nat Univ Chin Yi Technology The state telemetry technology and fault diagnosing system in large-scale wind power farms
CN114034481A (en) * 2021-11-15 2022-02-11 燕山大学 Fault diagnosis system and method for rolling mill gearbox
TW202208746A (en) * 2020-08-20 2022-03-01 國立勤益科技大學 Fault diagnosis system and method for wind turbine wherein the hidden faults in the operation process can be found in time through the real-time monitoring and trend prediction of the operating status of wind power generation equipment
CN114659790A (en) * 2022-03-14 2022-06-24 浙江工业大学 Method for identifying bearing fault of variable-speed wind power high-speed shaft

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
TW201120310A (en) * 2009-12-09 2011-06-16 Nat Univ Chin Yi Technology The state telemetry technology and fault diagnosing system in large-scale wind power farms
TW202208746A (en) * 2020-08-20 2022-03-01 國立勤益科技大學 Fault diagnosis system and method for wind turbine wherein the hidden faults in the operation process can be found in time through the real-time monitoring and trend prediction of the operating status of wind power generation equipment
CN114034481A (en) * 2021-11-15 2022-02-11 燕山大学 Fault diagnosis system and method for rolling mill gearbox
CN114659790A (en) * 2022-03-14 2022-06-24 浙江工业大学 Method for identifying bearing fault of variable-speed wind power high-speed shaft

Also Published As

Publication number Publication date
TW202411532A (en) 2024-03-16

Similar Documents

Publication Publication Date Title
US11514567B2 (en) On-line real-time diagnosis system and method for wind turbine blade (WTB) damage
van Dinter et al. Predictive maintenance using digital twins: A systematic literature review
Kumar et al. Latest developments in gear defect diagnosis and prognosis: A review
CN107941537B (en) A kind of mechanical equipment health state evaluation method
CN106555788B (en) Application based on the deep learning of Fuzzy Processing in hydraulic equipment fault diagnosis
KR102171491B1 (en) Method for sorting products using deep learning
Henriksson et al. Towards structured evaluation of deep neural network supervisors
CN107064170A (en) One kind detection phone housing profile tolerance defect method
CN114778112A (en) Audio identification and fault diagnosis method for mechanical fault of wind turbine generator system
Zhao et al. Multiscale graph-guided convolutional network with node attention for intelligent health state diagnosis of a 3-PRR planar parallel manipulator
CN112907562A (en) MobileNet-based SMT defect classification algorithm
TWI807985B (en) Fault detection method of wind turbine gearbox
Choung et al. Automatic discontinuity classification of wind-turbine blades using A-scan-based convolutional neural network
CN114627074A (en) Ground shooting fan blade real-time monitoring method based on deep learning
Memari et al. Review on the advancements in wind turbine blade inspection: Integrating drone and deep learning technologies for enhanced defect detection
Pu et al. AutoRepo: A general framework for multimodal LLM-based automated construction reporting
CN116258062B (en) Carton packaging machinery die cutter abrasion prediction method and device based on quality detection image
CN115508652A (en) Method and system for positioning and detecting defects of electric power system
CN111703557B (en) Virtual test system for intelligent ship swaying performance analysis
Dighvijay et al. A Faster R-CNN implementation of presence inspection for parts on industrial produce
CN113052296A (en) Power grid power transmission defect intelligent management system based on deep learning convolutional neural network technology
Zhang Reducing uncertainty in wind turbine blade health inspection with image processing techniques
Amin et al. An Uncertainty-Aware Health Monitoring Model for Wind Turbine Drivetrains Based On Bayesian Neural Network
Schmidt et al. An automated optical inspection system for PIP solder joint classification using convolutional neural networks
TWI845272B (en) Wind turbine gearbox fault diagnosis method