TWI742976B - Structure diagnosis system and structure diagnosis method - Google Patents

The present disclosure provides a structure diagnosis system and a structure diagnosis method. The structure diagnosis system includes: a lidar scanner scanning a structure to generate a point cloud data; an input interface receiving the point cloud data; and a processor receiving the point cloud data and generating a point cloud data set. The processor executes a surface degradation and geometry abnormal coupling diagnosis module to: marking a first point cloud range of a surface degradation area according to color space value of the point cloud data set; marking a second point cloud range of a geometry abnormal area according to coordinate value of the point cloud data set; when an abnormal area includes the first point cloud range and the second point cloud range at least partially overlapping each other, determining surface degradation or geometry abnormal occurring at the abnormal area and mark the abnormal area with a predetermined mode.

結構體診斷系統及結構體診斷方法Structure diagnosis system and structure diagnosis method

本揭露是有關於一種結構體診斷系統及結構體診斷方法，且特別是有關於一種自動診斷結構體損傷的結構體診斷系統及結構體診斷方法。The present disclosure relates to a structure diagnosis system and a structure diagnosis method, and particularly relates to a structure diagnosis system and a structure diagnosis method for automatically diagnosing structure damage.

全球工業設備及公共設施每年因力學損傷(或稱為幾何異常)及腐蝕損傷(或稱為表面劣化)所造成之損失超過千億美元。油槽洩漏、斷橋事件及太陽能板支撐架腐蝕等事件，皆造成嚴重損失及影響產業鏈供輸穩定。結構設施規模廣大且數量繁多且力學損傷及腐蝕損傷需仰賴不同診斷方法。由於專業人力不足，造成結構設施檢查效率低並難以滿足國內各領域設施的安全管理需求。The global industrial equipment and public facilities lose more than 100 billion U.S. dollars each year due to mechanical damage (or geometric anomaly) and corrosion damage (or surface degradation). Incidents such as oil tank leaks, broken bridges, and corrosion of solar panel support frames have caused serious losses and affected the stability of supply and transmission in the industrial chain. The structure and facilities are large in scale and numerous in number, and mechanical damage and corrosion damage need to rely on different diagnostic methods. Due to the lack of professional manpower, the inspection efficiency of structural facilities is low and it is difficult to meet the safety management needs of domestic facilities in various fields.

在力學損傷方面通常以人力局部安裝應變計進行幾何形狀變異量計算，但這難以察覺法規規定之缺陷大小。在腐蝕損傷方面通常以人力近距離目視進行表面色澤外觀檢查，並依照嚴重程度評級決定處理順序。上述方法都非常仰賴專業人員的現場判斷因此需要花費大量的時間與人力來進行。因此，如何能更有效率地診斷結構體損傷是本領域技術人員應致力的目標。In terms of mechanical damage, it is usually calculated by manually installing strain gauges locally, but it is difficult to detect the size of the defects stipulated by laws and regulations. In terms of corrosion damage, the surface color and appearance are usually inspected by human close-up visual inspection, and the processing order is determined according to the severity rating. The above methods rely heavily on the on-site judgment of professionals and therefore require a lot of time and manpower to proceed. Therefore, how to diagnose structural damage more efficiently is a goal that those skilled in the art should work on.

有鑑於此，本揭露提供一種結構體診斷系統及結構體診斷方法，能自動診斷結構體損傷。In view of this, the present disclosure provides a structure diagnosis system and a structure diagnosis method, which can automatically diagnose structure damage.

本揭露提出一種結構體診斷系統，包括：光學雷達掃描器，掃描結構體以產生點雲數據；輸入介面，耦接到光學雷達掃描器並接收該點雲數據；以及處理器，耦接到輸入介面，處理器接收點雲數據並產生點雲數據集。處理器執行表面劣化及幾何異常耦合診斷模組用以：根據點雲數據集的色彩空間值來標示表面劣化區域的第一點雲範圍；根據點雲數據集的座標值來標示幾何異常區域的第二點雲範圍；當異常區域包括至少部分重疊的第一點雲範圍及第二點雲範圍時，判斷異常區域發生表面劣化或幾合異常並以預定模式標示異常區域。This disclosure proposes a structure diagnosis system, including: an optical radar scanner, which scans the structure to generate point cloud data; an input interface, which is coupled to the optical radar scanner and receives the point cloud data; and a processor, which is coupled to the input In the interface, the processor receives point cloud data and generates a point cloud data set. The processor executes the surface degradation and geometric anomaly coupling diagnosis module to: mark the first point cloud range of the surface degradation area according to the color space value of the point cloud data set; mark the geometric anomaly area according to the coordinate value of the point cloud data set The second point cloud range; when the abnormal area includes the first point cloud range and the second point cloud range that at least partially overlap, it is determined that the abnormal area has surface degradation or multiple abnormalities and the abnormal area is marked with a predetermined pattern.

本揭露提出一種結構體診斷方法，包括：掃描結構體以產生點雲數據；接收點雲數據並產生點雲數據集；根據點雲數據集的色彩空間值來標示表面劣化區域的第一點雲範圍；根據點雲數據集的座標值來標示幾何異常區域的第二點雲範圍；以及當異常區域包括至少部分重疊的第一點雲範圍及第二點雲範圍時，判斷異常區域發生表面劣化或幾合異常並以預定模式標示異常區域。This disclosure proposes a structure diagnosis method, including: scanning the structure to generate point cloud data; receiving the point cloud data and generating a point cloud data set; and marking the first point cloud of the surface deterioration area according to the color space value of the point cloud data set Range; Mark the second point cloud range of the geometric abnormal area according to the coordinate value of the point cloud data set; and when the abnormal area includes the first point cloud range and the second point cloud range that at least partially overlap, it is judged that the abnormal area has surface degradation Or Jihe is abnormal and the abnormal area is marked in a predetermined pattern.

基於上述，本揭露的結構體診斷系統及結構體診斷方法會以光學雷達掃描器掃描結構體以產生點雲數據集，根據色彩空間值來標示表面劣化區域的第一點雲範圍並根據座標值來標示幾何異常區域的第二點雲範圍。當第一點雲範圍及第二點雲範圍至少部分重疊時，則可判斷此異常區域發生表面劣化或幾合異常並以預定模式標示異常區域。如此一來，結構體損傷機制的診斷可更有效率地進行。Based on the above, the structure diagnosis system and structure diagnosis method of the present disclosure scan the structure with an optical radar scanner to generate a point cloud data set, mark the first point cloud range of the surface deterioration area according to the color space value, and according to the coordinate value To mark the second point cloud range of the geometric anomaly area. When the range of the first point cloud and the range of the second point cloud at least partially overlap, it can be determined that the abnormal area has a surface deterioration or an abnormality, and the abnormal area is marked in a predetermined pattern. In this way, the diagnosis of structural damage mechanism can be performed more efficiently.

圖1為根據本揭露一實施例的結構體診斷系統的方塊圖。FIG. 1 is a block diagram of a structure diagnosis system according to an embodiment of the disclosure.

請參照圖1，本揭露一實施例的結構體診斷系統100包括光學雷達(Lidar)掃描器110、輸入介面120、處理器130、記憶裝置140及輸出介面150。光學雷達掃描器110例如是三維光學雷達掃描器並耦接到輸入介面120。輸入介面120耦接到處理器130。記憶裝置140耦接到處理器130。輸出介面150耦接到處理器130。輸入介面120例如是乙太網路介面或其他可傳輸光學雷達點雲數據的介面。處理器130例如是中央處理器(Central Processing Unit，CPU)或其他類似裝置。記憶裝置140例如是揮發性記憶體及/或非揮發性記憶體並用以儲存或暫存光學雷達點雲數據。輸出介面150例如是高畫質多媒體介面(High Definition Multimedia Interface，HDMI)、顯示埠(Display Port，DP)、或其他類似介面。Please refer to FIG. 1, the structure diagnosis system 100 of an embodiment of the present disclosure includes an optical radar (Lidar) scanner 110, an input interface 120, a processor 130, a memory device 140 and an output interface 150. The optical radar scanner 110 is, for example, a three-dimensional optical radar scanner and is coupled to the input interface 120. The input interface 120 is coupled to the processor 130. The memory device 140 is coupled to the processor 130. The output interface 150 is coupled to the processor 130. The input interface 120 is, for example, an Ethernet interface or other interfaces capable of transmitting optical radar point cloud data. The processor 130 is, for example, a central processing unit (CPU) or other similar devices. The memory device 140 is, for example, a volatile memory and/or a non-volatile memory and is used to store or temporarily store optical radar point cloud data. The output interface 150 is, for example, a high-definition multimedia interface (HDMI), a display port (DP), or other similar interfaces.

處理器130可執行資料前處理模組131、表面劣化及幾何異常耦合診斷模組132、異常區域追蹤預測模組133來處理光學雷達掃描器110掃描結構體(例如，管線、桶槽、風機塔架、橋梁、支撐架等)所產生的三維點雲數據以進行結構體異常區域的診斷及追蹤預測，並由輸出介面150輸出異常區域的診斷及追蹤預測結果。異常區域可包括表面劣化區域及幾何異常區域。表面劣化區域可包括塗層起泡、剝落及銹蝕的缺陷。幾何異常區域可包括變形、傾斜及真圓度變異的缺陷。舉例來說，輸出介面150輸出的診斷資訊可為座標群(5.17,-3.38,1.00)~(5.17,-3.38,-0.98)變形異常，變形量3.79 mm；座標群(4.89,-1.32,-3.97)~(4.92,-1.26,-3.93)腐蝕異常，面積12cm ²，深度3mm；座標群(4.97,-1.15,-3.84)~(4.98,-1.15,-3.82)塗層起泡異常，面積50cm ²等資訊。上述座標群又可稱為三維座標群。輸出介面150還可輸出例如「yyyy/mm/dd診斷本結構體具有2處腐蝕、7處塗層起泡及8處變形異常點」的事件通知。此外，輸出介面150還可利用三維圖形顯示結構體及其腐蝕、塗層起泡及變形異常點。 The processor 130 can execute the data pre-processing module 131, the surface degradation and geometrical anomaly coupling diagnosis module 132, and the abnormal area tracking prediction module 133 to process the optical radar scanner 110 scanning structures (for example, pipelines, barrels, wind turbine towers, etc.). The three-dimensional point cloud data generated by the frame, bridge, support frame, etc.) can be used for the diagnosis and tracking prediction of the abnormal area of the structure, and the output interface 150 outputs the diagnosis and tracking prediction results of the abnormal area. The abnormal area may include a surface deterioration area and a geometric abnormal area. Areas of surface deterioration may include defects such as blistering, peeling, and corrosion of the coating. The geometrically abnormal area may include defects such as deformation, tilt, and roundness variation. For example, the diagnostic information output by the output interface 150 can be the coordinate group (5.17, -3.38, 1.00) ~ (5.17, -3.38, -0.98) abnormal deformation, the deformation amount is 3.79 mm; the coordinate group (4.89, -1.32,- 3.97)~(4.92,-1.26,-3.93) abnormal corrosion, area 12cm ² , depth 3mm; coordinate group (4.97,-1.15,-3.84)~(4.98,-1.15,-3.82) coating blistering abnormal, area 50cm ² and other information. The above-mentioned coordinate group can also be called a three-dimensional coordinate group. The output interface 150 can also output an event notification such as "yyyy/mm/dd diagnoses that the structure has 2 corrosion, 7 coating blistering, and 8 abnormal deformation points". In addition, the output interface 150 can also use three-dimensional graphics to display the structure and its corrosion, coating blistering and abnormal deformation points.

在一實施例中，資料前處理模組131、表面劣化及幾何異常耦合診斷模組132、異常區域追蹤預測模組133以軟體或韌體實作並由處理器130來執行，但本揭露不限於此。在另一實施例中，資料前處理模組131、表面劣化及幾何異常耦合診斷模組132、異常區域追蹤預測模組133也可實作為處理器130中或耦接到處理器的硬體電路。下文中將對資料前處理模組131、表面劣化及幾何異常耦合診斷模組132、異常區域追蹤預測模組133進行詳細描述。In one embodiment, the data pre-processing module 131, the surface degradation and geometrical anomaly coupling diagnosis module 132, and the abnormal area tracking prediction module 133 are implemented by software or firmware and executed by the processor 130, but the present disclosure is not limited to this. In another embodiment, the data pre-processing module 131, the surface degradation and geometric abnormality coupling diagnosis module 132, and the abnormal area tracking prediction module 133 can also be implemented as hardware circuits in the processor 130 or coupled to the processor. . Hereinafter, the data pre-processing module 131, the surface degradation and geometric abnormality coupling diagnosis module 132, and the abnormal area tracking prediction module 133 will be described in detail.

[資料前處理][Data pre-processing]

在一實施例中，光學雷達掃描器110(即，三維光學雷達掃描器)可發出雷射光並接收結構體所反射的雷射光來掃描結構體以產生點雲數據。輸入介面120可接收點雲數據並將點雲數據提供給處理器130。處理器130可執行資料前處理模組131以接收點雲數據並根據結構體尺寸縮減或不縮減點雲數據並進行結構體的幾何重建，對點雲數據進行拓墣整齊操作，並對不同時間點的拓墣整齊操作後的點雲數據進行座標配準操作以產生具有時間序的點雲數據集。拓墣整齊操作例如是根據點雲數據的拓墣資訊來達到正確描述結構體幾何特徵的效果。座標配準操作可將不同時間點的點雲數據依照辨識出來的特定特徵(例如，桶槽固定於地面的釘子)或預定的多個基準點座標來進行座標配準，如此可減少同一個位置的點雲數據在不同時間點有位置誤差的情況。值得注意的是，具有時間序的點雲數據集為同一掃描站點且不同時間點的點雲數據集。透過本揭露的資料前處理流程，可獲得更精準的結構體診斷及預測結果。In an embodiment, the optical radar scanner 110 (ie, a three-dimensional optical radar scanner) can emit laser light and receive the laser light reflected by the structure to scan the structure to generate point cloud data. The input interface 120 can receive point cloud data and provide the point cloud data to the processor 130. The processor 130 can execute the data pre-processing module 131 to receive the point cloud data, reduce or not reduce the point cloud data according to the size of the structure, and perform geometric reconstruction of the structure, expand and clean the point cloud data, and perform different times The point cloud data after the point expansion and neat operation is subjected to coordinate registration operations to generate a time-series point cloud data set. The neat operation of the extension is, for example, based on the extension information of the point cloud data to achieve the effect of correctly describing the geometric features of the structure. The coordinate registration operation can perform coordinate registration of the point cloud data at different time points according to the identified specific features (for example, the nails of the barrel fixed to the ground) or predetermined multiple reference point coordinates, so that the same position can be reduced The point cloud data has position errors at different points in time. It is worth noting that the point cloud data set with time sequence is the point cloud data set of the same scanning site and at different time points. Through the data pre-processing process disclosed in this disclosure, more accurate structure diagnosis and prediction results can be obtained.

[結構體異常區域診斷][Diagnosis of structure abnormal area]

在一實施例中，處理器130可執行表面劣化及幾何異常耦合診斷模組132以進行以下運算。處理器130可讀取點雲數據集的色彩空間值(R,G,B)並根據點雲數據集的色彩空間值來標示表面劣化區域的點雲範圍(或稱為第一點雲範圍)。具體來說，處理器130可利用神經網路分割點雲數據集的色彩空間值的表面劣化區域並計算表面劣化區域的面積及深度。舉例來說，從點雲數據集可得知座標群(4.89,-1.32,-3.97)~(4.92,-1.26,-3.93)發生腐蝕異常，因此處理器130可從座標群(4.89,-1.32,-3.97)~(4.92,-1.26,-3.93)數據計算出腐蝕異常的面積為12cm ²且深度為3mm。處理器130可利用異常區域點雲數量相對於結構體點雲數量的百分比來進行表面劣化區域面積的計算。神經網路例如是卷積神經網路(Convolutional Neural Network，CNN)、全卷積網路(Fully Convolutional Networks，FCN)、編碼解碼網路(Encoder-decoder)、循環神經網路(Recurrent Neural Network，RNN)、生成對抗網路(Generative Adversarial Network，GAN)等神經網路。 In one embodiment, the processor 130 can execute the surface degradation and geometric abnormality coupling diagnosis module 132 to perform the following operations. The processor 130 can read the color space values (R, G, B) of the point cloud data set and mark the point cloud range of the surface deterioration area (or called the first point cloud range) according to the color space value of the point cloud data set . Specifically, the processor 130 may use a neural network to segment the surface deterioration area of the color space value of the point cloud data set and calculate the area and depth of the surface deterioration area. For example, it can be known from the point cloud data set that the coordinate group (4.89, -1.32, -3.97) ~ (4.92, -1.26, -3.93) has corrosion anomalies, so the processor 130 can learn from the coordinate group (4.89, -1.32). , -3.97) ~ (4.92, -1.26, -3.93) data calculates that the corrosion anomaly area is 12cm ² and the depth is 3mm. The processor 130 may use the percentage of the number of point clouds in the abnormal area to the number of point clouds in the structure to calculate the area of the surface deterioration area. Neural networks are, for example, Convolutional Neural Network (CNN), Fully Convolutional Networks (FCN), Encoder-decoder, Recurrent Neural Network, RNN), Generative Adversarial Network (GAN) and other neural networks.

處理器130還可讀取點雲數據集的座標值(X,Y, Z)並根據點雲數據集的座標值來標示幾何異常區域的點雲範圍(或稱為第二點雲範圍)。具體來說，處理器130以沿結構體受力方向與垂直結構體受力方向)分割點雲數據集的座標值的幾何異常區域並計算幾何異常區域的位置及變形量。The processor 130 may also read the coordinate values (X, Y, Z) of the point cloud data set and mark the point cloud range (or called the second point cloud range) of the geometric abnormal area according to the coordinate value of the point cloud data set. Specifically, the processor 130 divides the geometric abnormal area of the coordinate value of the point cloud data set along the force direction of the structure and the direction perpendicular to the force direction of the structure, and calculates the position and the amount of deformation of the geometric abnormal area.

當異常區域包括至少部分重疊的第一點雲範圍及第二點雲範圍時，處理器130判斷異常區域發生表面劣化或幾合異常並以預定模式標示異常區域。具體來說，當表面劣化區域的第一點雲範圍大於幾何異常區域的第二點雲範圍時，處理器130判斷異常區域發生表面劣化並以預定模式標示異常區域的第一點雲範圍。當表面劣化區域的第一點雲範圍小於幾何異常區域的第二點雲範圍時，處理器130判斷異常區域發生幾何異常並以預定模式標示異常區域的第二點雲範圍。舉例來說，以預定模式來標示異常區域、第一點雲範圍或第二點雲範圍可為使用預定顏色、箭頭標示、框選方式等模式來標示異常區域、第一點雲範圍或第二點雲範圍讓使用者明確識別。本揭露不限制預定模式的實作方式。When the abnormal area includes the first point cloud range and the second point cloud range that are at least partially overlapping, the processor 130 determines that the abnormal area has surface degradation or an abnormality and marks the abnormal area in a predetermined pattern. Specifically, when the first point cloud range of the surface deterioration area is larger than the second point cloud range of the geometric abnormal area, the processor 130 determines that the abnormal area has surface deterioration and marks the first point cloud range of the abnormal area in a predetermined pattern. When the first point cloud range of the surface deterioration area is smaller than the second point cloud range of the geometric abnormal area, the processor 130 determines that the abnormal area has a geometric abnormality and marks the second point cloud range of the abnormal area in a predetermined pattern. For example, marking the abnormal area, the first point cloud range, or the second point cloud range in a predetermined pattern may be using a predetermined color, arrow marking, frame selection method, etc., to mark the abnormal area, the first point cloud range, or the second point cloud range. The point cloud range allows the user to clearly identify it. This disclosure does not limit the implementation of the predetermined mode.

[結構體異常區域追蹤預測][Tracking and Prediction of Structure Abnormal Area]

在一實施例中，處理器130可執行異常區域追蹤預測模組133以透過機器學習演算法根據連續時間點的發生幾何異常的異常區域的點雲來預測異常區域在下一時間點的變異狀態。機器學習演算法例如是線性回歸(linear regression)、支持矢量回歸(Support Vector Regression，SVR)、集成學習(ensemble learning)等機器學習演算法。也就是說，處理器130可追蹤同一異常區域是否持續存在或有擴大的趨勢，並利用具有不同時間標記的連續點雲透過機器學習演算法來學習變異趨勢以預測下一時間點的變異狀態。舉例來說，不同結構體在多個時間點的大量變異趨勢數據可被輸入機器學習演算法來進行訓練。以凹陷深度為簡單範例，當異常區域在T0時間點的凹陷深度為1mm且在T1時間點的凹陷深度為2mm，則機器學習演算法的依照訓練結果判斷T2時間點的凹陷深度為3mm，其中T0時間點到T1時間點的時間間隔可相同或不同於T1時間點到T2時間點的時間間隔。值得注意的是，機器學習演算法可通過大量數據(例如，多組隨著時間變化的凹陷深度資訊)的訓練，從而對T2時間點的凹陷深度作出更精確的預測。In one embodiment, the processor 130 can execute the abnormal area tracking prediction module 133 to predict the mutation state of the abnormal area at the next time point based on the point cloud of the abnormal area where the geometric abnormality occurs at consecutive time points through a machine learning algorithm. The machine learning algorithms are, for example, linear regression (linear regression), Support Vector Regression (SVR), ensemble learning (ensemble learning) and other machine learning algorithms. In other words, the processor 130 can track whether the same abnormal area persists or has a tendency to expand, and uses continuous point clouds with different time stamps to learn the mutation trend through a machine learning algorithm to predict the mutation state at the next time point. For example, a large amount of variation trend data of different structures at multiple time points can be input into a machine learning algorithm for training. Taking the depression depth as a simple example, when the depression depth of the abnormal area at time T0 is 1mm and the depression depth at time T1 is 2mm, the machine learning algorithm judges that the depression depth at time T2 is 3mm according to the training results, where The time interval from T0 time point to T1 time point may be the same or different from the time interval from T1 time point to T2 time point. It is worth noting that the machine learning algorithm can be trained with a large amount of data (for example, multiple sets of depression depth information that changes over time), so as to make a more accurate prediction of the depression depth at the T2 time point.

圖2為根據本揭露一實施例的表面劣化及幾何異常耦合診斷的範例。FIG. 2 is an example of surface degradation and geometric abnormal coupling diagnosis according to an embodiment of the present disclosure.

請參照圖2，本範例取得大小10x15cm 的PU塗裝鋼板，以電化學放電法及鹽霧試驗製作表面劣化缺陷(即，塗裝起泡及銹蝕)並以外力撞擊鋼版製作局部變形區域。接著，以光學雷達距離鋼版5公尺掃描鋼板以取得點雲數據，並以表面劣化及幾何異常耦合診斷模組132進行鋼板損傷診斷。Please refer to Figure 2. This example obtains a PU coated steel sheet with a size of 10x15cm, uses electrochemical discharge method and salt spray test to produce surface deterioration defects (ie, coating blistering and corrosion) and impacts the steel plate with external force to produce local deformation areas. Then, the optical radar is used to scan the steel plate at a distance of 5 meters from the steel plate to obtain point cloud data, and the surface degradation and geometric abnormality coupling diagnosis module 132 is used to diagnose the damage of the steel plate.

[T0時間點][T0 time point]

鋼板有六個起泡缺陷210，並由表面劣化及幾何異常耦合診斷模組132診斷出表面劣化集合220及幾何異常集合230。表面劣化集合220包括六個表面劣化點雲範圍(即，第一點雲範圍)，且幾何異常集合230包括兩個幾何異常點雲範圍(即，第二點雲範圍)。表面劣化集合220可根據點雲數據集的色彩空間值(R,G,B)來標示且幾何異常集合230可根據點雲數據集的座標值(X,Y,Z)來標示。經過表面劣化集合220及幾何異常集合230的座標耦合之後可產生表面劣化集合220及幾何異常集合230座標至少部分重疊的異常區域241及異常區域242。也就是說，表面劣化集合220及幾何異常集合230的座標耦合是用於辨識表面劣化集合220的三維座標群及幾何異常集合230的三維座標群的至少部分重疊的區域。座標耦合不同於本揭露在資料前處理過程中對點雲數據進行的座標配準操作。異常區域241中表面劣化點雲範圍大於幾何異常點雲範圍因此被識別為表面劣化起泡。異常區域242中表面劣化點雲範圍大於幾何異常點雲範圍因此被識別為表面劣化起泡。The steel plate has six blistering defects 210, and the surface deterioration and geometric abnormality coupling diagnosis module 132 diagnoses the surface deterioration set 220 and the geometric abnormality set 230. The surface degradation set 220 includes six surface degradation point cloud ranges (ie, the first point cloud range), and the geometric anomaly set 230 includes two geometric anomaly point cloud ranges (ie, the second point cloud range). The surface degradation set 220 may be marked according to the color space values (R, G, B) of the point cloud data set, and the geometric anomaly set 230 may be marked according to the coordinate values (X, Y, Z) of the point cloud data set. After the coordinate coupling of the surface degradation set 220 and the geometric anomaly set 230, an abnormal area 241 and an abnormal area 242 in which the coordinates of the surface degradation set 220 and the geometric anomaly set 230 at least partially overlap can be generated. In other words, the coordinate coupling of the surface degradation set 220 and the geometric anomaly set 230 is used to identify at least partially overlapping regions of the three-dimensional coordinate group of the surface degradation set 220 and the three-dimensional coordinate group of the geometric anomaly set 230. The coordinate coupling is different from the coordinate registration operation performed on the point cloud data in the data pre-processing process of the present disclosure. The range of the surface deterioration point cloud in the abnormal area 241 is larger than the range of the geometric abnormal point cloud, and therefore it is recognized as surface deterioration blistering. The range of the surface deterioration point cloud in the abnormal area 242 is larger than the range of the geometric abnormal point cloud, and therefore it is recognized as a surface deterioration blistering.

[T1時間點][T1 time point]

鋼板有四個起泡缺陷250、兩個由起泡缺陷變成的銹蝕251及一個新產生的變形252，並由表面劣化及幾何異常耦合診斷模組132診斷出表面劣化集合260及幾何異常集合270。表面劣化集合260包括七個表面劣化點雲範圍(即，第一點雲範圍)，且幾何異常集合270包括兩個幾何異常點雲範圍(即，第二點雲範圍)。表面劣化集合260可根據點雲數據集的色彩空間值(R,G,B)來標示且幾何異常集合270可根據點雲數據集的座標值(X,Y,Z)來標示。經過表面劣化集合260及幾何異常集合270的座標耦合之後可產生表面劣化集合260及幾何異常集合270座標至少部分重疊的異常區域281及異常區域282。異常區域281中表面劣化點雲範圍大於幾何異常點雲範圍因此被識別為表面劣化起泡。異常區域282中表面劣化點雲範圍小於幾何異常點雲範圍因此被識別為幾何變形。The steel plate has four blistering defects 250, two rust 251 from blistering defects, and one new deformation 252. The surface deterioration and geometric abnormality coupling diagnosis module 132 diagnoses the surface deterioration set 260 and the geometric abnormal set 270 . The surface degradation set 260 includes seven surface degradation point cloud ranges (ie, the first point cloud range), and the geometric anomaly set 270 includes two geometric anomaly point cloud ranges (ie, the second point cloud range). The surface degradation set 260 may be marked according to the color space values (R, G, B) of the point cloud data set, and the geometric anomaly set 270 may be marked according to the coordinate values (X, Y, Z) of the point cloud data set. After the coordinate coupling of the surface degradation set 260 and the geometric anomaly set 270, an abnormal area 281 and an abnormal area 282 in which the coordinates of the surface degradation set 260 and the geometric anomaly set 270 at least partially overlap can be generated. The range of the surface deterioration point cloud in the abnormal area 281 is larger than the range of the geometric abnormal point cloud, and therefore it is recognized as surface deterioration blistering. The range of the surface deterioration point cloud in the abnormal area 282 is smaller than the range of the geometric abnormal point cloud, and therefore is recognized as a geometric deformation.

圖3為根據本揭露一實施例的神經網路分割表面劣化區域及面積深度估算的範例。FIG. 3 is an example of the neural network segmentation of the surface deterioration area and the area depth estimation according to an embodiment of the present disclosure.

請參照圖3，本範例在170片鋼板樣本上製作700顆尺寸分別為5mm/10mm/20mm的銹蝕及起泡缺陷，以光學雷達距離樣本5m進行三維點雲影像掃描並以神經網路進行訓練。舉例來說，三維點雲數據可被輸入卷積神經網路交錯排列的卷積層及池化層以進行特徵擷取並經過全連接層以輸出特徵分類結果。卷積層可採用整流線性單位(Rectified Linear Unit，ReLU)函數或其他函數作為激勵函數。最後，對97個銹蝕及起泡缺陷影像進行辨識率的測試。在20mm起泡樣本310及10mm銹蝕樣本320中，可見命中(hit)區域341、未命中(miss)區域342及錯誤警告(false alarm)區域343。最後，表面劣化面積深度估算圖330會被產生。深度估算圖330的三軸例如是X軸、Y軸及Z軸且單位為mm。Please refer to Figure 3. In this example, 700 rust and blistering defects with dimensions of 5mm/10mm/20mm were produced on 170 steel plate samples. The 3D point cloud image was scanned with the optical radar at a distance of 5m from the sample and the neural network was used for training. . For example, the three-dimensional point cloud data can be input to the convolutional layer and the pooling layer of the convolutional neural network to perform feature extraction and pass through the fully connected layer to output the feature classification result. The convolutional layer may use a rectified linear unit (ReLU) function or other functions as the excitation function. Finally, the recognition rate of 97 images of rust and blistering defects were tested. In the 20mm blister sample 310 and the 10mm rust sample 320, a hit area 341, a miss area 342, and a false alarm area 343 can be seen. Finally, the surface degradation area depth estimation map 330 is generated. The three axes of the depth estimation map 330 are, for example, the X axis, the Y axis, and the Z axis, and the unit is mm.

表一為本範例缺陷辨識率的比較表。Table 1 is a comparison table of defect recognition rate in this example.

表一 掃描距離5m 辨識率測試 缺陷類型 尺寸 召回率(Recall) 總召回率 塗層起泡 5mm 79% 78% 10mm 73% 20mm 80% 銹蝕 5mm 82% 85% 10mm 88% 20mm 85% Table I Scanning distance 5m Recognition rate test Type of defect size Recall Total recall Coating blistering 5mm 79% 78% 10mm 73% 20mm 80% Rust 5mm 82% 85% 10mm 88% 20mm 85%

從表一可得知，在本揭露的實驗中對於塗層起泡缺陷類型5mm、10mm及20mm的總召回率為78%，且對於銹蝕缺陷類型5mm、10mm及20mm的總召回率為85%。It can be seen from Table 1 that the total recall rate for the coating blistering defect types 5mm, 10mm and 20mm in the experiment of this disclosure is 78%, and the total recall rate for the corrosion defect types 5mm, 10mm and 20mm is 85%. .

表二為本範例量測結果與真實結果的比較表。Table 2 is a comparison table between the measurement results of this example and the real results.

表二 量測結果 真實結果 銹蝕面積 249mm ² 237mm ² 銹蝕深度 4.41mm 4.76mm Table II Measurement result Real result Rust area 249mm ² 237mm ² Corrosion depth 4.41mm 4.76mm

從表二可得知，在本揭露實驗中銹蝕面積的量測結果與真實結果相差了12mm ²，且銹蝕深度的量測結果與真實結果相差了0.35mm。 It can be seen from Table 2 that the measurement result of the rust area in this disclosure experiment differs from the real result by 12 mm ² , and the measurement result of the rust depth differs from the real result by 0.35 mm.

圖4為根據本揭露一實施例的結構體幾何變形量估算的範例。FIG. 4 is an example of estimating the amount of geometric deformation of a structure according to an embodiment of the present disclosure.

請參照圖4，以一桶槽結構體進行實驗，掃描所得的點雲數據集400經處理後可用於桶槽現況缺陷識別，且不同時間點的點雲數據集400可用於桶槽缺陷趨勢預測。舉例來說，桶槽缺陷識別可包括垂向變形範圍識別、壁鈑徑向變形識別及壁鈑真圓度識別。垂向變形範圍識別及壁鈑徑向變形識別的趨勢預測目標可為每8英吋內竟像變形量與所占範圍是否超過法規限制(例如，3.6英吋)。壁鈑真圓度識別的趨勢預測目標可由各角度垂向變形組合後進行計算。本揭露的結構體幾何變形量估算可涵蓋0~29.8%的變形範圍。Please refer to Figure 4, experiment with a barrel tank structure. The point cloud data set 400 obtained by scanning can be used for barrel tank status defect identification after processing, and the point cloud data set 400 at different time points can be used for barrel tank defect trend prediction . For example, the barrel groove defect recognition can include vertical deformation range recognition, wall radial deformation recognition, and wall plate roundness recognition. The trend prediction target for vertical deformation range recognition and wall radial deformation recognition can be the amount of deformation in every 8 inches and whether the occupied range exceeds the legal limit (for example, 3.6 inches). The trend prediction target of the roundness recognition of the wall sheet can be calculated after the combination of the vertical deformation of each angle. The geometric deformation estimation of the structure disclosed in the present disclosure can cover a deformation range of 0-29.8%.

圖5為根據本揭露一實施例的結構體變形趨勢預測的範例。FIG. 5 is an example of structural deformation trend prediction according to an embodiment of the disclosure.

請參照圖5，無外撐受力的桶槽510經由光學雷達掃描與資料前處理模組131的運算可獲得桶槽的點雲數據集520。異常區域追蹤預測模組可根據點雲數據集520預測出施加3mm外撐後的桶槽壁板點雲座標530。舉例來說，桶槽540的側面可進行圓柱展開成長方型550並每30度擷取點雲(即，以沿桶槽510受力方向與垂直桶槽510受力方向分割點雲數據集520)檢視以產生桶槽壁的多個點雲子集560。點雲570為實測無外撐點雲。點雲580為實測外撐後點雲。點雲590為預測外撐後點雲。本範例實測外撐後點雲與預測外撐後點雲的均方誤差(MSE)為1.13%，且平均絕對誤差(MAE)為1.66mm。Please refer to FIG. 5, the bucket 510 without external support force can obtain the point cloud data set 520 of the bucket through the optical radar scanning and the calculation of the data pre-processing module 131. The abnormal area tracking prediction module can predict the point cloud coordinates 530 of the barrel wall after the 3mm external support is applied based on the point cloud data set 520. For example, the side surface of the barrel 540 can be expanded into a cylindrical shape 550 and the point cloud is captured every 30 degrees (that is, the point cloud data set 520 is divided along the force direction of the barrel 510 and the vertical direction of the barrel 510. ) View to generate multiple point cloud subsets 560 of the barrel wall. The point cloud 570 is a measured point cloud without external support. The point cloud 580 is the measured point cloud after the external support. The point cloud 590 is the predicted point cloud after the external support. In this example, the mean square error (MSE) of the measured point cloud after the external support and the predicted point cloud after the external support is 1.13%, and the average absolute error (MAE) is 1.66mm.

圖6為根據本揭露一實施例的結構體診斷方法的流程圖。FIG. 6 is a flowchart of a structure diagnosis method according to an embodiment of the disclosure.

請參照圖6，在步驟S601中，掃描結構體以產生點雲數據。Referring to FIG. 6, in step S601, the structure is scanned to generate point cloud data.

在步驟S602中，接收點雲數據並產生點雲數據集。In step S602, point cloud data is received and a point cloud data set is generated.

在步驟S603中，根據點雲數據集的色彩空間值來標示表面劣化區域的第一點雲範圍。In step S603, the first point cloud range of the surface deterioration area is marked according to the color space value of the point cloud data set.

在步驟S604中，根據點雲數據集的座標值來標示幾何異常區域的第二點雲範圍。In step S604, the second point cloud range of the geometric abnormal area is marked according to the coordinate values of the point cloud data set.

在步驟S605中，當異常區域包括至少部分重疊的第一點雲範圍及第二點雲範圍時，判斷異常區域發生表面劣化或幾合異常並以預定模式標示異常區域。當表面劣化區域的第一點雲範圍大於幾何異常區域的第二點雲範圍時，處理器判斷異常區域發生表面劣化並以預定模式標示異常區域的第一點雲範圍。當表面劣化區域的第一點雲範圍小於幾何異常區域的第二點雲範圍時，處理器判斷異常區域發生幾何異常並以預定模式標示異常區域的第二點雲範圍。In step S605, when the abnormal area includes the first point cloud range and the second point cloud range that are at least partially overlapping, it is determined that the abnormal area has surface degradation or multiple abnormalities, and the abnormal area is marked in a predetermined pattern. When the first point cloud range of the surface deterioration area is larger than the second point cloud range of the geometric abnormal area, the processor determines that the abnormal area has surface deterioration and marks the first point cloud range of the abnormal area in a predetermined pattern. When the first point cloud range of the surface deterioration area is smaller than the second point cloud range of the geometric abnormal area, the processor determines that the abnormal area has a geometric abnormality and marks the second point cloud range of the abnormal area in a predetermined pattern.

綜上所述，本揭露的結構體診斷系統及結構體診斷方法會以光學雷達掃描器掃描結構體以產生點雲數據集，根據色彩空間值來標示表面劣化區域的第一點雲範圍並根據座標值來標示幾何異常區域的第二點雲範圍。當第一點雲範圍及第二點雲範圍至少部分重疊時，則可判斷此異常區域發生表面劣化或幾合異常並以預定模式標示異常區域。如此一來，結構體損傷機制的診斷可更有效率地進行。In summary, the structure diagnosis system and structure diagnosis method of the present disclosure scan the structure with an optical radar scanner to generate a point cloud data set, and mark the first point cloud range of the surface deterioration area according to the color space value. The coordinate value is used to indicate the second point cloud range of the geometric anomaly area. When the range of the first point cloud and the range of the second point cloud at least partially overlap, it can be determined that the abnormal area has a surface deterioration or an abnormality, and the abnormal area is marked in a predetermined pattern. In this way, the diagnosis of structural damage mechanism can be performed more efficiently.

雖然本揭露已以實施例揭露如上，然其並非用以限定本揭露，任何所屬技術領域中具有通常知識者，在不脫離本揭露的精神和範圍內，當可作些許的更動與潤飾，故本揭露的保護範圍當視後附的申請專利範圍所界定者為準。Although the present disclosure has been disclosed in the above embodiments, it is not intended to limit the present disclosure. Anyone with ordinary knowledge in the technical field can make some changes and modifications without departing from the spirit and scope of this disclosure. Therefore, The scope of protection of this disclosure shall be subject to those defined by the attached patent scope.

100:結構體診斷系統 110:光學雷達掃描器 120:輸入介面 130:處理器 140:記憶裝置 150:輸出介面 210:起泡缺陷 220:表面劣化集合 230:幾何異常集合 241、242:異常區域 250:起泡缺陷 251:銹蝕 252:變形 260:表面劣化集合 270:幾何異常集合 281、282:異常區域 310:起泡樣本 320:銹蝕樣本 330:表面劣化面積深度估算圖 341:命中區域 342:未命中區域 343:錯誤警告區域 400:點雲數據集 510:桶槽 520:點雲數據集 530:桶槽壁板點雲座標 540:桶槽 550:長方型 560:點雲子集 570、580、590:點雲 S601~S605:步驟100: Structural Diagnosis System 110: Optical radar scanner 120: input interface 130: processor 140: memory device 150: output interface 210: Blistering defect 220: Surface Deterioration Collection 230: Geometric anomaly collection 241, 242: abnormal area 250: Blistering defect 251: Corrosion 252: Deformation 260: Surface Deterioration Collection 270: Geometric Anomaly Collection 281, 282: Abnormal Area 310: Foaming sample 320: rust sample 330: Depth estimation map of surface deterioration area 341: hit area 342: Miss Area 343: Error warning area 400: Point cloud data set 510: Barrel Slot 520: Point Cloud Data Set 530: Barrel slot wall point cloud coordinates 540: Barrel Groove 550: rectangular 560: Point Cloud Subset 570, 580, 590: point cloud S601~S605: steps

圖1為根據本揭露一實施例的結構體診斷系統的方塊圖。 圖2為根據本揭露一實施例的表面劣化及幾何異常耦合診斷的範例。 圖3為根據本揭露一實施例的神經網路分割表面劣化區域及面積深度估算的範例。 圖4為根據本揭露一實施例的結構體幾何變形量估算的範例。 圖5為根據本揭露一實施例的結構體變形趨勢預測的範例。 圖6為根據本揭露一實施例的結構體診斷方法的流程圖。 FIG. 1 is a block diagram of a structure diagnosis system according to an embodiment of the disclosure. FIG. 2 is an example of surface degradation and geometric abnormal coupling diagnosis according to an embodiment of the present disclosure. FIG. 3 is an example of the neural network segmentation of the surface deterioration area and the area depth estimation according to an embodiment of the present disclosure. FIG. 4 is an example of estimating the amount of geometric deformation of a structure according to an embodiment of the present disclosure. FIG. 5 is an example of structural deformation trend prediction according to an embodiment of the disclosure. FIG. 6 is a flowchart of a structure diagnosis method according to an embodiment of the disclosure.

210:起泡缺陷 210: Blistering defect

220:表面劣化集合 220: Surface Deterioration Collection

230:幾何異常集合 230: Geometric anomaly collection

241、242:異常區域 241, 242: abnormal area

250:起泡缺陷 250: Blistering defect

251:銹蝕 251: Corrosion

252:變形 252: Deformation

260:表面劣化集合 260: Surface Deterioration Collection

270:幾何異常集合 270: Geometric Anomaly Collection

281、282:異常區域 281, 282: Abnormal Area