10 minute read
A critical link
Josiah Lau, Novlum Inc., Canada, outlines how 3D laser scanning works, and explains why the right software can significantly improve workflow efficiency.
Light Detection and Ranging (LiDAR), more commonly known as 3D laser scanning, has been around since the 1960s. Today, 3D laser scanners come in many forms and are used in applications ranging from large area 3D aerial surveys to collision detection on autonomous vehicles. Within the last several years, the use of 3D laser scanning to assess structural integrity and capacity of aboveground storage tanks has gained popularity. While 3D laser scanners can capture vivid 3D images of these large, tall, and internally dark structures in a short amount of time, the software required to transform the 3D data into useful information is lacking. This is the main reason that 3D tank inspection has not become the mainstream tank inspection method even though the information gleaned from this technology is extensive when compared to traditional inspection methods.
Scanning
In order to understand the software required to process laser scanning data, also known as point clouds, it is important to fi rst understand how 3D laser scanning works. At a high level, 3D laser scanners send out laser pulses that are refl ected back to the scanner when they hit objects. The measured distance is equal to half the round trip fl ight time multiplied by the speed of light (Figure 1).
The accuracy of each measurement depends on the ability of the scanner to resolve the peak intensity of the refl ected laser pulse, which is affected by how energy
propagates, absorbs, and refl ects off surfaces. Factors that affect the return intensity include ranging distance, ambient light, surface characteristics, and surface incidence angles. Since light travels in a straight line and cannot penetrate objects, physical obstructions in the scan area create shadow areas behind them (Figure 2). Because of these limitations, numerous scans are required in order to capture the entire structure of a storage tank and to meet the accuracy requirement for tank inspection (Figure 3).
Figure 1. 3D laser scanners send out laser pulses that are reflected back from objects.
Registration
Figure 2. Single scan showing shadows on shell caused by scaffolding, decreasing point density on floor with increasing scan distance, and decreasing point intensity on floor due to increasing incidence angles. A shadow area is always seen underneath the scanner.
Figure 3. 3D model created by registering multiple scans together. Complete coverage is achieved on shell and floor, including beneath individual scan locations.
Each scan is an independent 3D representation of the storage tank from the perspective of the scan location. To achieve a complete representation with suffi cient coverage of the entire tank, the scans need to be merged together to create a registered 3D model, a process called data registration (Figure 4). Many off-the-shelf software packages can be used to perform data registration, which is a well understood process because it is a common step for all 3D laser scanning work. Point cloud registration typically follows the same process regardless of whether the scan subject is a factory, a building, or a storage tank. The two most common data registration techniques are target-based registration and cloud-to-cloud, or targetless, registration. Target-based registration uses artifi cial targets during scanning such as checker targets. These targets serve as common points, or tie points, between the scans. The registration software uses these tie points to co-locate the scans. The speed and accuracy of registration depends on a software’s ability to accurately detect targets within each scan and correlate these targets between the scans. Because automatic algorithms often fail, effi cient manual tools to override target selection, placement, and correlation are necessary to improve accuracy and success. While this registration technique is most commonly used, it relies heavily on proper scan planning and preparation and requires longer in-fi eld time for target setup. Furthermore, there is a higher risk of failure if the number of targets used is inadequate, or the targets are not suffi ciently visible from all scans. Cloud-to-cloud registration uses physical features in the scans as natural targets, such as surfaces, hard edges, and corners to select tie points between scans. This technique is software intensive, but if done properly can yield a more accurate registered 3D model than target-based registration. The advantage Figure 4. Registration is the process of merging several independent 3D scans (left) into a single registered 3D model (right). Each colour represents points from a different 3D scan. of this technique is that targets
are not required, which reduces in-fi eld preparation and risk of registration failure. It can be more robust than target-based registration because an unlimited number of tie points can be used, or tie points can be localised to enhance accuracy in an area of interest. For example, if the upper levels of a tall structure is of interest, but scanning is done from the ground, selecting tie points in the upper levels as opposed to the ground level will ensure a better fi t in that area. When using targets, the target placement is limited to locations that an operator is able to physically access, which is typically close to the ground. While cloud-to-cloud registration has its advantages, the results are highly dependent on the algorithmic implementation in each software package. Sometimes, a hybrid approach can be used where artifi cial targets are used to automatically align the scans and natural targets are used to refi ne the registration.
Regardless of registration technique, the software of choice should have tools to quantify overall registration accuracy, visually review registration quality using cross-sections, and manually adjust the registration if needed. The interactivity of these tools makes a tremendous difference in workfl ow effi ciency. While fully automated approaches can save time, a trained operator should always review the registered model to ensure accuracy.
Figure 5. Registered data model (top), classified data model (centre), and isolated floor points (bottom).
Figure 6. Analysis results showing floor deformations from best-fit cone (left) and shell deformations from best-fit cylinder (right).
Classification
Once a registered 3D model has been created, each data point must be assigned to a specifi c class such as shell, fl oor, roof, columns, or other clutter. Accurate classifi cation of data points is required for proper analysis. For example, all shell data points must be identifi ed and extracted in order to analyse shell out-of-roundness. However, in order to analyse shell settlement, only fl oor points nearest the shell are required, but shell points must be excluded. Misclassifi cation of data points can lead to false anomalies in the results. For example, data points of columns or equipment (Figure 5) misclassifi ed as fl oor can result in false fl oor bulges, and data points of manways misclassifi ed as shell can result in shell roundness being exceeded. The accuracy of the analysis is highly dependent on the accuracy of point classifi cation.
Many software packages have tools to classify data because this is a standard process for all 3D laser scanning work. While software packages made for generic point cloud analysis will provide basic manual tools for classifi cation, they are time-consuming to use. However, software specifi cally designed for tank analysis can signifi cantly improve workfl ow effi ciency because it understands the basic components of tanks, such as cylindrical shells, cone fl oors, domed roofs, columns, girders/rafters, etc. Since the shell to fl oor weld area is
Figure 7. Analysis results showing shell out-of-roundness at bottom (left) and plumbness from top to bottom (right).
critical for tank inspection, it is not suffi cient to simply divide the tank as a vertical cylinder with top and bottom caps. Software that is specifi cally designed for tank analysis has the ability to cleanly differentiate shell points from fl oor points at the bottom weld, taking into account shell and edge settlement in the fl oor. Even though automated tools can signifi cantly speed up classifi cation, manual or semi-automated tools are necessary to review, cleanup, and reclassify data if needed.
In one of the projects that Novlum Inc. worked on, a storage tank was fl agged for further engineering review because it failed the shell settlement analysis. Upon review of the scan data, it was found that the code failure was not due to structural deformations, but rather misclassifi ed data. During scanning, the tank fl oor was wet, creating areas of void and high noise in the scans. These outliers caused by refl ections were misinterpreted as fl oor points, raised the fl oor elevations, and caused the shell settlement analysis to fail. Had the 3D data been interpreted properly, the engineering review and remediation costs would have been avoided.
Analysis and reporting
The last step in the workfl ow is analysis and reporting. Tank analysis is only possible after the diffi cult tasks of data registration and classifi cation have been completed. Although it is possible to extract basic metrics by comparing tank shells to cylinders and tank fl oors to planes, more comprehensive and detailed analysis requires software specialised for storage tanks. In order to compute tank capacities according to calibration standards, software specifi cally designed for tank inspections is required. Complete integration of governing standards into the software greatly reduces the effort required to manipulate the outputs and to present them meaningfully.
While the presentation of outputs may differ slightly across different software packages, the underlying analysis is generally the same. Tank shells are compared to vertical cylinders to show out-of-roundness and plumbness, and tank fl oors are compared to planes, cones, and domes to show local deformations. Floor elevations nearest the shell are extracted to compute shell settlement, and fl oor elevations along radial lines running inwards from the shell are extracted to calculate edge settlement. Floor and shell defl ections are often portrayed as heat maps showing deviations from nominal dimensions (Figure 6) while out-of-roundness and plumbness are shown as exaggerated radar plots (Figure 7).
While many software packages are able to generate the graphics and tables required for a tank inspection report, automatic report generation can not only reduce effort, but can also drastically reduce human error. While the tasks for assembling a report are not diffi cult, they are time consuming, repetitive, and susceptible to mistakes. To a client, simple errors in a tank report can put into question the integrity of the entire workfl ow and jeopardise a business relationship.
Workflow integration
Each of the steps in the workfl ow can be accomplished by off-the-shelf software. However, there is signifi cant advantage in being able to complete all of these steps within the same software package. For example, if a registration error is found during analysis, it is ideal if the registration error can be rectifi ed without having to reclassify the data afterwards. Often, software packages merge all of the scan layers into a single layer when exporting data, meaning it is not possible to go backwards and correct a registration error in the workfl ow, without having to reclassify all of the data.
While there is no magic solution for 3D tank inspections, the right software can signifi cantly improve workfl ow effi ciency. Automation can vastly speed up the process by completing repetitive tasks and reducing errors, but it cannot replace the role of a trained operator who is able to review and interpret the data and the results. The benefi ts of using 3D laser scanning for storage tank inspection are many, and depend heavily on the ability to accurately and effi ciently analyse the data.