Canadian provincial utility Hydro-Quebec has used on-site measurement techniques to determine the geometry of turbines and other components for more than 15 years. As the technology has evolved, so have the types of measurements that can be performed.
By Genevieve Gauthier, Robert Magnan and Stephan Beauregard
It is often valuable to have detailed geometric information on turbines installed in hydro facilities. However, the technical drawings provided by turbine manufacturers generally do not provide runner geometry, in particular the complex shape of the blades. This is mainly due to the difficulty of representing curved three-dimensional surfaces on a two-dimensional piece of paper but also may stem from manufacturers’ desire to protect their trade secrets.
To overcome this lack of knowledge, Hydro-Quebec has measured turbine runners in its plants for more than 15 years. This work was done to provide the 3D blade geometry needed for structural analysis and fluid flow simulations to assess the behavior of turbines, detect possible failures and improve efficiency. More recently, the utility has extended its measurements to other turbine components, such as casings, draft tubes, penstocks and even generators.
The rapid development of digitizing equipment and software has reduced turnaround time and improved overall accuracy. This, in turn, has opened up new possibilities for its use by an electric utility such as Hydro-Quebec.
The reverse engineering process
Central to the analysis of a turbine component is a computer model that represents the geometry. When drawings do not include dimensions, on-site measurements can provide this information. This is the heart of the reverse engineering (RE) process. For complex geometries involving curved shapes, a typical approach is to sample 3D points all over the object and build smooth surfaces. Depending on the complexity of the object and the accuracy level required, points sampled can vary from a few hundred to millions (point clouds).
When only sparse points are available, they are used to construct a set of curves acting as a skeleton for building surfaces. When a dense point cloud is available, a typical modeling technique consists of “shrink wrapping” a set of smooth surfaces over these points.
There are many challenges involved in measuring complex geometry on site without dismantling. Some concerns are access to the blades, calibration, installation of sensitive equipment in awkward places, sensor stabilization, registration of multiple 3D datasets, surface finish, axis location, edge recovery, data smoothing, model creation and overall accuracy assessment – are also critical to the success of an RE project.
Use cases
The below case studies show how the technology Hydro-Quebec uses has improved over the years.
Single point measurements of Francis turbine blades
Early in the 1990s, Hydro-Quebec chose a portable digitizing arm with a touch probe to measure one blade of a 5-meter-diameter Francis turbine. The arm had 6 degrees of freedom, a reach of about 1 meter and reasonable accuracy (0.1 mm).
There are no adequate flat surfaces in a Francis turbine to support the arm. Installation consisted of welding steel plates to the turbine and clamping the arm to these plates. Proper location of each plate involved defining its position, its orientation (they did not absolutely need to be laid horizontally), and the offset distance to allow room for the electrical connectors and unhampered rotations.
Covering the entire blade required moving the arm successively to each plate and recording the points in a common coordinate system using so-called leapfrog reference points to establish a geometric registration. A minimum of three noncolinear points is required to define a suitable transform between locations. To maximize the work envelope, these points are placed on the outskirts of the arm’s reach – where accuracy is at its worst. We generally used six points in a least-square fashion to account for lack of accuracy in measurement of the reference points.
The measurements start by laying out a grid on each side of the blade using a marker, after which individual points are digitized at each intersection (see Figure 1). To define the axis of the turbine, a surveying theodolite with an electronic distance meter was used to measure multiple points at the machined base of the shroud from the center of the runner. These were then used to compute a best-fit circle. The normal line through the center serves as the axis. One on-site blade measurement required about 300 man-hours.
The modeling work was done using Autocad and some in-house tools. The idea was to build principal curves through the digitized points and loft through them to create surfaces. This 3D curve-based method often results in bumpy surfaces. To fix that problem, a scanning technique was developed where a ball probe was swept across the blade’s surface. A new point gets created whenever the probe crossed a predefined cross-section plane. Compensation for the probe radius is handled in the modeling stage.
The 3D models were used for flow simulations in the entire turbine, which eventually led to the discovery of a recirculation flow in the draft tube at nominal operating conditions. Alteration to the blades’ trailing edge was designed and increased turbine efficiency by 1.5%.1
Non-contact measurement of a Francis turbine blade
To diagnose the source of a recurrent cracking problem with another Francis turbine, we wanted to build an FEA model to compute the stresses and a CFD model to compute the blade loading.
At the time (in 2000), we had bought a new digitizing arm with slightly less accuracy but greater reach (1.2 meters) and thinner tubes. A third-party laser line scanner was fitted to this arm, which allowed for the non-contact measurement of up to 10-cm-wide point stripes. This setup could be used to rapidly provide a high level of detail in critical areas like the trailing edge and fillets. All the modeling could be done using the points directly, without any compensation.
We also measure features at the junction between rotating and non-rotating parts to computed fitted circles (inlet and outlet band circles, inlet crown circle) in order to establish the unit’s axis.
Drawbacks included the laser’s sensitivity to surface finish and color. The blades had rusty spots, peeling paint, different levels of wear, cavitation pitting and some shiny areas. The power has to be adjusted to ensure proper contrast. In high-reflectivity areas, a strong laser line produces much noise in the data, and the point cloud thickness is unduly increased. Conversely, an underpowered laser fails to be captured over dark materials.
To handle the large point clouds produced by the new arm, major software upgrades were required. Using this technology, the power plant’s blades are still too large to scan completely; however, we stopped measuring points on a pre-established grid and started free scanning stripes all over the blade. We ended up with multiple stripes of points, more or less uniformly laid out (see Figure 2).
One difficulty with this technique involved the size of the laser scanner. It was bulky and required some 10 cm of stand back from the blade so that the triangulation system would work. In some parts of the runner passage (close to the trailing edge), it was difficult to fit the sensor between the blades.
The system was prone to frequent decalibration due to collision of the sensor with the blades. But it was only by rescanning the same region twice and comparing the before-and-after results that we became aware recalibration was required. Calibration was difficult and time-consuming. It involved moving the entire setup to a remote location (spiral case), which allowed for the various movements of the arm required. Furthermore, vibration during that process was a concern.
This new system did not improve the scanning time, but it provided more detailed information. It also had a profound effect on the way we build CAD surfaces from point clouds. We started to use a shrink-wrapping technique whereby a first roughly defined surface is constructed and then deformed so it passes through the point clouds underneath it. This method also smoothes out the data because the surface only approximates the point clouds.
Eventually, we purchased more powerful modeling software that allowed us to compute best-fit registrations using the overlapping portions of the point clouds. The best-fitting algorithms of the software iteratively adjust the transform between the point clouds to minimize discrepancies in a least-square fashion. This works well as long as there are corresponding 3D geometric features in both clouds, which enable the algorithm to find a unique match. This requirement can be hard to meet when there is only limited overlap in the blade area between two point clouds. When this happens, we sometimes resorted to adding a tiny object, such as a half-inch round magnet, on the blade to create a recognizable feature that can easily be edited out at the modeling stage.
Fourteen turbines were measured using this technology from 2000 to 2005.
Surface digitizing of a complete Kaplan turbine
Hydro-Quebec has used a white-light scanner since 2006. Composed of two cameras and a projector, it measures 3D coordinates by projecting white-light fringe patterns onto the surface. The cameras record the images of the patterns, which become distorted due to the relief of the surface. Using the two images, the software computes 3D coordinates for each pixel by triangulation. Data gathering for one area typically takes a few seconds and provides about 1 million data points without contact, while defining the shape of the measured area accurately and with much detail.
The control software allows the camera to be progressively moved to extend the measured surface. Registration is a two-step process, where first a rough alignment is computed using small circular markers affixed to the surface. The software automatically recognizes these markers and their random arrangement and computes a registration transform. Next, a best-fit registration using the overlapping point clouds fine tunes the transformation. In the end, all the points are automatically transformed in the global coordinate system linked to the first measurement.
The average spacing (resolution) of the digitized points and the measurement volume that can be digitized in one scan is controlled by switching lenses on the cameras and projector and changing camera spacing. The further the cameras are apart, the larger the view volume. A less obvious effect of changing view volume is the related change in depth of field and standoff distance. The large view volume (about 1 cubic meter) offers satisfactory accuracy, but the standoff distance requires the sensor to be about 1 meter away from the closest surfaces, which is not practical in a Francis turbine. Also, the sensor can only measure the points that are simultaneously in the field of view of both cameras and with a reasonable incidence angle. Having the cameras far apart makes it difficult to position the sensor between the blades.
Such a scanner is usually held using a tripod. In a turbine, where there is no horizontal flat surface and the passage between the blades is narrow and skewed, different tools had to be developed. We use an extensible pole that can be attached to steel surfaces using magnetic blocks. In installations where non-magnetic materials are used (such as cast iron or stainless steel), another type of extensible pole is stretched between the ceiling and floor of the distributor to support the first pole. This setup is not stable, but we found we could obtain a precise measurement by giving it time to stabilize (about 10 to 20 seconds). To facilitate positioning of the sensor and easily change its orientation, we used a hydrostatic camera ball head to mount the sensor to the pole. Such heads can handle the torque induced by awkward positioning.
We used the white light scanner to investigate a vibration problem in a 5-MW Kaplan turbine. With this scanner, all five blades could be digitized in one day (see Figure 3). By comparing the five point clouds, the shape of all the blades was found to be similar; however, the angular positioning of one of them was off by 0.4 degrees.
Blade repair
In 2008, blade failure occurred in a 7.5-MW Francis turbine: part of the blade near the trailing edge was lost. By digitizing the broken area of the blade and its neighbor with the white-light scanner, we were able to design and manufacture a replacement part. The runner was back in service within five weeks. The challenge here was linked to the small size of the runner. The turbine has 17 blades and a throat diameter of 2.1 meters, leading to narrow blade passages. Access was limited to a tiny balcony below the blade.
Due to the limited accessibility, we chose to digitize only the suction side of the blades and to do manual thickness measurements with a caliper in the failure area. The data was combined in a CAD modeler to obtain the shape of the missing region. A corresponding part was manufactured and successfully welded on site.
Complete Francis turbine measurements
The need for accurate CFD results puts more stress on the geometry measurement process. Until recently, a single blade was deemed representative of the entire runner. We frequently observed discrepancies between the various blades of the same runner, but only recently have we been able to assess their impact on overall machine performance.
Measurement of an entire runner is now possible. This was confirmed by measuring the blades of a 5-meter turbine. To help understand a vibration problem in one Francis runner, we studied the disparity between its 13 blades. We scanned the visible part of the suction side of all the blades using the white-light scanner in a large view volume setup. This gave us a skeleton upon which to attach more detailed measurements of individual blades, including the pressure sides. When the scanner is installed in one passage, both adjacent blade surfaces were measured. This helps reduce the time required to go around the runner. The full measurement took nine days and resulted in 640 scanned surface patches for a total of 200 million points.
The shape and position of every blade were compared, and we found significant variations in the blade shape near the trailing edge, local defects like cavitation zones, and vertical and azimuthal alignment defects. The net effect of these differences on overall hydraulic behavior has not yet been determined.
Modeling process
The data points can sometimes be used directly to study a scanned geometry. However, engineering software often requires a CAD model made up of curves and surfaces. This is what the modeling process is all about.
As mentioned, the first step in building a 3D model is to record all the point clouds in a common coordinate system. The data are then processed to build a fine triangle mesh to obtain representation of the surfaces. Average points are calculated in overlapping areas. To improve the mesh, erroneous points can be filtered, missing regions interpolated, and noisy patches smoothed.
To obtain a CAD representation of the geometry, curves are laid on the mesh to define the boundaries of a patch network. Surfaces are created on the curve network and iteratively deformed until they fit the triangle mesh. This iterative fitting process is stopped when the deviation error between the set of surfaces and the scanned points is acceptable, typically between 1 and 3 mm depending on location and turbine size.
To reduce the modeling effort, we are trying to use the triangulated mesh directly in the simulation software. This type of 3D model represents all the scanned details but also conveys surface imperfections and measurement noise.
Runner, spiral case and draft tube measurements
We recently used a long-range (LR) laser scanner to help reduce scanning time. This device has a rotary head that can make a 360-degree turn around the vertical axis and 320 degrees around the horizontal one, leaving out only a conical region at the base of the scanner. It measures the distance to all the surrounding surfaces with a 3-mm accuracy. Depending on the point resolution, scanning can take 1 to 50 minutes.
With a maximum reach of more than 70 meters, the LR scanner is perfectly suited to digitize large components like penstocks and draft tubes but can also be used for smaller objects provided details are not needed. We successfully digitized all the surfaces of a large (8-meter-diameter) propeller turbine in a few hours.
In a Francis turbine, the LR scanner can rapidly digitize the visible surfaces of each blade from underneath the runner. The resulting point cloud is noisier than the one obtained with the white-light scanner, but they match up well on average. This allows the two datasets to be combined to build a 3D model. Data from the LR scanner is used to position all the blades in the same coordinate system and to identify the axis location. The white-light scanner is used to add detail measurement and the surfaces in the blade-to-blade channel that are invisible to the LR scanner.
Discrepancies in efficiency between different units of a six-unit power plant brought up questions about the homology of the waterway geometries. To find out if they comply with technical drawings, we digitized all the peripheral components (penstock, spiral casing and draft tube) using the LR device (see Figure 4). A few blades from three different units were also digitized with the white-light system, and the visible surfaces from the runner center were scanned with the LR device. Measurements showed blade differences between the units in the trailing edge area. Detailed comparisons of the spiral case and draft tube geometries have yet to be done.
Penstock and semi-spiral case measurements
Efficiency testing using Gibson’s pressure-time technique requires that the “shape factor” of the waterway be worked out in the pressure (piezometer) measurement section to estimate the total water flow. This is traditionally established using surveying equipment to measure a few points around the penstock.
In 2008, such penstock dimensioning was conducted using our digitizing equipment. This 60-meter-long tunnel, 5 meters in diameter, can be described with a few hundred points. We elected to use a laser tracker to follow a spherical reflector through space. As long as a line of sight is maintained between the tracker and the sphere’s corner mirror, the tracker provides 3D coordinates with high accuracy (±0.03mm) over 70 meters. The digitized point is the center of the target sphere; hence, the real surface must be deduced by normal offsetting. The newer trackers make it possible to simply “pick up” the laser beam when visual contact is broken. The older ones need to physically sync by moving the sphere back to its seating on the tracker.
In 2009, still in the context of efficiency testing, it was necessary to take measurements in the semi-spiral casing of another plant. Eight piezometers are located in the middle of 12-inch by 12-inch steel plates attached to the walls, floor and ceiling of the large (20-meter-long) concrete casing. The LR scanner was used to digitize the casing. Using the cloud data, piezometer positions were extracted and section areas calculated. The same measurements were also made with the traditional method using a total station. A comparison between piezometer positions obtained using both methods shows differences of less than 5 mm. That study showed that the LR measurement method can be an alternative to the traditional one.
Conclusion
A range of 3D measurement sensors/scanners is available, and they are becoming more precise and easier to set up and calibrate. The availability of fast computers, high-precision electronics and laser equipment, and dedicated software, as well as the steadily decreasing price of sensors, all contribute to making this technology widely accessible. For occasional needs, independent contractors can provide required tools and expertise.
The field of applications for 3D models is constantly growing. We have demonstrated the atypical use of high-precision equipment for digitizing the various parts of hydraulic turbine units. In our power plant environment, every measurement is unique and no single type of measuring equipment is ideal for every application. Accuracy, accessibility, resolution, range, visibility, and environmental conditions all have a role in selecting which sensor to use. Often, the solution involves combining data from multiple sensors.
There are many challenges still to be overcome, including:
- Inferring a global rotation axis for the unit;
- Ability to deal with the high incidence angle surfaces that are often found, especially in Francis turbines; and
- Trailing edge definition, especially in the modeling stage. The reality often takes is a soft, slightly rounded edge
With all of these challenges, even though we use sensors with much higher accuracy, runner measurement will have an overall accuracy in the 1-2 mm range. It can reach 5 mm in larger components.
Modeling has is own set of issues. The most notable is the time required to build a complete surface model.
Finally, we are looking into new ways of carrying out complex measurements, with greater speed and more accuracy.
Authors’ note:
In consideration for the intellectual property of the manufacturers, the turbine geometries shown in this article have been distorted.
Note
1Masse, B., M. Page, A.-M. Giroux, and R. Magnan, “Improving Efficiency of a 195 MW Francis Turbine Using Numerical Simulation Tools,” Proceedings of the 20th IAHR Symposium on Hydraulic Machinery and Cavitation, International Association for Hydro-Environment Engineering and Research, Madrid, Spain, 2000.
Reference
Gauthier, Genevieve, Robert Magnon and Stephan Beauregard, “On-Site Turbine Geometry Measurement at Hydro-Quebec,” Proceedings of HydroVision International 2011, PennWell Corporation, Tulsa, Okla., 2011.
This article has been evaluated and edited in accordance with reviews conducted by two or more professionals who have relevant expertise. These peer reviewers judge manuscripts for technical accuracy, usefulness, and overall importance within the hydroelectric industry.
Genevieve Gauthier and Robert Magnan, P.E., are R&D engineers, and Stephan Beauregard is a mechanical technician at Hydro-Quebec’s Research Institute (IREQ). Gauthier focuses on digitizing and three-dimensional modeling of complex geometries, such as turbine components. Magnan works on numerical simulations of hydraulic flows. Beauregard performs experimental measurement, including digitizing and thermography.
More HR Archives Issue Articles
