See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/224190318

In-situ robotic interventions in hydraulic turbines

Conference Paper · November 2010

DOI: 10.1109/CARPI.2010.5624460 · Source: IEEE Xplore

CITATIONS READS
11 1,117

4 authors, including:

Bruce Hazel Patrick Mongenot

Hydro-Québec Hydro-Québec
58 PUBLICATIONS 559 CITATIONS 8 PUBLICATIONS 170 CITATIONS

SEE PROFILE SEE PROFILE

All content following this page was uploaded by Bruce Hazel on 20 January 2014.

The user has requested enhancement of the downloaded file.

2010 1st International Conference on Applied Robotics for the Power Industry
Delta Centre-Ville
Montréal, Canada, October 5-7, 2010 1

In-situ Robotic Interventions in Hydraulic Turbines

B. Hazel, J. Côté, Y. Laroche and P. Mongenot,
Institut de Recherche, Hydro Québec, Varennes, Canada

Abstract -- This paper presents the development and circular track, the first joint controlled variable is no longer d1
implementation of a robotic technology designed to perform in- but θ1.
situ interventions in hydroelectric turbines. A new manipulator TABLE 1 Denavit Hartenberg parameters of the Scompi manipulator for a
was designed with a unique, track-based kinematics well suited to straight track.
access turbine blades in a confined space. As most work is done Joint θi αi ∂i di Range
on curved surfaces, the robot relies on a curvilinear space model 1 90 0 0 d1 -5m < d1 < 5m
for trajectory generation. Several processes such as gouging,
2 0 90 ∂2 0 -207 < θ2 < 12
welding, grinding and hammer-peening have been integrated into
the robot to facilitate the maintenance of turbines. The robots 3 90 0 ∂3 0 - 95 < θ3 < 130
have been extensively employed by Hydro-Québec (HQ) for 4 0 90 0 d4 - 53 < θ4 < 214
cavitation and crack repairs in its turbines. Recently, the robots 5 -90 90 0 d5 -356 < θ5 < -4
were used to perform interventions in turbines based on fluid 6 0 -90 0 0 -90 < θ6 < 270
flow numerical analysis. For these new applications, a technology
capable of reshaping the surface’s profile with high precision was
developed. More than 30 successful field interventions involving
Fig. 1 shows the robot's reachable sub-envelope for a
up to three robots working simultaneously have been performed given position on the track. This sub-envelope is displaced
in HQ turbines over the last 15 years. along the curved or flat track to obtain the full envelope. The
sub-envelope shown includes collision avoidance between the
Index terms – Grinding, Hammer-Peening, Maintenance, Robot, robot and itself or the track. The condition number of the
Turbine, Welding robot’s Jacobian matrix [3] is limited to 15 to show the actual
attainable Cartesian space far from the robot's singularity.
I. INTRODUCTION
Considering the down-time required to dismantle a
turbine unit, most utilities perform in-situ inspections and
interventions to address issues such as cavitation and cracking.
Traditionally, these operations were completed manually
which implied considerable health and safety risks.
Productivity and quality of the work completed in the confined
space of the turbine have always been a challenge. In addition,
as most repairs involve welding, distortion and residual stress
concerns limit the problems that could be solved in-situ.

II. THE SCOMPI TECHNOLOGY

Over the last 15 years, HQ researchers have developed
and tested a new technology to robotize these interventions
and broaden the range of applications. At the heart of this
technology is a small 6-axis, track-based, portable robot Fig. 1. Scompi robot's sub-envelope for a given position on the track, with
named SCOMPI (Super COMPact robot Ireq). This multi- constraints on the Jacobian to avoid singularities.
process robot has been specifically designed to work within The envelope resembles two spheres where each
the confined space of axial and radial flow turbines under represents the left or right arm configuration of the robot. The
harsh environmental conditions [1]. inner sphere around the robot's first joint is primarily due to
A. Track-Based Kinematics collision avoidance and is bigger than that in front of the robot
The unique kinematics configuration of the robot and its which is due to a combination of collision avoidance and the
track allows it to work between the closely-spaced turbine condition number. The center hole is a consequence of the
blades. In addition, the robot’s high payload-to-weight ratio typical condition number limits chosen for the majority of the
allows for easy manual installation while providing sufficient processes. The robot can be configured with a greater
force capacity to perform grinding and peening tasks. condition number, however this results in greater instability.
The manipulator’s last five degrees of freedom (d.o.f.) are B. Simulation
rotary joints. Its first d.o.f. is track-based. The track can be In typical industrial applications, small components
straight, circular or a piece-wise sequence of circular track requiring work are brought to a large robotic cell and the robot
sections. Table 1 shows the Denavit Hartenberg (DH) is programmed offline. A number of similar small components
parameters [2] of the manipulator for a straight track. For a are required to justify the initial investment. With in-situ
interventions, a different paradigm is required. Small robots

978-1-4244-6634-4/10/$26.00 ©2010 IEEE

 2
are brought to the location of a large component requiring one where r(u,v) denotes a general 6 d.o.f. location on a bicubic
type of intervention and the robots must be programmed patch, γ, β, α are Euler’s roll, pitch and yaw angles, and
online. Prior to field work, a 3D simulation is run offline to F (u ) = [β1 (u ) β 2 (u ) β 3 (u ) β 4 (u )] (2)
optimize task-dependent parameters such as track curvature,
⎡1 0 0 0⎤
tooling fixtures and robot configurations. Fig. 2 shows such a ⎢0 0 ⎥⎥
simulation for an intervention within the confined space of a
Francis turbine.
[
= 1 u u2 u3 = ⎢
⎢− 3
] 0 1
3 − 2 − 1⎥
= UC
⎢ ⎥
⎣2 −2 1 1⎦
⎡ r (0,0 ) r (0,1) rv (0,0) rv (0,1) ⎤
⎢ r (0,1) r (1,1) r (1,0) rv (1,1) ⎥⎥
Q=⎢ v
⎢ru (0,0) ru (0,1) ruv (0,0 ) ruv (0,1)⎥
⎢ ⎥
⎣ ru (1,0) ru (1,1) ruv (1,0) ruv (1,1)⎦
⎡ β1 (v )⎤ ⎡1 0 − 3 2 ⎤⎡ 1 ⎤
⎢ β (v )⎥ ⎢
0 0 3 − 2⎥⎥ ⎢⎢ v ⎥⎥
F T (v ) = ⎢ 2 ⎥ = ⎢ = C TV
⎢ β3 (v )⎥ ⎢0 1 − 2 1 ⎥ ⎢v 2 ⎥
⎢ ⎥ ⎢ ⎥⎢ ⎥
⎣ β 4 (v )⎦ ⎣0 0 − 1 1 ⎦ ⎣v3 ⎦
r (u, v ) = UCQC T V
the u and v subscripts denote partial differentiation. The Q
matrix specifies the control knot at the corners of the patch. A
composite surface can be obtained from an N x M grid of such
patches. In order to ensure curvature continuity in all
Fig. 2. 3D simulation of an intervention on the upper section of the high
pressure side of the blades of a Francis turbine.
directions, the derivative at each knot must satisfy the
conditions given in (3).
C. Spatial Representation ru , m −1, n + 4ru , m, n + ru , m +1, n = 3(rm +1, n − rm −1, n ) (3)
As most work is done on arbitrarily-curved surfaces, the m = 1,2,...., M − 1
Scompi robot relies on a unique and powerful curvilinear
spatial representation to generate trajectories. In short, a 6 rv, m, n −1 + 4rv, m, n + rv, m, n +1 = 3(rm, n +1 − rm, n −1 )
d.o.f. parametric curvilinear coordinates frame is built to n = 1,2,...., N − 1
ruv, m −1, n + 4ruv, m, n + ruv, m +1, n = 3(rv , m +1, n − rv , m −1, n )
represent a curved region such as the surface of a turbine
blade. Fig. 3 shows the 2D parametric surface corresponding
to a 3d.o.f./6d.o.f. Cartesian space. m = 1,2,...., M − 1
The three orientation d.o.f. at the control knots can either
be given, or computed from the tangent vectors of the surface.
In the latter case, the surface normal n (blue arrows in Fig. 3)
is defined by (4)
r ×r
n= u v (4)
ru × rv
The last d.o.f. can be obtained by aligning the red or green
arrows with the ru or rv tangent vector. The w dimension is a
projection along the blue arrows as expressed in (5).
P(u , v, w) = r (u, v ) + w n(u , v ) (5)

Typically the u and v parameters are adjusted so as to match

the arc length along one side of the surface.
Fig. 3. 2D parametric curvilinear representation of a 3 d.o.f./6 d.o.f. The International Electrotechnical Commission (IEC) [5]
environment. specifies waviness requirements for turbine blades. The
The curvilinear space is composed of a grid of sample resolution of the parametric curvilinear workspace (distance
knots that define a composite parametric bicubic surface. between knots) can be chosen to correspond to the IEC
Within this bounded curvilinear space, the robot is free to specifications and the deviation from a smooth profile can be
move without collision and a tool can be added and aligned readily obtained in parametric coordinates. Fig 4 shows a
normal to the surface. A Ferguson parameterization [4] is region of a turbine blade that was measured with a high
used, as defined in (1). density of points. A composite parametric spline surface, with
r (u , v ) = [x(u , v ) y (u , v ) z (u , v ) γ (u , v ) β (u , v ) α (u , v )] (1) an inter-knot distance of approximately 150mm, was fitted to
the data with the goal of minimizing the sum of the square
r (u , v ) = F (u )QF T (v ) distances between the points and the surface.
 3
The color of the data points indicates the deviation from a when welding, or force when grinding). A diagram of such a
smooth surface. controller is shown in Fig. 6.

Fig. 6. Hybrid curvilinear process controller. Pd is the generalized process

variable to be regulated and Pm is the value obtained from the measurement
Fig. 4. Waviness expressed as the deviation from a smooth composite and process models.
parametric surface.

D. Trajectory Generation III. CAVITATION REPAIR

Once the curvilinear space has been created, most Cavitation in hydroelectric turbines is a degradation
trajectories can be defined as a family of straight lines in phenomenon that greatly damages the turbines in addition to
parametric coordinates. Fig. 5 shows grinding trajectories used adversely affecting the productivity of the turbine/alternator
to reshape a turbine blade. Software has been developed to group. In certain cases, the cavitation is so severe that it can
generate offline the grinding or welding tool paths required to pierce the blade of a turbine wheel. The cavitation [6],
transform a given measured surface into the desired target meaning the formation of pockets or bubbles of vapour at the
profile. centre of a liquid medium, manifests itself very differently
depending on the flow configuration where it develops, the
form and movement of the walls and the liquid's pressure. The
water vapour bubbles subjected to high pressure explode. This
results in the formation of a micro jet which hits the surface of
the blade, removing metallic particles from the surface.
Repeated at the same location, this phenomenon leads to
significant deterioration of the surface characterized by weight
loss and a surface finish of very poor quality.
The turbines affected by cavitation must be repaired
regularly. For the large majority of the utilities, the repair of
the cavitation damage is performed manually. Scompi robots
have been extensively employed by HQ for in-situ cavitation
repairs in turbines. Since 1992, at HQ, large-scale repairs [7]
have been robotized thanks to the introduction of the Scompi
robot: a technology specifically developed for the repair of
Fig. 5. The yellow lines are grinding tool paths generated offline for a section cavitation damage. In the past few years, universities and
of a turbine blade. research laboratories have gotten involved in this sector of
E. Multi-Process Robot activity, in particular Brazil [8] and Australia [9]. It is
however a laboratory prototype.
The maintenance of metal components such as turbines The robotic repair method consists of teaching the robot
involves processes such as gouging, welding, grinding and the work zone in which the welding and grinding will be
hammer peening. These processes have been integrated into completed. The operator first teaches the robot the points
the robot's controller. In typical repair work, a series of tasks which define the zone (see Fig. 7). The operator then
is generated that automatically applies or removes material to determines the contour corresponding to the exact location
or from a contour on a surface. The robot moves the tool along where the welding deposit must be made. It is possible to
a set of juxtaposed trajectories defined so as to fill the contour, program several contours of different dimensions in the same
while controlling the process parameters to achieve the desired work zone. The robot then performs the welding, filling the
layer depth or thickness. To minimize the effect of spatial specified contour. If needed, robotic grinding is performed
programming errors, the robot uses a hybrid control strategy prior to depositing the subsequent welding layer. The final
where five of the curvilinear space's d.o.f. are controlled in welding deposit is grinded by the robot to blend with the
position to follow the trajectories, and the sixth d.o.f., that surrounding area.
normal to the surface, is regulated using process parameters
(ex. Range finder when sensing, torch-to-work-piece distance
 4
surface is put under compression, which increases its
resistance to fatigue and cracking. Robotic hammer-peening
with the Scompi robot was an important part of several
applications [11-12], in particular the reinforcement of the
wheel crown of a Francis turbine [13] where a ring was
created from successive layers of welding. Each welded layer
was hammer-peened in order to counter the deformation
caused by the welding. Fig. 9 presents the hammer-peening of
the reinforcement ring. The two Scompi robots work
simultaneously over 180°, both when hammer-peening and
when welding. Controlling the hammer-peening force keeps
the force command constant. Pneumatic hammer-peening is a
process which is integrated into the Scompi technology, and is
hence controlled 100% by the robot. As with the other
processes integrated into the Scompi robot, the operator can
Fig. 7. An operator teaches the robot the control knots of the curvilinear work
space inside a Francis turbine. react in real-time to modify the hammer-peening parameters.

A. Twin-Torch Welding
With the purpose of increasing welding productivity, an
interesting option of twin-torch welding was integrated into
the Scompi robot to perform a large-scale cavitation damage
repair on a turbine draft tube (see Fig. 8). The height of the
draft tube was 1.37m (54 in). Welding was performed until the
two welding deposits met. The height of torch #1 is adaptively
controlled directly [10]. For torch #2 however, adaptive height
control is achieved by sacrificing one d.o.f. on one angle of
torch #1. This innovative approach allows the height of each
torch mounted on the same tool support and maintained by a
single Scompi robot to be controlled independently. The repair
was performed with three robots, each equipped with a twin-
torch support, working simultaneously. Fig. 9. Hammer-peening of the wheel crown of a Francis turbine. Two
Scompi robots are installed on back-to-back vertical tracks.

IV. TURBINE MODIFICATION

Recently, the robots were employed to perform low-cost
interventions in turbines of old design based on fluid flow
numerical analysis [14]. The hydraulic behaviour of the
existing equipment is analyzed, after which interventions are
designed to optimize the operation of the power plants.
Although the performance gains are small (0.5% to 1.5%
increase in efficiency, or 5% increase in power output), the
projects are very profitable. For these applications, new
capabilities are required to shape the turbine blades in order to
meet the target geometry specified by CFD (Computational
Fluid Dynamics). Specifically, a technology capable of
reshaping the surface's profile with high precision in the
turbine's coordinate system is required.
Fig. 8. Welding overlay on a draft tube with the twin-torch system mounted
on the robot. A. Measurement
B. Hammer-Peening Modern turbine blades are manufactured from castings
An innovative method was developed for high-volume that are shaped to within tight tolerances with very stiff and
welding repairs to resolve distortion and residual stress issues. accurate 5-axis milling machines. Portable robots, well suited
This method involves controlled cold hammer-peening each for in-situ operations, have neither the stiffness nor the
welded layer and has been employed to perform crown accuracy required for the traditional milling approach. To
reinforcements on five turbines at the Beauharnois power avoid these limitations, the robot is coupled with an accurate
plant. measurement system and the strategy employed is to control
Hammer-peening is a widely-known, yet rarely-used either the metal removal or deposition rate.
process, especially robotic hammer-peening. It allows the Fig. 10 shows a laser tracker [15-16] that follows a
constraints to be reduced and the deformation effect which retroreflector fixed close to a laser range finder [17], which
necessitates welding to be countered. The hammer-peened itself is mounted on the robot’s end-effector. Let {W} be the
 5
World fixed frame, {T} another fixed frame in which the laser surface. The phase shift in the oscillation pattern indicates the
tracker returns the position of the retroreflector, {R} a mobile trigger error to be compensated. Using precision triggers, the
frame attached to the center of the retroreflector, {L} another robot's vibrations during measurement are minimized if not
mobile frame in which the laser range finder returns its value completely eliminated.
and {M} the mobile frame at the intersection between the laser
and the surface. The kinematics chain that allows the location
of the laser spot on the surface to be expressed in the World
frame is given by (6).
W W T W ˆ R L
M P = T T R P + P R LT M P (6)
A
where BP denotes the position of the origin of frame {B}
A
expressed in frame {A}, B R the orientation (3 x 3 orthogonal
A
matrix) and BT the linear transform (homogenous 4 x 4
R
marix). If MP is sufficiently small, then the inaccuracy of
W ˆ W
PR has a negligible effect on MP .

Fig. 12. Phase shift correction between the laser tracker, robot and laser range
finder.

B. Target Profile Generation

The target geometry cannot be directly applied to any
turbine blade since each is different. Instead, the difference
between the target profile and the average original profile is
applied to each blade. Tests were carried out on a sample
blade in the laboratory to quantify the deformation which
results from performing both welding and grinding processes
on the same blade. Based on these results, the final target
Fig. 10. To minimize the measurement error which is due to the robot’s geometry and welding patch are modified to minimize the
angular error, D must be minimal.
deformation. Fig. 13 shows how this technique was applied to
The precision obtained on a flat marble surface (600mm x
modify a blade. The blade is shown in parametric coordinates
400mm) is approximately 50 microns when the laser tracker is
before and after welding. The target profile lies in between the
located within two meters. Fig. 11 shows the measurement of
two and can be obtained by grinding off the excess.
a turbine blade section at the Manic-3 power plant.

Fig. 13. A target profile is generated taking into account the deformation
Fig. 11. Faro laser tracker combined with an Acuity proximity sensor caused by welding. The overlay is done in three layers. The extremity of the
mounted on the Scompi robot inside a Francis turbine. extension plate is not welded, however due to the deformation of the blade, it
Since the measurements are taken dynamically, a trigger is displaced by more than 10mm.
signal is sent to each device. Trigger synchronization between Reshaping the blades into a relative target shape rather
devices is adjusted with a resolution of one millisecond in the than an absolute one prevents the turbine from becoming
robot's measurement controller. The synchronization is tuned unbalanced. The blade is measured after each step and the
from an oscillation pattern generated by moving the robot target shape can be adjusted in-situ if an unforeseen situation
back and forth like a pendulum over a precision marble arises.
 6
C. Field Trials [4] J. Ferguson. “Multivariable curve interpolation” J ACM, vol. 11, no. 2,
pp. 221-228, 1964.
Precision grinding was employed to reshape two runners [5] Hydraulic turbines, storage pumps and pump-turbines – Model
at the Manic-3 power plant with a precision of 0.2 mm with acceptance tests, IEC Standard 60193, 1999
respect to the target surface. In this first case, only the [6] F. Avalan, La Cavitation, mécanismes physiques et aspects industriels,
Presse Université de Grenoble, 1995.
grinding process was necessary since the work required was to [7] Y. Laroche, B. Hazel, J.-L. Fihey, “On site Repair of Hydraulic Runners
shorten and thin the blades. At the Bersimis-2 power plant, a at Hydro-Quebec”, Proceedings of the Int. Forum on Situ Robotic
more complex intervention was necessary. The blades of two Repair of Hydraulic Turbines, pp. 18-26, Harbin, China, 11-12 Nov.,
turbines had to be lengthened and thickened to achieve the 2000.
[8] R. D. Leal, “Impactos Sociais e Economicos da Robotizacao: Estudo de
desired surface. In this second case, both welding and grinding Caso do Projeto Roboturb“,Universidade Federal de Santa Catarina,
processes were required. Fig. 14 shows a Scompi robot Brazil, Mar. 2005.
grinding inside a turbine at the Bersimis-2 power plant. The [9] D. A. Rodriguez, J. Norrish, A. Nicholson, “Robot programming for non
robot is equipped with a custom-designed electrical grinder repetitive repair operations using vision systems“, trends in welding
research, proceeding of the 8th International conference, pp. 671-675,
that uses a permanent magnet synchronous motor. The grinder Australia, 2008.
weighs only 3kg, but can deliver up to 2.5kW of continuous [10] H. Fujimura, E. Ide and H. Inoue, “Joint Tracking Control Sensor of
power at 7000 rpm. GMAW,“ Transaction of the Japan Welding Society, vol. 18, no. 1, Apr.
1987.
[11] D. Thibault, R. Simoneau, J. Lanteigne and J.-L. Fihey, “Residual
stresses induced by robotized hammer-peening”, ICRS 7, Xi’an, China,
June 2004, Materials Science Forum, Vols. 490-491, pp. 352-357, Jul.
2005.
[12] J.-L. Fihey, R. Simoneau, J. Lanteigne, D. Thibault and Y. Laroche,
“Controlled Hammer-Peening on a restrained A514 (S690Q) weldment.”
High Strength Steels for Hydropower Plants, Graz, Austria, Jul. 2005.
[13] Y. Laroche, P. Mongenot et B. Hazel, “Multi-process portable robot
used in Francis hydraulic turbine at Beauharnois power plant to
reinforce the crown”, Canadian Welding Association Journal –
International Institute of Welding Special Edition, pp. 33-35, Aug. 2006.
[14] A.-M. Giroux, S. Houde, Y. Laroche, R. Dubois, “Improving the
performance of a 20 years-old Francis Turbine using numerical
simulations and robotized intervention,” 24th IARH Symposium on
Hydraulic Machinery and Systems, Iguassu, Brazil, 2008.
[15] K. Lau, R.J. Hocken, and W.C. Haight, “Automatic Laser Tracking
Fig. 14. Scompi robot grinding the upper area of the high pressure side of a Interferometer System for Robot Metrology”, Precision Engineering,
blade to smooth its profile after welding. vol. 8, no. 1, pp 3-8, 1986.
[16] K. Lau, R.J. Hocken, “Three and Five Axis Laser Tracking System”,
U.S. Patent 4,714,339, Dec. 22, 1987.
V. CONCLUSION [17] R. G. Dorsch, G. Haulsler and J. Herrmann, “Laser Triangulation:
Approximately 30 Scompi robot prototypes, of three Fundamental Uncertainty in Distance Measurement”, Applied Optics,
vol. 33, no. 7, Mar. 1994.
different generations, have been manufactured to date. More
than 30 successful field interventions, involving up to three
robots working simultaneously, were performed in HQ
turbines over the last 15 years. The robotic technology has
also been tested in the field by other utilities, such as
Électricité de France (EDF). In-situ robotic intervention
technology is well-proven and offers a viable alternative to the
manual repair of hydroelectric turbines.

VI. ACKNOWLEDGMENT
The authors wish to extend their gratitude to the numerous
colleagues and collaborators who contributed to the work
completed at the research institute, as well as the employees of
HQ’s Atelier Mécanique in Shawinigan, Québec who
performed the field tests.

REFERENCES

[1] Fihey, J.-L., Hazel, B. and Laroche, Y., "The Scompi Technology",
Proceedings of the Int. Forum on Situ Robotic Repair of Hydraulic
Turbines, Harbin, China,11-12 Nov., 2000.
[2] J. Denavit and R.S. Hartenberg. "A kinematic notation for lower-pair
mechanisms based on matrices." Trans ASME J. Appl. Mech, vol. 23, pp.
215-221, 1955
[3] R. Buckingham, P. Brett and K. Khodabandehloo, “Correlation of
Condition Number with Forces Using an Experimental Two Arm
Robot”, ISART, Tokyo, Japan, pp. 301-306, Mar. 1991.

View publication stats