A Clmbng Robot based on Under Pressure Adheson for the Inspecton of Concrete Walls K. Berns, C. Hllenbrand, Robotcs Research Lab, Department of Computer Scence, Techncal Unversty of Kaserslautern P.O. Box 3049 Kaserslautern, Germany http//agrosy.nformatk.un-kl..de Emal berns@nformatk.un-kl.de Abstract-- The nspecton of large concrete walls wth autonomous systems s stll an unsolved problem. One of the man dffcultes s the development of a very flexble platform, whch s able to move at horzontal and vertcal surfaces. The platform also has to overcome small obstacles and cracks. In ths paper a prototype of a clmbng robot wll be presented. Keywords-- Clmbng robot, mechatroncs concept, concrete wall nspecton As adheson prncples for the locomoton on a concrete wall only under pressure can be used. In the followng the requrements mposed on a clmbng robot for the nspecton of concrete walls are gven. Then based a thermo dynamcal model of adheson prncples rules for the mechatronc desgn of such a machne are derved. The sensor system necessary to support the locomoton and the basc control behavour s ntroduced. At the end of the paper frst tests wth the machne are descrbed. I. MOTIATION A very nterestng applcaton area for clmbng robots s the nspecton of concrete walls lke brdges, coolng towers or dams. E.g. almost 90% of all motorway brdges n Germany are made from renforced concrete or prestressed concrete. The total number of these brdges s 32 300, the total length s 1197 km. The averaged age s 33 years. To mantan the stablty, safety of traffc and durablty of these brdges regular nspectons are prescrbed by law. In Germany the test procedures and nspecton ntervals are fxed n the German standard DIN 1076. Accordng to ths standard nspectons have to be performed every three years, man nspectons have to be performed every sx years and nspectons after specal occurrences have to be performed after earthquakes, crashes wth vehcles, fres and other unusual stresses. To perform the nspecton tasks complcated access devces are needed at the moment (see fg.1). When scannng lterature for these knds of applcatons one can fnd dfferent types of clmbng robots wth dfferent adheson prncples. The 3 man classes whch can be dstngushed by ther locomoton ablty are [1]: wheeled-drven or chan-drven machnes, legged locomoton, locomoton based on arms and grppers. Fgure 1: Devce to access undersdes of brdges. II. REQUIREMENTS FOR A CLIMBING ROBOT For applcatons as descrbed above, the vehcle should be able to move horzontally, vertcally and over sde on slghtly bended surfaces (radus 5 m). For the nspecton task t s necessary to move contnuously (not possble wth sldng frame mechansms). The machne should also be able to drve over steps of up to 2cm and a wdth of 2 cm. It should carry a maxmum payload of 10kg and
reach a speed of 10 m/mn. The total weght of the machne should not exceed 30kg. Based on the requred moblty the robot should have a turnng cycle of less than 0.5 m and the dameter should not exceed 0.8m. For the use of such a machne a safety cable must be fxed on the robot. Ths safety cable can also be used for energy supply. Because normally an operator s not able to drve the machne from a base staton usng a joystck the machne must be autonomous n a way that the selecton of control behavors s done by tself. III. THEORETICAL BACKGROUND FOR THE DESIGN Based on the frst law of thermodynamcs we dervate the followng equaton whch descrbes the relaton between the change of pressure n a chamber and the volume of the chamber, the area of leakage and the flow rate area to the system whch generates the under pressure. In the equaton p represents the pressure, A L s the area of the leakage, A s the area of flow rate, and s the volume of the chamber. The terms p a -p s the dfference between the pressure outsde the chamber; p-p R s the dfference between the chamber pressure and the pressure generated by the under pressure system. κrt p& = 2ρ [ A p p A p p ] L a R Fgure 2: Smulaton of the 7 chamber system clmbng robot when movng on a vertcal wall over a crack. The gray areas show those chambers whch lost under pressure. It was calculated that n stuaton III, and I the robot s fallng down. The 7 chambers are connected to a under pressure reservor, n whch the vacuum engne s nstalled. As vacuum engne a vacuum cleaner motor s used, whch s able generate a maxmum under pressure of 250 mbar and maxmum volume flow of 0.05m 3 /s (see fg.3). It was calculated that for our above descrbed robot an under pressure of 100 mbar s adequate to hold the machne on vertcal walls as well as upsde down. Wth the help of thermodynamcs smulaton t was shown that the under pressure s changed durng 10ms to normal pressure when a crack s crossed wth a leakage area of 2cm 2. Therefore a one chamber system s not adequate for such type of clmbng robot. More chambers make the robot more robust regardng the roughness of the surface but at the same tme ncrease the techncal effort for ts constructon. As a compromse a 7 chamber system as shown n fg. 2 s selected. Fgure 3: Mechancal desgn of the under pressure system. The 7 chambers are connected to a reservor of 50l volume. In fg. 4 we calculate the necessary under pressure of the reservor n 4 dfferent smulatons. The calculatons are done based on the estmaton that the robot has a weght of 20 kg (no payload s consdered), and the gven dmensons of the robot. The robot s movng over a crack. Those chambers whch are over the crack loose under pressure. In stuaton III the under pressure whch s necessary to hold the machne on the wall, can not be generated by the under pressure system.
large leakage area are closed. One can see that the under pressure of the reservor s nearly constant. Fgure 4: The necessary reservor under pressure when movng over a crack n the drecton shown above. The stuatons I to I are those shown n fgure 2. Up to now a testbed s constructed (see fg.5) to verfy the results whch are calculated n our dynamc smulaton. Fgure 5: The testbed to examne the under pressure systems and the seals of the clmbng robot. As shown n fg. 2 there are several stuaton n whch the under pressure system s not able to generate the necessary under pressure of about 20mbar. Because all chambers are connected to each other va the reservor chamber the under pressure n the chambers and the reservor must be control va valves. In stuatons n whch the leakage area s very low the vacuum engne generates a very hgh under pressure, whch makes the movement of the clmbng machne mpossble. Therefore, one valve s used to adapt the under pressure n the reservor to the normal pressure. Each ndvdual chamber s equpped wth a valve, whch s used to close the connecton to the reservor n case the leakage area s large. In fg. 6 the behavour of the under pressure of the reservor and each chambers s shown when the robot moves over a crack and valves of those chambers wth a Fgure 6: Under pressure over tme of the chamber 1, 2 and the reservor. In (A) a large leakage area crosses chamber 1. The under pressure of the reservor s decreased tll the valve of chamber 1 s nearly closed. Because the valves are not totally closed the control can detect (wth the help of an under pressure sensor) f the leakage area s reduced n a way that the under pressure system s able to generate an adequate under pressure (100mbar for the robot descrbed above). In (B) the leakage area crossed chamber 2 and 6. Also n ths stuaton the reservor under pressure s not changed. In fg. 7 the closed-loop control s presented for the whole under pressure system. The control consders the nformaton of the system dentfcaton unt. Ths unt estmates the leakage area based on the valve areas and the measured pressures nsde the chambers. The strategy of the closed-loop control s to keep the under pressure of each chamber as hgh as possble under consderaton of the maxmum power of the vacuum engne. In normal stuaton (very small leakage areas) the under pressure s set to 100mbar wth the help of a PID controller. Open loop control Closed loop control plant Fgure 7: Closed loop control system for the under pressure system. Based on the strategy of surmountng a crack there exst dfferent stuatons n whch the clmbng robot wll fall down or reman on the wall. Fg. 8 shows a possble
strategy to overcome a crack of a wdth of 2cm and a depth of 2cm. The drven wheels are placed n the corners of the trangular. The forces whch hold the machne on the wall and whch are generated by the under pressure n the chambers can be summarzed to a force vector n a specfc poston (red dot n fg.8). In ths poston there s no torque around the mddle axs of the robot. Only f ths pont s nsde the trangular bounded by the drven wheels the robot s able to move over the crack. Therefore, t s necessary to detect cracks or obstacles 1m ahead to select an adequate moton strategy. For more nformaton about the theoretcal background see [2]. Fgure 8: Optmal strategy of moton to overcome a crack. On the left fgure only 2 chambers are generatng the adheson. After rotatng the robot n the worst case 3 chambers are stll generatng the under pressure. devaton of 2 degree per mnute. As accelerometers ADXL105 from Analog Devce are used. The output of the accelerometer s the sum of dfferent actve forces. There are the gravty force, the acceleraton of the robot and centrfugal forces. If the robot s not movng, only the gravty force s actve. If the robot s movng wth a constant velocty the gravty force and the centrfugal force s actve. It s very mportant to detect the correct drecton of gravty for a vehcle whch s drvng at vertcal planes as ths gves the drecton of the locomoton. Addtonally the accelerometer values can be ntegrated to determne the velocty n the navgaton frame. More precse nformaton about the velocty can be calculated from the odometre measurements. If the robot s movng on vertcal surface the robot can slp n drecton of the gravty force. The dstance of slppng can be determned by a dead reckonng. The developed dead reckonng wheel s shown n fgure 10. I. BASIC SENSOR SYSTEM For the above descrbed closed-loop control of the under pressure of the chambers as well as for the selecton of a moton strategy several sensors are used. The under pressure s measured by the pressure sensor FPM- 120PGR of the company Fujkura. Ths sensor has a resoluton of 0.1mbar whch s suffcent for the control. Even f there s a good adheson t possble for the robot to slp or rotate. For the detecton of those stuatons an nertal system and a dead reckonng wheel are developed. The nertal sensor system conssts of three accelerometers and smple pezo gyroscopes [5]. In fg. 9 the three bg components are the pezoelectrc vbratng gyroscope Seres EN called Gyrostar from Murata, whch are placed orthographcally. After calbratng the offset and ntegratng the measure value, the sensor has a Fgure 10: The dead reckonng wheel The passve wheel conssts of a rotaton unt nsde of the cylnder. The stck pontng out of the cylnder s connected wth a lnear bearng to adjust rough surface levels. At ts end a wheel s mounted on the center of the rotaton unt. The angle of the rotaton unt and the rotaton of the wheel are measured. The calculaton algorthm s based on followng equatons: Fgure 9: The nertal sensor system
n = 1 n ( ) x W = d * cos ϕ = 1 ( ) y W = d * sn ϕ These equatons are calculatng the poston of the wheel (Index w - contact pont of the wheel to surface). The nput s the torson of the cylnder ϕ and the drven dstance of the wheel d. Ths calculaton s a sequence based on the last calculated values. Ths output must be converted to the poston of the sensor (Index s): x = x * cos ϕ y ( ) L ( ) L S W * S = yw * sn ϕ * Insde ths equaton L s the dstance between the center of the cylnder and the contact pont of the wheel on the surface. Wth a buld up test assembly experments were carry out. The path whch the sensor had to measure was a rectangle wth edge dstance of 2 and 2 meter. The measured devaton was less than 10 mm. The results are acceptable for a dead reckonng measure system. Based on the sensors descrbed above the determnaton of the poston and the orentaton wth an accuracy of 1cm n each drecton and 1 n each rotaton axs was possble. Also slppng can be detect by ths sensor system.. SUMMARY In ths paper the theoretcal background for the adheson of our clmbng machne and the control concept s presented. It s shown that a mult-chamber system must be used to ensure safety whle clmbng on concrete walls. The mult-chamber adheson system was realzed wth one reservor chamber, n whch the vacuum engne s nstalled. The reservor s connected va valves to the 7 under pressure chambers, whch hold the machne on the wall. It was shown that by controllng the valves t s possble even f half of the chambers have normal pressure the machne can stll move on a vertcal wall. In normal stuaton (wthout cracks or obstacles) the under pressure n each chamber can be controlled n a way that on one hand safety and on the other hand a contnuous moton wth a fxed velocty can be guaranteed. gravty forces can be used for safer navgaton. At the moment several tests wth our test-platform are done. In parallel an omn-wheel, a manpulator arm and an nspecton sensor system are developed [3][4]. At the moment dams and brdges are envsaged to be target structures to be nspected by our robot system. In future several other structures lke coolng towers, tunnels and others, preferably the ones made from renforced and prestressed concrete, could be also nspected by our system. REFERENCE [1] K. Berns, C. Hllenbrand, T. Luksch Clmbng robots for commercal applcatons, Clawar 2003, Catana, Italy, 2003 [2] H. Hetzler Physkalsche Modellerung und Smulatonsuntersuchung zur Regelung enes radgetrebenen, sch mt Unterdruckkammern haltenden Kletterroboters, Studenarbet, FZI, Karlsruhe March 2002. [3] K. Berns, C. Hllenbrand A Clmbng Robot for Inspecton Tasks n Cvl, 1 st Internatonal Workshop on Advances n Serves Robotcs (ASCER) Bardolno, Italy, 2003 [4] F. Wese, J. Köhnen, H. Wggenhauser, C. Hllenbrand, K. Berns: Non Destructve Sensors for Inspecton of Concrete Structures wth a Clmbng Robot, Clawar 2001, Karlsruhe, Germany, 2001 [5] O. Cosmo, Entwcklung enes nertalen Messsystems zur Lage und Bewegungsbestmmung be gewchtsoptmerten Servcerobotern, Dplomarbet Forschungstentrum Informatk, Karlsruhe, Germany 2003. Also the basc sensor system s ntroduce whch s on one hand sde necessary for the close-loop control and on the other sde for the determnaton of the poston and the orentaton of the robot. It was shown that the knowledge of the orentaton of the robot manly accordng to the