Monitoring of Concrete Bridges with Long-Gage Fiber Optic Sensors DaNteLe INAUDI* AND SAMUEL VURPILLOT IMAC— Stress Analysis Laboratory, Swiss Federal Institute of Technology, Department of Civil Engineering CH-1015 Lausanne, Switzerland and SMARTEC SA, via al Molino 6, CH-6916 Grancia, Switzerland ABSTRACT: In many bridges, the vertical displacements are the most relev monitored in both the short and long term. Current methods (su mechanical exiensometers, . .) are often tedious to use and require the intervention of specialized ‘operators. The resulting complexity and costs limit the temporal frequency of these traditional mes surements, The spatial resolution obtained is in general low and only the presence of anomalies in the _slobal structural behavior can be detected and warrant a deeper and more precise evaluation. To measure bridge vertical displ quently in time, one solution consists of installing a network of fiber optic sensors during concrete pouring or installing them on the surface of the structure. By subdividing the whole structure into structural elements and those elem cells that are analyzed by the sensors, itis possible to obtain information about the average cell d formation (e.g., mean curvature) that can then be combined to obtain the global structural displace ment field. In 1996, a concrete highway bridge near Geneva (Switzerland) was instrumented with more than 100 low-coherence fiber optic deformation sensors. The Versoix Bridge is a classical concrete bridge consisting in two parallel pre-stressed concrete beams supporting a 30-cm concrete deck and two ‘overhangs. To enlarge the bridge, the beams were widened and the overhang extended, In order to in- crease the knowledge on the interaction between the old and the new concrete, we choose low-coher tence fiber optic sensors to measure the displacements of the fresh conerete during the setting phase and to monitor the long term deformations of the bridge. The aim is to retrieve the spatial displace- nents ofthe bridge in an earth-bound coordinate system by monitoring its internal deformations. The t parameters to be 2h as triangulation, water levels or Is into vertical and horizontal curvatures of the bridge are measured locally at multiple locations along the bridge span by installing sensors at different positions in the girder cross-section, By taking the double integral of the curvature and respecting the boundary conditions, it is then possible to retrieve the deformations of the bridge. This measurement methodology was also applied to the Lutrive Highway Bridge in Switzerland in order 1o measure the variation in Vertical bridge displacements due to a static load test, The results obtained using the low coherence interferometric sensors of the SOFO system were then compared with the displacements obtained through an optic leveling system. In the case of this cantilever bridge of 60 meters half-span equipped with 30 fiber optic sensors, a discrepancy of less than 7% was obtained between the two measuring systems. INTRODUCTION measurements often spaced over long periods of time (Hassan, Burdet and Favre, 1995; Burdet, 1998) The monitoring of a new or existing structure can be ap- proached either from the material or from the structural point of view. In the first case, monitoring will concentrate oon the local properties of the materials used in the construc~ tion (¢.g., concrete, steel, and timber) and observe their be- Tit security of bridges requires periodic monitoring maintenance and restoration, Excessive and non- stabilized deformations are often observed and although they rarely affect the global structural security, they can lead to serviceability deficiencies. Furthermore, accurate know!- edge of the behavior of bridges is becoming more important as new structures become lighter and as an inereasing number of existing bridges is required to remain in service beyond their theoretical service life, Monitoring, both in the long and short term, helps to increase the knowledge of the real behav- ior of the bridge and in the planning of the maintenance inter- ventions. In the long term, static monitoring requires an accurate and very stable system, able to relate deformation “Auror trom correspondence sould be aes havior under load or aging. Short base-length strain sensors are the ideal transducers for this type of monitoring ap- proach. Ifa very large number of these sensors are installed at different points in the structure, it is possible to extrapo- late information about the behavior of the whole structure from these local measurements. In the structural approach, the structure is observed from a geometrical point of view. By using long gage length deforma- tion sensors with measurement bases of the order of one toa few meters, itis possible o gain information about the deformations of the structure asa whole and extrapolate on the global behav- 280 Jouswat OF INTELLIGENT MareRtat SrsTeMs AND STRUCTURES, Vol. 10—April 1999 1045-389X,99104 0280-13 $10,010 {6 2000 Technomic Publishing Co., ne.ior of the construction materials, The structural monitoring ap- proach will detect material degradation like cracking or flow only if they have a direct impact on the form of the structure. This approach usually requires a reduced number of sensors when compared to the material monitoring approach, The availability of reliable strain sensors like resistance strain gages or, more recently, fiber Bragg gratings (Vohra et al., 198) has historically concentrated most research efforts, in the direction of material monitoring rather than structural monitoring. This latter has usually been applied using exter- nal means like triangulation, dial gages and invar wires. In- terferometric fiber optic sensors offer an interesting means of implementing structural monitoring with internal or em- bedded sensors (Inaudi et al. 1998a), THE SOFO YSTEM The functional principle of the SOFO system (Short for “Surveillance d’Ouvrages per Fibers Optiques”) is schema tized in Figure 1. The sensor consists of a pair of single- mode fibers installed in the structure to be monitored and having a length between 0.2 m and 10 m. One of the fibers. called measurement fiber, is in mechanical contaet with the host structure itself, while the other, the reference fiber, is placed loose in a neighboring pipe. All deformations of the structure will then result in a change of the length difference between these two fibers (Inaudi et al., 1997). ‘To make an absolute measurement of this path unbalance, a low-coherence double Michelson interferometer in tandem configuration is used. The first interferometer is made of th measurement and reference fibers, while the s tained in the portable reading unit. This second interferome- ter can introduce, by means of a scanning mirror, a well-known path unbalance between its two arms, Because of the reduced coherence of the source used (the 1.3 micron radiation of an LED), interference fringes are de- tectable only when the reading interferometer compensates the length difference between the fibers in the structure. scond is con- smc under test Couser sescng Unt SP A fy Phato- LED, Diode Figure 1. Setup of the SOFO system. If this measurement is repeated at successive times, the evolution of the deformations in the structure can be fol- lowed without the need of a continuous monitoring. This means that a single reading unit can be used to monitor several fiber pairs in multiple structures. The signal detected by the photodiode is pre-amplified and demodulated by a band-pass filter and a digital envelope filter. The precision and stability obtained by this setup have been quantified in laboratory and field tests to 2 micron, inde- pendently from the sensor length over more than four year. Even a change in the fiber transmission properties does not affect the precision, since the displacement information is, encoded in the coherence of the light and not in its intensity, Figure 2 shows a typical sensor for length up to 10m. This sensor is adapted to direct concrete embedding or surface mounting on existing structures. The passive region of the sensor is used to conneet the sensor to the reading unit and can be up to a few kilometers long. The reading unit is portable, waterproof and battery powered, making it ideal for dusty and humid environments as the ones found in most building sites. Fach measurement takes about 10 seconds and all the results are automatically analyzed and stored for further interpretation by the external, laptop computer. The measurements can either be performed manually, by connecting the different sensors one after the other, or automatically by means of an optical switch. Since the ‘measurement of the length difference between the fibers is, absolute, there is no need to maintain a permanent con- nection between the reading unit and the sensors. A single Figure 2. Standard SOFO Sensor.unit can therefore be used to monitor multiple sensors and structures with the desired frequene; STATIC DATA ANALYSIS ALGORITHMS In beams, slabs, vaults and domes, itis possible to measure the local curvature and the position of the neutral axis by measuring the deformations on the tensile and compressive sides of a given element. In many cases, the evolution of the curvature can give interesting indication on the state of the structure, For example, a beam, which is locally cracked, will tend to concentrate its curvature at the location of the cracks. Furthermore, by double integration of the curvature function, it is possible to retrieve the displacements perpendicular to the sensor direction (Vurpillot, Inaudi and Seano, 1996: Perregaux et al.. 1998). This is particularly interesting since in many cases the engineers are interested in deformation that are at a right angle to the natural direction in which the fiber sensors are installed. For example: in a bridge fibers are in- stalled horizontally, but vertical displacement are more inter- esting. In a tunnel the fibers are placed tangentially to the vault, but measurement of radial deformation is required. Ina dam the fibers are installed in the plane of the wall but dis- placements perpendicular to it have to be measured. Monitoring Principles In a structure undergoing complex deformations, it is often necessary to install a large number of sensors to obtain information. This presents two major difficulties: it is first necessary to decide the number, size and position of the sensors to be installed and then to analyze the huge data flow that results from the measurements, The deformations mea- sured by each single sensor rarely give useful information. Only an appropriate correlation between the values obtained by a much larger number of sensors and sometimes all of them generates data that can be correlated to observable quantities like the vertical displacement of a bridge or the crack widths in a conerete beam, The analysis method pre- sented here helps both in the engineering of the sensor network and in the analysis of the resulting measurements. The main principle of this method consists in subdividing the structure in to a number of macro-elements that undergo relatively simple deformations. These sections are further segmented into ce/ls containing only a few sensors or even a single one. To get more insight into this process we will con- centrate on beam-like structures that are much larger in one dimension than in the other (wo. Examples of these strue- tures include bridges, trusses, towers, frames or arches. The principle can however be extended to plate-like structures like shells, domes, shell-dams (Kronenberg et al., 1997), slabs (Poh and Baz, 1996) or wings. The first step requires the structure to be subdivided into sec- tions. Each section is supposed to have a constant or continu ously varying inertia, a constant load across its length and so ¥ Bsuipor secton secion2 secton3 entre cal Figure 3. Structure subdivision in Sections and Cells. introduction of local forces and supports only at its ends (see Figure 3). If the behavior of the materials in a section can be considered as homogeneous (¢.g,,1n0 local cracking), the poly- nomial degree that best approximate its deformation is deter mined either analytically or using finite-element programs. Ifa degree N+2 is found to approximate satisfactorily the defor mation of the section, this will be subdivided into N cells. Fora beam with constant inertia the deformation is polynomial of fh degree and three cells are therefore necessary (Vurpillot, Inaudi and Scano, 1996). For sections with variable inertia, itis sometimes useful to use cells with variable size. If local variations of the material behavior are expected, more cells are needed. In this case the number N and the size of the cells will be determined in such a way that the material properties inside each single cell ean be considered reason- ably constant. The deformation of the section will in this case be approximated by a polynomial of at most N+2 degree. Cell Analysis The sensors inside a single cell will be used to determine the local behavior of the materials and the global behavior of the cell itself. The sensors will generally be positioned paral- lel to the beam axis at different heights (see Figure 3). The bbase-length of the sensors can be shorter that the cell length, but the best precision is usually obtained with sensors having the same length as the cells. The most relevant parameters that can be analyzed at the cell level are the following: Mean Curvature The radius of curvature of the cell gives an indication of its bending, To obtain this value it is necessary to install one or more sensors at different distances from the neutral axis. Considering the Bernoulli’s law on an element dx of the beam, for the simple flexion, the equation of curvature is ex- pressed as: a)Monitoring of Conerete Bridges with Long-Gage Fiber Optic Sensors 283 where r is the curvature radius, x the curvilinear abscissa, © the strain in the x direction and y the distance from the neutral axis. A displacement gauge, installed parallel to the neutral axes, measures the deformation over a gage length L. The i y im yk where ra is the mean curvature radius and AL is the defor- ‘mation measured by the sensor after loading. Equation ( 2) shows that @ displacement gauge, placed parallel to the neutral axes measures the mean curvature H — of the cell. In the case of composed flexion and tempera- ture variations, itis easy to show that a pair of displacements, gauges, placed at different distances parallel to the neutral axes are necessary to measure the mean curvature of a cell In this case the curvature will be given by: Aly — ALy 1 — yak @) where AL, and AL; are the deformations measured by the two sensors, L their gage length andy,» their vertical po tion inside the cell, all relative to a common arbitrary origin (for example the bottom of the beam), It is interesting to notice that the thermal effects due to uniform temperature variations disappear by calculating the difference between the upper and lower sensors. This calculation can be generalized for two-dimensional cases using matrix calculus. Neutral Axis Position ‘The position of the neutral axis in a cell can be an indie: tion of the local cracking state (Abdunur, 1994) of the beam. of of its pre-stressing. It can be calculated by: _ vila ~ yoAL 4 68 Aly = AL i where e is the neutral axis position. This formula supposes however that the structure under- goes only mechanical actions and is not subject to simult neous temperature variatio Strain Diagram If more than two sensors are placed at different vertical positions inside the cell, it is possible to draw a strain diagram. This diagram shows the measured mean strain (ob- tained by dividing the measured deformation by the sensor length) as a function of the height in the beam. It is espe cially useful for composite or mixed structures where the co- hesion between different materials has to be evaluated. This diagram also gives the ratio between the elastic modules of the materials. If the different materials are supposed to work in traction or compression, the strain diagram will show if these suppositions are verified in the reality. The position of the neutral axis can also by determined looking at this diagram, ion Analysis After analyzing each cell separately we obtain an estima tion of the curvature and the position of the neutral axis as a function of the curvilinear abscissa along the beam length, From the local curvatures itis possible to obtain the section’s curvature function by fitting the cell's curvature values to a polynomial of the appropriate degree. For a simple beam section the curvature function is a second degree polynomial of the form Py(x) = ax? + by + 6) where the lower index refers to the degree of the polynomial jince the polynomial P2(x) has three unknowns, only three independent measurements (i¢., cells) are necessary to retrieve it for a single beam section, The values a, b, care then expressed by a trivial linear equation system: fi “@2tbetode || (6) i= 1.23 where — is the mean curvature on [x',:x/,] (the beginning. and the end of the ith cell), By solving this system for a, b,c we obtain the continuous curvature of the section. If the number of cells exceeds the polynomial degree chosen for the curvature function, the system should be solved by the least squares method. The curvature function can give interesting information ‘on the section behavior. For example, in the case of local cracking, the curvatures will concentrate at the position of the cracks. The curvature function also indicates if a ‘moment is present at the section ends and gives the relative strength between the forces acting on the structure. Structure Analysis By combining the curvature function of each can now calculate the curvature of the whole structure and obtain its deformation by double integration ction weEquations (5) are the curvature funetions of adjacent sec- tions. We retrieve the displacement functions by integrating them, Furthermore, the continuity of the displacement fune tion and its first derivative has to be guaranteed at the borders. The displacement functions are expressed by: Privy = SJPhorrde? + ax + B (e) where a! and fare the constants of the double integration that satisfy the system: Path) = PLO” Dlictin— ay 8) Py AP de OO ge lita) 0 Plex!) = OP4(x3) = 0 i) where n is the number of sections. Equation (8) guarantees the continuity of the deformation, Equation (9) the continu- ity of the slopes while Equation (10) ensures the absence of displacement at the extremities of the structures. This system has 2 unknowns for each of the 1 sections. The system has ((2—[) + (r=1)) + 2) = 2m equations: there~ fore, the solution is unique. We have determined that only three displacements sensors, set at different places in each beam section, are suf- ficient to determine the exact displacement of the whole beam, The whole beam is subdivided into sections of con- stant inertia and elastic response, uniformly loaded and with additional loads (force, moment, support. . .) only at the extremities. The general hypotheses are the satisfaction of the Bernoulli conservation law (stating that plain sections remain plain after loading) and the knowledge of the border condition at the extremities of the whole structure. The first hypothesis is satisfied in most cases, while the second is generally irrelevant, Engineers are interested in the struc- ture’s internal efforts. These efforts are only generated by ative internal displacements of the structure. This means that the model is able to calculate the deformation of the beam but not its rigid-body displacements in space. This jus lifies the arbitrary choice of Equation (10). To obtain infor- mation about the rigid-body displacements, internal sensors, are obviously useless and other measurements relative to fixed external points or absolute sensors (like GPS, incli- nometer, hydrostatic leveling system. . .) should be carried out, The minimal number of sensors to be placed in the structure depends on the number of parameters needed to re trieve its curvature, One sensor is sufficient for the simple flexion, a pair of sensors are necessary for the composed flexion and one additional sensor per section of beam is suf ficient for a triangular load, An increased number of sensors, ustally increases the measurement precision while adding a certain redundancy useful in the case of sensor failures. DYNAMIC DATA ANALYSIS ALGORITHM The current SOFO system is capable of measuring defor- mations with a precision of 2 microns over measurement bases between 20 em and 10m. The reading unit delivers a deformation reading each 10 s. If the measurement is per- formed on a structure undergoing dynamic deformations, the reading unit will record a snapshot of its instantaneous deformations each 10 s. If the deformations of the structure are statistically repeatable, as in the case of periodic loading or traffic loading, it is possible to gain interesting informa- tion about the structural dynamics by statistically analyzing a large number of these snapshots (Inaudi et al. 1998b). If the structure undergoes a time-dependent deformation AL(t) it is possible to associate a probability function P(AL) de- g the relative probability of finding the structure in a state of deformation AL at any given time. This can be done by counting the deformation measurements that fall into a given deformation range and divide by the total number of measurements. The obtained function, known as Rain-llow diagram for metallic structures, can be used to quantify the dynamic amplification factor of the monitored structure and to evaluate the residual fatigue life expectation. seribi ERSOIX BRIDGE The North and South Versoix bridges are two parallel twin bridges (see Figure 4). Each one supported two lanes of the Swiss national highway A9 between Geneva and Lausanne. The bridges are classical ones consisting in two parallel pre-stressed conerete beams supporting a 30 cm conerete deck and two overhangs. In order to support a third traffic lane and a new emergency lane, the exterior beams were widened and the over- extended. The construction progressed in two phase the interior and the exterior overhang extension. The first one began by the demolition of the existing internal overhang, followed by the reconstruction of a larger one. The second phase consisted to demolish the old external overhang, to widen the exterior web and to rebuild a larger overhang supported by metallic beams, Both phases were built by 14 m stages. Because of the added weight and pre-stressing, as well as the differential shrinkage between new and old concrete, the bridge bends (both horizontally and vertically) and twists during the construction phases. In order to increase the knowledge on the bridge behavior and performance and to optimize the concrete mix, the engineer choose low- coherence SOFO the fresh concrete during the setting phase and to monitor the long term deformations of the bridge. The bridge was in- strumented with more than hundred of these sensors. ‘ors to measure the displacements of Fiber Optic Instrumentation To obtain a good representation of the bridge deforma- tions, it was necessary to find a mean to measure theFigure 4. The Versoix Bridge during the rehabilitation works. horizontal, vertical and torsion displacements of the bridge during the different construction phases and in the long term. ‘To measure them, we decided to use an algorithm retrieving the displacements from curvature measurements, In order to guarantee a sufficient redundancy and to follow the concrete pouring stages of 14 m, | section of sensors was placed each 7 1m, Because of budget limitations, only 2 of the 6 spans were instrumented. A preliminary study by finite element simula- tion showed that the deformations of the two spans could be approximated by two Sth degree polynomial functions. At least 4 curvatures measurements for each span are therefore necessary to obtain the bridge spatial displacements. In each section, five sensors have been embedded in the new concrete and one sensor has been installed on the surface of the old concrete (see Figure 5). To obtain a good representation of the ‘mean curvature, sensors with four-meter active length were chosen. Two additional sensors were installed to give infor- mation about the differential shrinkage between the old and the new concrete, These sensors are rigidly connected to the existing conerete and their measurements can be compared with parallel sensors placed in the newly added concrete. This, installation scheme is repeated 12 times, 5 times in the first span and 8 in the second, The horizontal and vertical curvature is calculated sepa- fh of the 12 sections usin; spatial displacements are then calculated by integrating the mean curvature of the 12 sections. Sensor Network Installation To facilitate the implantation of automatic and remote monitoring, the whole sensor network had to be measured from one single and easily accessible location: the abutment. The main sensor network is composed by 96 fibers optics de- formations sensors with a 4 m active length and 2-10 m passive length, 14 optical cables with 10-100 m length, 14 local connections boxes and one central box. The central box also allows the installation of a reading unit, optical switches, portable PC and modem to measure the bridge remotely. The sensor installation followed the bridge widening schedule. The installation was very rapid; 2 hours were suf- Naw Concrete Figure 5. The Versoix Bridge cross-section and placement of the286 DANIELE INAUDI AND SAMUEL VuRPILLOT Figure 6. Sensor installation in concrete, ficient to place 4 sensors in each concrete pouring stage (for the interior overhang widening). Sensors were placed in the framework just after its completion and the building yard schedule was not delayed, Sensors were only held with plastic rings (and not fixed) to the re-bars (see Figure 6). Connection boxes were placed at the same time as the sensors in order to protect the optical connector, to check each one during the installation and to measure the bridge during construction, Finally, before the catwalks allowing the access to the local boxes were removed, all boxes were linked to th exterior sensors were attached to the surface of the old interior web using L-shaped metal- lic adapters (see Figure 7). During concreting, the workers were not aware that sensors were present in the framework and worked like every day, pouring the concrete directly on the sensors and vibrating as usual. central one. TI Figure 7. Surface sensor installation Al but nine of the sensors are working properly two years afier the end of the installation. Two sensors were lost during installation (one of them cut with an acetylene torch!); three during concreting and other two were damaged in the passive region. The death of the other two still waits for an explanation. Thanks to the redundancy built in the curvature analysis algorithms, these losses do not affect the perfor- mance of the sensor network as a whole. VERSOIX BRIDGE MEASUREMENTS Measurements on Single Sensors Figure 8 shows the typical concrete deformations mea- sured in one sensor during the first 8 months. All the optical fiber sensors of a same concrete pouring stage indicate about +9, 8.8, Figure 8. Typical deformation measurement over 8 months.Monitoring of Concrete Bridges with Long-Gage Fiber Optic Sensors 287 Date a Pe tigi ad Zod Ghowv Hive Zino ote Ean . Soo a Bows Figure 9. Comparison between cracked and un-cracked sections. the same behavior. On the graph, four phases are distin- guishable: the first is the drying shrinkage (phase 1), fol- lowed by a stabilization phase (phase 2) and finally there is a zone of variation (phase 3) corresponding to the thermal clongation of the bridge. Phase 4 is due to the decrease of the bridge temperature during the month of November 1996, These variations are consistent with a temperature variation of about 10°C that was actually observed. Figure 9 show measurement performed on two successive concrete pouring stages. The sensors in the same stage give similar results, but the two groups show a different behavior. Stage A has a shrinkage of about 0.02% while the other one has an apparent shrinkage of only about 0.005%. This differ ence is explained by the use of a different concrete mix between the 2 stages and the different climatic conditions. Stage B showed cracking while stage A did not. In the case of a cracked region, the 4m long sensor will typically bridge several cracks and the overall shrinkage of concrete is almost entirely compensated by the crack openings. This ex- plains the lower values and allows an easy and early detec- tion of the crack onset. ferential Shrinkage Measurements Figure 10 shows an example of the measurement of shrinkage for two parallel sensors, Al4 (new conerete) and A13 (old concrete). The two curves overlap almost perfectly, indicating an overall differential shrinkage close to zero, an index of a good cohesion between the two concretes ote ome ; ; ; : olor ~ ster moadiy’ adr go 2 3 an = i ore 7 Figure 10. Differential shrinkage between the old and new concrete. Distance from the abutment Abutment Pile 1 Pile? Figure 11. Horizontal displacement of the bridge. Global Displacement Measurements ‘The horizontal and vertical displacements of the first two spans of the bridge were calculated using the double- integration algorithm previously described. Figure 11 shows the horizontal displacement of the two spans of the bridge as calculated by algorithm, for different times and relatively to the line Abutment-Pile 2. The ob- served “banana” effect is due to the shrinkage of the concrete of the new exterior overhang, This eflect stabilizes to a value of5 mm of horizontal lateral displacement afier one month. During a load test, performed in Mai 1998 after the end of construction works, the vertical displacement of the bridge was also monitored using the fiber optic sensors, Figure 12 shows the measurement with SOFO sensors (Vertical Displacement Calculated) compared to those ob- tained with dial gages (invar wires under the bridge). This load pattern (Case A) consists in 6 trucks placed on the second span of the bridge (position 73 in the graph). The error of the algorithm is estimated from the deviation from a flat surface of the section deformations (the algorithm is based on the assumption that plan sections remain plain under load). The algorithm (Vertical Displacement Caleu- lated) retrieves within in the error interval the position of the sone me pee Le 5 il eae | “| Eugeeee Position bridge Figure 12. Vertical displacement, load case A288 DaNiete [NaubI AND SAMUEL VURPILLOT first pile (not entered gration) and matches the vert a boundary condition for the inte- displacement measured with the dial gages, Figure 13 shows the measurements during the load test for the case B. The case B consists in 6 trucks on the thitd span of the bridge (position right, out of the graph). In this case the error is very small because the torsion is not significant, the deformations are small and the condition “the plain sections remain plain” for the algorithm is better respected. Only the position 19 of the mechanical gages doesn't correspond well: this because this particular me- chanical gage is placed on the overhang of the bridge where torsion is more important instead than on the beams, The overall precision in the horizontal and vertical displace- ments of the two spans can be evaluated to 0.2 mm, a remark- able figure for a total integration length of over 100m, This shows that it is indeed possible to retrieve the global deforma- tions of a bridge using internal horizontal measurements. Ifthe deformations have to be known in an absolute referential and since the internal sensor can non give information on the rigid- body motion of the structure, it is necessary to complement the sensor network with at least two absolute measurements systems, for example inclinometers or GPS systems, ‘THE LUTRIVE BRIDGE The Lutrive (Switzerland) North and South bridges are ‘wo parallel twin bridges (see Figure 14). Each one supports two lanes of the Swiss national highway A9 between Lau- sanne and Vevey. Built in 1972 by the corbelling method with central articulations, the two bridges are gently curved (r = 1000 m) and each bridge is approximately 395 m long. on four spans. The two bridges have the same cross-section. It consists of a box girder of variable height (from 2.5m to 8.5m) and two slightly asymmetric cantilevers meant to reduce the effect of torsion in the curved bridges. Sensor Network The fourth span of the South bridge, fitted with an hydro- static leveling system measuring vertical displacements Vertal Displacement im) Psion in he bridge Figure 13. Vertical displacement, load case 8. Figure 14, The Lutrive Bridge, since 1988, was instrumented with 30 6m long SOKO sensors (see Figure 15), To measure the curvature variations, the sensors are installed in pairs at the interior of the box girder. Curvatures are measured with sensors placed near to the top and the bottom of bridge web and the vertical dis- placements can be retrieved by double integration of the cur- vatures, The sensors were used for quasi-static testing under thermal loading and under a static load-charging test, as well as for statistical analysis of the dynamical deformations. LUTRIVE BRIDGE STATIC MEASUREMENTS, Stat Measurements under Thermal Loading The static measurements under thermal loading extended during 24 hours between the 6th of July 1996 at 14h00 to the 7th of July 1996 at 14h00. The temperature difference mea- sured between the upper and the lower slab oscillated between 2.4°C and 5°C, Measurements were performed 2 hours during 20 minutes without stopping the traffic, sometimes necessitating a few trials before the bridge re- gained its calm after the passing of heavy vehicles, The curvature evolution during 24 hours presents a eyele correlated to the temperature differe jations in the bridge. The temperature of the bridge is slightly higher after 24 hours, This non-periodicity is retrieved on the curvature measurements, Taking into account the measurements ob- tained by an inclinometer placed at 10 m from the articula- tion and a null displacement on the pile as the border conditions for the double integration, the shape deformation V(x) is calculated for each temporal measurement. This function V(x,t) presents the vertical displacement of a point with co-ordinate x for each discrete measure at a time t Figure 16 shows the temporal comparison between the verti- cal displacements calculated with the fiber optic sensors and the hydrostatic leveling system for a point placed at quarter span. The hydrostatic leveling precision can be estimated to about -5 mm, The precision on the vertical displacement determined by the mathematical model can be obtained by a eachMonitoring of Concrete Bridges with Long-Gaxe Fiber Optic Sensors 289 LONGITUDINAL SECTION LONGITUDINAL SECTION OF THE ANALYSED BEAM er pte ess SFO ‘TRANSVERSAL SECTION OF THE ANALYSED BEAM + fer opie seers SOO ww le Lae Figure 18. Lutrive Bridge, sensor coniiguration. sensibility study of the model. Considering a precision of 10 um for the reading unit, 2 em for the vertical placement of the sensor, 2 cm for the sensors length and 10-5 radians for the inclinometer precision, a standard deviation of 20 wm is obtained. This example shows the possibility of obtaining, precise information about the vertical displacement of a single bridge span using only a reduced number of displace- ment sensors. A precision much higher than the one ob- tained by external measuring systems like optical or hydrostatic leveling can be achieved. Static Loading Test The 20th of November 1997, the Reinforced and Pre- stressed Concrete Institute (IBAP) of the Swiss Federal In- stitute of Technology of Lausanne (EPFL) performed a static as Fv opte Sons Vertical displacement mm] ® ° 2 tae teak 1912 28 00) 224 Aue ED B38 HOD 1628 Time Figure 16. Verical displacement at quarter span under thermal loading. load test on the Lutrive highway bridge. The loads were two trucks of 25 tons. They were placed in four different load cases: A & B for bending and C & D for torsion with the possibility of superposition for testing the linear behavior of bridge: situation A + situation B = situation C + situa- tion D. Every load case was performed three times with frequent base measures, measures without load that allowed for de- tecting the presence of permanent deformations. IBAP measured the deflections using optical leveling process, as well as inclinometer measurements on five cross sections equally spaced on the half span, The deformation measurement values serve as reference points for the mea- surement with the deformation fiber optic sensors. ‘The rotations measured by the inclinometers were used as boundary conditions in the curvature measurement algo- rithm, The results are good, with maximum 6.5% error (see Figure 17), We can make the same calculation with 3 sensors ‘Vettea detormaton fn Figure 17. Calculated and measured vertical diplacement using in- clinometer boundary conditions.290 DANIELE INAUDI AND SAMUEL VURPILLOT Figure 18. Calculated and measured vertical diplacement without boundary conditions, by cell and one inclinometer: in this case, the precision is slightly decreased A calculation of vertical displacement was tried without Figure 18) the necessary boundary condition is a zero incline between the double pillar. The ad~ vantage of this second type of calculation is the possibility to make an estimation of the vertical deformations on long, term without the presence of an inclinometer, which consti- tutes a complication of the measuring procedure. inclinometer values (s« Traffic Loading Test In order to test the statistical analysis algorithm, two sensors placed at quarter-span where measured during 24 hours under traffic loading. Both sensors were measured with a single scan of the SOFO system by coherence multi- plexing them with an external coupler. The data from the ‘ovo sensors are therefore correlated since the two readings are obtained in less than 0.1 s. Deformation {mm} Time [h:m:s] Figure 19 shows the results obtained by the two senso: The drift in the measurements on both sensors is due to the bridge's temperature variations that were however very small in this cold winter day. The total drift is of only about 80 microns over a sensor length of 6 m. This corresponds to a temperature variation of the order of 1°C. By subtracting the two deformation values and dividing by their length and dis- tance it is possible to obtain the curvature variations of the bridge Figure 20 shows the curvature readings after removal of the bridge's temperature drift with a polynomial fit on the initial data. As expected most data points lic around a zero curvature corresponding to the mean static state of the curvature values indicate a deformation induced by the passage of a truck. It can be noticed that these points are concentrated during the day when truck cir- culation is allowed. Except for one reading around 21:00 no trucks were registered during the night. Most curvatures are positive and indicate a downward bending of the bridge Under the quasi-static loading of the truck. A few points lie outside the noise floor curve for negative curvatures. These are either rebounds of the bridge after a truck leaves the in- strumented span or deformations induced by trucks on neighboring spans. Figure 21 show the statistical analysis on the whole set of data (Day and Night, 3269 data points), the night data only (Night, 996 points) and of a Monte-Carlo simulation based on simulated deformation measurements with a standard de- viation of one micron (noise Floor, 3267 points). The good agreement between these two latter curves indicates that the noise floor of the SOFO system limited the measurements during the night. During the day the higher deformations due to the passage of the trucks are clearly visible. The vertical Figure 19. Deformation data obtained on two parallel sensors placed at quarter-span in the Lutrve bridge.Monitoring of Concrete Bridges Curvature [1/m] 291 with Lony-Gage Fiber Optic Sensors Time [h:m:s] Figure 20. Curvature data atter removing the bridge's temperature drt. Jine indicates the curvature that was obtained during the static loading test with a truck of 28 t placed at center span. Since the load limit on the Swiss highway network is 28 t, Figure 21 indicates that the dynamic effects on the bridge are very limited. A rebound of a passing truck could for example induce higher curvatures than this same truck statically placed on the bridge. The isolated event at 2.7 10~° 1/meither indicates an overweight truck (some 40 t trucks are also circu- lating) or two trucks on the bridge at the same time. This same figure can be used to quantify the probability of having a truck on the bridge at any give time. This is given by the ratio of the number of points of the Day and Night curve inside, respectively outside the Noise Floor curve. This ratio gives 90% and indicates the probability of obtain- ing a reliable static deformation value when the measure- ments are performed under traffic conditions. Therefore, it can be considered sufficient to perform three measurements, on each sensor and discard any statistically aberrant value to men ar aur iy an seo Figure 21. Relative curvature frequency of the data in Figure 20, obtain a reliable value. A theoretical analysis considering the truck traffic density, the mean truck speed, the mean transit time gives a slightly lower value of 80%. CONCLUSIONS The measurements on the Versoix Bridge allowed gather- ing precious information about the behavior of the bridge. In particular, it was possible to observe the effect of different concrete types on the hindered shrinkage and to anticipate and observe the apparition of cracks in some of the sections. ‘The curvature measurements showed that it is indeed possi- ble to retrieve the spatial displacement in both the horizontal and vertical planes. The horizontal measurements clearly show a bending induced by the differential shrinkage pro- duced by the asymmetric distribution of the added concrete. The vertical displacements were more difficult to interpreter during construction, because of the continuous changes in the live loads (crane, trucks, temporary structural elements) and support conditions (temporary scaffolding and shoring). On the other hand, the vertical displacement during the load test were in excellent agreement this those obtained with dial gages. The treatment of the data resulting from the 100+ sensors proved to be very time consuming and dedicated software ‘was created in order to represent and analyze the data. The efficient handling of such huge and complex data flows cer- tainly represents a challenge that must not be underesti- mated. At IMAC-EPFL, two Ph.D. theses are currently in preparation on this topic. This static load test on the Lutrive shows the possibility to reconstruct the displacements of a bridge from curvature292 Dantete INaUDE AND SAMUEL VURPILLOT measurement with error maximum of 6.5% for a compli- cated structure (variable inertia, curvatures beam, Moreover, the curvature measurement allows not only for the determination of the displacements, but also for the de- termination of the source and importance of a detected problem. Until now, the measurement of vertical displace- ment has been the most important factor used to determine the health of a bridge. With new monitoring systems mea- suring the internal displacement (like fiber optic sensor) it will be possible to take the curvature measurement into con- sideration too. The presented experiments also show the possibility of using a relatively slow data acquisition system to statistically analyze dynamic deformations of structures. The only re- quirements are that the deformation measuring system is able to take instantaneous snapshots of a moving structure and that a sufficient number of data points are available for analysis. The results also show that it is possible to obtain re- liable static data on a bridge under traffie conditions. In general, these applications show that the presented system and the related data analysis algorithms are well adapted to the monitoring of bridges. They can be installed in new bridges or on refurbished ones and provide reliable short- and long-term data on the bridge behavior. ACKNOWLEDGMENTS, ‘The authors are indebted to Prof. L. Pflug, N. Casanova, P. Kronenberg, S. Lloret, B. Glisic, Prof. I. Smith, Prof. J. Conte, Dr M. Pedretti, R. Passera, P Colombo, R. Delez, F. Klein, S. Marazzi, O. Burdet, Prof. R. Favre and the whole IMAC, DIAMOND, IBAP and IMM teams for their help and useful discussion, The application on the Versoix bridge was realized thanks to the financial support of the state of Geneva (Switzerland), the one one the Lutrive Bridge thanks to the state of Vaud (Switzerland), The SOFO re- search program is conducted under the financial aid of the Swiss CTI (Commission pour la Technologie et IInnova~ tion) and of the Board of the Swiss Federal Institutes of Technology. REFERENCES, Abdunur C. 1994 “Monitoring of bridges subject 1 transversal eracks.” Second European Cosference on Smart Structures and. Matra Glasgow, October 1994, SPIE Volume 2361, 156-159 Burdet 0.1998 “Automatic Defieton and Temperature Monitoring of a Balanced Cantilever Concrete Bridge” 5 Intemational Conferesce on Short and Medium Span Bridges (SMSB V). Hassan M., Burdt 0. Fare R. 1995 “Analysis and Evaluation of Bridge Behavior under Static Load Testing Leading o Better Design and Judg- ment Cote,” Fourth International Bridge Conference, TRB, San Fran- Inaudi D. etal, 1997 “Embedded and surface mounted sensors for civil stretural monitoring.” Smart Structures andl Materials, San Diego, SPIE Vol. 3084-23, InaudiD. etal. 19984 “SOFO: Monitoring of Conerete Structures with Fier Opie Sensors,” 5 International Workshop on Material Properties and Design, Weimar, Acificato Publishers, pp. 495-514 Inui Det al, 19986 "Statistical Analysis of Under-sampled Dynamic Dis placement Measurement" Symposium on Start Structures and Mat as 98, San Diego Kronenberg P eta 1997 “Dam monitoring with fiberoptic sensors.” Start ‘Structures and materials, San Diego Perregaus N-ctal. 1998 "Vertical Displacement of Bridges using the SOFO System: a Fiber Optic Monitoring Method for Stuctures"12" ASCE Engineering Mechanics. San Dic, La Jolla Poh S., Baz A. 19% “Distributed sensor for rectangular plates” Smart Structures and materials, San Dieg, SPIE Volume 2718 p. 355. Vohra ST ta, 1998 “Quasi-Stati Stain Monitoring During the “Push Phase ofa Bor-Girder Bridge Using Fiber Brage Grating Sensor,” Eve ropean Workshop on Optica Fibre Sensors, Peebls Hy, Seodand urpilltS., nau D. Seana A. 1996 “Mathematical model for the deter mination ofthe vertical displacement fom intemal horizontal masuze- ment of a bridge” smart sutures and materials, San Diego, SPIE Volume 2719-08, Vurpilot S, Inaudi D. 1998 “Object-Oriented Concept Clasifaton of Lage Measurement Daa Sets in Civil Structures” 12” ASCE Enginese- ing Mechanics. San Diego, La Jolla.