A novel fiber optics based method to measure very low strains in large scale infrastructures

A novel fiber optics based method to measure very low strains in large scale infrastructures
of 6
All materials on our website are shared by users. If you have any questions about copyright issues, please report us to resolve them. We are always happy to assist you.
Related Documents
  A novel fiber optics based method to measure very low strains in largescale infrastructures Bradley Regez a, * , Mohammad Sayeh a , Ajay Mahajan a , Fernando Figueroa b a College of Engineering, Southern Illinois University at Carbondale, Carbondale, IL 62901, United States b NASA John C. Stennis Space Center, Technology Development and Transfer, Code HA30, Building 8306, Stennis Space Center, MS 39529-6000, United States a r t i c l e i n f o  Article history: Received 18 September 2007Received in revised form 22 May 2008Accepted 25 May 2008Available online 12 June 2008 Keywords: Fiber opticsMicrostrainsSpeckle patternSmall deformationLarge infrastructure a b s t r a c t An optic fiber strain measurement sensing mechanism has been developed that can mea-sure extremely lowstrains, i.e. in the order of 0–25microstrains, with a resolution close to0.1microstrain.ItisanalternatetotraditionalfoilgaugesaswellasthemorerecentFabry–Perot sensors for some applications. It is particularly suited to measure very low strains inlarge scale infrastructures where portions of the infrastructure can be used to mountextended lengths of the fiber optic, thereby providing high resolution. Examples of suchapplications are the measurement of small strains or vibrations in large structures suchas rocket test stands, buildings, oil rigs, bridges, dams, etc. The sensing system includesall off-the-shelf components and the focus of this paper is to provide details on how todevelop the system, calibrate it and then use it for specific applications.   2008 Elsevier Ltd. All rights reserved. 1. Introduction Measurement of extremely small strains, i.e. a fewmicrostrains (say 0–25microstrains), still remains a chal-lenging task. Traditional foil strain gauges (FSGs) havethe resolution requirements and are often used in manysuchapplicationsbutunderharshconditionssuchasfoundduring test conditions on rocket test stands their signal tonoise ratio for very low strains is not very good. Fiber op-tics has been shown to have good potential to measurelowvaluesof strains. Thestateof theart technologyseemsto be Fabry–Perot Fiber Optic Strain Gauges (FOSG) [1]with about 0.8microstrain resolution. The authors testedthe traditional FSGs and their version of the FOSG, andfound that the FOSG performance was superior to theFSG in terms of linearity, immunity to electromagneticinterference and time based drift. The authors’ applicationwas the measurement of strain on the legs of the rockettest stands so as to estimate the mass of the liquid oxygenin the tanks. Their estimate of the maximum strain on theeach of the four legs of the stand was approximately25microstrains. This is also the basis of this paper, as thesensingmechanismpresentedinthispaperwasalsodevel-oped for the same application.The strategy used in this paper is quite different to theFabry–Perot principle. The strategy used in this paper is toquantify the changes in the speckle pattern created byusing a multimode optic fiber. There are some distinctadvantages of the multimode system. When an incoherentsource of light is used, the output is a smooth distributionof intensity within the pattern. However, when the light iscoherent, the output consists of a number of ‘‘speckles” of varying intensities. When this multimode optical fiber isperturbed or ‘‘strained” the projected output speckle pat-tern changes. The change in the output consists of somespeckles becoming brighter, some dimmer and some notchanging at all, with the net intensity of the pattern, how-ever, remaining unchanged [2–5]. An analysis of thechanges in the speckle pattern of the multimode optical fi-ber can then be used to obtain information about the per-turbation [6,7]. This forms the basis for the measurement 0263-2241/$ - see front matter    2008 Elsevier Ltd. All rights reserved.doi:10.1016/j.measurement.2008.05.005 *  Corresponding author. Tel.: +1 6184537007; fax +1 6184537658. E-mail address: (B. Regez).Measurement 42 (2009) 183–188 Contents lists available at ScienceDirect Measurement journal homepage:  of extremely low values of strain using simple interpreta-tion schemes [8]. The main barrier to constructing multi-mode sensors, which analyzed the speckle pattern, is thedifficulty of explaining the speckle pattern and taking ac-count for a reasonable number of imperfections in the fi-ber. There has been some attempt to theoretically solvethe multi-mode optic fiber output [9,10], but it has stillnot resulted in a robust device that canbe usedin the field[11].Also,onehastobecarefulthatatverylowmechanicalstrains,thethermallyinducedstrainsplayasignificantrole[12] further complicating the theoretical model. Hence, anempirical approach is proposed in this paper to solve theproblem and provide a simple and robust sensing mecha-nism using some of the unique characteristics of multi-mode optic fibers.Ahmadet al. [13] used artificial neural networks to cor-relate the speckle patterns with respect to the straincaused in the multimode fiber optics. They achieved reso-lutionsof 0.8microstrains over aneffectivelengthof 11m.The sensors operated within a temperature range of 20–35  C. The method consisted of taking samples of intensityof the speckle pattern at various places and mapping thesesignals to a known output using a back propagation neuralnetwork. This network when trained gives the outputwhich corresponds to any small strain. This work formsthebasisofthestudyconductedforthisproject.Thispaperextends the capabilities of the multimode fiber opticsmethod [13] to attain 0.1microstrain resolution at verylow strains (0–50microstrains) under realistic conditions,i.e. the optic fiber is actually fixed to the specimen understrain. The earlier work [13] the fiber optic was simplywound around a piezoelectric element.One of the greatest advantages of the sensing schemedeveloped in this work, as compared to other fiber opticbased sensors, is that there is no need for a reference fiber,as is the case with the Mach-Zehnder interferometer andother such phase detector based sensors. Since the ‘‘refer-encefiber”of thesensingmechanismdevelopedinthispa-per isinthe samelocationasthe sensingfiber, thereislessdependenceuponthetemperature. Anotheruniqueadvan-tage over the Mach-Zehnder interferometer would be inapplications where the sensing fiber is very long, or it isdifficult to isolate the reference fiber from all extraneousinfluence. This is certainly the case for the applicationslisted in this paper. 2. Underlying theory  The main assumption in the theory is that a single mul-timode optical fiber is modeled to be many single mode fi-bersgroupedcloselytogether.Therearemanymethodsfordetermining the speckle pattern from such fibers. Thesedeterminations may differ from experimental results dueto imperfections such as micro-scratches at the end of the fiber, non-uniform refraction index, hysteresis, etc.[14]. Hence, more empirical methods for analyzing thespeckle pattern are required. The interference effects arecalculated by determining the electric field on a screen.One can use ray tracing to analyze the output of a multi-mode fiber [14,15]. Consider two plane waves launchedintoanopticalfiberandtheresultingintensityattheplaneof detection is given by I  ð  x ;  z  ¼ const Þ¼  12 g o ð E  1 þ E  2 Þ  ð E   1 þ E   2 Þ E  1  and  E  2  are the respective electric fields at the plane of detection due to each plane wave E  1  ¼ E   0 e   jk ð  z  cos h 1 þ  x sin h 2 Þ  ffiffiffi 2 p   ;  E  2  ¼ E   0 e   jk ð  z  cos h 2 þ  x sin h 1 Þ  ffiffiffi 2 p  where  h  is the angle between the mode and the optic axisat the departure point of the fiber. After simplification theintensity of the unstressed fiber becomes I  ¼  E  0 2  ffiffiffiffiffiffiffiffi 2 g 0 p   4cos 2 kx cos  h 1 þ h 2 2   sin  h 1  h 2 2     Under stress the new relation for intensity variationbecomes I  ¼  E  0 2  ffiffiffiffiffiffiffiffi 2 g 0 p    4cos 2 kx cos  h 1 þ h 2 2   sin  h 1  h 2 2   þ k ð d 1  d 2 Þ    where  d  represents the length of the ray, and the variationcomes from the  d 1 – d 2  term, that can also be referred to asthe D d  term. The general relation for the intensity relationfor multiple rays, say n, under stress [13] is given by I  ¼  E  0 2  ffiffiffiffiffiffiffiffi 2 g o p   X n  2 i ¼ 0 X n  1  j ¼ i þ 1 4cos 2 kx cos  h i þ 1 þ h  j þ 1 2   sin  h i þ 1  h  j þ 1 2  " þ k ð d i þ 1  d  j þ 1 Þ   n 2  2 n  Theintensityvariationequationpresentedaboveistruefor n  >2, and indicates that the shape of the output intensitygraphissinusoidal.Asaresultoftheelongationoftheopti-cal path, thephasesof theserays at thedetectionplanearealtered. This was shown in [13] along with the fact that asthe number of rays is increased, the output of the multi-mode fiber optic is overlapping sinusoids. The variationoftheoutputintensityisaresultofdifferencesinthebeamphasesduetothedifferenceinthetraveledpaths,i.e.smallchanges in ‘ d ’. It is this change that is detected and can becorrelated to the state of strain of the optic fiber. Eventhough this change in the speckle pattern looks random,it is not, and is always repeatable for specific values of the strain. The equation also shows how an increasedlength of the fiber optic will lead to higher resolution, i.e.a longer ‘ d ’ leads to a greater change in the variation of   d ,i.e. D d .Hence, the strategy to determine very lowstrainentailshaving a systemthat can match the changes in the specklepattern to pre-determined states of such low strains. Thiscan be done by using neural networks or look up tablesforcertaincharacteristicsextractedfromthesignal.Neuralnetworkswerechosenin[13] toshowfeasibility, andhavebeen chosen for this work to model the correlation be-tween the speckle pattern changes to quantitative valuesof strain. 184  B. Regez et al./Measurement 42 (2009) 183–188  3. Approach used and prototype developed The main objective of the project was the developmentofanewtechniqueusingfiberopticstobuildaninnovativesensor to measure extremely small strains. The aim of thisstudywastodevelopastraingaugethatcanmeasurecloseto 0.1microstrain resolutions in the range of 0–50micro-strains. This sensor has a direct application at NASAsStennis Space Center where there is a need for the non-intrusive measurement of the mass of liquid oxygen(LOX) in a tank by measuring the strain on the steel col-umns supporting the LOX tank. A schematic of the pro-posed system is shown in the Fig. 1.A multimode optic fiber (100 l m diameter) is woundelliptically and glued to a flat piece of steel. A compressiveload  P   is then applied to the flat piece of steel. A lasersourceisaimedatoneendofthefiberopticwhiletheotherend is aimed at an array of photocell sensors. The outputvoltage from the photocell sensors is then acquired andconverted using an A/D system. The digital data is thenanalyzed using a neural network model where a strain va-lue can be obtained from the output of the model.A prototype was developed and mounted on a steelplate whose dimensions are 14.25  10.10  0.64cm. Thedimension of the plate was chosen to be the optimal sizewhich was able to fit in the 15.24cm drill press vice thatwas used to apply a small compressive load to the plate.The pitch of the vice was 0.65cm, hence 1 revolution leadto a strain of 126microstrains. The vice, with the help of amicrometer, was moved with a resolution of 1/100th of adegree, providing a strain resolution of 1.26microstrains.A template was usedto assist in laying out the fiber op-tic on the steel specimen and creating a symmetricalshape. An elliptical shape was used to optimize the length,and yet not resort to sharp bends or criss-crossing of thefiber on itself. The laser used for the prototype was a stan-dard 2.5mW (650nm) laser diode module bought off-the-shelf as shownin Fig. 2 with the finished fiber optic sensormounted on the steel plate. For reference, a traditional foilstrain gauge is also fixed to the specimen and its outputmeasured through a data-acquisition system. This pro-vided discrete data points for reference purposes.A laser alignment fixture was developed that allowedthe laser light to enter the fiber optic on the same plane.The main component of the fixture was a steel tube whoselength was 8.26cm and internal diameter (ID) was honedslightly greater than 0.95cm and the outer diameter(OD) was 1.43cm. The laser module was then inserted inone end of the tube and held in place with a set screw.Thefiber optic holder was heldinplace withsixset screwsthat allowed position adjustment to optimize the amountof laser light entering the fiber optic. The photocells usedintheprototypesystemwerepurchasedoff-the-shelf.Eachcell had a dimension of 2  3mm. The cells were set nexttoeachotherina12  1arrayasseeninFig.3.Asinglecellwas capable of producing 0–5Volts which was directlycorrelated to the intensity of the light that the cell was ex-posed to. A digital scope (150MHz) was used to obtaintheoutput voltage for each cell. 4. Prototype optimization and operation Once the fiber is mounted and hooked up to the laser, afocusing technique is needed to optimize the intensity atthe projection tail. This technique is largely trial and error.The optimal placement of the fiber optic is easilyrecogniz-able by the significant increase in light intensity emittedfrom the projection tail as shown in Fig. 4. In addition tothis, the whole fiber gauge will glow an eerie red uponoptimization. At this point, securely lock down the fiberoptic holder using the set screws. P P  LASER    P   H   O   T   O   C   E   L   L   A   R   R   A   Y MULTIMODE FIBER OPTICSTRAIN VALUENEURAL NETWORKMODELA / DLASER LIGHTFIBER OPTIC ENDSOURCE    S   T   R   U   C   T   U   R   E   U   N   D   E   R   G   O   I   N   G    D   E   F   O   R   M   A   T   I   O   N Fig. 1.  Multimode fiber optic strain sensor concept. B. Regez et al./Measurement 42 (2009) 183–188  185  The sensor in Fig. 4 is in operational mode where atypical speckle pattern can be seen glowing on the target.It is this speckle pattern that changes due to a change instrain and is extremely sensitive even to small strains,and forms the basis of this new type of strain sensor.The photocells are placed on the target, then as thespeckle pattern changes the different photocell sensorsin the sensor array experience different light intensities.These are recorded during a calibration procedure andthen used to estimate the strain when an unknown strainis applied. The photocells should be positioned at a dis-tance from the projection tail at which the multi-modespeckles should be approximately the size of the photo-cell. The distance between the two is one factor thatcan determine the physical size of the projected speckles.The longitudinal distance between the fiber and laserdiode is also a factor that can be adjusted to change thespeckle size slightly. 5. Experimental results This section gives the experimental results to demon-strate the feasibility of the approach. The apparatus de-scribed above was cycled through 0–25microstrains of compressive strain using 10m of the fiber optic and fourphotocells (out of a possible twelve). Measurements inthe change of intensities in the four photocells were re-corded automatically at increments of 1.26microstrains.A total of 250 measurements were taken at each pointand their average taken as the final value.The light conditions were held constant, and the entiredata was collected within a one hour time frame. A straingauge was used to collect data at a few discrete points.The photocell data was collected using a digital oscillo-scope and the data downloaded to a computer. All systemparameters were held constant during testing as much aspossible. The load was applied to the steel plate usingthe drill press vice. The actual measurements by the fourphotocells are plotted in Fig. 5.It is important to note here that it is the unique outputsof the four photocells that are mapped to the actual strainestimate. Oneshouldconsidertheentiresensingsystemasthe sumtotal of the four photocells plus the mechanismtocovert the four readings into a single strain estimate. Theexperiment was repeated numerous times and the changein the output was less than 1%, hence only one set is pre-sented here as an illustration of the type of measurementsobtained from the photocells.Different mathematical models were developed to cap-ture the trends including linear and nonlinear models, butultimately were rejected (for the time being) due to thedevelopment of a perfect  neural network  model. A 10-10-5-5-2-1 neural network model was developed that per-fectly captures the trends. This software model (developedinMATLAB)takeslessthanfifteensecondstorunandgivesthe final results within three tries. That means the systemcanbe calibrated at any timein less than a minute. Herein,also lies one of the limitations of the sensing system: thesensor will always need to be calibrated with at least twoknown points after installation. This is certainly differentthan other similar sensors used for measuring strain. Theadvantageofusingthissensorliesinapplicationsrequiringhigh resolution, and large portions of the structure itself  Fig. 2.  Prototype sensor with the laser fixture and laser diode module. Fig. 3.  Photocell array during setup. Fig. 4.  Picture of an optimized system, notice the brightness of the fiberoptic and the light emitted from the projection tail of the fiber optic.186  B. Regez et al./Measurement 42 (2009) 183–188  areavailablesothat extendedlengthsoftheopticfibercanbe directly mounted on them.Fig.6showstheopticfibersensorreadingsasestimatedby the trained neural network (continuous plot) and thestrain gauge readings at discrete points (asterisks).The strain gauge readings are a little higher but wellwithin 1–2microstrains. The readings were taken at strainintervals of 1.26microstrains, and the neural network wasable to fit an excellent model to the data. The plot showsthat resolution of at least 1microstrain is achievable, andwith interpolation, one could achieve 0.1microstrains. 6. Temperature calibration Temperatureplaysakeyroleintheworkingof anysen-sor at such low strains. This section highlights work donein using the sensor to obtain strain readings on the speci-men due to temperature effects. These readings wouldthenbeusedasalookuptabletocompensateforthestraindue to temperature effects, thereby decoupling the tem-perature effects from the mechanical strain effects. Thiswould also be part of the calibration routine wherebyone would have to collect data for the mechanical strainto train the neural network as well as strain data due tothe temperature effect when the structure is not stressed.In this set the specimen with the mounted optic fiber isfixedwithinthevice,andplacedwithinatemperaturecon-trolledenvironment. Due tothe changeintemperaturethesteel specimen undergoes a strain. A K-type thermocoupleis used to measure the temperature of the steel specimen,and the voltage from the K-type thermocouple was con-verted into temperature using an Omega DP462 tempera-ture sensor. The temperature was varied using athermostat controlled heater.In this case due to the large strains, the traditionalstrain gauge is used as the reference sensor. The neuralnetwork model from the previous section is used to con-vert the photocell voltages to a strain measurement.Table 1 given below shows the strain from the optic fi-ber sensor as well as the traditional strain gauge for differ-ent temperature readings (25–60  C). -18 -16 -14 -12 -10 -8 -6 -4 -2 00.020.0250.030.0350.040.0450.050.0550.06 Photocell sensor readingsStrains (microstrains)    A  m  p   l   i   t  u   d  e   (  v  o   l   t  s   ) Fig. 5.  Output of 4 photocells. 051015-18-16-14-12-10-8-6-4-202 Optic Fiber Sensor readings and Strain Gage readings at discrete points Measurements      M     i    c    r    o    s     t    r    a     i    n    s Fig. 6.  Actual strain and sensor readings.  Table 1 Temperature callibration look-up table Temperature in Celsius Strain in microstrainsOptic fiber sensor Strain gauge25 5.0 630 32.5 3335 87.2 8840 140.9 14345 187.0 18850 226.2 22855 239.0 24060 240.9 243 B. Regez et al./Measurement 42 (2009) 183–188  187
Similar documents
View more...
Related Search
We Need Your Support
Thank you for visiting our website and your interest in our free products and services. We are nonprofit website to share and download documents. To the running of this website, we need your help to support us.

Thanks to everyone for your continued support.

No, Thanks