River Rapids Conference Proceedings of the Society for Experimental Mechanics Series Shock & Vibration, Aircraft/ Aerospace, Energy Harvesting, Acoustics & Optics, Volume 9 Anders Brandt Raj Singhal Proceedings of the 34th IMAC, A Conference and Exposition on Structural Dynamics 2016 River Publishers
Conference Proceedings of the Society for Experimental Mechanics Series Series Editor Kristin B. Zimmerman, Ph.D. Society for Experimental Mechanics Bethel, CT, USA
River Publishers Anders Brandt • Raj Singhal Editors Shock & Vibration, Aircraft/Aerospace, Energy Harvesting, Acoustics & Optics, Volume 9 Proceedings of the 34th IMAC, A Conference and Exposition on Structural Dynamics 2016
Published, sold and distributed by: River Publishers Broagervej 10 9260 Gistrup Denmark www.riverpublishers.com ISBN 978-87-7004-933-7 (eBook) Conference Proceedings of the Society for Experimental Mechanics An imprint of River Publishers © The Society for Experimental Mechanics, Inc. 2016 This work is subject to copyright. All rights are solely and exclusively licensed by the Publisher, whether the whole or part of the material is concerned, specifically the rights of translation, reprinting, reuse of illustrations, recitation, broadcasting, or reproduction in any other physical way, and transmission or information storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar methodology now known or hereafter developed. The use of general descriptive names, registered names, trademarks, service marks, etc. in this publication does not imply, even in the absence of a specific statement, that such names are exempt from the relevant protective laws and regulations and therefore free for general use. The publisher, the authors, and the editors are safe to assume that the advice and information in this book are believed to be true and accurate at the date of publication. Neither the publisher nor the authors or the editors give a warranty, expressed or implied, with respect to the material contained herein or for any errors or omissions that may have been made. The publisher remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Preface Shock & Vibration, Aircraft/Aerospace, Energy Harvesting, Acoustics & Optics represent one of ten volumes of technical papers presented at the 34th IMAC: Conference & Exposition on Structural Dynamics, organized by the Society for Experimental Mechanics and held in Orlando, Florida, January 25–28, 2016. The full proceedings also include volumes on nonlinear dynamics; dynamics of civil structures; model validation and uncertainty quantification; dynamics of coupled structures; sensors and instrumentation; special topics in structural dynamics; structural health monitoring, damage detection and mechatronics; rotating machinery, hybrid test methods, vibro-acoustics and laser vibrometry; and topics in modal analysis and testing. Each collection presents early findings from experimental and computational investigations on an important area within structural dynamics. The topics represent papers on practical issues improving energy harvesting measurements, shock calibration and shock environment synthesis, and applications for aircraft/aerospace structures. The organizers would like to thank the authors, presenters, session organizers, and session chairs for their participation in this track. Copenhagen, Denmark Anders Brandt Montreal, QC, Canada Raj Singhal v
Contents 1 Improved Cutting Force Measurements in Milling Using Inverse Filtering...................... 1 Martin Magnevall and Tomas Beno 2 Use of a Depth Camera as a Contactless Displacement Field Sensor........................... 13 Jean Michel Franco, Johannio Marulanda, and Peter Thomson 3 Uncertainty of Digital Image Correlation with Vibrating Deformable Targets ................... 21 Alfredo Cigada, Navid Hasheminejad, and Emanuele Zappa 4 Physical Vibration Simulation of an Acoustic Environment with Six Shakers on an Industrial Structure........................................................... 29 Randall L. Mayes and Daniel P. Rohe 5 Developing Conservative Mechanical Shock Specifications .................................. 43 Matthew Baker, Kelsey Neal, Katrina Sweetland, Garrison Stevens, Dustin Harvey, and Stuart Taylor 6 Force Limited Vibration Using the Apparent Mass Method................................. 53 Paul Marchand, Raj Singhal, and Mark O’Grady 7 Harmonic Force Excitation Analysis of a Small-Body Asteroid/Satellite System.................. 67 Joshua Johnson, William H. Semke, Shankar Nag Ramaseri Chandra, and Ronald Fevig 8 A Study on the Dynamic Interaction of Shock Response Fixtures and Test Payload............... 77 Jesus M. Reyes and Peter Avitabile 9 Modal Analyses and Experimental Verifications of Joined-Wing Configurations ................. 87 Berkan Alanbay, Melin S¸ahin, and Gu¨venc¸ Canbalog˘lu 10 Modal Testing of James Webb Space Telescope (JWST) Optical Telescope Element (OTE) . . . . . . . . . 103 Douglas J. Osterholt, David Cloutier, Timothy Marinone, and Reem Hejal 11 Quantification of Dynamic Differences Between Boundary Conditions for Environment Specification Improvement ............................................. 117 Julie M. Harvie and Randy Mayes 12 Modal Updating of Tail of a Military Helicopter.......................................... 133 Kurtulus¸ Ersoy, Mert Atasoy, and Cem Genc¸ 13 Modeling of High Frequency Shock Tests ............................................... 145 Washington J. DeLima, Melanie N. Ambrose, and Richard Jones 14 A Novel Method to Correlate a Rocket Launcher Finite Element Model Using Experimental Modal Test Measurements and Identification Algorithms ................... 153 Ronald N. Couch, Eliott J. Radcliffe, and Rickey A. Caldwell vii
15 Numerical Studies on the Reduced Order Modeling of Frictionless Joint Contact Interfaces ............................................................ 167 M. Breitfuss and H.J. Holl 16 Structural Design with Joints for Maximum Dissipation.................................... 179 M. Stender, A. Papangelo, M. Allen, M. Brake, C. Schwingshackl, and M. Tiedemann 17 A Hybrid Piezoelectric and Electrostatic Vibration Energy Harvester......................... 189 H. Madinei, H. Haddad Khodaparast, S. Adhikari, and M.I. Friswell 18 Design of Scaled-Down Composite I-Beams for Dynamic Characterization in Subcomponent Testing of a Wind Turbine Blade....................................... 197 Mohamad Eydani Asl, Christopher Niezrecki, James Sherwood, and Peter Avitabile viii Contents
Chapter 1 Improved Cutting Force Measurements in Milling Using Inverse Filtering Martin Magnevall and Tomas Beno Abstract Accurate estimates of cutting forces in metal cutting are important in the evaluation of e.g. different cutting tool geometries and concepts. However, dynamic influences from the measurement system affect the measurement result and may make the obtained cutting force data erroneous and misleading. This paper presents a method to construct an inverse filter which compensates for the dynamic influences from the measurement system. Using the suggested approach, unwanted dynamic effects from the measurement system can be counteracted. By applying the inverse filter it is possible to retain information related to the cutting forces at higher frequencies than possible with unfiltered data. The advantage of using the proposed method is illustrated by comparing simulated, inverse- and low-pass filtered cutting forces to unfiltered forces at different cutting speeds. The results indicate that inverse filtering can increase the usable frequency range of the force dynamometer and thereby provide more accurate and reliable results compared to both low-pass and unfiltered force measurements. Keywords Metal cutting • Cutting force • Dynamometer • Inverse filter • Deconvolution 1.1 Introduction Cutting forces are one of the most important quantity in the metal machining process. The cutting forces govern power and torque requirements in the machine tool and drive heat generation which catalyzes tool wear and determines the magnitude and direction of residual stresses in the machined component. Cutting forces also cause deflection of the cutting tool, machine tool and work piece and may have a negative influence on the machined components quality. The cutting forces are therefore important parameters in the evaluation of different cutting tool geometries and concepts. However, dynamic influences from the measurement system affect the result and make it difficult to obtain accurate cutting force data. A commonly used method to remove unwanted dynamic effects from the measured cutting forces is low-pass filtering. Low-pass filtering removes all information above a specified cut-off frequency and may therefore also remove important information related to the true cutting forces contained in frequencies above the cut-off frequency. This is especially evident in milling with transient cutting conditions when the rise times are short and the cutting forces thereby have high frequency content. Therefore, it is difficult to get reliable estimates of the amplitudes and rise times by low-pass filtering transient cutting forces, especially at high cutting speeds. Accurate estimates of both rise times and force amplitudes are important, e.g., when evaluating and comparing different tools and insert geometries. An alternative approach, to low-pass filtering, that improves the cutting force estimates, i.e. increases the effective frequency range of the force dynamometer, is therefore of great interest. For example, Tlusty et al. used accelerometers to compensate for the inertia and structural damping of the dynamometer, thereby increasing the effective frequency range [1]. This method has been proven to work under certain conditions, but encounters difficulties around resonance frequencies when the system inertia or damping is large. The method also has problems handling systems with more than one dominating mode [2]. Park and Altintas used a Kalman filtering technique to compensate for unwanted dynamics and process and measurement noise of a spindle integrated force sensor, [3, 4]. The same method has also been applied to cutting force measurements in micro end milling, [5]. Jensen et al., [2], developed a M. Magnevall (*) AB Sandvik Coromant, SE-811 81 Sandviken, Sweden e-mail: martin.magnevall@sandvik.com T. Beno University West, SE-461 86 Trollh€attan, Sweden e-mail: tomas.beno@hv.se #The Society for Experimental Mechanics, Inc. 2016 A. Brandt, R. Singhal (eds.), Shock & Vibration, Aircraft/Aerospace, Energy Harvesting, Acoustics & Optics, Volume 9, Conference Proceedings of the Society for Experimental Mechanics Series, DOI 10.1007/978-3-319-30087-0_1 1
method to obtain an invertible filter of a non-minimum phase frequency response function (FRF) of a force dynamometer. The method was tested on experimental data with promising results. The basic principle of the method is to divide the system’s transfer function into one stable and one unstable causal IIR-filter. The unstable causal IIR-filter is transformed into a non-causal stable FIR-filter. Then, by using these two filters in series a stable inverse filter is obtained. Depending on the location of the non-minimum-phase zeros of the system’s transfer function, the length of the non-causal FIR-filter can become large, introducing long time delays. However, if the method is applied on large data sequences, this will not cause a problem. In the case when the force dynamometer can be considered linear and the cross-frequency responses between the directions x, y and z are negligible, a minimum-phase inverse filter can be constructed and used to counteract the dynamometer dynamics and thereby increase the usable bandwidth of the dynamometer. A procedure for creating a minimum-phase inverse filter is described in this paper and applied in both simulations and on experimental data from milling under various cutting conditions. 1.2 Inverse Filtering Consider the FRF H(ω) between the applied force FR(ω) and the force output from the dynamometer FD(ω), Eq. (1.1). H ωð Þ¼ FD ωð Þ FR ωð Þ ð 1:1Þ In an ideal case, i.e. when there are no dynamic influences from the dynamometer, the magnitude of H(ω) is equal to unity and the phase equal to zero for all frequencies. However, due to e.g., the mass and shape of the work piece, the stiffness and damping of the force dynamometer and the distance between the actual cutting position and the positions of the force transducers in the dynamometer, the magnitude and phase of the frequency response between applied force and force output will deviate from the ideal values and thus the measured force will differ from the applied force. By applying an inverse filter, describingH 1 ωð Þ, on the recorded force signal these unwanted dynamic effects can be counteracted. If H(ω) is stable and minimum-phase (all zeros of the system lies within the unit circle in the z-domain) then the system is directly invertible. Usually, mechanical systems are stable and mixed-phase (zeros both inside and outside the unit circle). If the system has zeros outside the unit circle it cannot be directly inverted, since the result will then have unstable poles and the filter output exponentially tends toward infinity. However, a mixed or maximum-phase FRF can be transformed into a minimum-phase FRF while still keeping the amplitude characteristics but changing the phase. An invertible filter describing the characteristics of the minimum-phase FRF can then be estimated as described in this section. 1.2.1 Transformation into Minimum-Phase A mixed- or maximum-phase transfer function can be transformed into a minimum-phase transfer function by, e.g., using the Hilbert transform [6] or cepstrum [7, 8]. In this paper, real cepstrum is used to transform the FRF of the force dynamometer into minimum-phase. Let h(n) be a real sequence with H(ω) as its Fourier transform. Its real and complex cepstrumcˆ (n) and hˆ (n) are defined as: ^C ωð Þ¼ H ω ð ð ÞÞ¼log H ω j ð Þj ^c nð Þ¼F 1 ^C ωð Þ ^H ωð Þ¼log H ω ðð ð ÞÞÞ ^h nð Þ¼F 1 ^H ωð Þ ð1:2Þ where log H ωð Þ refers to the natural logarithm of H ωð Þ and F 1 denotes the inverse Fourier transform. Some useful relations between minimum-phase and maximum-phase sequences and their complex cepstrums are, [8]: • If h(n) is a minimum-phase sequence, hˆ(n) will be a casual sequence. • If h(n) is a maximum-phase sequence, hˆ(n) will be an anti-causal sequence. 2 M. Magnevall and T. Beno
Let the minimum-phase counterpart to h(n) be denoted by hmin(n) and its complex cepstrum denoted by hˆmin(n). The Kramers–Kronig relations for a causal sequence states that the entire sequence can be described by its even part. The relationship between the even part of hˆ(n) and its Fourier transform is given in [9]: ^h e nð Þ¼F 1 H ω ðð ð ÞÞÞ¼^c nð Þ ð1:3Þ Since hˆ(n) is a causal sequence when h(n) is minimum-phase, hˆmin(n) can be estimated as: ^h min nð Þ¼ 2^c nð Þ ^c nð Þ 0 n >0 n ¼0 n <0 8< : ð1:4Þ According to Equation (1.2), the minimum-phase transfer function Hmin(ω) can be obtained by: Hmin ωð Þ¼eF ^h min nð Þ ð Þ ð1:5Þ 1.2.2 Fitting an Invertible Digital Filter to the Minimum Phase FRF The Steiglitz-McBride iteration algorithm is used to find the filter coefficients describing the given impulse response of the estimated minimum phase FRF, Hmin(ω), [10]. This method is based on non-parametric frequency response characteristics and describes the identified FRF using a polynomial model. The identification is performed using in-house written Python code, alternatively the MATLAB function invfreqz.m can be used to estimate the coefficents [11]. The identified model is a discrete representation of the minimum-phase transfer function of the force dynamometer, hmin(n), and can be represented as: Hmin zð Þ¼ B zð Þ A zð Þ ¼ b 1ð Þþb 2ð Þz 1 þ. . . þb nbþ1 ð Þz nb a 1ð Þþa 2ð Þz 1 þ. . . þa naþ1 ð Þz na ð 1:6Þ The result is an invertible IIR-filter described by the coefficients banda,wherenbandnaare the total number of coefficients in the numerator and denominator, respectively. nb and na are selected by visual inspection ensuring a satisfactory fit between the measured and estimated FRFs. 1.3 Simulations To verify the proposed method and identify usable frequency ranges for the inverse filters, simulations using a model of the force dynamometer obtained from experimental data are performed. The model is based on the FRF matrix of the force dynamometer, with the work piece, mounted in the machine tool. The FRF matrix was estimated using impulse excitation in x- andy-directions, results are shown in Figs 1.1 and 1.2. The cross-frequency response in the region 0 to 3000 Hz is small in both directions. Also, the coherence functions are close to unity up to approximately 3000 Hz, implying linear relationships between inputs and outputs. These results indicate that a linear model, with negligible cross frequency response, of the force dynamometer is valid for frequencies up to approximately 3000 Hz. Based on the FRF measurements of the dynamometer, inverse filters in both x and y-directions are estimated and evaluated with respect to both amplitude and phase correction. Comparisons between the measured FRFs and the minimumphase FRFs are shown in Figs 1.3 and 1.4. To clearly show the amplitude and phase characteristics of the inverse filters, the combined FRFs are also displayed; these are calculated as: 1. Inverse Fourier transform the measured mixed-phase FRF, H(ω), the result is the impulse response. 2. Apply the inverse filter to the obtained impulse response; the result is a unit impulse. 3. The combined FRF is the Fourier transform of the signal obtained from the inverse filter. If the inverse filter behaves perfectly, the magnitude of the combined FRF should be unity and the phase zero for all frequencies. 1 Improved Cutting Force Measurements in Milling Using Inverse Filtering 3
The magnitudes of Hxx(ω) and Hyy(ω) matches well with their respective minimum-phase FRFs Hxx(min)(ω) and Hyy (min)(ω). The combined FRFs indicate a good amplitude correction over the entire frequency range, even at dominant modes. Due to the minimum-phase transformation, the phase responses of measured and minimum-phase FRFs differ, which is clearly visible in the phase responses of the combined FRFs. However, the phase responses of the combined FRFs are dominated by linear trends which can be related to constant delays in the time domain and will not affect the amplitude responses of the inverse filtered signals. Thus, the phase responses affecting the inverse filtered signals are obtained by removing the linear phase trends from the combined FRFs. As seen in Figs 1.3 and 1.4, the phases of the combined FRFs have deviated approximately 7 degrees from zero at 2500 Hz. Therefore, it is expected that frequencies above 2500 Hz will be out of range for the inverse filter. To avoid influences from the resonances at 2550 Hz in the x-direction and 2700 Hz in the y-direction, the frequency limit of the inverse filters was set to 2400 Hz in both directions; higher frequencies are removed by low-pass filtering. Low-pass filtering also removes any high-frequency noise present in the inverse filtered signal. High frequency noise is common when performing inverse filtering due to the nature of the filter. The reason is that mechanical systems normally act as low-pass filters, attenuating high frequencies. When these systems are inverted they will instead act as high-pass filters and therefore respond badly to high frequency noise. The low-pass filter design used in the simulations is a Butterworth filter of order 3. To remove any phase distortion caused by the low-pass filter, zero-phase filtering is performed by filtering the data in both forward and reverse directions. To test the behavior of the inverse filters, simulations are carried out for the cutting speeds and feed rates listed in Table 1.1. Down milling and 50 % radial immersion is used to excite the dynamometer with a high frequency transient signal and thereby clear effects from the dynamometer dynamics appear. The simulations are carried out using bothx andy as feed directions. Mechanistic modeled cutting forces are used as input (see Table 1.2), these are filtered through ETFEs of the measured mixed-phase transfer functions, hxx(n), hxy(n) andhyy(n), hyx(n). Each output is then inverse filtered and compared to the reference cutting force (simulated input force), Figs 1.5 and 1.6. The simulated output forces have contributions both from the point FRFs and the cross FRFs. Due to the phase distortion caused by the minimum-phase transformation, 0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000 0 0.2 0.4 0.6 0.8 1 Frequency [Hz] Coherence 0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000 0 2 4 6 8 10 Frequency [Hz] Magnitude [N/N] H xx H xy H xx H xy Fig. 1.1 Measured FRFs of the force dynamometer from impulse excitation. The dynamometer is excited in the x-direction and responses collected in x- and y-directions 4 M. Magnevall and T. Beno
Coherence 0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000 Frequency [Hz] 0 2 4 6 Magnitude [N/N] 0 0 0.5 1 500 1000 1500 2000 2500 3000 3500 4000 4500 5000 Frequency [Hz] H yy H yx H yy H yx Fig. 1.2 Measured FRFs of the force dynamometer from impulse excitation. The dynamometer is excited in the y-direction and responses collected in y- and x-directions 0 500 1000 1500 2000 2500 3000 −600 −400 −200 0 Phase [Deg] Frequency [Hz] 0 500 1000 1500 2000 2500 3000 0 2 4 6 8 10 Frequency [Hz] Magnitude [N/N] Measured FRF, H xx Estimated (ETFE), H xx(min) Combined FRF Measured FRF, H xx Estimated (ETFE), H xx(min) Combined FRF Combined FRF without linear phase Fig. 1.3 Magnitude and phase responses of Hxx(ω), Hxx(min)(ω) and the combined FRF, nb ¼200 and na ¼55: The dynamometer’s dominant modes are: 1110 Hz, 1725 Hz, 2700 Hz 1 Improved Cutting Force Measurements in Milling Using Inverse Filtering 5
superposition of e.g. hxx(min)(n) andhxy(min)(n) does not apply. Therefore onlyhxx(min)(n) or hyy(min)(n) are used in the inverse filters. Thus, for the results to be accurate, the effect of the cross FRFs have to be negligible. The similarity between the signals is estimated by calculating the ratio of the energy in the difference between reference and inverse filtered force, E f R tð Þ f I tð Þ ð Þ and the energy in the reference signal, E(fR(t)), according to: η ¼min 1þ RfIfI 0ð Þ f 2 I 2 RfRfI τð Þ fRfI RfRfR 0ð Þ f 2 R ! ð1:7Þ where RfRfR 0ð Þ and RfIfI 0ð Þ are the autocorrelations at zero time delay between reference and inverse filtered forces, respectively. RfRfI τð Þ refers to the cross correlation between reference and inverse filtered forces, where τ is the time delay 0 500 1000 1500 2000 2500 3000 −600 −400 −200 0 Phase [Deg] Frequency [Hz] 0 500 1000 1500 2000 2500 3000 0 2 4 6 8 Frequency [Hz] Magnitude [N/N] Measured FRF, H yy Estimated (ETFE), H yy(min) Combined FRF Measured FRF, H yy Estimated (ETFE), H yy(min) Combined FRF Combined FRF without linear phase Fig. 1.4 Magnitude and phase responses of Hyy(ω), Hyy(min)(ω) and the combined FRF, nb ¼200 and na ¼55. The dynamometer’s dominant modes are: 1120 Hz, 1430 Hz, 1630 Hz, 1750 Hz and 2550 Hz Table 1.1 Equipment and cutting data used in the cutting force measurements Machine tool MORI SEIKI NV5000 Cutting tool Sandvik R331.35-050A20EM100 Insert Sandvik N331.1A-08 45 08H-NL H10 Work piece material AISI 7075 Force dynamometer Kistler 9255B Charge amplifier Kistler 5011 Cutter diameter, Dcap 50 [mm] Radial cutting depth, ae 10; 25; 40 [mm] Axial cutting depth, ap 3 [mm] Number of teeth, zc 1 Cutting speeds, vc 200; 400; 800; 1200; 1490; 2000 [m/min] Tooth passing frequencies, ft 21.2; 42.4; 84.8; 127.3; 158.1; 212.2 [Hz] Feed rates, fz 0.05; 0.1; 0.15; 0.2; 0.25 [mm/tooth] 6 M. Magnevall and T. Beno
between the two signals, [12]. fR and fI are the average values of the reference and inverse filtered forces, respectively. The best fit between the signals is found at the time delay, τ, where the energy in the difference between the two signals has a minimum. If η ¼0 the two signals are identical. The simulation results show that the inverse filters are able to counteract the dynamic influences and reconstruct the reference cutting forces within a small error margin for all cutting speeds and feed rates listed in Table 1.1. As seen in Figs 1.5a and 1.6a the difference between reference and inverse filtered forces increases as the cutting speed increases. This is expected since the cutting speed affect ramp up times and thereby the frequency content in the reference force signal. Higher cutting speed leads to higher frequencies in the reference force. Thus, the frequency range of the inverse filter may not be enough to fully describe the transient behavior in the force signal at high cutting speeds. Comparisons between reference and inverse filtered forces using maximum feed rate and cutting speed are shown in Figs 1.5b and 1.6b. Additionally, simulations were performed without considering the effect of the cross FRFs on the dynamometer outputs. Neglecting the cross FRFs did not show any significant changes in the results confirming that the cross FRFs are negligible. 1.4 Experimental Tests The proposed method was evaluated in experimental cutting tests using different cutting speeds, feed rates and radial immersion. To be able to test the method over a large span of cutting speeds, the tests were performed in aluminum. Forces in both x- and y-directions were recorded and inverse filtered. The results are compared with unfiltered, low-pass filtered and simulated cutting forces. The simulated cutting forces are mechanistic modeled and the cutting coefficients are estimated from milling tests in the work piece used in the experimental tests, see Table 1.2, [13]. The test setup is illustrated in Fig. 1.7. Equipment and cutting data used are listed in Table 1.1. Table 1.2 Estimated cutting force coefficients from milling tests vc Ktc Kte Krc Kre Kac Kae 200 [m/min] 732.26 26.38 163.71 15.65 52.07 18.90 [N/mm2] 400 [m/min] 693.05 18.37 128.67 9.66 36.80 14.771 [N/mm2] 800 [m/min] 638.00 17.92 99.43 11.33 5.47 12.69 [N/mm2] 1200 [m/min] 649.03 21.78 42.14 14.67 23.76 4.61 [N/mm2] 1490 [m/min] 617.35 19.71 24.12 12.77 87.01 16.72 [N/mm2] 2000 [m/min] 597.66 17.40 24.44 11.61 105.45 15.61 [N/mm2] 500 0.05 0.1 0.15 Feed Rate, fz [mm/tooth] Force [N] 0.2 0.25 a b 0.5 1 1.5 2 2.5 3 0.5 1 1.5 2 2.5 3 1000 Cutting Speed, vc [m/min] 1500 2000 0 0.002 0.004 0.006 0.008 0.01 Time [s] 1500 1000 Reference force, fR(t) Inverse Filtered, fI(t) Dynamometer Output, fD(t) 500 0 Fig. 1.5 Simulation results with feed in x-direction; (a) The isolines of η [%], Eq. (1.7). (b) Comparison between dynamometer output, inverse filtered and reference input forces (Cutting speed, vc ¼2000 [m/min]; Feed rate, f z ¼0:25 [mm/tooth]) 1 Improved Cutting Force Measurements in Milling Using Inverse Filtering 7
The cut-off frequency used for direct low-pass filtering of the cutting forces was set by visually inspecting the FRFs in Figs 1.1 and 1.2. As seen, the frequency response of the force dynamometer is relatively flat up to 600 Hz. Information contained in the region above 600 Hz is expected to be effected by the force dynamometer dynamics and generate an erroneous output. Thus, the cut-off frequency used for direct low-pass filtering was set to 600 Hz. The cut-off frequency used for low-pass filtering the inverse filtered signals is the same as used in the simulations, 2400 Hz. The filter design and filtering techniques used in the simulations are also used in the experimental tests. Figs 1.8 and 1.9 show results from tests with different radial immersion and cutting speeds. The results show that inverse filtered cutting forces are able to predict both amplitude and ramp-up in a reliable manner. Compared with simulated forces the difference in amplitude is small for all cutting conditions tested. Both amplitude and ramp-up prediction is better for inverse filtered compared to low-pass filtered forces. The difference is especially clear for transient cutting conditions and high cutting speeds. Another issue using low-pass filtering is that changes in the estimated cutting forces with respect to cut-off frequency make it difficult to tell if and when the obtained results are accurate. 500 0 0.05 0.1 0.15 Feed Rate, fz [mm/tooth] Force [N] 0.2 0.25 a b 0.5 1 1.5 2 2.5 3 0.5 1 1.5 2 2.5 1000 Cutting Speed, vc [m/min] 1500 2000 −200 0 200 400 600 800 1000 1200 1400 0 2 4 Time [s] x10−3 6 8 Reference force, fR(t) Inverse Filtered, fI(t) Dynamometer Output, fD(t) Fig. 1.6 Simulation results with feed in y-direction; (a) The isolines of η [%], Eq. (1.7). (b) Comparison between dynamometer output, inverse filtered and reference input forces (Cutting speed, vc ¼2000 [m/min]; Feed rate, f z ¼0:25 [mm/tooth]) Fig. 1.7 The test setup showing force dynamometer, workpiece and cutter used in the measurements 8 M. Magnevall and T. Beno
Also, using low-pass filtering the cut-off frequency should be set to the highest possible value in order to preserve as much information as possible. This requires measurements of the dynamometers frequency response for each specific set-up. Thus, to properly select a cut-off frequency for low-pass filtering, the same number of frequency response measurements are needed as are needed to construct an inverse filter. Since the frequency range, in this case, was extended by 400 % using inverse filtering compared to low-pass filtering, the results become more reliable and less sensitive to changes in cutting conditions. 1.5 Conclusions A method to create an inverse filter for improved cutting force measurements based on minimum-phase FRFs has been presented. The method is based on the assumptions that the force dynamometer can be described by a linear model and that the cross FRFs of the system are negligible. These assumptions were verified by simulations, combining both measured mixed-phase and minimum-phase FRFs of the force dynamometer, confirming the inverse filters validity within the boundaries of the cutting conditions used in the measurements. The method successfully counteracted the dynamometer dynamics in experimental tests with different feed rates, cutting directions and cutting speeds. The results show that a more reliable estimation of the cutting forces can be obtained using the proposed method compared to traditional low-pass filtering, especially under transient cutting conditions and high cutting speeds. Since the dynamic corrections are performed Fig. 1.8 Comparison of unfiltered, simulated, inverse filtered and low-pass filtered cutting forces. Down milling; Cutting speed, vc ¼1200 [m/min]; Tooth passing frequency, f t ¼127:3 [Hz]; Feed rate, f z ¼0:2 [mm/tooth]; Radial immersion ae ¼25 [mm] 1 Improved Cutting Force Measurements in Milling Using Inverse Filtering 9
in the time domain, compensation can be performed on both stationary and non-stationary signals, which allows the method to be used under both constant and varying cutting conditions. The method can be used to study detailed force responses such as transient entering and/or exiting forces, which is important, e.g., for cutting tool design. References 1. Tlusty, J., Jang D., Tarng, Y.: Measurements of milling force over a wide frequency range (1987) 2. Jensen, S.A., Shin, Y.C., Davies, P.: Inverse filtering of unwanted system dynamics in cutting force measurement. In: American Society of Mechanical Engineers, Dynamic Systems and Control Division (Publication) DSC, vol. 58, pp. 167–174 (1996) 3. Park, S.S., Altintas, Y.: Dynamic compensation of spindle integrated force sensors with Kalman filter. J. Dyn. Syst. Meas. Control 126(3), 443–452 (2004) 4. Altintas, Y., Park, S.: Dynamic compensation of spindle-integrated force sensors. CIRP Ann. - Manuf. Tech. 53(1), 305–308 (2004) 5. Park, S., Malekian, M.: Mechanistic modeling and accurate measurement of micro end milling forces. CIRP Ann. - Manuf. Tech. 58(1), 49–52 (2009) 6. Hawksford M.J.: Minimum-phase signal processing for loudspeaker systems (1996) 7. Pei, S.-C., Lin, H.-S.: Minimum-phase FIR filter design using real cepstrum. IEEE Trans. Circ. Syst. II-Express Briefs 53(10), 1113–1117 (2006) 8. Oppenheim, A.V., Schafer, R.W.: Discrete-Time Signal Processing, 3rd edn. Pearson Education, Upper Saddle River (2010) 9. Proakis, J.G., Manolakis, D.G.: Digital Signal Processing: Principles, Algorithms and Applications, 3rd edn. Prentice-Hall, Upper Saddle River (1996) Fig. 1.9 Comparison of unfiltered, simulated, inverse filtered and low-pass filtered cutting forces. Down milling; Cutting speed, vc ¼2000 [m/min]; Tooth passing frequency, f t ¼212:2 [Hz]; Feed rate, f z ¼0:2 [mm/tooth]; Radial immersion ae ¼40 [mm] 10 M. Magnevall and T. Beno
10. Hayes, M.H.: Statistical Digital Signal Processing and Modeling. Wiley, New York (1996) 11. Magnevall, M., Lundblad, M., Ahlin, K., Broman, G.: High frequency measurements of cutting forces in milling by inverse filtering. Mach. Sci. Technol. 16(4), 487–500 (2012) 12. Bendat, J.S., Piersol, A.G.: Engineering Applications of Correlation and Spectral Analysis, 2nd edn. Wiley, New York (1993) 13. Altintas, Y.: Manufacturing Automation: Metal Cutting Mechanics, Machine Tool Vibrations, and CNC Design, pp. 35–42. Cambridge University Press, Cambridge (2000) 1 Improved Cutting Force Measurements in Milling Using Inverse Filtering 11
Chapter 2 Use of a Depth Camera as a Contactless Displacement Field Sensor Jean Michel Franco, Johannio Marulanda, and Peter Thomson Abstract During experimental tests, optical displacement measures can provide reliable data about the behavior of structural elements without altering key parameters, such as damping, stiffness, or mass, with low cost and high spatial density of measurements. Motion capture Systems are used in different application from medicine to cinematography, involving different types of image processing techniques, but its application to measure the response of civil structures is costly and of limited value in terms of real implementations. Range/Depth Cameras, on the other hand, can provide a 3-D imaging Solution to capture motion and displacements at an affordable cost. These cameras are widely available and used in the videogames industry. This paper presents the first steps for the implementation of a large-displacement measurement methodology and its application. Keywords Artificial vision • Instrumentation • Displacement measures • Depth camera 2.1 Introduction Kinect for Xbox360™[1] is essentially a set of sensors which comprises a triaxial accelerometer, an RGB camera and an infrared camera, initially developed for detecting human features in three dimensions, with a primary application in the field of video games Through a pattern generated by an infrared laser a Range/Depth camera is achieved for three-dimensional scene detection where lighting stops playing an important role as it is in other artificial vision systems, making it a sensor with good performance [2–4] and remarkably low cost. (<150 US $) [5, 6]. The methodology is proposed based on improvements over the Kinect for Xbox360™raw data using 3D interpolations and a 3D correspondence technique [7] for the measurement of the actual displacement field at a certain time. Test were made using two acquisition methodologies, one based on a continuous 3D reconstruction using Kinect Fusion that is included on the Microsoft Kinect Framework; other extracting raw Kinect depth using Matlab through the Image Acquisition Toolbox that supports Kinect for Xbox360™devices. Once an acquisition is made, it is performed a 3d interpolation for a normalization of the scattered data in aims to provide normalized data and a posterior processing with a 3D correspondence technique to improve results. Due to the measuring characteristics and low cost, has become a multipurpose sensor, in different areas from surveys of complex three-dimensional scenes [8], applications focused on improving and reducing costs in augmented reality systems [5], to characterization of turbulent flows using multiple Kinect’s [9]. There are studies of the use of the Kinect for the realization of whole plant phenotypes [10], evaluation of postures in the human body and on-line medical evaluation using Kinect generated point clouds from a clinical environment [3, 6, 11]. There are applications for tracking objects in 3D space [16], machine vision applications in robotics for automated three-dimensional survey [12], also there are multiple calibration approaches [13–15] and even improvements to sensor characteristics [16] as are further comparisons with instruments such as laser scanners commonly used for three-dimensional surveys [13] demonstrating the versatility of this sensor. J.M. Franco (*) • J. Marulanda • P. Thomson Escuela de Ingenier´ıa Civil y Geoma´tica, Universidad del Valle, Cali, Colombia Grupo de Investigaci on en Ingenier´ıa S´ısmica, E olica, Geote´cnica y Estructural (G-7), Cali, Colombia e-mail: jean.franco@correounivalle.edu.co #The Society for Experimental Mechanics, Inc. 2016 A. Brandt, R. Singhal (eds.), Shock & Vibration, Aircraft/Aerospace, Energy Harvesting, Acoustics & Optics, Volume 9, Conference Proceedings of the Society for Experimental Mechanics Series, DOI 10.1007/978-3-319-30087-0_2 13
2.2 Static Displacement Approach A foam piece with three straight perpendicular segments, each one in a different coordinate direction, is deformed in a complex arbitrary displacement form as an initial setup. On the surface of the piece, reflective markers used on the OPTITRACK motion capture system are located, composed of six Near Infrared Cameras to obtain a precise 3D location of each marker on the surface of the foam (Fig. 2.1). The OPTITRACK camera array location is calibrated following the instructions provided by the developer and the positions of each marker are captured. Using Kinect Fusion software the geometry of the foam is measured, moving the Kinect until a full survey of the piece is completed and a three-dimensional point cloud is saved. The process is repeated for another arbitrary displacement form as a final setup of the piece (Fig. 2.2). Using a CAD software, marker positions where used to compare the Euclidean displacements between the initial and finals position of the piece. A total of 21 points were evaluated on four faces of the piece. It is found that the differences between displacements found using OPTITRACK motion capture system and the Kinect are well related (Table 2.1). Using the Coherent Point Drift algorithm [7] to establish the correspondence between point clouds it is found a good correspondence visually, as seen on Fig. 2.3. The CPD algorithm fails to achieve a good assessment of the displacement at the six positions showed on Fig. 2.3, but provides a good guess of the Euclidean distances (Table 2.2). 2.3 Dynamic Displacement Approach A Dynamic shaker APS 400 ELECTRO-SEIS ® is instrumented using a UNIMEASURE LX-PA linear position transducer and a Kinect for Xbox360™(Fig. 2.4) to measure a frequency sweep between 0.1 and 2.5 Hz using an average amplitude of 5 cm. Kinect Depth raw data is acquired using the Image Acquisition Toolbox for Matlab and the displacement data is acquired using a NI DAQ PAD60-15 data acquisition system. Pinhole camera calibration parameters are applied to the Kinect depth raw data. The shaker positions are extracted from the same point over time on the Kinect depth data. A comparison between the Kinect extracted positions over the measured positions using the show a correlation of 79.2 % (Fig. 2.5). A spectrogram of the Kinect retrieved displacement signal is shown in Fig. 2.6. Kinect for xbox360™ OPTITRACK IR cameras™ Fig. 2.1 Experimental setup for static displacement approach 14 J.M. Franco et al.
Fig. 2.2 Initial (green) vs. final (red) for one face Table 2.1 Kinect vs. Optitrack marker displacements Displacement Kinect (m) Displacement OPTITRACK (m) Difference (mm) % Difference Face 1 1 0.0306 0.0305 0.1 0.3 2 0.0724 0.0765 4.1 5.7 3 0.1503 0.1531 2.8 1.9 4 0.2496 0.2500 0.4 0.2 5 0.364 0.3574 6.6 1.8 6 0.4882 0.4813 6.9 1.4 Face 2 7 0.0089 0.0058 3.1 34.8 8 0.0605 0.0562 4.3 7.1 9 0.1459 0.1396 6.3 4.3 10 0.2560 0.2453 10.7 4.2 11 0.3733 0.3625 10.8 2.9 12 0.4967 0.4884 8.3 1.7 Face 3 12 0.4967 0.4884 8.3 1.7 13 0.5444 0.5363 8.1 1.5 14 0.5947 0.5897 5 0.8 15 0.6566 0.6513 5.3 0.8 16 0.7154 0.7133 2.1 0.3 Face 4 17 0.5747 0.5742 0.5 0.1 18 0.6110 0.6112 0.2 0.0 19 0.6560 0.6599 3.9 0.6 20 0.7070 0.7102 3.2 0.5 21 0.7599 0.7684 8.5 1.1 Mean 3.3 (The indicated bold values are for the maximun and minimun Difference %) 2 Use of a Depth Camera as a Contactless Displacement Field Sensor 15
Fig. 2.4 Experimental setup for dynamic displacement approach Table 2.2 Kinect CPD vs Optitrack marker displacements Displacement Kinect CPD (m) Displacement Optitrack (m) Difference (mm) % Difference Face 1 1 0.0376 0.0305 7.1 18.9 2 0.063 0.0765 13.5 21.4 3 0.1574 0.1531 4.3 2.7 4 0.2597 0.25 9.7 3.7 5 0.375 0.3574 17.6 4.7 6 0.4671 0.4813 14.2 3.0 Mean 9.1 Fig. 2.3 3-Dimensional correspondences for Face 1 16 J.M. Franco et al.
Using the same methodology, the response of a benchmark structure of 3 DOF with a first natural frequency of 2.67 Hz (Fig. 2.7) is measured during free vibration and the displacements are retrieved from the Kinect for Xbox360™as shownon Fig. 2.8. A Power Spectral density of the retrieved displacements is shown in Fig. 2.9 where a predominant frequency of 2.69 Hz is identified Fig. 2.6 Spectrogram for the Kinect retrieved displacements Fig. 2.5 ENCODER vs. Kinect retrieved displacements 2 Use of a Depth Camera as a Contactless Displacement Field Sensor 17
Fig. 2.8 Benchmark structure retrieved displacements Fig. 2.7 Benchmark structure Fig. 2.9 Kinect displacement power spectral density
2.4 Conclusions and Perspectives Kinect Fusion acquisition methodology showed a poor performance on low magnitude displacements in comparison to a motion capture system, but has the advantage of being able to perform a 3-dimensional survey and can provide a good guess of the displacements on elements with complex geometry. Kinect depth raw data calibrated using a pinhole model parameters showed high accuracy assessing the displacements of a dynamical shaker and the dynamic measures shows a good agreement up to 2.5 Hz when large displacements are present. On a benchmark structure it is found a good correlation on the identified first natural frequency. It is expected to improve the displacement measurements using the proposed methodology and a Kinect for Windows v2 sensor with better performance and a Time of Flight technique to obtain depth data. References 1. Microsoft®. Kinect™for Xbox360™Available from: http://www.xbox.com/en-us/kinect/ (2011) 2. Che´ne´, Y., et al.: On the use of depth camera for 3D phenotyping of entire plants. Comput. Electron. Agric. 82, 122–127 (2012) 3. Jing, T., et al.: Scanning 3D full human bodies using s. IEEE Trans. Vis. Comput. Graph. 18(4), 643–650 (2012) 4. Nakamura, T.: Real-time 3-D object tracking using Kinect sensor. In: IEEE International Conference on Robotics and Biomimetics (ROBIO), pp. 784–788. (2011) 5. Placitelli, A.P. and Gallo L.: Low-cost augmented reality systems via 3D point cloud sensors. In: Signal-Image Technology and Internet-Based Systems (SITIS), 2011 Seventh International Conference on, 2011 6. Clark, R.A., et al.: Validity of the Microsoft Kinect for assessment of postural control. Gait Posture 36, 372–377 (2012) 7. Myronenko, A., Song, X.: Point set registration: coherent point drift. IEEE Trans. Pattern Anal. Mach. Intell. 32(12), 2262–2275 (2010) 8. Varadarajan, K.M., Vincze M. (2011) Surface reconstruction for RGB-D data using real-time depth propagation. In: Computer Vision Workshops (ICCV Workshops), 2011 I.E. International Conference on, 2011 9. Berger, K., et al.: The capturing of turbulent gas flows using multiple Kinects. In: Computer Vision Workshops (ICCV Workshops), 2011 I.E. International Conference on, 2011 10. Che´ne´, Y., et al.: On the use of depth camera for 3D phenotyping of entire plants. Comput. Electron. Agri. 82, 122–127 (2012) 11. Placitelli, A.P., Gallo, L.: 3D point cloud sensors for low-cost medical in-situ visualization. In: Bioinformatics and Biomedicine Workshops (BIBMW), 2011 I.E. International Conference on, 2011 12. Nakamura, T.: Real-time 3-D object tracking using Kinect sensor. IEEE International Conference on Robotics and Biomimetics (ROBIO), pp. 784–788. (2011) 13. Chan-Soo, P., et al.: Comparison of plane extraction performance using laser scanner and Kinect. In: Ubiquitous Robots and Ambient Intelligence (URAI), 2011 8th International Conference on, 2011 14. Dutta, T.: Evaluation of the Kinect™sensor for 3-D kinematic measurement in the workplace. Appl. Ergon. 43(4), 645–649 (2012) 15. Khoshelham, K.: Accuracy Analysis of Kinect Depth Data. GeoInf. Sci. 38(5), 6 (2010) 16. Suttasupa, Y., Sudsang, A., Niparnan, N.: Plane detection for Kinect image sequences. In: Robotics and Biomimetics (ROBIO), 2011 I.E. International Conference on, 2011 2 Use of a Depth Camera as a Contactless Displacement Field Sensor 19
Chapter 3 Uncertainty of Digital Image Correlation with Vibrating Deformable Targets Alfredo Cigada, Navid Hasheminejad, and Emanuele Zappa Abstract Digital Image Correlation (DIC) is a full-field, non-contact optical technique to measure the contour, deformation, vibration and strain on the surface of an object. Studies on this method and its uncertainty are mostly focused on motion in static conditions. In dynamic conditions, as the target is moving with respect to the camera, the images acquired during the experiments have a considerable amount of motion effect (blurring). This motion effect is the main cause of uncertainty in measurement with DIC method. This work experimentally investigates the effect of the shutter time and the motion of the target on the uncertainty of DIC method. Tests were done on an aluminum cantilever beam, with first natural frequency of 7.6 Hz and a random speckled pattern on its surface. An 8 MP camera acquired the decay free vibration of the full length of the beam (0.93 m) with 50 frames per second while a laser triangulation sensor was used as a reference transducer in a section of the beam. Effect of exposure time, sensor gain and vibration amplitudes on measuring the uncertainty is explored. A software was developed to synchronize the laser and camera, while DIC analysis was done with Vic-2D software. The results show that the uncertainty of DIC is proportional to the product of the exposure time and the initial displacement of the beam (i.e. the stripe length). Afterwards, deconvolution method, which is a processing technique to estimate the displacement of an object and the motion effect that exist in an acquired image, was used to improve the results in case of having a deformable target in dynamics conditions. Keywords Digital image correlation • Uncertainty • Motion effect • Deconvolution • Deformable targets 3.1 Introduction Digital image correlation is a popular vision-based measurement method to track the motion of an object. DIC was first developed in the 1980s by Sutton et al. [1]. This technique has been used mostly in static applications [2], but recently, its application has been extended to dynamic applications too [3–5]. It is known that in static conditions the measuring uncertainty of a vision-based measurement system is dependent on image resolution, contrast, processing algorithm, noise, etc. but in dynamic condition there are also other factors like exposure time of the camera or relative motion between the camera and the object [6]. Therefore, setup of the camera and specifying its different parameter can be important for a DIC analysis in dynamic conditions. In this study, experiments were done on a deformable object (cantilever beam) to investigate the important parameters of the camera that can affect the uncertainty of DIC method in dynamic conditions. In dynamic conditions, the motion of the target causes blurring in the acquired images. This blurring is an important source of uncertainty in DIC measurement. To solve this issue, one can decrease the exposure time of the camera. However, it is not possible to arbitrarily decrease the exposure time, because decreasing the exposure time leads to darker images. Therefore, a higher light intensity is required to have images with proper brightness, which is not always possible in real applications [7]. Thus, deconvolution method to create new reference images was suggested to solve this issue for rigid objects [8]. In this method, the motion effect that exists in an acquired image is estimated by deconvolution method. Then this motion is simulated on a reference image. Using this technique, DIC accuracy in dynamic conditions is improved, as both the acquired image and the reference image have the same blurring [8, 9]. In this research, after finding the important parameters of the camera and motion on DIC uncertainty and comparing two different image processing method to estimate the displacement, deconvolution method to generate new reference images was used to improve the accuracy of DIC method for deformable targets in dynamic conditions. A. Cigada • N. Hasheminejad • E. Zappa (*) Department of Mechanical Engineering, Politecnico di Milano, Via La Masa 1, Milan, Italy e-mail: emanuele.zappa@polimi.it #The Society for Experimental Mechanics, Inc. 2016 A. Brandt, R. Singhal (eds.), Shock & Vibration, Aircraft/Aerospace, Energy Harvesting, Acoustics & Optics, Volume 9, Conference Proceedings of the Society for Experimental Mechanics Series, DOI 10.1007/978-3-319-30087-0_3 21
3.2 Experiments The experiments are done on a beam made of aluminum. The cross section of the beam is a 10*30 mm rectangular and the length of the beam is 993 mm. First, using stencil technique, a speckle pattern was painted on one side of the beam. The speckle pattern was painted on the 10*993 mm face of the beam by a frame with circle shape holes. The nominal diameter of these circles was chosen 1.2 mm, which corresponds to speckles with a diameter of 4 px in the images for the specific setup condition, of these experiments. This is the optimal size according to the literature [10, 11]. A portion of the painted side of the beam can be seen in Fig. 3.1. Then one side of the beam was fixed to a grip and an initial displacement was applied to the tip of the cantilever beam. The camera was in front of the beam, capturing images during the motion. The variables of the experiments are the amount of applied initial displacement to the beam and the exposure time of the camera. Before applying the excitation, a picture of the beam in static conditions was acquired; this image will be used as the reference image during the DIC analysis. The beam has free vibration and stops after a few seconds. Target’s motion was recorded by a digital camera with a frame rate of 50 fps. The shutter speed of the camera was changed from 1000 to 2000, 3000, 4000 and 5000 ms. Meanwhile, the gain was 0, 3, 6 and 9 depending on the camera’s shutter speed. The vertical displacement of the beam, at a section 800 mm apart from the constraint, was measured simultaneously with a laser triangulation sensor. A software was developed to synchronize the camera and laser sensor. The camera used during the experiment is a GX3300 produced by Allied Vision Technologies with 3296*2472 px resolution and the laser triangulation sensors is an ILD 1400-20(00) produced by Micro Epsilon company with a range of 20 mm. 3.3 Digital Image Correlation The acquired images during the experiments were analyzed by Vic-2D software. Vic-2D uses DIC algorithms to calculate the displacement of each portion of the images acquired during the test with respect to the reference image acquired in static conditions. The first test of the experiment was analyzed with different subsets (15*15, 21*21 and 27*27 pixels) and it was proven that the results are not considerably affected by this parameter. The rest of the analysis is done using a 15*15 px subset and a step of 3 px. The uncertainty associated with the estimated displacement of each subset depends on different parameters, including the speckle quality in the subset, the local lighting conditions and the noise. Sigma is an output of Vic-2D, which shows standard Camera Laser Speckled Beam Speckle Pattern 933 mm 800 mm Laser Sensor Camera 466.5 mm 466.5 mm Fig. 3.1 (Left) Measurement setup, (Right) Schematic configuration of the experiment, front view (up), top view (down) 22 A. Cigada et al.
RkJQdWJsaXNoZXIy MTMzNzEzMQ==