Fig. 17. Detailed measured deformations. (a) Test 6: measured εm45, εmand εmrr. (b) Test 7: measured εm45, εmand εmrr.kand k(used for both normal and shear stresses) are morecritical for the evaluation of shear stress than for normal stress.Thus, the order of truncation determined for the regularization ofthe inverse method is smaller for shear stress than for normal stress.It should be noted that in this work the automatic procedure of thetruncation process has not been used and should be developed forfuture industrial applications.Since the inverse method is based on truncated sums, it is log-ical to obtain unwanted large oscillations just before and afterthe stress peak (alike the reconstruction with Fourier series of arectangular signal). However, this classical issue is overcome byapplying a filtering window in order to focus on contact stressesonly. The contact length lCis evaluated by using the strip entryand exit thicknesses t0and t1, the rolling force and the approxi-mate deformed roll radius as established classically by Hitchcock(1935). Thus, the filtering window is set such as the contact isselected. Contact stresses presented in Figs. 15 and 16 are evaluatedwith the inverse method developed in this paper and the mea-sured strains presented in Fig. 13. Very classical peak of pressurerr(Rs, ) and sign change of shear stress r(Rs, ) are obtained inthe roll gap. Amplitudes of shear stress and contact pressure donot have the same order of magnitude along the contact length,which was predicted by numerical simulations. Furthermore, thehigh-speed data acquisition frequency enables the evaluation ofcontact stresses at semi-industrial rolling speeds (test 5) with-out any significant quality loss. The shear stress reconstruction issomewhat a bit poorly reconstructed compared to normal stress,due to insufficient number of harmonics used for reconstruction;however this difficulty may be overcome by increasing the acqui-sition frequency and by correcting the radial strain measurementissue. Present results are anyway promising for future adaptationon high speed industrial mills. Pilot tests presented in this sec-tion can be modeled numerically using the software LAM3 basedon Hacquin (1996) as mentioned in the introduction. The stripis modeled by FEM with elasto-visco-plastic behavior and theroll remains elastic. The coupling between strip and roll defor-mations are taken into account as well as rolling force, rotationspeed, entry and exit tension applied to the strip, entry and exitFig