You can use this kind of setup:

http://www.mdpi.com/materials/materials-07-01422/article_deploy/html/images/materials-07-01422f1-1024.png

If you would like to glue, a lot will depend on the strength your expect your concrete will have. Tec7 can glue concrete and has a tensile strength of 2,6Mpa. Otherwise you can use epoxy glues, thee can easy go up to 17 Mpa. For example  "VersaChem® 5 Min. Epoxy System" can be used on concrete and has a tensille strength of 10Mpa

http://www.vanhoorenpolyester.be/epoxy_lijm.htm

That is probably why most testing of concrete is done in compression.  

You can embed a reinforcing bar concentrically from both sides of the specimen (top and bottom) protruding a sufficient length outside the specimen to be fixed in the machine jaws. Determine the diameter and embedded length of the bar (from both sides) to ensure that pull-out failure will not occur and tension failure will commence in concrete.  You can use the following formula to estimate the bond strength (u in MPa) of the bars with the surrounding concrete:

u = 20 sqrt(fc')/db

fc' is the concrete compression strength in MPa, and db is the bar diameter in mm.

Hi Alex, I wold recommend resurfacing the inserts or purchasing a new grip set with a higher clamping force so the grip faces clamp the samples securely. The blog post below includes commonly experienced issues with test fixtures and grips and recommended solutions:

http://www.admet.com/5-things-know-buying-tensile-testing-grips/

Alex,

Your problem is bigger than you think. Concrete is not like metals. Not only do you have to worry about getting a grip on the sample, because concrete is such a brittle substance that gets more brittle as it gets stronger, you also have to take care not to introduce extra moments through misalignments, potentially in both x and y, that can lead to inconsistent or erroneous results. Something like the opposite of loading a compression sample through a ball is needed. I suggest this can be done through chains or cables depending on loads, attached to plates glued or bolted to your specimens. Be very careful because such attachments can fly around at failure. Even with such a system, aligning the top and bottom along the intended axis is also necessary. For this I sometimes used two lasers.

Direct tension testing of concrete is very hard. If you are looking for loading curves going from zero through the peak and back to zero, it is even harder than that. As always, the question: why do you want this value? Maybe the question is already answered. Maybe there is another test to get you the information you need, Brazilian or flexural test perhaps.

Thanks Mr. Miquel Joseph for your answer which I consider as practical.

Thanks Mr. Luke for explaining the problem.

i introduced a notch , a few mm either side of specimen center hence inducing failure in that position. I use the reduced area of the notched section to calculate the direct tensile stress after failure