Data Mining in Tensile Testing

Pierre L. Douillet, Xavier Legrand¹ and Gilles Pellan²

Abstract:

The standard traction test used for tensile testing of materials (metallic as well as composite) leads, in a first time, to a raw report, that records about a thousand of points $\left( x, y\right)$ for each test piece. In a second time, several quantities are inferred from the raw report, that characterize the mechanical properties of the test piece, such as Young's modulus and proof strength.

The final test report contains only these inferred quantities. Our papers shows that the raw report contains not only the requested mechanical quantities and a lot of noise due to the measurement uncertainties, but also a residual quantity of knowledge concerning the test piece as well as the testing machine itself.

By the way, we describe the protocol that will be used in a near future to determine which points of the raw record are relevant for the determination of Young's modulus. A consensus exists about the choice of the bounds, but the way these bounds are inferred is a key point for the determination of the uncertainty interval around the best value.