The automatic recognition of a metal component or work-piece currently relies on optical techniques and image matching. It is not possible to distinguish work-pieces with different materials. In this paper, a novel electromagnetic inductive sensor array similar to those used in the electromagnetic tomography (EMT) has been designed to address this problem. Furthermore, instead of reconstructing the full magnetic polarizability tensor, we have proposed a partial tensor approach, which shows that a 2D tensor is capable of distinguishing the material difference and recognising the geometric dominance of work-pieces with experimental data. In addition, it has been found that the phase of the tensor is strongly linked to the materials properties while the magnitude of the tensor eigenvalues implies the basic geometry of work-piece.