The deduction of mechanical properties from experimental data is essentially an inverse analysis where the data are compared to the predictions of a model by modifying the input parameters for the model until a satisfactory match is attained. Often times, this is done

manually via trial-and-error. There are, however, rigorous mathematical methods that offer robust inverse analyses with enhanced accuracy. In the present study, artificial neural networks

(ANN) are employed to deduce the in-situ constitutive laws of tungsten (W) fiber reinforced bulk metallic glass (BMG) matrix composites.

Experimental data consist of lattice strain from the W fibers obtained by neutron diffraction and total composite strain measured by an extensometer. The mechanical behavior of the composites is modeled via finite element analysis (FEA). The constitutive behavior of the fibers and the matrix are described using the Voce and power laws, respectively. The effect of thermal residual stresses is also included as a function of freezing temperature below which, residual stresses start to build up during cooldown.

The goal of the present inverse analysis via ANN is to optimize the values of the Voce and power law parameters plus the freezing temperature a total of seven parameters. First, a forward ANN is constructed that attempts to match the predictions of FEA which is run multiple times via a random selection of the seven parameters. Next, inverse ANN are constructed to optimize the values of the seven parameters, i.e., as inverse models. Finally, the optimized

parameters are input to the forward ANN and compared to the experimental data.

This is the first application of ANN in the analysis and interpretation of engineering diffraction data. It demonstrates the power of ANN in conducting robust inverse analysis of such data. The approach developed and presented here can also be employed in the optimization of engineering diffraction experiments to increase their accuracy and efficiency.