Neural Network Estimation of Material Mechanical Properties

The patent performs preliminary action by training the neural network model in advance using a large database of simulated indentation curves with known mechanical properties. This pre-training phase allows the model to learn the complex nonlinear relationships between indentation responses and material properties, so that when actual experimental data is input, the model can directly provide accurate predictions without being sensitive to small experimental errors.

The patent introduces a neural network model as an intermediary between the indentation experimental data and the mechanical property extraction. This intermediary model acts as a bridge that processes the experimental data through learned patterns, reducing the direct sensitivity to experimental errors that would otherwise affect traditional inversion methods.

The patent uses copying by creating numerous synthetic indentation curves through computational simulations rather than requiring an equivalent number of expensive physical experiments. The neural network is trained on a large database of simulated data that copies the essential characteristics of real indentation behavior, allowing accurate model training without proportionally large numbers of high-fidelity experimental data sets.

The patent applies parameter changes by systematically varying material properties, indentation conditions, and geometric parameters in the simulation process to generate diverse training data. This approach creates a comprehensive training database that covers a wide range of scenarios without requiring physical experiments for each case, thereby reducing the need for large numbers of high-fidelity data sets.