Popis: |
Due to demand for lower emissions and better crashworthiness, the use of boron ultra high strength steel (UHSS) has greatly increased in manufacturing of automotive components. However in many cases an idealized component has got different mechanical properties in different regions. For example in an automotive structural component such as B-pillar, which may undergo impact loading, it is desirable that there are certain regions in it which are softer and more ductile so that component's overall energy absorption is improved. The innovative process of tailored hot stamping allows for this by controlling the localized cooling rates, through actively dividing the tooling into heated and cooled zones. A barrier to optimal application of the technique is that a reliable phase distribution model is required to predict the distribution of different phases which occur in the different regions of a tailored hot stamped component. Currently most of the existing physical models for phase distribution prediction in boron steel after hot stamping only take into account the thermal history of the region while not accounting for the effect of deformation and thus have had only limited success so far. This research has developed a novel state-of-the-art Artificial Neural Network (ANN) based phase distribution prediction model for 22MnB5 boron UHSS steel, which is able to successfully take into account both the thermal and mechanical history while making final phase distribution predictions during tailored hot stamping. The model was developed and validated using data generated from extensive tailored hot stamping thermo-mechanical physical simulation experiments and scanned surface instrumented nanoindentation based phase quantification method. For the development of the ANN based model, the backpropagation algorithm was deployed on the available experimental data from 40 thermo-mechanical physical simulation experiments to learn the complex multivariate functional relationship between the thermal and mechanical history of the samples and the final resulting phase distributions in them. Advanced statistical techniques were used for preventing overfitting in the ANN based model while learning, for making the optimal use of limited available experimental data and for quantification of uncertainties in the predictions made by the model. After the ANN based model had been developed, its prediction performance was rigorously measured and analyzed. During measuring its prediction performance over the data used for its development, it had a prediction root mean square error of just 5.4% over 120 phase volume fraction predictions. During its validation over the completely new independent experimental data, the ANN based model had root mean square prediction error of just 7.7% over 30 phase volume fraction predictions. This excellent prediction performance of the developed ANN based model demonstrated its reliability and robustness and established the potential for ANN model to be used in future computer aided engineering applications for tailored hot stamping process. |