PhD Oral Exam - Farshid Hajializadehkouchak, Mechanical Engineering

When studying for a doctoral degree (PhD), candidates submit a thesis that provides a critical review of the current state of knowledge of the thesis subject as well as the student’s own contributions to the subject. The distinguishing criterion of doctoral graduate research is a significant and original contribution to knowledge.

Once accepted, the candidate presents the thesis orally. This oral exam is open to the public.

Abstract

Additive manufacturing (AM) processes are among the manufacturing methods that are implemented in various industries. Direct metal deposition (DMD) is part of AM processes that uses the laser heat source to deposit the metallic material in form of powder or wire onto a substrate and build a component in a layer-by-layer scheme. DMD process is known to be cost-effective and easily adoptable to building complex structures without the need for preparing tools and dies. Therefore, it makes the process as one of the possible solutions for the products in the design process. The AM processes that are applied on the metals often involve with the formation of residual stresses and distortions. The heating and cooling cycles impose residual stresses and distortions to the fabricated part and affect its performance in the service. Hence, minimizing the residuals stresses and distortions of the fabricated part seems to be one of the major challenges that a designer needs to address.

There are several methods and techniques used for measuring the residual stresses of metallic components that are accurate and reliable. However, application of these methods can result in partial or complete damages to the fabricated parts and also requires considerable time and tooling for the experiment. Alternative solutions are developed to predict the residual stresses without damaging the parts including numerical-based approaches such as finite element (FE) analysis. Application of the FE in assessing the residual stress distribution of the parts prior to production step saves a lot of time and it is cost-effective. The results of the detailed FE analysis offer any desirable output such as all six components of the residual stress, strains, and distortions. The FE analysis of DMD process includes thermomechanical analysis; temperature history of the elements is obtained by performing a pure heat transfer analysis, then it is applied to the mechanical model to calculate the structural response of the part. The mechanical analysis is thought to be the time-consuming one and it determines the computational time of the FE analysis of the AM part. Therefore, several techniques and approaches were developed in the literature to address this issue and improve the computational efficiency of the FE method.

Through out this thesis, a novel approach of integrating the FE analysis with artificial neural networks (ANNs) is presented. ANNs are part of machine learning (ML) algorithms that function as the real neural network in the brain and tries to determine the logical relationship between the given inputs and the associated output(s). Therefore, a comprehensive dataset should be generated to reflect whole features of a system. Thereafter, the ANN is constructed and trained with the provided data. After the training, the network is capable of predicting the outputs without running any high demanding computations. A feed forward ANN with gradient decent backpropagation is developed in Keras (Tensorflow) with two hidden layers with different neurons for each structure. The gradient decent approach with backpropagation is used in Keras as the training method. The rectified linear unit (RELU) is also used as the nonlinear activation function of the network. To evaluate the error in the training step, the mean squared error (MSE) is utilized in the constructed network. The network is trained by minimizing the error function and adjusting the weights assigned to each layer/neuron. In the present study, several metallic structures with 12-layers and 18-layers deposition made from AISI 304L are considered. For each structure, the detailed thermomechanical FE analysis is performed and temperature history of the elements along with their dimensional features are extracted as the inputs and the residual stress components are recorded as the outputs. ABAQUS/STANDARD is used to perform the FE analysis. Several subroutines are developed to equip ABAQUS with the certain tools to account for the element addition (i.e. UMAT, UMATHT and UHARD) and also for generating certain laser path movements and heat source model (i.e. DFLUX).

The results of the 12-layers structures are utilized to train the constructed ANN and achieve an acceptable training accuracy. The input data of the 18-layers are then fed into the trained network and the residual stresses are predicted. The results of the integrated ANN-FE are compared with the results of the residual stresses of 18-layers obtained from the detailed thermomechanical analysis. The errors are calculated and shown in the form of 3D contours and scattered errors in 2D and 3D. Finally, the histogram analysis is performed for each 18-layers structures to better present the fraction of the elements with the error ranges. Furthermore, the computational times are recorded and compared to evaluate the efficiency of the proposed novel ANN-FE method.

The results showed that for almost all of the structures and all the stress components, the predicted pattern of the residual stress distribution is consistent with of the detailed FE analysis. Some of the regions with high errors are associated with the low-stress state zone in which the actual stress magnitude is very low and the high error poses no critical condition. Although there are some predictions showing higher errors in some regions, majority of the elements in the structures show the prediction error of less than 15% supported by the histogram analysis. Significant improvement (6 times as an average) in the computational time is also achieved.