Search for: All records

PurposeThe purpose of this study is to develop a deep learning framework for additive manufacturing (AM), that can detect different defect types without being trained on specific defect data sets and can be applied for real-time process control. Design/methodology/approachThis study develops an explainable artificial intelligence (AI) framework, a zero-bias deep neural network (DNN) model for real-time defect detection during the AM process. In this method, the last dense layer of the DNN is replaced by two consecutive parts, a regular dense layer denoted (L1) for dimensional reduction, and a similarity matching layer (L2) for equal weight and non-biased cosine similarity matching. Grayscale images of 3D printed samples acquired during printing were used as the input to the zero-bias DNN. FindingsThis study demonstrates that the approach is capable of successfully detecting multiple types of defects such as cracks, stringing and warping with high accuracy without any prior training on defective data sets, with an accuracy of 99.5%. Practical implicationsOnce the model is set up, the computational time for anomaly detection is lower than the speed of image acquisition indicating the potential for real-time process control. It can also be used to minimize manual processing in AI-enabled AM. Originality/valueTo the best of the authors’ knowledge, this is the first study to use zero-bias DNN, an explainable AI approach for defect detection in AM. 

The interface characteristics of the matrix and fibers significantly influence the evolution of residual stress in composite materials. In this study, we provide a methodology for reducing the residual stress in laminated composites by modifying the thermomechanical properties at the fiber–matrix interface. A hydrothermal chemical growth method was used to grow Zinc Oxide nanowires on the carbon fibers. We then utilized a novel digital image correlation approach to evaluate strains and residual stresses, in situ, throughout the autoclave curing of composites. We find that interface modification results in the reduction of residual stress and an increase in laminate strength and stiffness. Upon growing ZnO NWs on the carbon fibers, the maximum in situ in-plane strain components were reduced by approximately 55% and 31%, respectively, while the corresponding maximum residual stresses were decreased by 50.8% and 49.33% for the cross-play laminate [0°/90°] layup in the x and y directions, respectively. For the [45°/-45°] angle ply layup in the x-direction, the strain was decreased by 27.3%, and the maximum residual stress was reduced by 41.5%, whereas in the y-direction, the strain was decreased by 166.3%, and the maximum residual stress was reduced by 17.8%. Furthermore, mechanical testing revealed that the tensile strength for the [45°/-45°] and [0°/90°] laminates increased by 130% and 20%, respectively, with the interface modification. 

Pedestrian dynamics is an approach for modeling the fine-scaled movement of people. It is finding increasing application in the analysis of infection risk for directly transmitted diseases during air travel. A parameter sweep is often needed to evaluate infection risk for a variety of possible scenarios to account for inherent variability in human behavior. A low discrepancy parameter sweep was recently introduced to reduce the computational effort by one to three orders of magnitude. However, it has the following limitations: (i) a low overhead parallelization leads to significant load imbalance, and (ii) the convergence rate worsens with dimension. This paper examines whether pseudorandom and hybrid sequences can overcome these defects and whether the convergence criteria can be changed to yield accurate solutions faster. We simulate the deplaning process of an airplane using different parameter sweep strategies and evaluate their relative computational efficiencies. Our results show that hybrid and pseudorandom parameter sweeps are advantageous for moderate accuracy, while a low discrepancy sweep is preferable for high accuracy. Our results also show that the convergence criteria could be relaxed substantially to yield accurate solutions around a factor of 20 faster. They promise to help a variety of applications that employ large parameter sweeps for modeling infection risk.