publications
publications by categories in reversed chronological order. generated by jekyll-scholar.
2022
- Multi-event model updating for ship structures with resource-constrained computingJason Smith, Hung-Tien Huang, Austin Downey Jr, and 3 more authorsIn Sensors and Smart Structures Technologies for Civil, Mechanical, and Aerospace Systems 2022 2022
Naval structures are subjected to damage that occurs on short-term (i.e. impact) and long-term (i.e. fatigue) time scales. Digital twins of ship structures can provide real-time condition assessments and be leveraged by a decision-making framework to enable informed response management that will increase ship survivability during engagements. A key challenge in the development of digital twins is the development of methodologies that can distinguish the various fault cases. Moreover, these methodologies must be able to operate on the resource-constrained computing environments of naval structures while meeting real-time latency constraints. This work reports on recent progress in the development of a multi-event model updating framework specially designed to meet stringent latency constraints while operating on a system with constrained computing resources. The proposed methodology tracks structural damage for both impact and fatigue damage through a swarm of particles that represent numerical models with varying input parameters with set latency and computational restraints. In this work, numerical validation is performed on a structural testbed subjected to representative wave loadings. Results demonstrate that continuous fatigue crack growth and plastic deformation caused by impact can be reliably distinguished. The effects of latency and resource constraints on the accuracy of the proposed system are quantified and discussed in this work.
- Audio-Based Wildfire Detection on Embedded SystemsHung-Tien Huang, Austin RJ Downey, and Jason D BakosElectronics 2022
The occurrence of wildfires often results in significant fatalities. As wildfires are notorious for their high speed of spread, the ability to identify wildfire at its early stage is essential in quickly obtaining control of the fire and in reducing property loss and preventing loss of life. This work presents a machine learning wildfire detecting data pipeline that can be deployed on embedded systems in remote locations. The proposed data pipeline consists of three main steps: audio preprocessing, feature engineering, and classification. Experiments show that the proposed data pipeline is capable of detecting wildfire effectively with high precision and is capable of detecting wildfire sound over the forest’s background soundscape. When being deployed on a Raspberry Pi 4, the proposed data pipeline takes 66 milliseconds to process a 1 s sound clip. To the knowledge of the author, this is the first edge-computing implementation of an audio-based wildfire detection system.
2023
- Real-time structural validation for material extrusion additive manufacturingYanzhou Fu, Austin R.J. Downey, Lang Yuan, and 1 more authorAdditive Manufacturing 2023
Material extrusion additive manufacturing is a technology that produces a part by controlling the melting and extrusion of a thermoplastic filament. A void defect in a specific position can significantly impact the whole product’s structural quality and mechanical properties. Different defect detection systems have been built and have achieved good detection accuracy. However, a challenge remains in determining if the structural performance of a part with defects will fail to meet design requirements. This research develops a real-time product structural quality validation system using a multi-dimensional accumulation-threshold-based decision-making approach. The proposed system is validated on a consumer-grade 3D printer with an optical camera. The layer-wise damage information is output from a trained convolutional neural network. The developed structural quality validation system links the obtained defect information with the decision boundary developed by the accumulation-threshold-based decision-making approach to evaluate the effect of the defect. The accumulation-threshold-based decision-making approach is trained on a novel component health index that links the location and size of defects across layers. Experimental results show that the decision boundary obtained from the accumulation-threshold-based decision-making approach performs well on test data with 96% recall and a 91% F1-score, which means the decision boundary can provide good overall results while being biased to limit false negative results. The proposed real-time structural quality validation system is validated online for a dog-bone test specimen. Results show that the computational time of the structural quality validation system requires at most 688 ms, while a dog-bone layer takes 75 s to print, thus demonstrating that the proposed system can meet required real-time constraints. The capability to run online enables users to cancel a print mid-process for specimens that will not meet structural loading requirements. The developed system is demonstrated through a video, which is provided in the supplemental materials. The dataset for this work is published as a public repository containing 450 samples with 221 failure classes.