93887 research outputs found
Sort by
Smart Security System Based on Edge Computing and Face Recognition
No full textPhysical security is one of the most basic human needs. People care about it for various reasons; for the safety and security of personnel, to protect private assets, to prevent crime, and so forth. With the recent proliferation of AI, various smart physical security systems are getting introduced to the world. Many researchers and engineers are working on developing AI-driven physical security systems that have the capability to identify potential security threats by monitoring and analyzing data collected from various sensors. One of the most popular ways to detect unauthorized entrance to restricted space is using face recognition. With a collected stream of images and a proper algorithm, security systems can recognize faces detected from the image and send an alert when unauthorized faces are recognized. In recent years, there has been active research and development on neural networks for face recognition, e.g. FaceNet is one of the advanced algorithms. However, not much work has been done to showcase what kind of end-to-end system architecture is effective for running heavy-weight computational loads such as neural network inferences. Thus, this study explores different hardware options that can be used in security systems powered by a state-of-the-art face recognition algorithm and proposes that an edge computing based approach can significantly reduce the overall system latency and enhance the system reactiveness. To analyze the pros and cons of the proposed system, this study presents two different end-to-end system architectures. The first system is an edge computing-based system that operates most of the computational tasks at the edge node of the system, and the other is a traditional application server-based system that performs core computational tasks at the application server. Both systems adopt domain-specific hardware, Tensor Processing Units, to accelerate neural network inference. This paper walks through the implementation details of each system and explores its effectiveness. It provides a performance analysis of each system with regard to accuracy and latency and outlines the pros and cons of each system
Energy Efficient Hardware for Neural Network Applications
No full textWith the explosion of AI in recent years, there has been an exponential rise in the demand for computing resources. Although Moore’s law has so far kept up with conventional computational demands in the past, it has become evident that the efficiency and area gains with transistor scaling are no longer exponential, but rather incremental. The standard Von Neumann architecture imposes a limit on efficiency and latency as data is shuttled repeatedly between the compute and memory units. On the other hand, AI workloads rely heavily on matrix-vectormultiplication which get exponentially expensive with vector widths. In-memory and nearmemory computing have come up as promising alternatives that addresses both these issues elegantly while reducing energy requirements.A variety of NN models rely on fast and repetitive evaluation of exponential transcendental functions. In many cases, this is done by range reduction technique and math tables. For optimal energy efficiency and throughput, it is best if these tables reside as close as possible to the circuit where it is consumed. We propose a mixed-signal macro with dual functionality: ability to do matrix vector multiplication as well evaluate exp(x) for 32-bit IEEE 754 floating point number. The said macro consists of 64x64 array of special 8T cells that stores the math tables without hindering normal SRAM functionality. The charge based MVM engine uses two ADCs with reconfigurable precision, allowing faster throughput for sparse inputs. As the outputs of these operations are separate, it allows for high flexibility to use the macro in any neural-network hardware that needs either or both the functions.Spiking Neural Networks (SNN) can perform sequential learning tasks efficiently by using the inherent recurrence of membrane potential (Vmem) accumulation over several timesteps. However, the data movement of Vmem creates additional memory accesses, which becomes a bottleneck operation. Additionally, SNN input spikes are highly sparse in nature, which can be exploited for efficient hardware implementation. We propose an SNN accelerator based on inmemory processing that addresses these. The said accelerator consists of 9 compute macros and 3 neuron macros, which can be programmed to work either serially (9 compute, 1 neuron) or in 3 parallel sets (3 compute, 1 neuron) to support different layer sizes. Peripheral logic computes the membrane potential and stores it in the same compute macro, thus avoiding unnecessary data movement. The neuron macro keeps track of final membrane potential and generates output spikes. This accelerator was designed to run at 200Mhz at 1.2v in TSMC 65nm node
The Application of Lorawan as an Internet of Things Tool to Promote Data Collection in Agriculture
No full textInformation about the conditions of specific fields and assets is critical for farm managers to make operational decisions. Location, rainfall, windspeed, soil moisture, and temperature are examples of metrics that influence the ability to perform certain tasks. Monitoring these events in real time and being able to store historical data can be done using Internet of Things (IoT) devices such as sensors. The abilities of this technology have previously been communicated, yet few farmers have adopted these connected devices into their work. A lack of reliable internet connection, the high annual cost of current on-market systems, and a lack of technical awareness have all contributed to this disconnect. One technology that can better meet the demand of farmers is LoRaWAN because of its long range, low power, and low cost. To assist farmers in implementing this technology on their farms the goal was to build a LoRaWAN network with several sensors to measure metrics such as weather data, distribute these systems locally, and provide context to the operation of IoT networks. By leveraging readily available commercial hardware and opens source software two examples of standalone networks were created with sensor data stored locally and without a dependence on internet connectivity. The first use case was a kit consisting of a gateway and small PC mounted to a tripod with 6 individual sensors and cost close to $2200 in total. An additional design was prepared for a micro-computer-based version using a Raspberry Pi, which made improvements to the original design. These adjustments included a lower cost and complication of hardware, software with more open-source community support, and cataloged steps to increase approachability. Given outside factors, the PC architecture was chosen for mass distribution. Over one year, several identical units were produced and given to farms, extension educators, and vocational agricultural programs. From this series of deployments, all units survived the growing season without damage from the elements, general considerations about the chosen type of sensors and their potential drawbacks were made, the practical observed average range for packet acceptance was 3 miles, and battery life among sensors remained usable after one year. The Pi-based architecture was implemented in an individual use case with instructions to assist participation from any experience level. Ultimately, this work has introduced individuals to the possibilities of creating and managing their own network and what can be learned from a reasonably simple, self-managed data pipeline
Applied Ergonomics
Get PDFThe ergonomic injuries and solutions have been extensively studied in the construction industry; however, the prevalence of work-related injuries, risky activities, and effective solutions in the transportation industry are not understood. This study aims to explore the prevalence of work-related injuries, risky activities, and potentially ergonomic solutions among transportation workers. The approach to this study included exploration of worker type, injury types, and activities of top concern through historical injury data and an online survey, and proposal and evaluation of ergonomic solutions through onsite observations and field experiments. Results from this study found that back injuries were the most common type of injury sustained. Performing lifting and pushing/pulling activities have caused the most injuries. Back exoskeletons and ergonomic handles were identified as potential solutions to help reduce the risk of injury. Additionally, higher platforms were also suggested to help prevent workers from being forced to perform activities by exerting their back excessively
Reduced Order Modelling of Mistuned Integrally Bladed Rotors
No full textThis work aims to study the mode localization behavior of the mistuned rotor, which is the root cause of the unexpected premature fatigue failure. The unsteady loading, flow separations, tip leakage flows, vortex shedding, and acoustic instabilities induce nonlinear blade vibrations and responses. An accurate finite element model can help predict the maximum dynamic response and shed light on the dynamics of the mistuned system, which can lay out guidelines for design and manufacturing processes. Cyclic symmetry structures are generally simplified as Finite Element (FE) models of single sectors for analysis purposes to reduce computational costs. However, inherent blade mistuning breaks the cyclic symmetry, and often the full blisk must be modeled, which has millions of degrees of freedom (DOF), making it computationally too expensive. These simulations are often coupled with Monte Carlo simulations (MCS) and Latin hypercube for the probabilistic analysis of random mistuning, which requires a large sample set, further increasing the computational costs.Previous research and aeromechanical analysis used lumped mass and beam frame assembly models, which were very robust but had a low order of accuracy. This paved the way for developing FE-based Reduced Order Models (ROM). These high-fidelity complex models can capture the simplified nonlinearities in reduced-order models. The CMM (Component Mode Mistuning) and FMM (Fundamental Mode Mistuning) models were studied on the embedded stage of the Purdue 3-Stage axial compressor to understand the accuracy and usability of these methods for regions of interest.A brief comparison between the ROM models is made in this study. Although the FMM model is a simple, accurate model for determining the impact of mistuning on forced response when we have an isolated family of blade modes, the accuracy decreases considerably in cases with strong modal participation from other families. The more complex CMM model is required to study mistuned responses in veering regions, regions with high modal density, and instances of disk-dominated modes. The FMM model estimates the amplification well for mistuning cases with low deviations and high nodal diameters. The CMM model captures the intricate details of the response well and converges rapidly with the increasing number of tuned system modes. The modal participation in the veering regions was also captured reasonably well by CMM. The forced response for cases with small standard deviation was predicted well by both the reduced order models. The effect of the arrangement of the deviations was also explored, which showed significant amplification reduction.This study will guide the future to predict forced response incorporating frequency mistuning and aerodynamic coupling, which would be validated with the experimental data
Challenging the Notion of Role Models in Engineering Outreach Programs for Youth
Get PDFEngineering outreach programs often portray outreach educators as role models for youth. It is widely believed that introducing youth, especially girls, to potential engineering role models will broaden participation in engineering majors and careers. Based on interviews with and surveys of fourth- and fifth-grade girls participating in an engineering outreach program, we question whether youth are looking for career role models, and we challenge the assumption that youth will take up an adult as a role model simply because the adult is presented as such. We question what role these ‘‘models’’ play in the minds and lives of youth and argue that it may differ from what we expect. To be clear, we are not arguing that engineering role models are not important or not influential. Rather, we think it is important to gain a better understanding of how youth, particularly girls, view these potential engineering role models, which will allow us to optimize the significance of these adults to the youth participating in engineering outreach
Crowdsourcing/Winter Operations Dashboard Upgrade
Get PDFINDOT has recently completed the deployment of Parsons telematics-based dash-cameras, automatic vehicle locator (AVL) positions, and spreader rate monitoring across their winter operations fleet. The motivation of this study was to develop dashboards that integrate connected vehicle data into the real-time monitoring and after-action review of winter storms. Each month approximately 13 billion connected vehicle records are ingested for the state of Indiana and almost 99 billion weather data records are ingested nationwide in 15-minute intervals. This study developed techniques to utilize this connected vehicle data and weather data to monitor real-time mobility of interstates and post storm after-action assessments to identify improvement opportunities of winter operations activities. In multiple instances, these agile reviews have influenced operational changes in snow removal and maintenance around the state, leading to a marked improvement in observed mobility and safety
