WEPPE: Wireless Edge-Computing Personal Protective Equipment for Large-Scale Health Monitoring

该提案的研究目标是提供一种通用的、可扩展的边缘计算架构，以支持下一代个人防护设备（PPE）技术。传感器和无线传感器网络（WSNs）的激增导致大量数据生成，增加了中央数据中心的计算负担，造成数据传输瓶颈，并阻碍了实时决策过程。这些挑战是由于物联网设备在计算能力、内存和无线带宽（BW）分配方面的现有限制而出现的。选择个案研究作为开发这种系统的框架，是因为最近迫切需要更好地跟踪可传播疾病在大面积地区的传播。WEPPE项目将采取两阶段办法来应对上文提到的挑战。在第一阶段，该项目将研究一种低成本的可喷墨打印的非线性元件，并在柔性基板上开发一个机器学习平台，用于低级传感器数据处理或原位计算。在第二阶段，该项目将集成一个高效的基于模拟脉冲的数据编码和解码方案，将第一阶段处理后的传感器数据无线中继到数据中心，而无需扩展网络带宽。拟议的WEPPE项目预计将产生一个独特的机器学习框架，该框架取决于水库计算的基本原理，新型喷墨打印传感器和非线性元件，以及具有安全通信的无线数据遥测方案。定制的硬件和底层计算将实现原位边缘计算，同时保持高质量的数据抽象，以实现实时网络级或大数据处理，从而实现快速决策。教育目标是扩大女性、少数民族和非洲裔美国学生的参与，并为下一个工程挑战时代培养和教育他们。这个拟议的项目将研究如何通过基于硬件的机器学习和数据加密/解密方案的边缘计算可以有效地解决有限带宽、安全数据传输、高密度数据吞吐量和节能原位计算等物联网问题。该项目主要针对四个研究目标-（i）用于传感器网络分析的水库计算架构研究，（ii）用于传感和物理计算的喷墨打印设备研究，（iii）研究节能正交脉冲和多比特数据映射，以及（iv）基于正交模拟脉冲的数据压缩和解压缩研究。一个基于水库计算架构的机器学习平台，特别是回声状态网络（ESN），将研究其简单性，较少的训练时间，相对减少的训练数据量，以及易于部署。作为这项工作的一个组成部分，该项目还将研究喷墨打印的低成本非线性元件，这将是在柔性基板上开发机器学习平台的核心构建模块。储存器将生成状态向量，其是原始数据的超维度化加密表示，并且因此将提供数据压缩和安全性。故障检测和传感器融合将通过训练库和合并状态向量来进行。然后，来自储存器的状态向量将在“可穿戴集线器”中通过使用模拟正交脉冲（AOP）的k位编码方案被进一步加密和频谱压缩。在“本地服务器”，来自所有可穿戴集线器的编码AOP将通过n脉冲压缩技术进行压缩并传输到“数据中心”。“数据中心”的安全接收器将使用安全的读出神经元解码状态向量，提供预测，并将其发送回最终用户，以便通过深度学习和其他机器学习方法进行监控或大规模处理。该奖项反映了NSF的法定使命，并通过使用基金会的智力价值和更广泛的影响审查标准进行评估，被认为值得支持。

The research objective of this proposal is to provide a general-purpose, scalable edge-computing architecture critically needed to support the next generation of personal protective equipment (PPE) technology. The proliferation of sensors and wireless sensor networks (WSNs) results in high-volume data generation, increases the computational burden at the central data center, creates data transmission bottlenecks, and hinders the real-time decision-making process. These challenges arise due to the existing limits of IoT devices on computational power, memory, and wireless bandwidth (BW) allocation. The case study chosen as a framework for developing such a system is motivated by the recent and urgent need for better tracking of the spread of transmittable diseases over large areas. The WEPPE project will resort to two-phase approaches to address the challenges mentioned earlier. In the first phase, the project will investigate a low-cost inkjet-printable nonlinear-element and develop a machine-learning platform on a flexible substrate for low-level sensor data processing or in-situ computation. In the second phase, the project will integrate an efficient analog pulse-based data encoding and decoding scheme to wirelessly relay the processed sensor data from the first phase to a data center without requiring extended network bandwidth. The proposed WEPPE project is expected to produce a unique machine learning framework that hinges on the fundamentals of reservoir computing, novel inkjet-printed sensors and nonlinear elements, and wireless data telemetry scheme with secure communication. Customized hardware and low-level computing will enable in situ edge computing while maintaining quality data abstraction for real-time network-level or big data processing for rapid decision-making. The education goal is to broaden the participation of female, minority, and African-American students and train and educate them for the next era of engineering challenges.This proposed project will investigate how edge computing via hardware-based machine learning and data encryption/decryption schemes may effectively resolve the IoT problems of limited bandwidth, secure data transmission, high-density data throughput, and power-efficient in-situ computation. The project has targeted mainly four research goals - (i) Research on Reservoir Computing Architectures for Sensor Network Analysis, (ii) Research on Inkjet-Printed Devices for Sensing and Physical Computing, (iii) Investigate Energy-Efficient Orthogonal Pulses and Multi-bit Data Mapping, and (iv) Research on Orthogonal Analog Pulse Based Data Compression and Decompression. A reservoir computing architecture-based machine learning platform, especially the Echo State Network (ESN), will be investigated for its simplicity, less training time with relatively reduced training data volume, and ease of deployment. As an integral part of this effort, the project will also investigate an inkjet-printed low-cost nonlinear element, which will be a core building block for developing a machine-learning platform on a flexible substrate. The reservoir will generate a state vector, which is a hyper-dimensionalized encrypted representation of the raw data, and as a result, will provide data compression and security. Fault detection and sensor fusion will occur by training the reservoir and merging the state vectors. The state vectors from the reservoirs will then be further encrypted and spectrally compressed in the "Wearable Hub" by a k-bit encoding scheme using analog orthogonal pulses (AOP). At the "Local Server," the encoded AOPs from all the wearable hubs will be compressed by an n-pulse compression technique and transmitted to the "Data Center." The secured receiver at the "Data Center" will decode the state vectors using secured read-out neurons, providing predictions to be sent back to the end users for monitoring or large-scale processing by deep learning and other machine learning methods.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.