Enabling Fast Deep Learning on Tiny Energy-Harvesting IoT Devices

Internet-of-Things (IoT) devices are becoming both intelligent and green. On the one hand, Deep Neural Network (DNN) compression techniques make it possible to run deep learning applications on devices equipped with low-end microcontrollers (MCUs). By performing deep learning on IoT devices, in-situ decision-making can be made, which can improve the responsiveness of such devices to the environment and reduce data uploading to edge servers or clouds to save valuable network bandwidth. On the other hand, many IoT devices in the future will be powered by energy harvesters instead of batteries to reduce environmental pollution and achieve permanent service free of battery maintenance. As the energy output of energy harvesters is tiny and unstable, energy harvesting IoT (EH-IoT) devices will experience frequent power failures during their execution, making the software task hard to progress. The deep learning tasks running on such devices must face this challenge and, at the same time, ensure satisfactory execution efficiency. We believe deploying deep learning on EH-IoT devices that execute intermittently will be a challenging yet promising research direction. To motivate research in this direction, this paper summarizes existing solutions and provides an in-depth discussion of future challenges that deserve further investigation. With IoT devices becoming more intelligent and green, DNN inference on EH-IoT devices will generate a much more significant impact in the future in academia and industry.

Water distribution networks (WDNs) are a critical component of modern society, responsible for delivering water to local communities and industries while maintaining continuous, stable flow within the system, which carries low-grade hydropower potential. These systems are vulnerable to disruptive events, such as pipeline failures and water contamination. To mitigate the risk of disruption events, recent EU policies have focused on the digitalisation of critical infrastructure, including WDNs and wastewater treatment plants (WWTPs), through the integration of Internet of Things (IoT) technologies. Digitalisation enables system operators to collect real-time data and perform predictive maintenance, thereby improving resilience and operational efficiency. Thus, system electricity requirements will increase to support IoT and digitalisation retrofitting.This creates an opportunity to deploy energy-harvesting (EH) devices within WDNs and WWTP infrastructure that harness and utilise the resonant and kinetic energy inherent in water flow. Although the benefits of EH devices are often considered minimal and their energy outputs are low, EH devices are considered a viable power source for low-power sensors, such as pressure, contamination, and temperature sensors.This study proposes and applies a feasibility assessment framework to evaluate EH devices as an alternative, decentralised power source for IoT-enabled monitoring in WDNs. The framework includes technical performance analysis, economic viability, and environmental impact assessments, combined with probability-based resilience modelling. It uses case-specific hydraulic, operational, design, and economic data to quantify EH power outputs, assess the costs of design and deployment, evaluate the environmental footprint of the devices, determine sensor energy requirements, and assess system resilience across different deployment scenarios.Results demonstrate that the EH device has a mechanical power-generation capacity potential ranging from 0.049 to 36 watts, depending on the study location, flow conditions, and design characteristics. This EH generation capacity is sufficient to power pressure, contamination, and temperature sensors. The resilience modelling indicates that the detection probability of WDNs exhibits the most significant gains from adding sensors at low deployment levels. Thus, most of the improvement in detection levels occurs between deploying the first five to ten sensors; beyond this threshold, adding more sensors exhibits diminishing marginal returns to detection probability and system resilience. In addition, adding more sensors can significantly reduce system resilience by increasing power requirements, thereby placing excessive load on the power supply of the EH devices. Thus, it increases the risk of failure rather than enhancing resilience.Overall, the results underscore the importance of strategic sensor allocation over high-density deployment and of balancing monitoring coverage with energy availability. Furthermore, the results show that EH devices are suitable power-generation technologies for supporting the digitalisation of WDNs and informing the design of new monitoring systems and sensor placement to enhance system reliability, enable cost-effective monitoring, and maximise mitigation benefits. In addition, the proposed framework provides decision-makers with a structured approach to assessing EH integration applications for digitalised WDNs, focusing on enhancing resilience through monitoring, conducting technical performance and cost-benefit analyses, assessing environmental trade-offs, and designing monitoring strategies aligned with sustainability and resilience objectives.

AlN piezoelectric thin films for energy harvesting and acoustic devices

Energy harvesting from structure vibration, human motion or environmental source has been the focus of researchers in the past few decades. This paper proposes a novel design that is suitable to harvest energy from human motions such as dancing or physical exercise and use the device to engage young students in Science, Technology, Engineering and Math (STEM) fields and outreach activities. The energy harvester (EH) device was designed for a dominant human operational frequency range of 1–5 Hz and it can be wearable by human. We proposed to incorporate different genres of music coupled with energy harvesting technologies for motivation and energy generation. Students will learn both science and art together, since the energy harvesting requires understanding basic physical phenomena and the art enables various physical movements that imparts the largest motion transfer to the EH device. Therefore, the systems are coupled to each other. Young people follow music updates more than robotics or energy harvesting researches. Most popular videos on YouTube and VEVO are viewed more than 100 million times. Perhaps, integrating the energy harvesting research with music or physical exercise might enhance students’ engagement in science, and needs investigation. A multimodal energy harvester consisting of piezoelectric and electromagnetic subsystems, which can be wearable in the leg, is proposed in this study. Three piezoelectric cantilever beams having permanent magnets at the ends are connected to a base through a slip ring. Stationary electromagnetic coils are installed in the base and connected in series. Whenever the device is driven by any oscillation parallel to the base, the unbalanced rotor will rotate generating energy across the stationary coils in the base. In another case, if the device is driven by an oscillation perpendicular to the base, a stress will be induced within the cantilever beams generating energy across the piezoelectric materials.

One of the major concerns in using IoT devices is life of the battery, especially in remote places where power resources are limited. By employing certain low powered sensors, the IoT devices with higher power consumption can be activated only when required. In this paper, a methodology is proposed for reducing the consumption of power in an IoT device which sends data from a RFID tag to an output file. The consumption is reduced by implementing sleep mode when, the device is not required to be active, and a wakeup signal and studying the impact it has on the power consumption. An energy harvesting device is also integrated to make the device self-sustainable.

In this era of the Internet of Things (IoT), the increasing number of IoT devices benefit from energy harvesting (EH) technology which enables a sustainable data acquisition process, including data sensing, communication, and storing for promoting the well beings of the society. However, intermittent and low EH power confines the data acquisition process. Specifically, due to frequent harvesting power outages and depletion of energy for expensive data transmission to the IoT edge server, insufficient energy is allocated for data sensing resulting in the missing of key information. To address this issue, this article proposes a sparsity-aware spatiotemporal data sensing framework for EH IoT devices to minimize the data sensing rate/energy while acquiring comprehensive information and reserving sufficient energy. In this framework, the IoT devices sample critical sparse spatiotemporal data, and then the sparse data are sent to the edge server for reconstruction. To maximize the reconstruction accuracy subject to the limited power supply and intermittent work patterns of EH devices, we first propose the QR-based algorithm QR-ST to initiate a sensing scheduling for each EH device. Due to the unstable and intermittent work pattern, the schedule needs to be dynamically fine-tuned based on environmental inputs. Therefore, we further propose a multiagent deep reinforcement learning-based method named S-Agents for the IoT edge server to globally select the sensing devices at each time slot, where the spatial and temporal features of reconstructed data are guaranteed. Experimental results show that the proposed framework reduced the reconstruction error by 66.30% compared with baselines.

An indoor photovoltaic energy harvesting device technology has been developed that achieves high efficiencies and high power per area at low light levels. A 50 mm x 20 mm energy harvesting device comprising a compound semiconductor material is demonstrated. At 200 lux white LED illumination (CCT 2,831 K), its operating power at maximum power point is 170.0 μW with Voc = 4.61 V and Isc = 43.8 μA, corresponding to an active area power density of 21.8 μW/cm2. Based on this high-efficiency energy harvesting device, an indoor photovoltaic (PV) energy harvesting power module (IPEHPM) has been developed for autonomously powering IoT sensor nodes and other devices which ordinarily would require disposable or user-rechargeable batteries to operate. The IPEHPM can deliver 3.6 ~ 4.2 V output voltage, with 100 mA pulse current for up to 600 ms. An average energy harvester to storage efficiency of 88.7% has been demonstrated at 200 lux. We demonstrate that the IPEHPM is suitable for perpetually powering a wireless building automation CO2 sensor node at 200 lux, with CO2 measurement and data transmission every 150 s. Applications of high-efficiency energy harvesting devices in wearables and in wireless beacons are also discussed.

Joint beamforming and phase-shifting design for energy efficiency in RIS-assisted MISO communication with statistical CSI

A low-cost printed organic thermoelectric generator for low-temperature energy harvesting

In this paper, a design for an energy harvesting device using cantilever beam will be investigated and experimental results will be presented to validate the design. The energy harvesting device in the study is 31-unimorph piezoelectric which was used to convert small amplitude mechanical vibration from a specific machine application into an electrical energy source that could be used for electronic devices with low power requirements. The primary purpose of the design is to illustrate a method to design a cantilever beam that is optimized for attached position of piezoelectric by Experiment and FEM. From the given vibration data a range of frequencies where the energy harvesting device will generate the greatest amount of energy is determined. The device is then designed specifically targeting that frequency range with sinusoidal wave about resonant frequency. And results of this study show the change trend of output voltage according to changing circuit elements. This approach is presented as part of a more general approach to designing energy harvesters for any application. Also, it will be shown how attached position of piezoelectric used for cantilever beam were chosen.

Energy harvesting devices which convert low frequency mechanical energy sources such as human motions and ocean waves into electricity are attractive for powering portable devices and for green‐energy generation. To date the state‐of‐the‐art mechanical energy harvesting devices can only work efficiently at high vibration frequencies. Here, a biocompatible and flexible mechanical energy harvesting device is reported utilizing ionic diode as the transducer. This device utilizes the redistribution of cations and anions at the two hydrogel electrodes under stress to convert mechanical energy to electricity. It is shown that the device can be operated at low frequencies with high output current, e.g., 13.5 µA cm−2, owing to the high ion concentration and unique working mechanism of the device. Moreover, the output current density and power density can be improved further by employing a multilayer configuration. By stacking five units with parallel structure, the hydrogel diode device can generate an output current of 64.3 µA cm−2 and power density of 0.48 µW cm−2. Considering the very high electric energy density of ionic devices, the hydrogel energy harvesting device demonstrated herein paves a way for efficient mechanical energy harvesting from many common low frequency sources.

Vibration energy harvesting is increasing in popularity due to potential applications such as powering wireless sensors and ultra low power devices. For efficient energy harvesting, matching the device frequency to the source frequency is a major design requirement. Since mechanical vibrations differ in characteristics (frequency and acceleration amplitude), it is difficult to design an individual energy harvesting device for every source. Recently, several groups have pursued techniques to tune the resonance frequency of the vibrating structure through active and passive methods. In this paper, work has been done to attain a self-tunable energy harvesting device, which utilizes a magnetic force resonance frequency tuning technique to tune the device. The device is successfully tuned with in a bandwidth of ± 27% of its untuned resonance frequency, considering root mean square of the peak power output as the cutoff for frequency bandwidth. Since the technique is semi-active, energy is only consumed to tune the resonance frequency and is not required to remain at that specific frequency. The device consists of a piezoelectric cantilever beam array which is displaced to the desired distance to induce magnetic stiffness and to match the source frequency using a DC motor. The device has a power output of approximately 0.7 mW to 1 mW in the designed cutoff frequency range. The amount of energy consumed by the actuator to displace the beam is approximately 3.5 W to 4.5 W, which requires approximately 150 minutes to reclaim the expended energy.

In wind turbine tower (WTT) bolt health monitoring, traditional energy harvesting devices have low efficiency, insufficient energy conversion efficiency, high energy consumption of monitoring systems, and insufficient system integration and optimization. This paper proposed a new energy self-powered technology to achieve long-term and continuous health monitoring of WTT. It designed an efficient energy harvesting device that used natural energy such as wind turbine tower vibration and temperature change to achieve energy harvesting, optimized materials and structures, improved energy harvesting efficiency, and ensured a long-term stable supply of energy sources. The paper used an efficient energy conversion module to convert the collected mechanical energy and thermal energy into electrical energy, and uses high-capacity energy storage devices for storage, so that the system can operate stably without an external power supply. Energy self-power supply technology can be effectively integrated with the health monitoring system, and low-power sensors and wireless data transmission modules can be designed to reduce the system’s overall energy consumption and ensure the real-time collection and transmission of health monitoring data. Intelligent scheduling algorithms optimize energy usage strategies so that energy resources can be reasonably allocated, avoiding excessive consumption of limited energy by the system and realizing long-term stable operation of the WTT monitoring system. The experimental results show that the energy conversion rate of the designed energy harvesting device is 81.7% at a frequency of 100Hz and 79.4% at a high temperature of 100°C, showing obvious advantages in efficiency under vibration frequency and temperature changes. The power consumption of the wireless data transmission module of the self-powered system increases from 15 kW in 6 hours to 22 kW in 24 hours, and the power consumption of the energy management module increases from 12 kW to 18 kW. The power consumption remains low, which is suitable for long-term monitoring tasks.

This paper deals with a unique principle of energy harvesting technologies. An energy harvesting device generates electric energy from its surroundings using some kind of energy conversion method. Therefore, the considered energy harvesting device does not consume any fuel or substance. The presented energy harvesting system is used forenergy harvesting of electrical energy from mechanical shocks. The presented energy harvesting system uses a very sensitive vibration energy harvester, which was developed for an aeronautic application at Brno University of Technology. This energy harvesting system is a complex mechatronic device, which consists of a precise mechanical part, an electromagnetic converter, power electronics (power management) and a load (e.g., wireless sensor). The very sensitive vibration energy harvester is capable of usingthe mechanical energy of mechanical shocks and it can harvest useful energy. This energy harvesting system is used with a wireless temperature sensor and measured results are presented in this paper.

With the rapid development of the Internet of Things (IoT) and communication technology, deep neural network (DNN) applications, such as medical imaging, speech transcription, handwritten text recognition, have been widely used in IoT devices. However, due to resource constraints on these devices, e.g., limited memory capacity, weak computing capacity, and low battery capacity, IoT devices cannot support complicated DNN operation effectively and, thus, fail to fulfill the requirements of Quality of Service of mobile users. One promising approach is to migrate the DNN model to a remote cloud server to reduce the computing burden on IoT devices. Unfortunately, it still suffers from high delay and low bandwidth when communicating with cloud servers. Although the transmission delay of the edge server is low, its computing capacity lacks scalability and elasticity.To make matters worse, in the real world, the wireless connection between IoT devices and the cloud is intermittent, which can cause offloading failures during large-scale DNN data transmission. In this article, we describe a DNN model migration framework to overcome the abovementioned challenges, which consists of three parts: DNN model preprocessing, partition-offloading plan, and partition-uploading plan. Accordingly, we introduce the operation of the DNN migration and the available methods for each part. In addition, we improve the DNN partition-uploading plan in a multiuser edge-cloud collaborative computing environment. Finally, we highlight the important challenges of achieving more efficient DNN migration and point out the unresolved issues of DNN migration, which may shed light on future research directions.