The range of operations of electric vehicles (EVs) is a critical aspect that may affect the user's attitude toward them. For manned EVs, range anxiety is still perceived as a major issue and recent surveys have shown that one-third of potential European users are deterred by this problem when considering the move to an EV. A similar consideration applies to aerial EVs for commercial use, where a careful planning of the flying range is essential not only to guarantee the service but also to avoid the loss of the EVs due to charge depletion during the flight. Therefore, route planning for EVs for different purposes (range estimation, route optimization) and/or application scenarios (terrestrial, aerial EVs) is an essential element to foster the acceptance of EVs as a replacement of traditional vehicles. One essential element to enable such accurate planning is an accurate model of the actual power consumption. While very elaborate models for the electrical motors of EVs do exist, the motor power does not perfectly match the power drawn from the battery because of battery non-idealities. In this paper, we propose a general methodology that allows to predict and/or optimize the operation range of EVs, by allowing different accuracy/complexity tradeoffs for the models describing the route, the vehicle, and the battery, and taking into account the decoupling between motor and battery power. We demonstrate our method on two use cases. The first one is a traditional driving range prediction for a terrestrial EV; the second one concerns an unmanned aerial vehicle, for which the methodology will be used to determine the energy-optimal flying speed for a set of parcel delivery tasks.

Neural Architecture Search (NAS) is quickly becoming the go-to approach to optimize the structure of Deep Learning (DL) models for complex tasks such as Image Classification or Object Detection. However, many other relevant applications of DL, especially at the edge, are based on time-series processing and require models with unique features, for which NAS is less explored. This work focuses in particular on Temporal Convolutional Networks (TCNs), a convolutional model for time-series processing that has recently emerged as a promising alternative to more complex recurrent architectures. We propose the first NAS tool that explicitly targets the optimization of the most peculiar architectural parameters of TCNs, namely dilation, receptive-field and number of features in each layer. The proposed approach searches for networks that offer good trade-offs between accuracy and number of parameters/operations, enabling an efficient deployment on embedded platforms. Moreover, its fundamental feature is that of being lightweight in terms of search complexity, making it usable even with limited hardware resources. We test the proposed NAS on four real-world, edge-relevant tasks, involving audio and bio-signals: (i) PPG-based Heart-Rate Monitoring, (ii) ECG-based Arrythmia Detection, (iii) sEMG-based Hand-Gesture Recognition, and (iv) Keyword Spotting.Results show that, starting from a single seed network, our method is capable of obtaining a rich collection of Pareto optimal architectures, among which we obtain models with the same accuracy as the seed, and 15.9-152× fewer parameters. Moreover, the NAS finds solutions that Pareto-dominate state-of-the-arthand-tuned models for 3 out of the 4 benchmarks, and are Pareto-optimal on the fourth (sEMG). Compared to three state-of-the-art NAS tools, ProxylessNAS, MorphNet and FBNetV2, our method explores a larger search space for TCNs (up to 1012×) and obtains superior solutions, while requiring low GPU memory and search time. We deploy our NAS outputs on two distinct edge devices, the multicore GreenWaves Technology GAP8 IoT processor and the single-core STMicroelectronics STM32H7 microcontroller. With respect to the state-of-the-art hand-tuned models, we reduce latency and energy of up to 5.5× and 3.8× on the two targets respectively, without any accuracy loss.

The market of small drones has been recently increasing due to their use in many fields of application. The most popular drones are multirotors, in particular quadcopters. They are usually supplied with batteries of limited capacity, and for this reason their total flight time is also limited.As a consequence of the non linear characteristics of batteries, estimation of the real flight time may become an issue, since most battery models do not include all the non idealities. Consequently, applications such as delivery service, search and rescue, surveillance might not be accomplished correctly because of inaccurate energy estimations.This paper describes a battery-aware model for an accurate analysis of the drone energy consumption; this model is then applied to a scenario of drone delivery. Results show an accuracy greater of about 16% with respect to the traditional estimation model.

Metal Additive Manufacturing (AM) is a pillar of the Industry 4.0, with many attractive advantages compared to traditional subtractive fabrication technologies. However, there are many quality issues that can be an obstacle for mass production. The in-situ camera-based monitoring and detection of defects, taking advantage of the layer-by-layer nature of the build, can be an effective solution to this problem. In this context, the use of Computer Vision and Machine Learning algorithms have a very important role. Nonetheless, they are up to this date limited by the scarcity of data for the training, as well as by the difficulty of accessing and integrating the AM process data throughout the fabrication. To tackle this problem, this paper proposes a system for in-situ monitoring that analyses images from an off-axis camera mounted on top of the machine to detect the arising defects in real-time, with automated generation of synthetic images based on Generative Adversarial Network (GAN) for dataset augmentation purposes. The computing functionalities are embedded into a holistic distributed AM platform allowing the collection, integration and storage of data at all stages of the AM pipeline.

The global shift towards electric vehicles (EVs) represents a critical strategy for fighting climate change and reducing both reliance on fossil fuels and $CO_{2}$ emissions. In this scenario, battery management systems (BMS) become crucial in EVs to ensure battery safety, reliability, and efficiency. Recently, data-driven estimation techniques have been proposed to estimate battery-related metrics, such as remaining capacity or aging condition. These techniques emerged as an answer to batteries' intrinsic variability, but they rely primarily on full charge/discharge cycles, overlooking the nuances of partial charging, which is prevalent in real-world usage. This paper presents a novel BMS architecture for EVs, leveraging digital twin technology and data-driven modeling to address these challenges. The proposed dynamic dual-model approach seamlessly integrates real-time monitoring with cloud-based analytics to continuously evaluate and predict battery capacity degradation. Key innovations include sophisticated feature engineering and segmentation strategies, which enable precise state of health (SoH) estimation across a wide range of driving conditions. Additionally, the architecture incorporates a dynamically retraining State of Charge/Energy (SoC/SoE) model that adapts to battery aging, thereby maintaining high accuracy throughout the battery's life cycle. Extensive validation using datasets from public institutions demonstrates the effectiveness and robustness of the proposed system.