Mid Sweden University

Enabling broad use of electromagnetic energy harvesting in wireless sensor screws requires robust systems that work with weak vibrations and varying frequency profiles. This contribution presents an energy management circuit incorporating two cooperating DC-DC converters controlled by self-powered MOSFET switches and a passive voltage multiplier enabling low-voltage start-up. The circuit operates effectively over an acceleration range of 0.07-0.21 g. It consistently harvests energy across a wider frequency range than energy management circuits based on single DC-DC converters. For example, at an acceleration of 0.21 g, the frequency range is 20–30 Hz. Thereby it realizes, e.g. at 25 Hz, an efficiency of 72%. The proposed circuit enables robust energy harvesting in a wide frequency range, supporting wireless sensor operation even under low-vibration conditions typical of industrial predictive maintenance. 

Keyword spotting (KWS) is a key component of smart devices, enabling efficient and intuitive audio interaction. However, standard KWS systems deployed on embedded devices often suffer performance degradation under real-world operating conditions. Resilient KWS systems address this issue by enabling dynamic adaptation, with applications such as adding or replacing keywords, adjusting to specific users, and improving noise robustness. However, deploying resilient, standalone KWS systems with low latency on resource-constrained devices remains challenging due to limited memory and computational resources. This study proposes a low computational approach for continuous noise adaptation of pretrained neural networks used for KWS classification, requiring only 1-shot learning and one epoch. The proposed method was assessed using two pretrained models and three real-world noise sources at signal-to-noise ratios (SNRs) ranging from 24 to -3 dB. The adapted models consistently outperformed the pretrained models across all scenarios, especially at SNR≤18 dB, achieving accuracy improvements of 4.9% to 46.0%. These results highlight the efficacy of the proposed methodology while being lightweight enough for deployment on resource-constrained devices.

Human Activity Recognition using wearable sensors has applications in health monitoring, entertainment, and industrial settings. However, the performance of Human Activity Recognition models in real-life settings is usually lower than in laboratory settings due to the reduced quantity and quality of the sensors available in the former. Here, we propose using a suitable shared representation space to incorporate the information of additional sensors available during training time to address these limitations. We evaluate two representation spaces: one created using Feature Agglomeration and the other using Uniform Manifold Approximation and Projection (UMAP) under three conditions to evaluate their performance and robustness to noise: clean data, Gaussian noise, and Magnitude Warping noise using three datasets: Opportunity, Cooking, and PAMAP2. Our results consistently show that the representation spaces enhances performance relative to the conventional single-sensor method. The UMAP approach outperforms Feature Agglomeration, achieving up to a 14% improvement in the F1-Score metric when using clean data. In the presence of Gaussian noise, the UMAP representation space not only improves classification performance but also exhibits resilience to noise in the Opportunity and PAMAP2 datasets. While the UMAP method exhibits lower robustness to noise in the Cooking dataset, it still achieves the highest performance. When experimenting with Magnitude Warping noise, the UMAP representation space shows varying levels of robustness across datasets but still enhances performance to some extent. Using shared representations, we leverage the higher number and quality of sensors available in laboratory settings for training HAR models, while releasing the usual requirement of using the same number of sensors at the final deployment. 

The mining sector harnesses advancements in automation, digitalization, and interconnected technologies from Industry 4.0 to enhance efficiency, safety, and sustainability. Conveyor belts play a critical role in mining operations, facilitating the continuous and efficient transport of bulk materials over long distances, directly impacting productivity. While anomaly detection in specific conveyor belt components has been extensively studied, continuous monitoring to identify root causes of failures remains in its early stages. Existing methods for anomaly detection in mining conveyor belt duty cycles rely on supervised machine learning (ML) to classify internal machine modes as an intermediate step. While these approaches offer high explainability, they are constrained by the need for extensive labeled data for internal machine modes. This study proposes a novel pattern recognition approach combining unsupervised and supervised ML models for real-time anomaly detection in conveyor belt operational cycles. By evaluating combinations of TinyML models, the approach achieved average F1-scores of 83.2% for abnormal cycles and 97.0% for normal cycles, surpassing the state-of-the-art by 11.4% and 3.3%, respectively. Deployed on low-power microcontrollers, the proposed methods demonstrated efficient, real-time operation, reducing energy consumption by up to 84.5% (4.1 μJ per inference) and program memory usage by up to 72.1%. These results provide valuable insights for detecting early mechanical failures and enabling targeted preventive maintenance. 

Energy harvesting enables the supply of low-power embedded sensor systems that perform important tasks, such as condition monitoring. In applications with rotating elements, like bearings, variable reluctance energy harvesters (VREHs) can provide a significant amount of power to the system; however, a power conditioning system is needed to extract the maximum power, manage an energy storage element, and regulate the output voltage. Previous works focus either on the optimum design of the harvester that maximizes the output power level with given space constraints or concentrate on the optimum design of the power conditioning system for a fixed harvester design. This work proposes co-designing both elements to optimize the power delivered to the energy storage element. The results show that the main variable connecting both systems is voltage. On the harvester side, it is demonstrated that different designs can provide different voltage levels while maintaining the maximum output power, while on the power conditioning side, a trade-off between rectification and conversion losses must be considered. This analysis is presented with simulations and validated with experimental results. Finally, the limitations of commercial power management units (PMUs) are exposed.

Keyword spotting (KWS) is an essential function that enables interaction with ubiquitous smart devices. However, in resource-limited devices, KWS models are often static and can thus not adapt to new scenarios, such as added keywords. To overcome this problem, we propose a Continual Learning (CL) approach for KWS built on Binary Neural Networks (BNNs). The framework leverages the reduced computation and memory requirements of BNNs while incorporating techniques that enable the seamless integration of new keywords overtime. This study evaluates seven CL techniques on a 16-classuse case, reporting an accuracy exceeding 95% for a single additional keyword and up to 86% for four additional classes. Sensitivity to the amount of training samples in the CL phase, and differences in computational complexities are being evaluated. These evaluations demonstrate that batch-based algorithms are more sensitive to the CL dataset size, and that differences between the computational complexities are insignificant. These findings highlight the potential of developing an effective and computationally efficient technique for continuously integrating new keywords in KWS applications that is compatible with resource-constrained devices.

Efficient firmware distributions in energy-harvesting (EH) LoRa networks require that EH LoRa sensors simultaneously receive data fragments from a server without facing power failures. This requirement is difficult to satisfy due to the impact of EH rates and LoRa transmission parameters on the efficiency of firmware distributions. We present FioRa+, a novel energy neutrality-aware multicast firmware distribution framework for EH LoRa networks. It gradually distributes a firmware image to EH LoRa sensors in an energy-neutral manner according to their future energy availability predicted using embedded machine learning models. Consequently, the need for additional firmware distributions caused by unsuccessful firmware image reconstructions is reduced. Through one-hop neighbor discovery, on-demand relay, flexible energy query, and coverage assessment mechanisms, FioRa+ ensures that all EH LoRa sensors can receive data fragments from the server at the scheduled time using high data rates. Equipped with a relay scheduling algorithm, it circumvents the collision of data fragments relayed by EH LoRa sensors using identical data rates. The experimental results show that FioRa+ renders up to 113 × shorter distribution time and 22.7 × less distribution overhead than the state of the art.

Wireless condition monitoring of critical rotating components, such as shafts, bearings, and gears, can be efficiently powered by energy harvesting technologies, supporting improved operation and maintenance. However, conventional energy harvesters struggle to meet the energy demands of wireless condition monitoring operating at low rotational speeds, severely limiting their practical deployment in industrial applications. To address this challenge, this paper proposes a new Halbach-enhanced variable reluctance energy harvester (HE-VREH) to improve the power density under low-speed conditions. A theoretical model based on magnetic circuit analysis is established to predict the voltage response of the proposed HE-VREH, while finite element simulations are performed to analyze the magnetic flux distribution and benchmark its performance against existing variable reluctance harvesters. Moreover, experimental measurements validate the theoretical model in predicting the output performance. The experimental results indicate that the proposed HE-VREH generates a power of 210.32 mW at a rotational speed of 55 rpm, achieving a normalized power density of 1.83 mW/(cm3∙Hz2). Furthermore, an autonomous wireless sensing system powered by the HE-VREH is able to successfully capture vibration and torque signals during rotation. The combination of detailed modelling and experimental validation confirms the potential of the novel HE-VREH to enable self-powered condition monitoring in industrial rotating machinery, thereby bridging the critical gap between energy harvesting efficiency and low-speed operational requirements. 

Near-sensor diagnosis has become increasingly prevalent in industry. This study proposes a lightweight model named LD-RPMNet that integrates Transformers and Convolutional Neural Networks, leveraging both local and global feature extraction to optimize computational efficiency for apractical railway application. The LD-RPMNet introduces a Multi-scale Depthwise Separable Convolution (MDSC) module, which decomposes cross-channel convolutions into pointwise and depthwise convolutions while employing multi-scale kernels to enhance feature extraction. Meanwhile, a Broadcast Self Attention (BSA) mechanism is incorporated to simplify complex matrix multiplications and improve computational efficiency. Experimental results based on collected sound signals during the operation of railway point machines demonstrate that the optimized model reduces parameter count and computational complexity by 50% while improving diagnostic accuracy by nearly 3%, ultimately achieving an accuracy of 98.86%. This demonstrates the possibility of near-sensor fault diagnosis applications in railway point machines.

Accurate localization of acoustic emission (AE) sources in composite materials remains a significant challenge due to the material’s anisotropy, complex wave propagation paths, and environmental noise. This study proposes a Long Short-Term Memory-optimized Time Difference Mapping method (LSTM-TDM) to address these challenges. By leveraging deep learning to analyze the time difference features of AE signal propagation, the proposed method significantly enhances the localization accuracy of the TDM method. Comparative experiments were conducted using three methods: the TDM method, the General Regression Neural Network optimized TDM method (GRNN-TDM), and the proposed LSTM-TDM method. The results demonstrate that the GRN-NTDM method improves localization accuracy by 15.38% compared to the TDM method, while the LSTM-TDM method achieves the best performance with an average error of 14.26 mm, a 50.63% reduction in error relative to the TDM method.The findings indicate that the LSTM-TDM method offers significant advantages in AE source localization for composite materials, providing critical insights and potential applications for structural health monitoring in such materials.