Accurate remaining useful life (RUL) prediction is essential for the safe and cost-effective operation of safety-critical systems such as electronic components and engines. While data-driven machine learning approaches have demonstrated strong performance for RUL estimation, their effectiveness is limited by the lack of full run-to-failure data and by strict privacy and intellectual property constraints in industrial settings. Federated learning (FL) enables collaborative model training across multiple data owners without direct data sharing, but it does not, by itself, provide formal privacy guarantees and remains vulnerable to information leakage. This paper presents a privacy-preserving DP-enhanced FL setup for RUL prediction that combines federated learning with differential privacy (DP). We describe an end-to-end implementation based on the Opacus DP library, highlight practical challenges arising from the integration of DP into recurrent neural network architectures, and propose solutions to address them. Using two representative RUL datasets (CMAPSS and SiC MOSFET), we analyze the effect of DP noise on prediction performance and on the functional dependence between the predicted RUL and the already lived life feature. The results demonstrate that differential privacy can be integrated into federated RUL prediction with limited degradation in predictive performance, providing practical insights for deploying privacy-aware collaborative models in industrial environments

Predicting the remaining useful life of electronic components is a crucial aspect for predictive maintenance and system reliability across multiple fields and applications. Data-driven approaches, particularly those methods based on machine learning, are currently being used due to their ability to model complex degradation patterns without the need for explicit physical modeling. However, several challenges remain, including the availability and quality of data, as well as the uncertainty quantification of the results. To tackle these obstacles, this work explores the use of synthetic data for augmenting datasets, as well as feature selection and the assessment of different neural network architectures, including models with recurrent layers and probabilistic output. The proposed approach was evaluated on a silicon carbide metal-oxide-semiconductor field-effect transistor dataset. The best results for modeling the remaining useful life of these devices were obtained with a model trained on an augmented dataset that included synthetic data. This model’s probabilistic output allows building a confidence interval for the predictions, which is helpful to identify outliers. This model outperformed other state-of-the-art algorithms using only 4 out of 22 features, demonstrating the effectiveness of the feature selection procedure, the data augmentation method, and the neural network architecture for this case study.

Federated Learning (FL) is a decentralized training framework that allows participants to build shared models without exchanging raw data, effectively addressing privacy constraints in industrial use cases. FL has been increasingly investigated in power electronics for Remaining Useful Life (RUL) estimation to enhance predictive maintenance. This paper explores the key challenges associated with FL for RUL estimation, including data heterogeneity, communication constraints, and model aggregation complexities. Furthermore, we discuss future directions, such as advanced aggregation techniques, hybrid physics-informed models, and standardization of FL benchmarks to improve their applicability. Insights are drawn from recent research, including a case study on federated RUL estimation in power electronics [6].

To generate data used for developing schemes and models for CM, PHM, and for estimating RUL of power electronic devices, accelerated aging experiments in the form of power cycling are often performed. In these experiments, a set current is passed through the power devices and is turned on and off in regular cycles. Due to the mismatch in CTEs of the materials in the devices, the on/off cycles will generate thermally induced stress in the various material interfaces, which is the main cause of failures. Most of the power cycling setups that are currently used can only manage a single set on-state current level and fixed on/off times (which is also the common standard for lifetime testing); a condition that is very far from most real applications. The experimental setup described here is based on a Gamry Reference 3000AEpotentiostat/galvanostat/ZRA working with a Gamry 30k Booster, which can be programmed to generate a variable load current profile and will thus enable the application of more realistic conditions for accelerated aging of power electronic devices in the lab. This will improve prognostics model development and provide excellent use cases for evaluating the capabilities of the prognostics algorithms for generalization to field conditions. The application of variable load profiles from the field, instead of the regular on/off cycles traditionally used, is not compatible with the commonly used method of using the chip itself as a temperature sensor. Instead, we here present a novel method of estimating the junction temperature using a device specific derivation of thermal parameters from the measured cooling block temperature, case temperature, and dissipated power in conjunction with simulations using the PySpice simulation package implemented in Python. The setup coupled with the new junction temperature estimation is an important step in enabling predictive maintenance of power devices that is currently missing from the power electronics community. © 2024 IEEE.

Remaining useful life prediction models are a central aspect of developing modern and capable prognostics and health management systems. Recently, such models are increasingly data-driven and based on various machine learning techniques, in particular deep neural networks. Such models are notoriously “data hungry”, i.e., to get adequate performance of such models, a substantial amount of diverse training data is needed. However, in several domains in which one would like to deploy data-driven remaining useful life models, there is a lack of data or data are distributed among several actors. Often these actors, for various reasons, cannot share data among themselves. In this paper a method for collaborative training of remaining useful life models based on federated learning is presented. In this setting, actors do not need to share locally held secret data, only model updates. Model updates are aggregated by a central server, and subsequently sent back to each of the clients, until convergence. There are numerous strategies for aggregating clients’ model updates and in this paper two strategies will be explored: 1) federated averaging and 2) federated learning with personalization layers. Federated averaging is the common baseline federated learning strategy where the clients’ models are averaged by the central server to update the global model. Federated averaging has been shown to have a limited ability to deal with non-identically and independently distributed data. To mitigate this problem, federated learning with personalization layers, a strategy similar to federated averaging but where each client is allowed to append custom layers to their local model, is explored. The two federated learning strategies will be evaluated on two datasets: 1) run-to-failure trajectories from power cycling of silicon-carbide metal-oxide semiconductor field-effect transistors, and 2) C-MAPSS, a well-known simulated dataset of turbofan jet engines. Two neural network model architectures commonly used in remaining useful life prediction, long short-term memory with multi-layer perceptron feature extractors, and convolutional gated recurrent unit, will be used for the evaluation. It is shown that similar or better performance is achieved when using federated learning compared to when the model is only trained on local data.

Multi-purpose ports’ efficient and sustainable operation relies on seamless coordination and decision-making among multiple organizations. This paper underscores the critical importance of forecasting resource and infrastructure utilization for informed operational, tactical, and strategic decision-making. The proposed approach draws on digital twin technology to enable collaborative decision-making by modeling complex port environments to enable shared situational awareness among stakeholders. Illustrated through a collaborative project involving the RISE Research Institutes of Sweden, National University of Singapore, Grieg Connect, Umeå University, Kvarken Ports Umeå, and INAB, we propose a digital twin design to empower the port as a decision- maker in multi-organizational settings to proactively plan and optimize its utilization of present and future resources. 

Prognostics and health management (PHM) is an engineering discipline that aims to maintain system behaviour and function and ensure mission success, safety and effectiveness. Addressing the challenges in prognostics and health management for modern intelligent systems, especially automated driving systems, is complex due to the contextual nature of faults. This complexity necessitates a thorough understanding of spatial, and temporal conditions, and relationships within operational scenarios and life-cycle stages. This paper introduces a framework designed to automatically recognize driving scenarios in automated driving systems using graph neural networks (GNNs). The framework extracts relational data from image frames, constructing graph-based models and transforming unstructured sensory data into structured data with diverse node types and relationships. A specific graph neural network processes the graph model to reveal and detect operational conditions and relationships. The proposed framework is evaluated using the KITTI dataset, demonstrating superior performance compared to conventional feed-forward networks such as MLP, particularly in handling relational data. 

As deep learning methodologies progress, convolutional neural networks (CNNs) are seeing growing application in image recognition, especially within the domain of bearing fault diagnosis. Utilizing CNN for the automatic identification of features in vibration-related images can enhance both accuracy and efficiency in bearing fault recognition. However, the industry has developed a number of pre trained CNN models and makes it demanding to select models for specific tasks of bearing fault diagnosis. Therefore, this paper aims to compare three of the main-trend CNN models’ prediction accuracy and computational performance comprehensively. First, three pretrained CNN models-VGG16, ResNet and SqueezeNet, were trained to classify the bearing vibration signal dataset which had been converted to 2-D scalogram images. Transfer learning was applied to all models in this process. Then, the prediction accuracy and training time were examined and compared. Results showed that the accuracy of all CNN models were acceptable but SqueezeNet has the lowest runtime and highest accuracy. ResNet showed slightly lower performance and VGG16 experienced overfitting and produced the lowest accuracy with the longest runtime. Overall, for similar tasks of image-classification-based bearing fault diagnostics, SqueezeNet is a better candidate compared to other main stream pretrained CNN models, which can bring great accuracy accompanied with better efficiency. The findings of this work are good reference for future model selection in bearing fault diagnosis and related tasks.

In engineering, prognostics can be defined as the estimation of the remaining useful life of a system given current and past health conditions. This field has drawn attention from research, industry, and government as this kind of technology can help improve efficiency and lower the costs of maintenance in a variety of technical applications. An approach to prognostics that has gained increasing attention is the use of data-driven methods. These methods typically use pattern recognition and machine learning to estimate the residual life of equipment based on historical data. Despite their promising results, a major disadvantage is that it is difficult to interpret this kind of methodologies, that is, to understand why a certain prediction of remaining useful life was made at a certain point in time. Nevertheless, the interpretability of these models could facilitate the use of data-driven prognostics in different domains such as aeronautics, manufacturing, and energy, areas where certification is critical. To help address this issue, we use Local Interpretable Model-agnostic Explanations (LIME) from the field of eXplainable Artificial Intelligence (XAI) to analyze the prognostics of a Gated Recurrent Unit (GRU) on the C-MAPSS data. We select the GRU as this is a deep learning model that a) has an explicit temporal dimension and b) has shown promising results in the field of prognostics and c) is of simplified nature compared to other recurrent networks. Our results suggest that it is possible to infer the feature importance for the GRU both globally (for the entire model) and locally (for a given RUL prediction) with LIME. 

Residual stresses created during the packaging process can adversely affect the reliability of electronics components. We used incremental hole-drilling method, following the ASTM E 837-20 standard, to measure packaging induced residual stresses in discrete packages of power electronics components. For this purpose, we bonded a strain gauge on the surface of a Gallium Nitride (GaN) power component, drilled a hole through the thickness of the component in several incremental steps, recorded the relaxed strain data on the sample surface using the strain gauge, and finally calculated the residual stresses from the measured strain data. The recorded strains and the residual stresses are related by the compliance coefficients. For the hole drilling method in the isotropic materials, the compliance coefficients are calculated from the analytical solutions, and available in the ASTM standard. But for the orthotropic multilayered components typically found in microelectronics assemblies, numerical solutions are necessary. We developed a subroutine in ANSYS APDL to calculate the compliance coefficients of the hole drilling test in the molded and embedded power electronics components. This can extend the capability of the hole drilling method to determine residual stresses in more complex layered structures found in electronics.