https://www.mdu.se/

In response to increasing demands for reliability and uptime, organizations are progressively monitoring more of their mission-critical assets through various sensing and data collection devices. The accumulated data enables several emerging technologies, particularly data-driven approaches such as machine learning, which are becoming more viable in industrial contexts. These technologies have the potential to enhance the effectiveness and efficiency of asset management and maintenance. A key framework for realizing this potential is prognostics and health management, an engineering approach that deals with the identification and prognostication of system degradation. A major aspect of prognostics and health management is remaining useful life prediction, which develops models to forecast the remaining operational time of systems. Accurate prediction of future system state provides useful insight that aids maintenance planning. This thesis addresses challenges and aspects of data-driven remaining useful life prediction with a focus on deep learning-based approaches. The research proposes solutions to key challenges in remaining useful life prediction, including limited access to complete run-to-failure trajectories, data sharing constraints, and decentralized training requirements. Additionally, this thesis investigates remaining useful life predictions for discrete power electronics, components used in safety-critical high-power applications such as automotive systems -- an area that remains understudied within prognostics and health management. The findings demonstrate that remaining useful life prediction is a viable technology in these domains, with models benefiting from self-supervised pretraining and decentralized training through federated learning. Furthermore, the research establishes that discrete power electronics can be effectively monitored using data-driven remaining useful life prediction methods.

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 shortterm 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.

Industrial heating significantly contributes to global greenhouse gas emissions, accounting for a substantial portion of annual emissions. The transition to fossil-free operations in the heating industry is closely linked to advancements in industrial electrical heating systems, especially those using resistance heating wires. In this context, Prognostics and Health Management is crucial for enhancing system reliability and sustainability through predictive maintenance strategies. The integration of machine learning technologies into Prognostics and Health Management has significantly improved the precision and applicability of Remaining Useful Life modeling. This improvement enables more accurate predictions of component lifespans, optimizes maintenance schedules, and enhances operational efficiency in industrial heating applications. These developments are essential for reducing greenhouse gas emissions in the sector. This paper serves as a guide for conducting Remaining Useful Life modeling for industrial batch processes. It evaluates and compares two methodologies: deep learning-based approaches using full time-series data, such as recurrent neural networks and their variants, and feature-engineering-based methods, including random forest regression and support vector machines. Our results show that the feature-oriented approach performs better overall in terms of predictive accuracy and computational efficiency. The study includes a detailed sensitivity analysis and hyperparameter estimation for each method, providing valuable insights into developing robust and transparent Prognostics and Health Management systems. These systems are crucial in supporting the heating industry’s move towards more sustainable and emission-free operations. The findings reveal that feature-oriented methods are both performant and robust, particularly excelling in handling outliers. The random forest regression model, in particular, demonstrated the highest performance on the test dataset according to the chosen evaluation metrics. Conversely, trajectory-oriented methods exhibited less bias across varying levels of degradation, a helpful characteristic for Prognostics and Health Management systems. While feature-oriented methods tend to systematically underestimate Remaining Useful Life at high true values and overestimate it at low actual values, this issue is less pronounced in trajectory-oriented models. Overall, these insights highlight the strengths and limitations of each approach, guiding the development of more effective and reliable predictive maintenance strategies.

Robust and accurate prognostics models for estimation of remaining useful life (RUL) are becoming an increasingly important aspect of research in reliability and safety in modern electronic components and systems. In this work, a data driven approach to the prognostics problem is presented. In particular, machine learning models are trained to predict the RUL of wire-bonded silicon carbide (SiC) metal-oxide-semiconductor field-effect transistors (MOSFETs) subjected to power cycling until failure. During such power cycling, ON-state voltage and various temperature measurements are continuously collected. As the data set contains full run-to-failure trajectories, the issue of estimating RUL is naturally formulated in terms of supervised learning. Three neural network architectures were trained, evaluated, and compared on the RUL problem: a temporal convolutional neural network (TCN), a long short-term memory neural network (LSTM) and a convolutional gated recurrent neural network (Conv-GRU). While the results show that all networks perform well on held out testing data if the testing samples are of similar aging acceleration as the samples in the training data set, performance on out-of-distribution data is significantly lower. To this end, we discuss potential research directions to improve model performance in such scenarios.