Linköping Electronic Conference Proceedings
Not a member yet
1113 research outputs found
Sort by
Evaluating Modelling Performance: Sensitivity Analysis of Data Volume in Industrial Batch Processes
Get PDFThe iron and steel industry, a cornerstone of global industrial development, is accountable for a significant environmental footprint, contributing to 7.2% of global greenhouse gas emissions. This significant portion underscores the sector's substantial impact on climate change. The projected increase in steel production by an estimated 30% by the year 2050, further accentuates the urgent need for innovation and sustainable practices within this industry. Given these considerations, prioritising the development of more efficient and environmentally friendly production methods becomes not only a matter of environmental responsibility but also a crucial aspect of ensuring the industry's long-term viability. This work presents an investigation that evaluate the impact of product series and modelling complexity regarding the prediction of products downstream properties for industrial batch processes. The system under observation is the production of thermocouple wire-rod materials, starting from the smelt-shop and concluding after the hot-rolling mill. The first perspective considered is how to model processes with more than one product of the same product series, in this case different alloy products that are of the same product series, namely thermocouples. In addition, models of escalating complexity are being implemented. This involves examining whether the successful generalisation of simpler models necessitates the adoption of a more sophisticated approach
Enhanced Anomaly Detection in Aero-Engines using Convolutional Transformers
Get PDFGas turbines are vital in power generation and propulsion systems. However, these engines are exposed to complex and variable operating conditions, which makes early and accurate fault detection essential for predictive maintenance and minimizing unplanned downtime. This paper proposes a novel approach that combines convolutional neural networks (CNNs) with transformer architectures to address these challenges. The proposed Convolutional transformer model aims to enhance the accuracy and robustness of turbofan fault classification by integrating the feature extraction capabilities of CNNs with the contextual learning strengths of transformers. Through rigorous experiments, we seek to demonstrate our approach's performance in classification accuracy and generalization across different operating conditions. We utilize a comprehensive synthetic dataset, C- MAPSS, derived from multiple aircraft engine units as the benchmark for this study. The results for the proposed model show an accuracy of 99.6% on the test dataset. The outcome has the potential to be extended and fine-tuned for different types of gas turbines for diverse applications
A Deep-Unfolding Approach to RIS Phase Shift Optimization Via Transformer-Based Channel Prediction
Get PDFReconfigurable intelligent surfaces (RISs) have emerged as a promising solution that can provide dynamic control over the propagation of electromagnetic waves. The RIS technology is envisioned as a key enabler of sixth-generation networks by offering the ability to adaptively manipulate signal propagation through the smart configuration of its phase shift coefficients, thereby optimizing signal strength, coverage, and capacity. However, the realization of this technology's full potential hinges on the accurate acquisition of channel state information (CSI). In this paper, we propose an efficient CSI prediction framework for a RIS-assisted communication system based on the machine learning (ML) transformer architecture. Architectural modifications are introduced to the vanilla transformer for multivariate time series forecasting to achieve high prediction accuracy. The predicted channel coefficients are then used to optimize the RIS phase shifts. Simulation results present a comprehensive analysis of key performance metrics, including data rate and outage probability. Our results confirm the effectiveness of the proposed ML approach and demonstrate its superiority over other baseline ML-based CSI prediction schemes such as conventional deep neural networks and long short-term memory architectures, albeit at the cost of slightly increased complexity
Numerical Methods for the Flow Fields; A Comparative Review
Get PDFThis paper provides a comparative overview of four numerical methods widely employed in computational fluid dynamics and related fields: Finite Volume (FV), Lattice Boltzmann Method (LBM), Smoothed Particle Hydrodynamics (SPH), and Spectral Methods. FV discretizes the domain into control volumes, emphasizing conservation laws and flux integrals across cell faces. It's renowned for its robustness, particularly in complex geometries. LBM is a mesoscopic approach simulating fluid dynamics through particle interactions on a lattice grid. Its intrinsic parallelism and ability to handle complex boundary conditions make it suitable for multiphase flows and porous media simulations. SPH represents fluids as a set of particles, where properties are smoothed over neighboring particles using a kernel function. SPH excels in free surface flows, astrophysical simulations, and fluid-structure interaction due to its Lagrangian nature and adaptive resolution. Spectral Methods discretize functions using orthogonal basis functions, such as Fourier or Chebyshev polynomials, enabling high-order accuracy and spectral convergence. They are preferred for problems with smooth solutions and periodic boundary conditions, like turbulence simulations and wave propagation
Position Paper: New Views of Shots - Towards Measures of Net Visibility and Reachability
Get PDFIn this position paper, we define two new metrics: net visibility (the fraction of the net that can be seen from the perspective of the puck) and net reachability (the fraction of the net that could be reached by the puck). Reachability is slightly different from visibility because even though there might be a small portion of the net visible in a certain area (a hole), that hole may not be large enough for the puck to pass through and reach the net. We describe a framework for computing our metrics using a combination of puck and player tracking (PPT) data and video analysis (image processing). We use data and video from an NHL game to provide a proof of concept for computing net visibility and reachability. We also describe areas where more work can be done to improve the accuracy of the results and allow the computations to be fully automated. Our position is that these metrics would be valuable in studying shooter decisions and skills, goaltender and player locations and that the technologies could be used to create virtual reality images or videos
Life Cycle Assessment of Floating Offshore Wind Farms: The Case of Hywind Tampen in Norway
Get PDFTo address climate change and energy security issues from fossil fuels, wind power is a promising renewable energy source, projected to grow significantly by 2050. Offshore wind energy, especially floating offshore wind farms shows great potential due to higher and more consistent wind speeds at sea. However, these turbines have negative environmental burdens throughout their life cycle. This The present study focuses on a comprehensive cradle-to-grave life cycle assessment of the Hywind Tampen floating offshore wind farm in Norway. The assessment covers all stages from manufacturing, transportation, installation, operation, and maintenance to decommissioning, utilizing openLCA® software and ecoinvent 3.9 database with the ReCiPe 2016 impact assessment method. Key findings indicate that manufacturing is the primary contributor to total emissions, followed by operation and maintenance. The study emphasizes the necessity of developing more sustainable manufacturing methods, designing turbines that are more efficient and versatile, and better maintenance forecasting and planning in order to minimize the environmental impact of these turbines
Using an advanced simulation tool for successful conversion of reheating furnace to full oxyfuel operation
Get PDFOxyfuel combustion complements decarbonization efforts by reducing the energy needs in high-temperature industries. Steel reheating furnaces are good candidates for full oxyfuel operation since this can lead to up to 30% energy savings. Linde uses an in-house tool to simulate reheating furnaces for airfuel to oxyfuel conversion. This paper follows a real customer case, starting with an airfuel simulation setup used to analyze the furnace, followed by oxyfuel simulations for burner design and energy savings estimations. These simulations lead to a successful installation of oxyfuel burners for the reheating furnace located at Ovako Imatra site. After the commissioning is completed, performance evaluation is done by comparing a reference airfuel operation period with an oxyfuel combustion period. Full oxyfuel conversion results in 27% energy savings for hot charge and high production rate periods thanks to significantly lower flue gas losses. Removing nitrogen from the oxidizer decreases the flue gas volume, reducing the total heat capacity of the off-gas stream. The savings are around 30% for cold charge and average production rate periods
Model Predictive Control for Integrated Photovoltaic (PV) and Electrolysers System
Get PDFThe European Union (EU) has set an ambitious target to reach carbon neutrality by 2050, prompting industries to develop roadmaps to achieve this goal. In this context, hydrogen and hydrogen-based fuels play a crucial role in achieving net-zero emissions. Instead of relying on hydrogen production from steam reforming natural gas systems, electrolysers offer a sustainable alternative to address climate and energy challenges. The integration of solar energy systems with electrolysers can further diminish carbon emissions and enhance sustainability. Typically, these processes are simulated using process simulation software platforms that employ first-principle models based on the mass and energy balances within the system. The adoption of Model Predictive Control (MPC) algorithms not only benefit from improves advance control methods and optimization but also facilitates the automation and efficient operation of these processes. This study aims to mathematically model and simulate an integrated photovoltaic (PV) and Proton Exchange Membrane (PEM) water electrolyser system for hydrogen production. Additionally, it assesses the impact of MPC algorithms on the system's efficiency. The study undertakes modeling of PV systems incorporating a maximum power point tracking algorithm to capitalize on optimal power generation and ensure a consistent direct current supply to the electrolysers. Mathematical modeling of PEM water electrolysers is performed to establish the current-voltage curve in steady-state mode and to assess water and gas permeability through the membrane in dynamic mode. Finally, the study identifies input variables, such as electrolyser temperature, and evaluates their effects on key indicators like system efficiency through performance analysis
Data Center Resource Usage Forecasting with Convolutional Recurrent Neural Networks
Get PDFEnergy efficiency, scalability, and reliability are increasingly important for sustainable data centers. In this paper, we focus on forecasting real-world resource usage using neural network time series models, specifically utilizing convolutional recurrent long short-term Memory (LSTM) and gated recurrent unit (GRU) architectures. In our analysis, we compare LSTM and GRU in terms of forecasting accuracy and computational complexity during model training. We demonstrate that recurrent neural networks are more accurate and robust compared to the traditional autoregressive integrated moving average (ARIMA) time series model in this complex forecasting problem. GRU achieved a 9% reduction and LSTM a 5% reduction in forecasting mean squared error (MSE) compared to ARIMA. Furthermore, the GRU architecture with a 1D convolution layer outperforms LSTM architecture in both forecast accuracy and training time. The proposed model can be effectively applied to load forecasting as part of a data center computing cluster. In this application, the proposed GRU architecture has 25% fewer trainable parameters in the recurrent layer than the commonly used LSTM
