35 research outputs found
Thermal characterisation and reliability analysis of power electronic devices in wind and solar energy systems
No full textPower electronic converters (PECs) are used for conditioning the flow of energy between renewable energy applications and grid or stand-alone connected loads. Insulated gate bipolar transistors (IGBTs) are critical components used as switching devices in PECs. IGBTs are multi-layered devices made of different coefficient of thermal expansion (CTE) based materials.
In wind and solar energy applications, IGBT’s reliability is highly influenced by the operating conditions such as variable wind speed and solar irradiance. Power losses occur in switching transient of high current/voltage which causes temperature fluctuations among the layers of the IGBTs. This is the main stress mechanism which accelerates deterioration and eventual failures among IGBT layers due to the dissimilar CTEs. Therefore, proper thermal monitoring is essential for accurate estimation of PECs reliability and end lifetime.
Several thermal models have been proposed in literature, which are not capable of representing accurate temperature profiles among multichip IGBTs. These models are mostly derived from offline modelling approaches which cannot take operating conditions and control mechanisms of PECs into account and unable to represent actual heat path among each chip.
This research offers an accurate and powerful electro thermal and reliability monitoring tool for such devices. Three-dimensional finite element (FE) IGBT models are implemented using COMSOL, by considering complex heat interactions among each layer. Based on the obtained thermal characteristics, electro thermal and thermo mechanical models were developed in SIMULINK to determine the thermal behaviour of each layer and provide total lifetime consumption analysis. The developed models were verified by real-time (RT) experiments using dSPACE environment.
New materials, such as silicon carbide (SiC) devices, were found to exhibit approximately 20°C less thermal profile compared to conventional silicon IGBTs. For PECs used within wind energy systems, PECs driving algorithms were derived within the proposed models and by adjusting switching frequency PECs cycling temperatures were reduced by 12°C which led to a significant reduction in thermal stress; approximately 27 MPa. Total life consumption for the proposed method was calculated as 3.26x10-5 which is approximately 1x10-5 less compared to the other both methods. Effects of maximum power tracking algorithms, used in photovoltaic solar systems, on thermal stress were also explored. The converter’s thermal cycling was found approximately 3 °C higher with the IC algorithm. The steady state temperature was 52.7°C for the IC while it was 42.6 °C for P&O. In conclusion, IC algorithm offers more accurate tracking accuracy; however, this is on the expense of harsher thermal stress which has led to approximately 1.4 times of life consumption compared to P&O under specific operating conditions
Machine learning In Li ion Battery Health Prediction
No full textIn today’s rapidly evolving industrial landscape, the quest for efficiency, reliability,
and cost-effectiveness has spurred a paradigm shift towards proactive maintenance
strategies. At the forefront of this transformation lies predictive maintenance, offering
the promise of minimizing downtime and maximizing asset lifespan through
data-driven insights. This thesis focuses specifically on the forecasting of State of
Health (SOH) and Remaining Useful Life (RUL) for lithium-ion (Li-ion) batteries, a
critical component in numerous industrial applications.Leveraging the wealth of data
available, particularly in the realm of Li-ion battery performance, this research employs
advanced machine learning methods, including Deep Neural Networks (DNN)
and hybrid models like Convolutional Neural Network-Bidirectional Long Short-
Term Memory (CNN-BiLSTM), to develop accurate and reliable predictive maintenance
models. The optimized DNN model for SOH prediction showed a significant
improvement, with MAE, MSE, and RMSE reduced by 17.54%, 25.86%, and
13.85%, respectively, compared to the base model. For RUL prediction, the best
CNN-BiLSTM model (MC-SCNN-BiLSTM) achieved an RMSE of 0.0259, MAE of
0.0183, and MAPE of 1.1951, demonstrating its superior performance.Through systematic
experimentation and comprehensive analysis, this study not only improves
the predictive capabilities of these models but also advances the efficiency and intelligence of maintenance practices within diverse industrial sectors.Günümüzün hızla geli¸sen endüstriyel ortamında verimlilik, güvenilirlik ve maliyet
etkinli˘gi arayı¸sı, proaktif bakım stratejilerine do˘gru bir paradigma de˘gi¸simini te¸svik
etti. Bu dönü¸sümün ön saflarında, veriye dayalı içgörüler aracılı˘gıyla arıza süresini en
aza indirme ve varlık ömrünü en üst düzeye çıkarma vaadi sunan tahmine dayalı bakım
yer alıyor. Bu tez, özellikle çok sayıda endüstriyel uygulamada kritik bir bile¸sen
olan lityum iyon (Li-iyon) piller için Sa˘glık Durumu (SOH) ve Kalan Faydalı Ömür
(RUL) tahminlerine odaklanmaktadır. Li-ion pil performansı alanında, bu ara¸stırma,
do˘gru ve güvenilir tahmine dayalı bakım geli¸stirmek için Derin Sinir A˘gları (DNN) ve
Evri¸simli Sinir A˘gı-Çift Yönlü Uzun Kısa Süreli Bellek (CNN-BiLSTM) gibi hibrit
modeller dahil olmak üzere geli¸smi¸s makine ö˘grenme yöntemlerini kullanır. modeller.
SOH tahmini için optimize edilmi¸s DNN modeli, temel modelle kar¸sıla¸stırıldı˘gında
MAE, MSE ve RMSE’nin sırasıyla %17,54, %25,86 ve %13,85 oranında azalmasıyla
önemli bir geli¸sme gösterdi. RUL tahmini için en iyi CNN-BiLSTM modeli (MCSCNN-
BiLSTM), 0,0259’luk bir RMSE, 0,0183’lük MAE ve 1,1951’lik MAPE’ye
ula¸sarak üstün performansını ortaya koydu. Sistematik deneyler ve kapsamlı analiz
yoluyla, bu çalı¸sma yalnızca Bu modellerin öngörücü yetenekleri aynı zamanda çe-
¸sitli endüstriyel sektörlerdeki bakım uygulamalarının verimlili˘gini ve zekasını da geli
¸stirir.M.S. - Master of Scienc
GÜÇ ELEKTRONİĞİ CİHAZLARININ TERMAL STRES PARAMETRELERİNİ SİNİR AĞI ALGORİTMALARINDA KULLANARAK ÖMÜR SÜRESİ TAHMİNİ MODELLEMESİ
No full textThe demand for high power rating applications is increasing rapidly. The essential
components needed to fulfill these demands are power electronic devices and
circuits such as IGBTs, diodes, rectifiers, inverters, and DC-DC converters.
However, these components usually are sensitive to parameter changes and can
face significant failures if we don’t examine our system component’s reliability
well. The traditional approaches to calculating the failure-tolerant capability in
power electronic systems are redundancy designs, which select the individual
components in the circuits with sufficient thermal and electrical stress margin, thus
expecting their low failure rates and, consequently, high reliability of the overall
reliability system.
This work will try to involve neural network technology in reliability topics by
building a model that can estimate the device's thermal stress as a function of the
design parameters and predict the remaining lifetime. The solar and wind profiles
of a solar-wind hybrid renewable system, which would be constructed in METU NCC campus, will be used to estimate the junction temperature of discrete IGBTs
automatically and use this estimation to enhance the lifetime of the inverters by
using a controller that prevents the system from working under high frequencies
when possible extreme junction temperatures may occur, the thing that can lead to
a four-times reduction in the lifetime consumption of power electronics inverters.Yüksek güç uygulamalarındaki artan talebe bağlı olarak, güç elektroniği cihazları
ve devreleri, bu talepleri karşılamak için günden güne yeni teknolojiler barındırmak
zorundadırlar. IGBT'ler, diyotlar, doğrultucular, invertörler ve DC dönüştürücüler
bu bileşenlere verilebilecek cihaz örnekleridir. Bu cihazlar, genellikle değişen
çalışma ortamı parametrelerine duyarlı olup, bu değişkenlerin tasarım sırasında
dikkate alınmaması büyük bozulmalara neden olabilir.
Bundan dolayı, sistem bileşenlerinin güvenilirliğini iyi incelenmelidir. Geleneksel
yaklaşımlarla güç elektroniği sistemlerinde arıza toleranslı yeteneğini hesaplamak
artık zorlaşmıştır. Tasarımlar ve devrelerdeki münferit bileşenlerin yeterli düzeyde
seçilmesi, termal ve elektriksel stres marjı analizleri, düşük arıza oranlarını
mümkün kılabilir. Bu çalışmada, bir model oluşturulup, sinir ağı teknolojisi
kullanımı ile bir güvenilirlik analizi yapılmıştır.
Tasarım parametrelerinin bir fonksiyonu olarak, cihazların termal stresini ve kalan
ömrünü tahmin edebilecek şekilde bir model geliştirilmişitir. Bu model, işletim
sisteminin rüzgar profilini Kullanacak ve ODTÜ-KKK’ nde kurulacak güneş rüzgar hibrit yenilenebilir sistem yerleşkesi için bir ömür süresi analizi çalışmasını
mümkün kılacaktır.M.S. - Master of Scienc
Towards More Reliable Renewable Power Systems - Thermal Performance Evaluation of DC/DC Boost Converters Switching Devices
No full text<span style="font-family: &quot;Times New Roman&quot;,serif; font-size: 10pt; mso-fareast-language: EN-US; mso-fareast-font-family: Calibri; mso-fareast-theme-font: minor-latin; mso-ansi-language: EN-GB; mso-bidi-language: AR-SA;">Power electronic converters (PECs) are one of the most important elements within renewable power generation systems. The reliability of switching elements of PECs is still below expectations and is a major contributor to the downtime of renewable power generation systems. Conventional technology based elements such as Silicon Insulated Gate Bipolar Transistors (IGBTs) operate as switching components in PECs. Recent topological improvements have led to new devices called Silicon Carbide (SiC) MOSFETs which, are also being used as switching elements for PECs. <span style="color: black; mso-themecolor: text1;">This paper presents detailed investigations into the performance of those switching devices with a focus on their reliability and thermal characteristics. Namely, trench gate NPT, FS IGBT topologies and SiC MOSFET are firstly modelled using 3-D multi-physics finite element modelling to gain clear understanding of their thermal behaviour. Subsequently, modelling outcomes are verified by using those devices as switching elements in operational boost converters. The purposely-developed test setups are utilised to critically assess the performances of those switching devices under different loading and environmental conditions. In general, </span></span><span style="font-family: &quot;Times New Roman&quot;,serif; font-size: 10pt; mso-fareast-language: EN-GB; mso-fareast-font-family: Calibri; mso-fareast-theme-font: minor-latin; mso-ansi-language: EN-GB; mso-bidi-language: AR-SA;">SiC device was found to exhibit<span style="mso-spacerun: yes;"> </span>about 20 °C less in its operating temperature and therefore expected to offer more reliable switching element. </span></jats:p
