Estimation of remaining useful life by using neural network method for lithium based batteries in aviation applications
Havacılıkta kullanılan lityum tabanlı bataryaların yapay sinir ağları ile ömür kestirmine katkılar
- Tez No: 600399
- Danışmanlar: DR. ÖĞR. ÜYESİ DERYA AHMET KOCABAŞ
- Tez Türü: Yüksek Lisans
- Konular: Elektrik ve Elektronik Mühendisliği, Electrical and Electronics Engineering
- Anahtar Kelimeler: Belirtilmemiş.
- Yıl: 2019
- Dil: Türkçe
- Üniversite: İstanbul Teknik Üniversitesi
- Enstitü: Fen Bilimleri Enstitüsü
- Ana Bilim Dalı: Elektrik Mühendisliği Ana Bilim Dalı
- Bilim Dalı: Elektrik Mühendisliği Bilim Dalı
- Sayfa Sayısı: 97
Özet
Uzun bir tarihe sahip olan bataryalar günlük hayatımızda pek çok kolaylık sağlamaktadır. Elektronik cihazlar, medikal uygulamalar, elektrikli araçlar, havacılık sektörü gibi pek çok alanda büyük öneme sahip olan bataryaların bir çok türü bulunmaktadır. Günümüzde yüksek enerji yoğunluğu, yüksek hücre gerilimi, düşük ağırlık gibi pek çok nedenden dolayı popüler hale gelen lityum iyon bataryalar, gelişen teknolojinin ihtiyacına göre gün geçtikçe daha fazla önem arz etmeye başlamıştır. Bu durum lityum iyon bataryalar üzerine daha çok Ar-Ge çalışmasının yapılmasına yol açmıştır. Havacılık sektöründe bataryalar, acil durum senaryolarında, kritik yükleri beslemek motorun çalışmayı durdurması durumunda tekrar başlatmak gibi önemli rol oynamaktadırlar. Havacılık tarihinde yaşanan bir takım ciddi hava aracı kazalarının sonucunda havacılık sektöründe batarya kullanımı zorunlu değilken FAA gibi otoritelerce zorunlu hale getirilmiş ve bir takım standartlar belirlenmiştir. Yaşanan tecrübeler doğrultusunda hava araçlarında MIL-E-7016F standartında da belirtildiği gibi, bataryanın, kapasitesinin %80'inin altına düşmesi durumunda değiştirilmesi gerekmektedir. Bu durumda kullanılan bataryanın tam kapasitesi üzerinden değil de bir miktar güvenlik payı bırakılarak yani kapasitenin %70-75 aralığında olduğu düşünülerek gerekli hesaplamalar ve modellemeler yapılmalıdır. Bu sayede bataryadan kaynaklanabilecek olası kazaların önüne geçmek mümkündür. Batarya sağlık takip sistemleri, gün geçtikçe bataryalara duyulan ihtiyaçla birlikte artmaktadır. Yapılan çalışmada ise sağlık takip sisteminin alt başlıklarında olan batarya ömür kestirimi ile havacılık sektöründe kullanılan lityum iyon bataryaların olası hata durumları engellenmeye çalışılmıştır. Batarya ömür kestirimi için yapay sinir ağları metodu kullanılmıştır. Çalışmada, batarya verilerinin elde edilmesi uzun bir süreç gerektirdiği ve ekipmanların yetersizliği nedeniyle Maryland Üniversitesi CALCE Araştırma Enstitüsü'nden hazır olarak CS-2 ve CX-2 isimli farklı yük senaryolarında şarj deşarj edilen lityum iyon batarya verileri kullanılmıştır. Ardından batarya verileri eğitim, validasyon ve test verileri olarak ayıklanmıştır. Verileri işlemek için Matlab programının Neural Net Fitting ürünü kullanılmış, ileri yayılım ve geri yayılım metotları kullanılarak bir makine öğrenimi metodu ile veriler analiz edilmiş ve ömür kestirimi sonucundaki hata payı uygun parametreleri seçerek en aza indirilmeye çalışılmıştır. Yapılan çalışmalar sonucunda ömür kestiriminde eldeki batarya verilerinin doğru bir şekilde anlamlandırılıp veriler içerisinden gerekli parametreler seçildiği takdirde kestirim sonuçlarındaki hata oranının oldukça düşü çıktığı gözlenmiştir.
Özet (Çeviri)
Today, there are many inventions use in industries to ease humans' life in many ways. Due to creative and advanced inventions, systems are getting more complex although productivity and capability are improved. Complexity is come with the price such as faults and failure on these advanced systems that can cause financial loss and casulities to people. To reduce the faults and failures cause pecuniary loss and intangible damages, people start to calibrate and maintain their systems. Therefore, health management systems become an important topic worldwide. Fault and anomaly can be detected and reported thanks to the health management systems. As well as these health management systems can predict remaining useful life of a system. Due to advantages of the health management system, it became a popular topic in science world and started to apply in many advanced systems such as aerospace battery applications. Prognostic and diagnostic are 2 important aspects of faults and anomaly detection in a system. Prognostic represents to detection of any fault or anomaly before it happens. Prognostic approach leads to set maintenance period of a system to reduce or prevent faults and anomalies. Diagnostic approach is identification of a fault condition by observation of parameters of a system. In aeronautics appliances, prognostic approach is too important and critic to prevent failures. In flight mode of an aircraft if a failure in any critical system, results of it can be catastrophic. With the prognostic approach in aeronautics applications, maintenance periods can be arranged by observing some parameters, therefore possibility of failure can be reduced or eliminated. Prognostic approach can be applied for many systems in aircrafts. One of these systems is energy storage system has an important role in case of emergency to supply critical loads or re-start aircraft engines. As a back-up power supply, batteries are the most common element as a part of energy storage systems. Battery which is a storage device that consist one or more cells inside whose chemical energy create a flow of electrons to produce electrical energy. Before the aircraft accident in 1969, there is no obligation to use battery in aircrafts. However, some of regulations are written in blood and aviation is an industry that are not accept negligence. Therefore, after the accident keeping battery in aircrafts would be an obligation. Some authorities about aviation such as FAA, make operative some standarts about keeping battery for back-up power in case of any emergency. Batteries have a lot of type, but most common ones are lead acid, nickel cadmium and lithium ion batteries. Between all of types of batteries, lithium ion batteries are preferred in these days due to their higher specific energy, lower weight, higher cell voltage. Lithium ion batteries are used in many applications such as mobile phones, computers, cameras, electrical devices, medical appliances, aviation. Lithium ion batteries provide many advantages compared to other battery types. Especially in the small electrical devices, lower weight and higher specific energy are strong reason to use lithium ion batteries. Besides, in aviation appliances, lithium ion batteries are important. In the aviation history, lead acid batteries would be used. After 1899, nickel cadmium batteries were invented. Then, in the late 1900s, lityum based batteries were invented. In these days, lithium batteries are used in some of aircrafts due to many advantages in field of weight, specific energy and low maintenance cost. Although lithium ion batteries have significant advantages compared to other type of batteries, they have a few disadvantages such as thermal runaway, overheating, degredation of battery capacity. These disadvantages are highly important especially when aircraft is flying or carrying passengers. Lithium ion batteries have a non-linear charge and discharge characteristics. Due to this non-linearity, measurement of the capacity degredation of a lithium ion battery which affects the state of health, is an important engineering problem. Determining the capacity degredation of a lithium ion battery gives a clue about battery remaining useful lifetime (RUL). Especially in a critical application which supplies from a battery unit for an emergency, end of life of a battery must be known to predict any pecuniary loss and intangible damages. Therefore, researches on state of health of batteries are common subject in these days. In this thesis, remaining useful life of lithium ion battery is estimated by neural network algorithm is one of the machine learning method. Battery parameters are taken from Maryland University Calce Battery Research Group because measuring parameters of lithium ion battery is a long process and there is technical impossiblity. For different load scenarios, CX-2 and CS-2 lithium cobalt oxide batteries are charged and discharged at constant amper rates. With obtained data set, capacity degradation graphs of batteries are plotted by using parameters as time and battery capacity measured in a period repeatedly. However, capacity degradation graphs were included too much noise due to measurement. Therefore, they are needed to be filtered. Besides, filtering process had to protect the reality of the data. So, a filtering factor is selected empirically to eliminate noise which is just come from measurement device. Results of filtered degradation graphs and other determined parameters such as initial capacity, rated capacity and discharge current are seperated for training data, validation data and test data. The data is analysed with Matlab Neural Net Fitting Toolbox. With feed forward and backpropagation methods using machine learning toolbox on Matlab. In the analysis, input parameters (capacity degradation, initial battery capacity, rated capacity and discharge current) and an output parameter which is remaining lifetime as number of remaining battery cycle, are used to predict an accurate remaining useful time of the battery. Neural network method is depended learning like a human by using a large data set. Also the variety of parameters and quantity are important points. Therefore, in this study, determining suitable parameters for remaining useful life prediction is the most important stage. Charge and discharge characteristics of batteries are not linear. Also, charge and discharge characteristics can change depends on load scenario excessively. For higher discharge current, battery temperature is rising swiftly than 1 C rate. Besides, high temperatures cause an increase on battery aging. The relation between discharge current and battery aging causes selecting discharge current as an input parameter. Another parameter that affects remaining useful life of battery is initial capacity. Although structures and electrical features are same of each CX-2 battery, capacities of these batteries have minor differences. Even small differences in battery capacity can be affect battery aging, so, as a second parameter, initial battery capacity is selected. Also, rated capacity is an important parameter for prediction of remaining useful life of batteries. Data set of CX-2 and CS-2 batteries are used to improve the accuracy of the prediction. In neural network applications, variety is an important factor to train the model to reach more accurate results. Therefore, different rated capacities are used as an input parameter for a better trained model. As a fourth but most important parameter is remaining capacity. Unlike constant parameters, this parameter shows how many cycles is left. Remaining capacity is decreasing when the battery is used to charge and discharge. Besides, the battery must be replaced with a new battery before maximum initial capacity is reached to %80 of its nominal capacity according to MIL-E-7016F Military Specification. The last parameter is remaining cycle of the battery is used as the output parameter. There is an output parameter is corresponding to each input parameters group. There are six batteries for each CX-2 and CS-2 batteries. To analyse the parameters of 12 batteries, Neural Net Fitting Toolbox of Matlab is used. Data set of one battery out of 12 batteries is selected as an independent test data. Other 11 batteries are used to create the model and all data set of 11 batteries is separated to three group as training, validation and test data. Model is trained by using training data and knows where it stops thanks to validation data. After model is created, model is controlled by test data. However, to avoid to unintended memorization, rates of training, validation and test data must be selected properly. Also, the independent test data is used for controlling of accuracy of the model results. These processes are repeated for 12 batteries and compared. In this study, it has been aimed that is used for aviation applications if this method is suitable to predict the remaining useful life of batteries with high accuracy. In conclusion, with the database and selecting suitable parameters of batteries, decreasing error rate is observed. Also, it has been seen that the larger database is, the more accurate results are predicted. For future work, it will be aimed to have a larger and various dataset for a higher capacity battery and collect real-time dataset by using an embedded system on aircraft central computer when the aircraft is in flight mode.
Benzer Tezler
- Lityum iyon bataryaların makine öğrenimi yöntemleri ile sağlık durumu kestirimi
State of health estimation for lithium-ion batteries using machine learning methods
ÇETİN ORAL
Yüksek Lisans
Türkçe
2021
Elektrik ve Elektronik Mühendisliğiİstanbul Teknik ÜniversitesiElektrik Mühendisliği Ana Bilim Dalı
DR. ÖĞR. ÜYESİ DERYA AHMET KOCABAŞ
- An adaptive data filtering model for remaining useful life estimation
Kalan faydalı ömür tahmini için uyarlanabilir veri filtreleme modeli
OĞUZ BEKTAŞ
Doktora
İngilizce
2018
Havacılık ve Uzay MühendisliğiUniversity of WarwickHavacılık ve Uzay Mühendisliği Ana Bilim Dalı
PROF. DR. JEFF JONES
- Time series analysis with deep learning approaches for industry 4.0
Endüstri 4.0 için derin öğrenme yaklaşımlarıyla zaman serisi analizi
CEREN NUR BAŞ
Yüksek Lisans
İngilizce
2019
Bilgisayar Mühendisliği Bilimleri-Bilgisayar ve KontrolGalatasaray ÜniversitesiBilgisayar Mühendisliği Ana Bilim Dalı
DOÇ. DR. ÖZLEM DURMAZ İNCEL
- Design of predictive maintenance model using artificial intelligence methods
Yapay zeka yöntemleri kullanılarak kestirimci bakım modelinin dizayn edilmesi
BEGÜM AY TÜRE
Yüksek Lisans
İngilizce
2021
Bilgisayar Mühendisliği Bilimleri-Bilgisayar ve Kontrolİstanbul Ticaret ÜniversitesiBilgisayar Mühendisliği Ana Bilim Dalı
PROF. DR. ABDÜL HALİM ZAİM
DOÇ. DR. AKHAN AKBULUT
- Siamese networks for predictive maintenance
Kestirimci bakım için siyam ağları
UĞUR CEYLAN
Yüksek Lisans
İngilizce
2022
Bilgisayar Mühendisliği Bilimleri-Bilgisayar ve KontrolGebze Teknik ÜniversitesiBilgisayar Mühendisliği Ana Bilim Dalı
DR. ÖĞR. ÜYESİ YAKUP GENÇ