Lityum iyon bataryalarda parça ömür kestirimi
Calculating remaining useful life of lithium-ion batteries
- Tez No: 550448
- Danışmanlar: PROF. DR. SALMAN KURTULAN
- 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: Kontrol ve Otomasyon Mühendisliği Ana Bilim Dalı
- Bilim Dalı: Kontrol ve Otomasyon Mühendisliği Bilim Dalı
- Sayfa Sayısı: 91
Özet
Çalışma kapsamında ömür kestirimi üzerine çalışılan Lityum İyon bataryalar kullanımı tehlikeli olmasına karşın, akıllı telefonlarda, elektrikli araçlarda, yeni nesil hava araçlarında tercih edilen enerji yoğunluğu yüksek bir depolama aracıdır. Endüstriyel bataryalar, bazı platformlarda harekete geçmek için ilk enerjiyi sağlarken, bazı sistemlerde ise acil durumlardaki enerji ihtiyacını karşılamak için kullanılmaktadır. Hata ve arızalar geçmişten günümüze can kayıplarına ve ekonomik zararlara sebep olmuştur. İnsanoğlu bu sebeple kaza riskini azaltmak, fonksiyonlarının devamını sağlamak ve güvenirliliği artırmak için makine ve araçlarda uzun yıllardır bakım yapmaktadır. Kalan ömür kestirimi, hatalar meydana gelmeden evvel ne zaman bozulacağını fiziksel ve matematiksel modellerle veya tecrübeyle kestiren sağlık takip sisteminin özel bir konusudur. Kalan ömür kestirimiyle, kullanıcı sistemdeki performans düşümü konusunda bilgilendirilir ve önceden bakım yapılarak ünitelerin ömrü artırılmış olur. Bakım standartlarında, hava aracı sistemleri gibi acil durumlardaki enerji ihtiyacını karşıladığı sistemlerde, bataryaların %80 kapasite düşümünde değiştirilmesi gerektiği önerilmektedir [17-18]. Toplam acil durum yük kapasitesi, bataryanın anma kapasitesinin % 72- %75'ine denk gelecek şekilde tercih yapılırsa, %5 ile %8 arasında güvenlik payı ile yetersiz batarya kapasite arızası engellenmiş olacaktır. Bataryaların kapasite düşümünü izlemek için, çok uzun süre gözlem yapmak ve kaydetmek gerektiği için öncelikle hipotez oluşturabilmesi adına hazır veri arayışına girilmiş ve Birleşik Devletler'de bulunan Maryland Üniversitesi Calce Araştırma enstitüsünden edinilmiştir. Veriler, 5 adet 1100 mAh kapasitedeki C-S2 ve 6 adet 1350 mAh kapasitedeki C-X2 tipi bataryadan elde edilmiştir. Aynı ortam şartlarında doldurulup boşaltılan bataryalar farklı zamanlarda yüzde 80 kapasite düşümüne eriştiğinden kestirici arayışına girilmiştir. Klasik kontrol teknikleriyle bataryanın kalan kullanım ömrünü kestirmek zor olacağından makine öğrenimi ailesinden regresyon modelleri araştırılmıştır. Sonuç olarak regresyon ağaçları modelinin eğitimi ve doğrulanması sağlandıktan sonra yüksek doğrulukta sonuç ürettiği görülmüştür. Farklı ortam koşullarını gösteren daha fazla veri girişi yapılırsa model başarısı artırılabilir. Bu çalışma kapsamında, regresyon ağaçları modeli sadece Lityum İyon bataryaların ömür kestiriminde kullanılmış olsa da gelecekte aynı bakış açısının farklı sistemlerin ömür kestiriminde de kullanılabileceği düşünülmektedir.
Özet (Çeviri)
In the past, many type of batteries were invented by scientists. Even some of them only experimented in laboratories, some others like Lead Acid, Nickel Cadmium and Lithium based batteries became very popular on industry because of their advantages. This thesis focused on life prediction of Lithium Ion batteries which are used in smart phones, electric cars and new generation air vehicle systems even its explosion can cause catastrophic failures on the platform. The main purpose of using industrial batteries is providing motive power, for actuation and fulfilling emergency power demands. Faults and failures result financial loss and casualties to people from past to present. Human calibrate and maintain their machines and equipment to reduce accident risk, make capable them to carry on their function, continue manufacturing and increase reliability. When a system comes into operation, frequency of breakdowns and malfunctions will be increased over time. It happens due to run above its capability, absence of required consumable materials such as water and oil, or human factor. Health management systems detect and report faults, failures, anomalies, calculate remaining useful life and remaining capacity. Even health management systems were considered as a patch for existing system, nowadays engineers consider to design it as a part of the plant. This provides system with economical servicing, manipulating users about maintenance, increasing overall reliability and reducing storage costs. Fault detection and anomaly detection are 2 important aspects of maintenance model. In order to understand whether a fault or anomaly some parameters of plant can be investigated. These parameters may be current, voltage or any sensor measurement. These parameters should normally reside between low and high set values however failures and anomalies can be indicated after in the event of digressing band. Prognostic is a special topic of health management system that detects failures before it happen with physical, mathematical models or past experience. Prognostic makes users know about degradation of system and increase components life since preventive maintenance can be done. In order to have high success rates for prognostic, fault detection success is a precondition since it's impossible to understand when a failure happens before knowing failure. According to some maintenance standards, it's recommended to change batteries when it achieves 80% of rated capacity especially when designed for providing emergency power to systems like air vehicle systems [17-18]. If batteries are chosen as 72% - 75% of rated capacity equal to total emergency demand, insufficient capacity failure of battery can be prevented by %5-%8 safety margin. Regression models from machine learning family were investigated for this study owing to fact that classical control techniques are difficult for calculating remaining useful life of a battery. Machine learning term is firstly used in 1959 and was studied by researchers during 1970's and 1980's. This is a technique use statistic for prediction even without deep knowledge about system. It is still popular topic even now because of being easier to record and share data. Regression problems like battery capacity degradation can be solved by machine learning algorithms. Having meaningful data set is a must for this models. Machine learning data set consists of three parts and these are training data set, validation data set and test data set. Monitoring and recording performance data for a long time is essential to create a prognostic model. Fortunately, this data was available from Calce Research Group of Maryland University instead of performing experiments [16]. Performance data consists of 5 piece of identical C-S2 battery (1100mAh rated) and 6 piece of identical C-X2 battery (1350 mA rated). Since all identical batteries reach %80 of rated discharge capacity differently on same ambient conditions, machine learning models are tried individually. Regression Tree is one of the machine learning model which is popularly used for regression and classification problems. It can offer 2 biggest advantages. First of all, it's too simple since algorithm behind model works like human brain and secondly it is easier to be shown by trees how data is interpreted than other techniques like Support Vector Machine or Neural Networks. ID3 and Cart are mostly known algorithms for building an optimal regression tree model. After finishing our analysis for this study, it has been decided to put discharge current, rated discharge capacity, battery initial discharge capacity and remaining discharge capacity of each cycle was defined as 4 inputs of regression tree model. Discharge current is important for battery aging because drawing relatively high currents can increase temperature of battery and high temperature can cause fast aging. On the contrary if connected unit draw low currents, battery can be used for longer periods. Remaining discharge capacity can be calculated by doing an integral from time versus discharge current graph. It is only input that changes dynamically for each cycles. Training data set involves 2 different battery samples. Reason of inserting rated discharge capacity is introducing this difference to model. Initial discharge capacity was added as input in addition to rated capacity starting because discharge capacity of same batteries can be different even 2 piece of battery are identical. Only output of desired regression tree model is remaining cycle to achieve %80 of rated capacity. Because of lithium ion battery characteristic, discharge capacity decreases dramatically by each cycle. So that maintenance standards suggest to replace with new one once achieve this point as mentioned above. Before using, data set was processed by specially developed algorithm for eliminating noises and calculating discharge capacity from discharge current for each cycle in Matlab software. Afterwards, battery prognostic model was constructed with“Regression Learner Tool”inside Matlab again. Thanks to“Regression Learner ”tool, same dataset only needs be introduced by one time. It offers different machine learning models such as Support Vector Machines, Linear Regression, Gaussian Process Regression, Ensembles of Trees and Regression Trees to be implemented for regression problems. To make sure that our model can generate consistent and highly accurate results for battery prognostics, each time one of eleven battery dataset allocated for testing and remaining ten batteries used for training. Goal of this master thesis is yielding 11 different prognostic models that give precise results each time to show consistency. Performance of all candidate regression models were measured and it has been seen that,“Regression Tree”model always gave acceptable results for each scenario when especially prediction point near to failure point. Some assumptions have been made when designing a model for prognostic in this study. First of all, our remaining useful calculation model can only be applicable on C-S2 and C-X2 batteries at 4 constant temperature levels. However, in reality connected units draw variable currents, and that's make batteries to provide variable discharge capacity. Also, batteries can work at extreme conditions, such as low and high temperature levels, and that's make again batteries to behave differently. So in case of adding new samples into dataset to demonstrate different working conditions, more precise outputs from model can be yielded. We believe that, regression trees can also be used for remaining useful life prediction of different electrical or mechanical systems in addition to lithium ion batteries in the future. According to some maintenance standards, it's recommended to change batteries when it achieves %80 of rated capacity when designed for providing emergency power to systems like air vehicle systems [17-18]. If batteries are chosen as %72 - %75 of rated capacity equal to total emergency demand, insufficuent capacity failure of battery can be prevented by %5-%8 safety margin. Monitoring and recording performance data for a long time is essential to create a prognostic model. Fortunately, this data was yielded from Calce Research Group of Maryland University instead of performing experiments. Performance data consist of 5 piece of identical C-S2 battery (1100mAh rated) and 6 piece of identical C-X2 battery (1350 mA rated). Since all identical batteries reach %80 of rated discharge capacity differently on same ambient conditions, a new predictor model was investigated. Because classical control techniques are inefficient for calculating remaining useful life of a battery, regression trees from machine learning family was used in this study. It has been seen in this study that model can generate highly accurate results after training and validation when especially prediction point near to failure point. We believe that, regression trees can be used not just only predicting remaining useful life of lithium-ion batteries but different electrical or mechanical systems too in the future.
Benzer Tezler
- Modifiye edilmiş grafitin lityum iyon pillerde anot malzemesi olarak kullanımının incelenmesi
Investigation of modified graphite as anode material for lithium ion batteries
NERGİS AKIN
Yüksek Lisans
Türkçe
2015
Enerjiİstanbul Teknik ÜniversitesiEnerji Bilim ve Teknoloji Ana Bilim Dalı
PROF. DR. NİLGÜN KARATEPE YAVUZ
- Bir feribot için yakıt hücresi batarya hibrit sisteminin modellenmesi ve simulasyonu
Modeling and simulation of fuel cell battery hybrid system for a ferry
RUKİYE GÜLMEZ
Yüksek Lisans
Türkçe
2022
Denizcilikİstanbul Teknik ÜniversitesiDeniz Ulaştırma Mühendisliği Ana Bilim Dalı
PROF. DR. CENGİZ DENİZ
- Development and characterization of new layered cathode materials for lithium ion batteries
Lityum iyon bataryalar için yeni tabakalı katot malzemelerinin geliştirilmesi ve karakterizasyonu
BERKE PİŞKİN
Doktora
İngilizce
2018
EnerjiOrta Doğu Teknik ÜniversitesiMetalurji ve Malzeme Mühendisliği Ana Bilim Dalı
PROF. DR. MEHMET KADRİ AYDINOL
- Contributions to equivalent circuit model parameter estimations of NiMnCo and LiFePO type lithium-ion batteries
NiMnCo ve LiFePO hücre tiplerine ait lityum-iyon bataryaların elektriksel eşdeğer devre parametre tahminine katkılar
MAHMUT KEREM EREN
Yüksek Lisans
İngilizce
2021
Elektrik ve Elektronik Mühendisliğiİstanbul Teknik ÜniversitesiElektrik Mühendisliği Ana Bilim Dalı
DR. ÖĞR. ÜYESİ DERYA AHMET KOCABAŞ
- Lityum-iyon bataryalarda sağlık durumu kestirimi için model geliştirilmesi
Model development for state of health estimation in lithium-ion batteries
YASİN TAŞAN
Yüksek Lisans
Türkçe
2019
Elektrik ve Elektronik MühendisliğiGebze Teknik ÜniversitesiElektrik Mühendisliği Ana Bilim Dalı
PROF. DR. ABDULKADİR BALIKÇI