1. Dr. Cairo Lúcio Nascimento Júnior Eng. Prof.M. José Affonso Moreira Penna Eng. M.Sc. Leonardo Ramos Rodrigues Instituto Tecnológico de Aeronáutica 1

2.  Introduction  PHM (Prognostics and Health Management)  Lithium-Ion Battery  Capacity Model  Health Monitoring Model  Model Simulations  Estimating Remaining Useful Life  Case Study  Conclusions 2

3. Study of lithium- ion’s battery Analyses of experimental data Models Simulation of battery discharges through the life cycle Monitoring Study and application of techniques for PHM RUL Estimation  Motivation  Reducing costs of operation and maintenance;  Flight safety improvement;  Development of techniques for prognosis.  Objective  Develop methodology for estimating the Remaining Useful Life (RUL) of lithium-ion’s aeronautical battery.  Methodology 3

4.  Present scenario: ◦ MTBF (Mean Time Between Failures);  Maintenance tasks are assigned based on hard times;  Decrease in the dispatch of the aircraft;  Insufficient data to predict failure;  Possible degradation in flight safety;  Current proposals : ◦ Data-Driven Methods; ◦ Model-Based Methods. 4

5.  The beginning: ◦ Discipline in process of maturation; ◦ Study of the mechanism of failure ↔ Life cycle management; ◦ 1970 - HUMS (Health & Usage Monitoring Systems): MH-47E Chinook (vibration monitoring to predict failures in the gears of the rotors of helicopters); ◦ 1980 - Manufacturers envision business opportunity and start fabrication of this system (Smiths Aerospace). 5

6.  Fundamental concepts: 1. All electromechanical systems age as a function of use, passage of time, and environmental conditions; 2. Component aging and damage accumulation is a monotonic process that manifests itself in the physical and chemical composition of the component; 3. Signs of aging (either direct or indirect) are detectable prior to overt failure of the component (i.e., loss of function); 4. It is possible to correlate signs of aging with a model of component aging and thereby estimate remaining useful life of individual components. 6

7.  Data-driven methods ◦ Capture and analyze multi-dimensional and noisy data containing a large number of variables related to component degradation; ◦ Management of uncertainty; 7

8.  Model-based methods ◦ Development of first-principles models of component use and damage accumulation; ◦ Use operational data to fine-tune model parameters; ◦ Model-based prognostics typically result in more accurate and precise RUL estimation; ◦ Advantages in validation, verification, and certification since the model response can be correlated with laws of nature. 8

9.  Why study this type of battery? ◦ Increasing application in the aerospace industry (Boeing 787, Airbus A380); ◦ Higher energy density, low self-discharge, long life in stock; ◦ Available experimental data at NASA Ames Prognostics Data Repository. 9

10.  Failure modes ◦ Over-voltage; ◦ Under-voltage; ◦ Low temperature operation; ◦ High temperature operation; ◦ Mechanical fatigue;  Life Cycle 10

11.  Data Repository  Source: NASA Ames Prognostics Data Repository;  34 lithium-ion batteries (C nominal =2Ah);  Repetitive cycles of discharge, recharge, and impedance measurement;  Archives ”.mat”. 11

12. ◦ Data treatment  Extrapolation of discharge curve. ◦ Effect of degradation over the life cycle  Reduced time of discharge;  Reduction of voltage. 12

13.  Data repository ◦ Battery capacity (C) calculated by ◦ State of Charge (SoC) calculated by 13

14.  Discharge model ◦ (PAATERO, 1997) e (SPERANDIO, 2010); ◦ Voltage U (I,T,SoC) calculated by 14

15.  Discharge model ◦ Determination of parameters x 1...x 17 :  First discharge curve of each selected batteries;  FMINSEARCH (MATLAB®) minimizing square error;  Error mean=0,0565 V (<1.8%);  Error variance= 0,0058 V 2. 15

16.  Capacity model ◦ Linear model  Capacity = f (T, I, nc) ◦ Determination of the parameters c0 and c1:  Selected five batteries with different discharge profiles;  Selected c0 and c1 models;  fminsearch (MATLAB®) minimizing square error;  Error mean=0,0324 Ah (<2,2%);  Error variance=0,0035 (Ah) 2 16

17.  Capacity model ◦ Capacity x electrical current  Low electrical current:  Higher initial capacity C 0 ;  Faster loss of capacity. ◦ Capacity x temperature  High temperature:  Higher initial capacity C 0 ;  Faster loss of capacity. 17

18. ◦ State of Health (SoH) ◦ Delta Health ◦ nc (C=0) ◦ Capacidade @SoH ◦ Relative Number of Cycles (ncr) ◦ Remaining Useful Life (RUL) 18

19. ◦ Battery Model  Capacity Model  SoC calc  Discharge Model 19

20. ◦ Health Monitoring System  Source 20

21. ◦ Health Monitoring Model  SoH calc 21

22.  rnc calc  SoH calc 22

23. ◦ Example of Simulation  Example of the evaluation of SoH, delta health and nrc at determinate operation profile throughout the life of the battery. At the cycle 210 the discharge profile change from I=4A and T=43°C to I=2A and T=24°C. 23

24. 24 Method proposed to estimate the remaining useful life (RUL min and RUL max ): 1.A linear regression of the SoH data available to date using the function REGRESS (Matlab R2010b); 2.Evaluation of the cycle number at which the battery reaches the minimum threshold of SoH (SoHmin) by extrapolating the line obtained by linear regression; 3.Addition of the uncertainty of the model and of the future operating profile to be performed.

25. 25 1.A linear regression: 2.Evaluation of nc failure and RUL:: 3.Addition of the uncertainty:

26. 26 ◦ Case A  electrical starting of the engines;  15 minutes discharge;  I=4A (exponential decay);  T=43ºC. Simulation failure at cycle 495  Even if the battery can execute the starting profile until cycle number 770, the battery cannot comply with the emergency requirement after cycle 495, as shown in Figure 20. In this case the failure of the battery is declared on cycle 495.

27. 27 ◦ Case A  RUL estimation  good accuracy;  good precision (approximately 34 cycles)

28. 28 ◦ Case B  Nominal operation of I=4A and T=24ºC;  Non-anticipated degradation;  Increase on the ambient temperature (T=43ºC) during 25 cycles. Simulation failure antecipated from cycle 1130 to cycle 1059

29. 29 ◦ Case B  RUL estimation  good response and accuracy even with a dynamical change;  good precision (approximately 79 cycles).

30. 30 Contact: Prof. Dr. Cairo Lúcio Nascimento Júnior cairo@ita.com Eng. M.Sc. José Affonso Moreira Penna zeaffonso@gmail.com Eng. M.Sc. Leonardo Ramos Rodrigues leonardo.ramos@embraer.com.br Instituto Tecnológico de Aeronáutica