# Introduction lectrical companies nowadays are facing a lot of pressure considering equipment maintenance or replacement on the one hand and reducing operating expenses on another hand. Maintaining old equipment can be an expensive task, and that's why power network operators should create a strategy of a most cost-effective method of equipment maintenance or replacement. The same situation about equipment maintenance is happening at the Power Industry of Serbia. Among other equipment circuit breakers (CB) are a matter of concern, because most of CB's that are currently in operation are installed during 70es and 80es (minimum oil CB's), which means that they are at the end of their life, which is period characterized by increased number of faults and consequently increased maintenance. Findings in this paper represent continuing of CB's RUL assessment [1]. After gathering recent data, it is useful to investigate results from previous research and try to make new conclusions. In previous research [1], using Weibull distribution we determine CB's probability of failure by analyzing voltage drop values on its contacts, and using Poisson distribution the probability of failure if the number of short circuit trips exceeds limit value. Both distributions were already used in literature and research for similar problems. In [2], Weibull distribution was used for statistical analysis of age or wear out related CB faults. In [3],Poisson distribution was used for modeling component faults in the power system with statistical data from maintenance and repairs. In [4] presents analysis of fault types and their consequences, with cost structure and maintenance strategies. During wear out period fault intensity of high voltage (HV) CB's follows Weibull distribution. In [5], analysis of SF6 and minimum oil, CB's faults were performed. Research includes totally 1546 CB's from the Swedish and Finland power systems. Weibull distribution is assessing the RUL of CB's components, which were the source of the fault. In [6] they use the same distribution for reliability, RUL, and fault intensity assessment of HV SF6 CB's. In [7], few modified models of Weibull distribution were purposed for equipment reliability assessment in the power system. Least Squares method estimates parameters of Weibull distribution. In [8] they use the same method for parameter estimation, where researchers are creating transformer lifetime model with Weibull distribution based on condition monitoring data. # II. # Weibull Distribution Assessment Basic recommendations when choosing distribution are following [9]: ? Choose distribution, which researchers most frequently use in the same field of work. ? Choose distribution, which gives the most conservative results. ? Choose a simpler type of distribution. For example, if two-parameter distribution gives similar results like three-parameter distribution, then two-parameter distribution should be used. Researchers deploy Weibull distribution very often when equipment aging and reliability has to be analyzed [10]. Weibull distribution can describe three types of equipment states (infant mortality, a period of normal work, wear out state) through the bathtub curve [11]. Weibull cumulative distribution function represents the probability of failure in a given period (1). In this case, two-parameter distribution was used, which consists of slop parameter (?) and shape parameter (?). One of the main goals is to conclude whether new data follow Weibull distribution and is it justifiable to use it for this type of RUL assessment. The calculation covers CB's in 5 different categories (considering feeder type and rated voltage) and with two subcategories (1. Normal voltage drop value, 2. Permissible voltage drop value is by 25% larger [12]), making that way ten different categories in total. By this categorization, we can observe RUL more clearly, and come to the conclusion what makes the greatest influence on CB's aging process. Minitab 17 software and least square method [13] calculates Weibull distribution function with rightcensored data (case when some devices didn't fail during the period of analysis) for all CB's categories. For old and new data following values were calculated and compared: Weibull parameters and correlation coefficient. By observing the results of calculated correlation coefficient, it is obvious that with an increased number of the data correlation coefficient is becoming greater, which means that data are becoming closer to Weibull distribution. Next, Weibull parameters (scale parameter and shape parameter) were calculated with new data and compared with old ones (Table 2). By observing Weibull parameters from table 2, two conclusions could be made (taking into account the results from a previous paper [1]); underground feeders (both criteria of voltage drop value limit) have the highest ? while overhead feeder has the lowest value. Considering ? parameter, 10 kV feeders (+25% limit voltage drop level) have a longer time to failure, while 35kV feeders have the lowest ? value (they will fail sooner than 10kV CB's). In Table 2, values are showing expected aging phenomena. The number of failed CB's is increasing, but on the other hand, with a greater number of data, a new insight could be perceived. Scale parameter (?) is, in most cases, slightly increased, which suggests that RUL is not as we were expecting with old data and that CB's survival time is slightly greater compared with previous research. # III. Poisson Distribution Assessment Poisson distribution [14, 15] is the discrete distribution used for modeling a number of events which are appearing in a specific period. Poisson distribution for calculation probability for a known number of past events (k) in the time interval (t) is [16]: ??(??) = (????) ?? ?? ????? ??! Where is: ?? -the number of faults in period (t) ?? -fault intensity ?? -a time interval ??(??) -the probability of appearing r number of faults in period t Cases of Poisson distribution use [17]: 1. Researcher can present an event with the whole number 2. The occurrence of an event doesn't depend on any other event 3. Mean value of event occurrence in a specific period is known 4. Number of events is countable In the power system, Poisson distribution can predict faults such as short circuit faults. The number of those faults depends on feeder type (underground or overhead) and also by the area configuration where power network is situated (residential area, forest). Another influencing factor is weather condition and power network quality. # Procedure for Chi-Square Goodness of Fit Test One of the methods for determining are date follow Poisson distribution is the Chi-squared test ( ?? 2 test). This method represents a test of statistical hypothesis and is used to determine a significant difference between expected and observed intensity Table 3 presents a number of short circuit trips on one 10kV feeder. The mean of the Poisson distribution is: ?? = (0 ? 5 + 1 ? 2 + 2 ? 0 + 3 ? 0 + 4 ? 0 + 5 ? 0 + 6 ? 0 + 7 ? 0 + 8 ? 0 + 9 ? 0 + 10 ? 0) 7 ?? = 0.2857(4) Example ( 5) and 6) are presenting expected fault intensity calculation, and the table 5 presents values of that calculation. ?? 0 = ??(?? = 0) = ?? ?0.2857 (0.2857 ) 0 0! = 0.7515(5) ?? 0 = 0.7515 ? 7 = 5.26 Degrees of freedom are ?? ? ð??"ð??" ? 1. In this case number of classes is?? = 11 (number of faults intensity), and from data we estimate one parameterð??"ð??" = 1(in this case one parameter, ?). In the end, degrees of freedom are equal to 11-1-1=9. Value of significance level is selected to be 0.05. That value means there is a 5% probability that the observed relationship between variables exists by coincidence [22]; in other words, data doesn't follow assumed distribution [23]. This value shows at which ?? 2 value H0 hypothesis is acceptable. [24] If ?? 2 < 16.92(illustratedin figure 2) than the H0 hypothesis is acceptable, which means there is no evidence that the data doesn't follow Poisson distribution. 7. Results are showing that in most cases (89%), data are following Poisson distribution. Analysis of short circuit faults will continue in the future periods to determine will the bigger amount of data increase fit to Poisson distribution. # IV. # Conclusion In this paper, new data are used to check the correctness of methods used in previous research. New This paper proves that it is justifiable to use Weibull and Poisson distribution for CB's remaining useful life estimation. With these two methods, CB's RUL could be calculated very fast and easy which could be later used for other studies such as risk assessment, power station reliability assessment, determining critical points in the power system, or justification of CB replacement. Research in this field will be continued by gathering data from other power operators in the Power Industry of Serbia to better understand the problem of the CB aging process by using voltage drop values and short circuit faults. ![the time at which 63.2% failure rate behavior and tells whether failures are decreasing or increasing. Shape parameter value has the following meaning: ?<1 indicates infant mortality, ?=1 period of normal work ?>1 wear-out failures. The higher value of beta indicates a greater rate of failure. a) Data analysis with Weibull distribution One power company (which is part of the Electrical Industry of Serbia) owns all CB's which are part of analysis, and the same specialized work force is maintaining and monitoring their work years back. Totally 427 CB's were part of the analysis, and their monitoring process started in 2007. This research consists of two separate periods. Data for the first research are including period of 10 years (2007-2017), after that, next research new data from the last three years (2017-2020).](image-2.png "") 1![Figure 1: Example of Weibull distribution graphs and values for 10kV feeders +25%](image-3.png "Figure 1 :") 23![[18].The Chi-squared test can test hypothesis do analyze data follow a certain distribution. It can also test Poisson distribution[19]. The calculation is carried out in the following way [20, 21]:?? 2 = ? (?????)Where is:?? 2 -Chi-squared value O -observed value E -expected value 1. Hypothesis formulation: a. Null hypothesis: ?? 0 : data Xis following Poisson distribution-??~?????????????? b. Alternative hypothesis: ?? 1 : data X doesn't follow a Poisson distribution 2. Calculation of Chi-squared test:](image-4.png "2 ?? ( 3 )") 1Feeder typeCorrelation coefficient until 2017 yr. until 2020 yr.Overhead +25%0.9850.986Overhead0.9930.997Underground +25%0.9760.984Underground0.9650.97710 kV feeders +25%0.9880.99510 kV feeders0.9890.99235 kV feeders +25%0.9720.97135 kV feeders0.9840.989All feeders +25%0.9890.988All feeders0.9900.993 2Old dataNew dataFeeder type??Failed \ suspensions??Failed \ suspensionsOverhead +25%39.09 5.147100/8739.425.069111/78Overhead37.08 4.797131/5637.424.935141/48Underground +25%41.54 6.05563/16944.525.26866/167Underground38.09 6.07097/13540.235.490101/13410 kV feeders +25%43.44 5.62787/22445.505.10097/21510 kV feeders40.39 5.071135/17642.004.918142/17235 kV feeders +25%35.24 5.59379/3135.785.41980/3035 kV feeders33.83 5.61596/1434.145.66299/11All feeders +25%40.37 5.582166/25541.775.206177/245All feeders37.98 5.281231/19039.165.134242/182 3YearNumber of trips20130201402015020160201702018120191Using values from table 3, we calculate eachfault intensity (Table 4). 5Numberof faultsPoissonObservedExpected (?? ?? )(k)00.751555.260310.214721.503020.030700.214730.002900.020440.000200.001550.000000.000160.000000.000070.000000.000080.000000.000090.000000.0000>100.000000.0000Calculation of Chi-squared value:?? 2 = ?(?? ? ??) 2 ??=(5 ? 5.2603) 2 5.2603+(2 ? 1.5030) 2 1.5030+(0 ? 0.2147) 2 0.2147+(0 ? 0.0015) 2 0.0015??? 2 = 0.4139 6 7Number of CB's where H 0Number of CB'shypothesis is acceptedwhere H 0(data follow Poissonhypothesis isdistribution)rejected1481989 %11 % F © 2020 Global Journals Analysis of Weibull and Poisson Distribution use in Medium Voltage Circuit Breakers RUL Assessment * Methodology for circuit breakers remaining useful life assessment AleksandarDragan Stevanovi ? DraganJanji? Tasi? Tehnika 0040-2176 5 2019 Union of Engineers and Technicians of Serbia * Reliability-Based Preventive Maintenance of Oil Circuit Breaker subject to Competing Failure Processes FIberraken RafMedjoudj Rab Medjoudj Dj KDAissani Haim International Journal of Performability Engineering 9 5 Sep. 2013 * Predicting the failures of transformers in a power system using the poisson distribution: a case study AÜnsal BMumyakmaz NSTunaboylu ELECO 2005 International Conference on Electrical and Electronics Engineering Bursa 2005 * Applications of High Voltage Circuit-Breakers and Development of Aging Models MPSc Choonhapran 2007 Technischen Universitat Darmstadt * Circuit breaker failure data and reliability modelling TMLindquist LBertling REriksson IET Gen., Transm. & Distrib 2 6 Nov. 2008 * Reliability estimation of high voltage SF6 circuit breakers by statistical analysis on the basis of the field data XZhang EGockenbach ZLiu HChen LYang El. Power System Research 103 Oct. 2013 * Combining Modified Weibull Distribution Models for Power System Reliability Forecast MDong ABNassif IEEE Transactions on Power Systems 34 2 March 2019 * Transformer lifetime modelling based on condition monitoring data DZhou International Journal of Advances in Engineering & Technology 2231-1963 May 2013 * Failure Rate Analysis of Power Circuit Breaker in High Voltage Substation TSuwanasri MTHlaing CSuwanasri GMSARN International Journal 8 2014 * Operating manual for medium voltage medium oil circuit breakers for internal assembly JZSikorska MHodkiewicz LMa Mech. Syst. and sign. Process 25 5 Jul. 2011. 12. 1984 Prognostic modelling options for remaining useful life estimation by industry * A Probabilistic Analysis Approach to Making Decision on Retirement of Aged Equipment in Transmission Systems WLi JZhou JLu WYan IEEE trans. on power delivery 22 3 July 2007 * Predicting the failures of transformers in a power system using the poisson distribution: a case study AÜnsal BMumyakmaz NSTunaboylu ELECO 2005 International Conference on Electrical and Electronics Engineering Bursa 2005 * The Chi-Square Goodness-Of-Fit Test for a Poisson distribution: Application to the Banking System MSulaimon OOlutayo Statistics: 1.4 Chi-squared goodness of fit test, Mathematics learning support centre Loughborough university, Leicestershire, UK Apr. 2016 4040 03 Issue 08 * Ani Grubi?i? -Chi-squared test and its application 2004 Faculty of Electrical Engineering and Computing, University of Zagreb