Excerpt

## Inhalt

CHAPTER INTRODUCTION

RESIDUAL LIFE PREDICTIONOFELECTRONIC COMPONENTS

RESIDUAL LIFE PREDICTION OF ELECTROLYTIC CAPACITOR

Lifetime Acceleration Factors

RESIDUAL LIFE PREDICTION OF FIXED RESISTOR

Effects of Temperature on Life

Effects of operating voltage on Life

RESIDUAL LIFE PREDICTION OF DIODE

Effect of temperature on life

Effect of voltage on life

ARTIFICIAL INTELLIGENCE TECHNIQUES USED FOR RESIDUAL LIFE PREDICTION

Artificial neural network technique

Fuzzy Inference System

Adaptive Neuro-fuzzy Inference System (ANFIS)

CHAPTER LITERATURE REVIEW

Salah-Al-Zubaidi, et.al:

Cherry Bhargava, et.al:

Salah Al-Zubaidi, et.al:

PrakashHPatil, et.al:

Zhigang Tian :

Seung Wu Lee, et.al:

Adithya Thaduri, et.al:

Garron KMorris, et.al:

Ajith Abraham, et.al:

Youmin Rong, et.al:

Prabhakar VVarde, et.al:

Vishu Madaan, et.al:

CHAPTER REVIEW OF BASIC CONCEPTS AND DEVELOPMENT OF

Introduction

Outline of proposed work

Selection of components for life estimation

Methods for estimating the remaining useful life of electronic components

Experimental method (Using ALT)

Analytical method

Residual life estimation using artificial intelligence techniques

Artificial neural network technique

Fuzzy Inference System

Adaptive Neuro-fuzzy Inference System (ANFIS)

Design of decision support system

Tools Used:

MATLAB tool

Scope of the study

Residual life estimation using artificial intelligence techniques

Artificial neural network model

Fuzzy Inference System

Adaptive Neuro-fuzzy Inference System (ANFIS)

Design of fuzzy based decision support system

Modeling of fuzzy based decision support system

CHAPTER

WORK DONE

Residual life estimation of capacitor

Life estimation of electrolytic capacitor using analytical method

Residual life estimation of capacitor using ALT (acceleration life testing) approach

Residual life estimationof resistor

Residual life estimation of resistor using Acceleration life testing

Residual life estimation of resistor using artificial intelligence modeling

Residual life estimation of Diode

Residual life estimation of diode using ALT (acceleration life testing) method

Life prediction of diode using expert artificial intelligence modeling

Design of decision support system

CHAPTER RESULTS AND DISCUSSION

Residual life estimation of capacitor using analytical method

Residual life of electrolytic capacitor obtained using artificial intelligence modeling

Residual life of electrolytic capacitor obtained using artificial neural network model

Residual life of electrolytic capacitor obtained using fuzzy model

Residual life of electrolytic capacitor obtained using ANFIS model

Comparison of output life obtained using various techniques

Life estimation of electrolytic capacitor using experimental method (using ALT)

Fuzzy based decision support system interfaces for electrolytic capacitor

Fuzzy based decision support system interfaces for resistor

Residual life estimation of diode using experimental method (ALT)

Fuzzy based decision support system interface for diode

CHAPTER CONCLUSION & FUTURE SCOPE

CHAPTER CONCLUSION & FUTURE SCOPE

CHAPTER REFERENCES

ANNEXURE

## CHAPTER

## INTRODUCTION

Residual life prediction is the technique which demonstrates how reliable a particular electronic system or component works under in specific operating conditions. The remaining useful life relies on the failure rate of a component and on the operating conditions of a device. This failure rate drifts for the duration of the life of the item with time. Life is an important aspect while choosing the electronic hardware. Residual life estimation and life prediction are two distinct terms [1] [2]. The importance of life estimation is to evaluate the remaining useful life (RUL) of a specific component under the different stress parameters.

As an increasing number of components are integrated on to a chip, the chances of failure increase, as the different parts have their own stress factors and different working conditions. So the condition monitoring strategies are utilized which enhances the reliability of a component and a suitable move to be made before any harmful breakdown happens. The electronic circuits need a failure estimation technique to protect the system from unavoidable failures [3].

### RESIDUAL LIFE PREDICTIONOFELECTRONIC COMPONENTS

Residual life estimation of electronic components is an important fact these days as electronic components and devices becomes a great need of society. Residual life prediction is predicting the remaining useful life of a component or device based on various failure factors of any component and it also depends on the operating conditions. Many methods for predicting the life of electronic components have been developed. The life of electronic components can be predicted by creating an intelligent system for the failure analysis [4] [5]. The capability to predict the life of electronic components is a key to prevent the sudden costly failure and it will increase the overall performance and reliability of a system [6]. So, remaining useful life prediction is an important factor for every active and passive electronic component such as resistor, capacitor and diode etc.

### RESIDUAL LIFE PREDICTION OF ELECTROLYTIC CAPACITOR

One of the major aspects for electronic engineers regarding capacitor is to predict its remaining useful life in order to protect it from sudden failures and prevent the complete system breakdown. The life of electrolytic capacitors is mostly relies on various environmental and electrical factors where environmental factors are temperature, humidity, atmospheric pressure and vibration and electrical factors are operating voltage, ripple current and dissipation factor. Out of these factors, temperature (ambient temperature) is the critical factor while estimating the life of aluminum electrolytic capacitors whereas; conditions such as vibration, shock and humidity have little impact on the actual life of the capacitor [7-8].

#### Lifetime Acceleration Factors

Electrolytic capacitors are by assessed by accelerated life tests. The accelerated life tests contain four components (one for temperature, voltage and ripple current) which are given by the accompanying equation [9-10]:

LP = LT * LV * LR* LVIB * LH

Where:

LP = Predicted lifetime

LT = Temperature acceleration factor

LV = Voltage acceleration factor

LR = Ripple current acceleration factor

LV = Voltage acceleration factor

LR = Ripple current acceleration factor

##### 1.2.1.1 Temperature factor

A capacitor is basically an electro-chemical gadget in which increased temperatures ranges maximize the chemical reaction rates inside the capacitor (normally with a 10°C rise in temperature, the chemical reaction rate become twice). The higher temperature ranges cause maximum changes in value of capacitance and dissipation factor because of the continuous dissipation of the electrolyte through the capacitor seal and the equivalent series resistance that is a measure of electrolyte loss also changes with change in temperature. So, the higher temperature changes affects the lifetime of a capacitor to a great extent [11].

##### 1.2.1.2 Voltage factor

The electrolytic capacitor have a specified rated voltage and while operation if the voltage more than the rated value is applied then it leads to more heat dissipation and degrades the life of electrolytic capacitor. The life of electrolytic capacitor is affected less by applied voltage than by operating temperature. Constant application of excessive voltage will quickly expand the ripple current and may cause instant damage to the capacitor [12].

##### 1.2.1.3 Ripple Current factor

Electrolytic capacitors have higher heat dissipation due to ripple current. To guarantee the capacitor's life, the greatest ripple current of the capacitor is specified [13]. At the point when ripple current flows through the capacitor and the internal heat is produced inside the capacitor with a rise in temperature. Internal heat dissipation produced by ripple current can be given by:

W = IR2 * RESR + V * IL

Where:

W = Internal heat dissipation

IR = Ripple current

RESR = Internal resistance (Equivalent Series Resistance)

V = Applied voltage

IL = Leakage current

##### 1.2.1.4 Humidity

Humidity is an important factor affecting the life of capacitor. If moisture absorbed by the sealing of the capacitor case then it leads to parametric changes (especially in the ripple current) and results in reduced lifetime and sometimes serious failure occurs due to more moisture penetration [14].

### RESIDUAL LIFE PREDICTION OF FIXED RESISTOR

The carbon film resistor is a kind of fixed resistor that uses carbon film to limit the electric current to flow through it. These resistors are generally utilized as a part of large electronic circuits. One of the significant worry for electronic specialists with respect to fixed resistor is to forecast their remaining useful life so as to preserve it from unpredictable failures and system shutdown [15]. The life of fixed resistor is relied on various environmental and electrical factors. Environmental element is temperature whereas electrical elements are working voltage and power dissipation. Out of these components, temperature is the most affected to the life of fixed resistor [16]. So, the conditions such as voltage and humidity have little impact on the resistor.

#### Effects of Temperature on Life

With the rise in operating temperature of resistor the performance of resistor starts degrading and after a specified temperature resistor stops working and may harm the complete circuit. Maximum temperature limit is defined in terms of maximum power called power rating and at highest temperature rate a resistor dissipates maximum power.

#### Effects of operating voltage on Life

The operating voltage also have a significant effect on life of a fixed resistor, When the operating voltage rise up more than the specified value, amount of power dissipation through the resistor increases and it will results in failure of resistor.

### RESIDUAL LIFE PREDICTION OF DIODE

The diode is a type of electrical component that is used to rectify the electric voltage to certain level. The diodes are used in many electronic circuits. One of the important aspect of diode for electronic industries is to predict their remaining useful life in order to save it from sudden failures and complete circuit shutdown [17]. The life of a diode is generally dependent on environmental and electrical parameters. Environmental parameters are temperature and humidity whereas electrical parameters are voltage and current.

#### Effect of temperature on life

As the ambient temperature of a diode increases the working performance of diode starts degrading and after a specified temperature diode fails and may harm the complete circuit.

#### Effect of voltage on life

The life of diode is affected less by applied voltage. When during operation at voltages above the rated voltage of a diode is applied, the internal current of diodes starts rising and more heat dissipation takes place through the diode and if the current rises to a great extent it will degrade the performance of diode completely and sometimes leads to failure of diode ct [18].

### ARTIFICIAL INTELLIGENCE TECHNIQUES USED FOR RESIDUAL LIFE PREDICTION

The actual operating life of every product depends on its real operating conditions. Hence these operating conditions must be taken into account while determining the total life and consumed life of any electronic component. A methodology is proposed for estimating the remaining useful life prediction under different operating conditions. The total life of a component is determined using different techniques for a set of operating conditions. Some of these techniques are:

#### Artificial neural network technique

Artificial Neural Network is an analogous system of human neural network which tries to mimic the functioning of actual brain. Input data along with target data has been fed to the network. Activation function has been provided to start the process where system learns by itself how output is coming. The system gets train with the number of epoch are specified. The system will train itself and reduce the error after every epoch. And hence after specific number of epoch we get the best result [20]. The number of neurons in the input layer consists of voltage, ripple current, temperature and dissipation factor which are used to obtain the life of electronic component The ANN Figure is shown below,

Abbildung in dieser Leseprobe nicht enthalten

**Fig.1.1 Artificial Neural Network**

#### Fuzzy Inference System

Fuzzy Inference System or Fuzzy Logic is used to handle Ambiguity and Uncertainty in Data. As the complexity increases we can’t make exact statement about the behavior of the system as in the traditional method we were using Binary Logic which says 0 or 1, i.e., YES or NO, but the real world problems are beyond as it can’t be TRUE or FALSE only. Taking water Problem than the possible answer could be HOT, COLD, SLIGHTLY COLD, SLIGHTLY HOT, EXTREMELY COLD, EXTREMELY HOT etc. For this purpose we deal with Linguistic variables in fuzzy which are user understandable. Entire input set is known as Crisp Set which after fuzzification converts into fuzzy sets [21]. Here we use the concept of Membership function, it defines the membership of particular input value in the fuzzy sets, and its range is from 0 to 1. If input value has complete membership, it is 1 otherwise it can be any value in this range.

In FIS we defined certain rules for fuzzification to defines crisp relation into Fuzzy arelation in IF, THEN, ELSE format, such as:

IF (F is x1, x2.an) THEN (G is y1, y2….yn) ELSE (H is z1, z2….zn)

This fuzzified data goes to decision-making unit which decides about the membership function and hence attached the related linguistic variable for that particular value. The fuzzy output from this block directly goes to defuzzifier Interface unit, which is reverse of Fuzzifier. And hence after this block we get proper output in crisp set form as defuzzifier converts fuzzy set back to crisp set. Fuzzification as well as defuzzification unit are assisted by knowledge base which has design base as well as rule base for making rules and modifying data [22]. The Block diagram of FIS is given below in Figure 1.2

Abbildung in dieser Leseprobe nicht enthalten

**Fig.1.2 Block Diagram of Fuzzy Inference System**

#### Adaptive Neuro-fuzzy Inference System (ANFIS)

ANFIS is a hybrid techniques comprises both ANN as well as fuzzy tool. It has advantage of both the technique as ANN has this self-learning mechanism but it doesn’t know how the hidden process is following to reach the particular target and the disadvantage is that the output is not that user understandable also we need very precise and accurate. It can’t handle ambiguity [23].

On the other hand the advantage with Fuzzy logic is that it can handle uncertain data and also, we use linguistic variable to have better understanding but no self-learning is there. Hence to omit each other advantages, these two techniques have been combined to formed third technique that is ANFIS (Adaptive neuro-fuzzy Inference system). Here the rules needed by fuzzy get self-updated through the self-learning mechanism possessed by ANN. That’s why less number of errors is shown by the predicted data of ANFIS [24]. The basic structure of ANFIS is shown in the figure below,

Abbildung in dieser Leseprobe nicht enthalten

**Figure.1.3 Adaptive Neuro-Fuzzy Inference System**

## CHAPTER

## LITERATURE REVIEW

### Salah-Al-Zubaidi, et.al:

In this paper, prediction as well as comparison has been made about the tool life in End Milling of Ti-6A1-4V Alloy by taking multiple regression technique along with artificial neural network technique. It has been discussed that the traditional methods for prediction could not handle the non-linearity in data which acts as drawbacks and then move towards using AI techniques which draws the non-linearity very well. Regression model is being generated using SPSS software which given the mathematical equation showing the link between multiple inputs and providing the output. It is a supervised learning method which is feed with input and output both and the system is trained to give desire output by training or learning. As a result, it is found that the ANN model is better than regression model as the accuracy for former is 92% and latter is 86.53%. Dealing with linear data is the prime advantage of this technique [25].

### Cherry Bhargava, et.al:

In this paper, need of reliability has been discussed that’s why it has become one of the ace research engineering area now a days. For electronics market low cost and high performance are the most desirable characteristics and they are somewhat highly dependent reliability for that particular system. Various techniques has been discussed for failure prediction like empirical method (based on MIL-HDBK-217, Bell core etc.), analytical methods (based on fault-tree analysis, finite element method etc.), theoretical methods (based on physics of failure model, Real time prediction model etc.), Based on soft computing (based on neural network, fuzzy Logic, genetic etc.) and analysis based on testing (here, the electrical, thermal, monte carlo analysis) are made. Brief introduction of all these techniques has been given along with examples. Conclusion has been made that empirical method MIL-HDBK still has been considered as a standard but they consider failure model under constant hazard rate only but component load profile is not considered [26].

### Salah Al-Zubaidi, et.al:

In this paper, an intelligent modeling has been done on milling tool Ti6Al4V using adaptive neuro-fuzzy inference system. Practical data of 14 sets has been extracted about cutting speed, feed rate and depth of cut. Surface roughness has been taken as output parameter to develop intelligent model. Different assumption are made when ANFIS parameters are taken for PVD coated carbide and another for uncoated carbide. Also, different number of epochs is taken to determine which will give best result based on their implementation. Obtained result says the model 1 with 100 numbers of epochs has given best result [27].

### PrakashHPatil, et.al:

In this paper, an intelligent greenhouse monitories set is proposed using GSM. It’s an effective and smart way to avoid or abolish manual labor from the field, which sometime become challenging due to large area. The parameters like temperature, humidity, light and CO2 has been taken for monitoring through sensors (Temperature sensor-LM35, Humidity sensor-HSM-20G, Light Dependent resistor (LDR) etc.) and its equivalent digital is taken through analog to digital converter and fed to microcontroller as input. Microcontroller is programmed to generate predefined specific output in terms switching ON Fan through relay 1 if temperature is high, glowing of bulb/LED through relay and if light intensity is low, starting sprinklers if humidity or moisture level is less amongst the plants through relay 3 and Switching ON Ventilators if CO2 level is high through relay4. Also, one of the inputs is connected to GSM module; it needs SIM card to generate an SMS for the user. Entire system is feed with predefined range of working parameter beyond which they raise the alarm [28].

### Zhigang Tian :

In this paper, a new approach has been taken to implement ANN model for finding RUL (remaining useful life) of any system. The data implemented is taken from pump bearings in the site. Traditionally only failure data was taken but here suspension data of any system is also taken as it greatly affect the operational life. Suspension data is the log which is been created when system is non-operational for maintenance and actually not fail. The real challenge in this paper was to generate data log for suspension data. The Artificial neural network has been used to train the system with failure data as input and the percentage of RUL as output. The two hidden layer with two neurons each has been used. The parameter for performance is not MSE (Mean square error) but validation data set is used to do so. In the entire data set two –third is used for training and one-third is used for simulation as sample (Validation-set). ANN network used is resilient back propagation network and for better performance, it has been trained 30 times [29].

### Seung Wu Lee, et.al:

In this paper, the reliability for electronic component has been discussed based on various aspects and failure models. Reliability is not only the function of components parts, or subsystem’s parameters but also depends on the working or operational environment. So far, many methods have come into limelight for predicting the reliability, but the main factor is still the failure model we are considering. On general purpose, there are many methods such as MIL-HDBK217 N2, PRISM, Telcordia, NTT procedure etc. MIL-HDBK217 is highly used for electronic components. Telcordia is used for the serially oriented blocks to predict reliability and PRISM mainly for military purpose equipment. Here, three methods have been taken for comparison over their result for MTBF i.e. Mean time between failures. It provides the reliability result to the user without any need to use any applied program just by having access to internet. Another is through using numerical method based on the model provided by the MIL-HDBK217, which is having genre like microcircuit, diodes etc. out of which MMIC from micro sub circuits category is taken. The third one is through software named Relex has been taken which is a commercial link available for reliability. The result through Developed Web, MIL-HDBK217 and Relex were almost same i.e., MTBF=0.169 failures/million hours on an average. Overall, the developed web system is expected to contribute more because of the consideration of small and medium sub circuit or component which is not possible in case of general purpose system due to financial aspect [30].

### Adithya Thaduri, et.al:

In this paper, two different reliability prediction models has been used for determining MTTF(Mean Time To Failure) for instrumentation amplifier and BJT and their comparison has been done. Also the analysis has been used to determine the estimate cost of reliability prediction using both methods. For traditional approach, RIAC 217+ has been used to determine reliability using numerical method and here temperature is the considerable parameter. All the constant values have been taken through this handbook only. Same method has been done for BJT Transistor also. Another method is physics of failure method where analysis of stressor parameter is needed to determine exact failure result. Testing is done at physical level (Accelerated stress testing) to record the minimum or maximum value which could lead to failure. In comparison, the better result is achieved using physics of failure method for both the components. For cost analysis, the parameter depends on the cost of reliability Handbook for traditional method and the circuit and setup cost for physics of failure method. It is a variable parameter depending on the characteristics of the system [31].

### Garron KMorris, et.al:

In this paper reliability of printed circuit board has been examined under Environmental condition. If it will be observed that working conditions for practical ground is very different from what has been specified theoretical. Proper analysis of printed circuit board is strictly needed in early ages as if it has been fabricated nothing can be undoing. Environmental temperature is one of the factors which highly deviates the operation of any practical system. Reliability analysis done here is through Weibull analysis and MIL-HDBK 217 F for predicting exact failure model for future uses also. Final prediction is done using Monte Carlo Analysis which gives error of 22%, for final reliability model of printed circuit board [32].

### Ajith Abraham, et.al:

In this paper, one of the soft computing technique termed as “Genetic Programming” has been taken to draw intelligent model for monitoring online failure system for electronics system. Stressor is the name given to the parameter whose variability tends to change the reliability or behavior of the system or may even failure. Susceptibility is the property of the system that how susceptible it is for stressor parameter. In this paper, stressor- susceptibility relation has been used to draw the data set for empirical model. Also, experimental value has been compared with genetic programming predicted values [33].

### Youmin Rong, et.al:

The aim of this paper is to optimize seam shape in this process using back propagation neural network (BPNN) and genetic algorithm (GA). After performing experiment through Taguchi method, the included parameter for input is welding speed, gap and welding speed rate. The experimental data set is fed to BPNN for predicting the output and that output is fed to genetic programming so that through optimization the error percentage may get reduce. The final output of this BPNN-GA technique is very effective [34].

### Prabhakar VVarde, et.al:

In this paper lifetime extension methods using early failure and early maintenance warning has been discussed. This early warning method also tend to increase the life cycle of the products as if before failure warning has been generated than necessary measures could be taken to extend the life like repair or replacement of any component. Approaches used here involve sensing the performance, operational environment, remaining useful Life (RUL), health diagnosis has been used. This kind of technique is very useful for manufacturing industries as it develop an intelligent as well as reliable system for maintenance [35].

**2.12 Yang Zhao, et.al. (2011):** In this paper, quantitative analysis has been discussed for reliability prediction and assessment of electronic system. Here analysis on reliability is done to implement better fault testing, health management and maintenance system. During analysis knowledge of hardware as well as software system is needed. Quantitative analysis is mainly a prognostic health analysis for reliability assessment of electronic systems [36].

### Vishu Madaan, et.al:

This paper illustrates about the fuzzy logic based medical expert system that analyses the diseases associated with blood. Specialists and doctor can use this framework to help them in their daily life. The proposed expert medical system also enables the doctor and specialists to seek after the best treatment to analyze the disease. The expert system was designed using MATLAB GUI in which the fuzzy expert system was designed first and used to implement the final expert system user interface. This system can be designed using neuro-fuzzy technique to obtain the more accurate results.

## CHAPTER

## REVIEW OF BASIC CONCEPTS AND DEVELOPMENT OF

**METHODOLOGY**

### Introduction

The goal of residual life prediction is to estimate the reliability and lifetime of electronic parts and electronic circuits. Due to advancements in technology, in the need of high integration and fast speed manufacturers are not concentrating much on the residual life estimation of small electronic components used in the design of complete system or circuit while designing an electronic gadget and device. Residual life estimation is a prediction methodology to prevent the system from accidental failure and discrepancies. Many parameters affects the behavior of electronic circuits and must be considered for improving actual performance for example, design parameters and working parameters. By observing the performance of a component under various operating conditions, real-time failure prediction of electronic components can be accomplished. Various methods can be used for collecting failure data of different active and passive components such as capacitor, resistor and diode by observing the output of an individual component at various operating conditions beyond and between the specified values.

Abbildung in dieser Leseprobe nicht enthalten

**Figure 3.1** Bath tub curve for component life

Residual life estimation is the general term in our day-to-day life which means how much we can rely over that particular thing. Reliability is the degree which tells how reliably a particular electronic system or component will work as it is expected to, in the specific or desired duration.”

Now operational reliability includes three parts:

a) Reliability, whether system is giving desired output or not.

b) Reliability, whether it’s working fine under different environment.

c) Reliability, whether different conditions are not fluctuating the output.

In the deep study of Reliability there comes the term “Life Estimation”. Life estimation is to estimate the remaining useful life of a component in the desired operating condition. In this work, the term residual life estimation is defined as actual life of a component under the working conditions in use. Here, residual life of an electronic component is estimated using different techniques under different operating conditions. Remaining useful life can be given as:

LRUL = LE - LU

Where, LRUL is remaining useful life

LE= Estimated life,

Lu= Used life

Abbildung in dieser Leseprobe nicht enthalten

**Figure 3.2** Remaining useful life estimation of a component

### Outline of proposed work

Residual life prediction of electronic components is a great need of electronic society as the electronic devices are becoming integrated and high speed. The remaining useful life of a electronic components or devices depends on various failure factors of any component and on the operating conditions. Various techniques for predicting the life of electronic components have been developed. The proposed methodology estimates the lifetime of active and passive electronic components such as capacitor using various techniques such as analytical method (using lifetime acceleration factors), experimental method (using acceleration life testing) and artificial intelligence techniques such as ANN, Fuzzy and ANFIS.

#### Selection of components for life estimation

There are many electronic components used in high speed electronic devices nowadays. All these .components have some residual life that depends on the operating conditions of a component. In this work, basic components such as electrolytic capacitor, fixed resistor and diode are selected as these components are the part of many high speed electronic devices. So, life estimation of these components is necessary in order to prevent the complete device from a serious breakdown and costly failures.

Abbildung in dieser Leseprobe nicht enthalten

### Methods for estimating the remaining useful life of electronic components

The life estimation model is developed using various methods such as experimental method, analytical method and artificial intelligent model. The experimental method is the proposed method to predict the remaining useful life of any electronic component.

#### Experimental method (Using ALT)

The numerical models are developed for the life estimation or failure estimation which provides better results and this mathematical model comes under analytical methods for example FEM stands for finite element methods where we look for the failure In this method, lifetime is estimated using the ALT (Acceleration life testing). ALT is the process of testing the component under the stress factor temperature in order to find the failure rate and life of a component. This method is mainly used by big manufacturing unit where large number of samples or electrical units is subjected under tests. These tests can be environmental, electrical and thermal. One of the examples of such testing is accelerated life testing (ALT) using temperature. The quality of a component during the life test was observed by regular checking of the value of the component by digital multimeter. The testing executes s of in following steps:

a) In the first step, the components are placed on the hotplate, the value of each component is measured and the desired temperature level is set in the hot plate. Wait till the temperature achieves the maximum rated temperature.

b) The trial was executed for specified time. This time length is chosen according to various temperature ranges. Time interval must be shorter at higher temperature as chances of failure of components are more as compared to lower temperature limit.

c) Measure and note the value of every components after few hours and check how many components got failed after few hours and calculation of the output life is done.

d) After collection of failure data, Life was calculated using Arrhenius equation given below:

** Life = 1/TDH * AF (3.1)**

Where, AF is failure rate which is given by:

Abbildung in dieser Leseprobe nicht enthalten

**Figure 3.4** Capacitors kept at different temperature ranges for ALT

#### Analytical method

In this method, many factors that affect the residual life of a capacitor such as environmental factors include temperature, humidity and vibration whereas electrical factors include voltage, ripple current and ESR has been analyzed and life was estimated based on these factors. In the previous methods only two or three factors were considered as reviewed in literature. In this work, all the factors are considered in order to predict the lifetime of a capacitor.

**3.3.2.1 Lifetime acceleration factors**

Electrolytic capacitors are assessed by different lifetime acceleration factors. These factors contain many components such as temperature, voltage, ripple current, frequency, humidity and vibration [9-10]. The residual life can be estimated using following equation:

** LE = LY * AT * AV * AR *AF * AVIB * AH (3.3)**

Where:

LP = Estimated lifetime

LY = Lifetime (datasheet)

AT = Temperature acceleration factor

AV = Voltage acceleration factor

AR = Ripple current acceleration factor

AF = Frequency acceleration factor

AVIB = Vibration acceleration factor

AH = Humidity acceleration factor

###### 3.3.2.1.1 Temperature factor

A capacitor is basically an electro-chemical gadget in which increased temperatures ranges maximize the chemical reaction rates inside the capacitor (normally with a 10°C rise in temperature, the chemical reaction rate become twice). The higher temperature ranges causes maximum changes in value of capacitance and dissipation factor because of the continuous dissipation of the electrolyte through the capacitor seal and the equivalent series resistance that is a measure of electrolyte loss also changes with change in temperature. So, the higher temperature changes affects the lifetime of a capacitor to a great extent [11]. Temperature acceleration factor is given by:

** (3.4)**

Where, is temperature acceleration factor

To is maximum temperature that is 85ºC

Tu is used temperature

###### 3.3.2.1.2 Voltage factor

The electrolytic capacitor have a specified rated voltage and while operation if the voltage more than the rated value is applied then it leads to more heat dissipation and degrades the life of electrolytic capacitor. The life of electrolytic capacitor is affected less by applied voltage than by operating temperature. Constant application of excessive voltage will quickly expand the ripple current and may cause instant damage to the capacitor [12]. Voltage factor is given by:

** (3.5)**

Where, Av is Voltage acceleration factor

Va is Applied voltage

Vm is Maximum voltage that is 50V

###### 3.3.2.1.3 Ripple current factor

Electrolytic capacitors have higher heat dissipation due to ripple current. To guarantee the capacitor's life, the greatest ripple current of the capacitor is specified [13]. At the point when ripple current flows through the capacitor and the internal heat is produced inside the capacitor and the performance of capacitor starts degrading, when the internal heat exceeds beyond the limit, the capacitor gets fail. The acceleration factor for ripple current is given by:

** (3.6)**

Where, Ai is empirical safety factor which is defined as:

If To is 85°, then Ai is 2

If To is 105°, then Ai is 4

Ia is ripple current in application

Io is ripple current at maximum temperature that is 4.7mA

ΔTo is rise in core temperature of capacitors

i.e. 5 K at T0 = 105 °C

10 K at To = 85 °C

###### 3.3.2.1.4 Humidity factor

Humidity is an important factor affecting the life of capacitor. If moisture absorbed by the sealing of the capacitor case then it leads to parametric changes (especially in the ripple current) that produce heat and results in reduced lifetime and sometimes serious failure occurs due to more moisture penetration. Humidity acceleration factor is given by:

** (3.7)**

Where, C is the constant

is the applied relative humidity

is the maximum specified relative humidity that is 85%

###### 3.3.2.1.5 Vibration factor

Vibration is another acceleration factor that affect the lifetime of capacitor. If a capacitor goes throw more vibration rate then the capacitor go through the complete breakdown otherwise the vibration have a little impact on life of capacitor. The vibration accelerator is given by:

** (3.8)**

Where, Vm is maximum vibration rate that is 30Hz

Va is applied vibration rate

βis fatigue parameter whose value is 8 or 9

###### 3.3.2.1.6 Frequency factor

The frequency factor also affects the life of capacitor. The frequency exceeds as the internal ripple current starts rising. The frequency acceleration factors are given by:

** (3.9)**

Where, ESRM is ESR at specified frequency (50/120 Hz)

ESRA is ESR at applied frequency

### Residual life estimation using artificial intelligence techniques

#### Artificial neural network technique

Artificial Neural Network is an analogous system of human neural network which tries to mimic the functioning of actual brain. Input data along with target data has been fed to the network. The system gets train with the specified number of epoch. The system will train itself and reduce the error after every epoch and hence after specific number of epoch we get the best result [20]. The number of neurons in the input layer consists of input parameters which are used to obtain the life of electronic component.

#### Fuzzy Inference System

Fuzzy inference system is used to handle ambiguity and uncertainty in data. As the complexity increases exact statement cannot be made about the behavior of the system as in the traditional method Binary Logic which says 0 or 1 is used that indicates YES or NO, but the real world problems are beyond as it cannot be TRUE or FALSE only. For this purpose we deal with Linguistic variables in fuzzy which are user understandable. Entire input set is known as crisp Set which after fuzzification converts into fuzzy sets [21]. Here we use the concept of Membership function, it defines the membership of particular input value in the fuzzy sets, and its range is from 0 to 1. In FIS certain rules are defined for fuzzification that defines crisp relation into fuzzy relation in IF, THEN, ELSE format, such as:

IF (F is x1, x2.an) THEN (G is y1, y2….yn)

These rules are used to obtain the output life.

#### Adaptive Neuro-fuzzy Inference System (ANFIS)

ANFIS is a hybrid techniques comprises both ANN as well as fuzzy tool. It has advantage of both the technique as ANN has this self-learning mechanism but it has the disadvantage that is the output is not that user understandable. It can’t handle ambiguity [23]. On the other hand the advantage with Fuzzy logic is that it can handle uncertain data and also, linguistic variable can be used to have better understanding but no self-learning is there. Hence to combine each other advantages, these two techniques have been combined to formed third technique that is ANFIS (Adaptive neuro-fuzzy Inference system). Here the rules needed by fuzzy get self-updated through the self-learning mechanism possessed by ANN. That’s why less number of errors is shown by the predicted data of ANFIS [24].

### Design of decision support system

The fuzzy based decision support system is designed. In this system, an interface is designed. Designing of graphical user interface (GUI) is the last phase of this work. Using the interface, a user can interact with the expert system to check the operating condition of capacitor. The failure parameters of capacitor are given to this expert system as an input and the expert system gives the output as the remaining useful life and also the interpretation according to the range of output life such as component is good or not.

### Tools Used:

#### MATLAB tool

MATLAB tool is a superior instrument for designing and simulating the artificial intelligence systems. It incorporates computation, perception, intelligence and programming condition. It has information structures, contains worked in altering and troubleshooting devices and backings protest arranged programming. These elements make MATLAB tool a magnificent device for instructing and research. It is an intelligent framework whose fundamental information component is a cluster that does not require dimensioning. It has capable inherent schedules that empower a wide assortment of calculations. Particular applications are gathered in packages alluded to as toolbox. There are toolboxes for data handling, typical calculation, control hypothesis, reproduction, enhancement, and a few different fields of connected science and designing. MATLAB tool involves the Test system and additionally programming window, utilized for calculation, recombination, modeling GUI (Graphical user interface). It is profoundly utilized as a part of the field like communication, machine learning, reliability prediction, digital signal processing etc. MATLAB simulink is an incorporated window contains numerous Interfaces like ANN, Fuzzy, Neuro-Fuzzy and so on. The programming Interface has editor Window to compose the program. In the proposed methodology machine learning techniques like ANN, Fuzzy and ANFIS are applied on formulated data that is obtained from experimental, and analytical analysis will be used for predicting the residual life of electronic components such as capacitor.

**[...]**

- Quote paper
- Cherry Bhargava (Author)Shivani Gulati (Author), 2017, Remaining Useful Life (RUL) Prediction of electrolytic Capacitor using Artificial Intelligence, Munich, GRIN Verlag, https://www.grin.com/document/427692

Comments