The following reasons are the most common causes for the failure of the frequency converter:
1. Improper installation:
This is often a combination of selecting inappropriate cable types, gauges or fuses and neglecting the instructions of the installation manual. Problems caused by the improper installation often become apparent early at the testing and commissioning phase.
2. Cooling fan wear:
Internal cooling fans are in constant stress in continuous use and are often the first part to fail in frequency converters. However, the failure of the cooling fan does not always mean the failure
of the whole device. The cooling fan can often be replaced with a new one but the failure always causes an unwanted downtime in the use of the drive.
3. Capacitor wear:
Capacitors wear electro-mechanically during the use. Especially electrolytic capacitors have a limited life time and age faster than dry components. Capacitors are also temperature sensitive.
High temperatures, often caused by a high current, can affect the life time of the component negatively.
4. Overuse:
The life time of the components that are used at a rating higher than its operating limit will decrease and eventually fail. Most frequently these components are located in the inverter bridge of the frequency converter.
5. Over- and undervoltage and current:
If either the voltage or the current is at the level that the frequency converter is not rated for, it is possible that the components will be damaged and eventually they will fail. Often the excess heat generated by the spikes in voltage or current is the reason for this damage. (Wilkins, 2014)
3. RELIABILITY ENGINEERING
When the equipment in the production starts failing, one way to search for the reason behind the failures is the statistical reliability analysis of the failure data. Word reliability often also includes terms availability, maintainability and safety. Availability tells how well the equipment keeps its functioning state in its environment. The more equipment is in use during its lifetime, the better the availability is. Maintainability tells how time consuming it is to maintain the equipment. The less time it takes, the better the maintainability. Safety stands for equipment’s ability not to harm anyone or anything in its lifetime. Together reliability, availability, maintainability and safety are abbreviated as RAMS. In the reliability engineering the main purpose is to develop ways and tools for assessing RAMS of components, equipment and systems. (Birolini, 2017) In pulp and paper industry all unnecessary production breaks are minimised to keep the production both cost and time effective. Integrating reliability engineering in the process helps the production to achieve this goal.
3.1. Reliability
Reliability expresses the probability for an item that ”it will perform its required function under given conditions for a stated time interval.” Reliability is normally expressed byR. Qualitatively reliability can be defined as the ability of the item to remain functional and quantitatively it tells the probability that no operational interruption will happen during a stated time interval. Reliability applies to both repairable and nonrepairable items. (Birolini, 2017) Frequency converters are sometimes repairable, for example if the smoothing capacitor fails. Sometimes the damage can be so severe that the frequency converter is beyond repair and it needs to be changed to a new one. In this thesis both repairable and nonrepairable frequency converters are researched. A common factor for inspected frequency converters in this thesis is that they have failed so that repairs or replacement are needed.
For reliability to make sense, a numerical presentation of reliability, for example R = 0.95, must be accompanied with a definition of required function, environmental, operation and maintenance conditions, mission duration and the state of the item at the beginning of the mission. (Birolini, 2017)
3.1.1. Characteristics of reliability
The simplest way to express reliability is to compare the amount of functional items with the entire item base. ReliabilityR(t) is the number of functional itemsI(t)until the momenttdivided by the whole item populationNaccording to
R(t) =I(t)
N . (1)
Reliability functionR(t)tells the probability if item, component or system will work at the timet, or that it has not failed by the timet. (Kiiski, 2012)
Failure function F(t) is a distribution function of failure probability. It tells the probability that component or system will break in a certain time. Failure function is integral of failure density within
a certain time as shown in (2).
F(t) =1−R(t) (2)
Failure density tells the statistical probability for failure within a certain time. If we test for example ten components until every component fails, and then mark the result up every day and during day three two of the components fail, the failure density for day three would be 0,2. Failure densityf(t) can be expressed as a failure functionsF(t)time derivate. Failure density is defined in accordance with (3).
f(t) =dF(t)
dt (3)
Failure rateλ(t)tells the frequency of failures in a system or a component. Failure rate can be used to deduce when systems or components lifetime is about to end so it works as a measure of the reliability of the item. Failure rate depends on the failure density f(t)and reliability functionR(t)according to (M¨akel¨a, 2017)
Exponential distribution is very popular mainly for its simplicity and it fits well in describing the reliability of complex systems especially during their operational phase. One of the exponential distribution model’s features is that item’s or system’s probability for failure is always the same regardless of its age. This means that in exponential distribution model the item’s failure rateλ is always constant. This is often not true and this is why the exponential model is suitable for describing only a small part of failure mechanisms.
Mean time to failure or MTTF gives items average lifetime expectancy before its first failure. MTTF is defined by the average value of the reliability functionR(t)which can be also be expressed as the expected value of the density function f(t)of time until failure. These formulas are shown in (6).
MT T F=
In exponential distribution model MTTF and failure rateλ are thus reciprocal numbers of each other and the failure rate is a constant. (M¨akel¨a, 2017)
MTTF is used in cases where the device is unrepairable. When the parts or components of the frequency converter are damaged, they are often just replaced with new ones.
MTBF is different from MTTF in a way that it takes into account the time used for repairs so it gives the mean time between a failure and a last time the device was repaired and put back into production. (Birolini, 2017) Repair times are often unknown for frequency converters, because they are just replaced with new ones. Failured units, if repairable, are then repaired and stored for later use.
MTBF can still give an indication of how often the devices fail on average. MTBF can be calculated if the time period (t) and the amount of failures are known (Nf) during the time period.
MT BF = t
Nf (7)
3.1.3. Weibull distribution
A wider range of different systems can be described by using the Weibull distribution which is a generalised version of exponential distribution. (Birolini, 2017) It can characterize all increasing, constant and decreasing failure rates and can be very helpful when decisions involving life-cycle costs and maintenance have to be done. With the help of Weibull analysis, the point where a certain percentage of items will have failed can be calculated. This helps to estimate when to replace the items and the developing maintenance schedules and inventories of replacement units. (Mraz, 2013) Weibull distribution has two parameters and the reliability functionR(t)is described
R(t) =e−(ηt)β. (8)
Parameterη describes the characteristic life of an item. It tells when 63.2% of the population has failed. Variablet is the time of interest for the item. When equation is solved, user can insert time into it to get the probability for the item to last that certain period of time. (Mraz, 2013)
Parameterβ is called the shape parameter because it defines the slope shape of Weibull distribution line that best fits the data points. Whenβ is smaller than one, items’ failure rate decreases with time.
This means that item will fail soon after commissioning. In this case Weibull resembles the gamma distribution. Ifβ equals one the failure rate is constant as in the exponential distribution model. If β is bigger than one, the failure rate increases with time meaning the older items are more likely to break than the new ones. In cases whereβ is two, analysis becomes the Rayleigh distribution and if β is bigger or equals three, data resembles a normal distribution. (Mraz, 2013) Examples of howβ defines the shape of Weibull reliability function are given in Figure (3.1).
One way to calculate these parameters is to linearize the Weibull distribution. This is done by taking a natural logarithm twice on both sides of (8). This is presented in (9).
ln(−ln(R(t))) =βln(t
η) =β(ln(t)−ln(η)) (9)
Figure 3.1:Different shapes of Weibull reliability function. The shape of the curve is defined by parameterβ, the shape parameter. (Reliawiki, 2019)
Selecting new variables asx=ln(t)andy=ln(−ln(R))gives a linear equation (10).
y=β(x+ln(η)) (10)
The Weibull model is then formed by calculating new variablesxandyfrom failure data. By using linear regression analysis slopeβ and constant termβln(η)can be calculated. Characteristic lifeη can be solved from constant term.
The accuracy of Weibull analysis depends on the quality, quantity and type of the data and it has several requirements to do a valid analysis. First, data must include item-specific failure data and time to failure has to be known for the population. Second, you have to have the information about the items that did not fail. Third, it is highly recommended to know failure mode root causes and separate them. However, Weibull analysis can be done even if all these points are not known. In this case analysis is not as accurate as with specific failure data but it can still be very valuable in analysing the reliability of an item.
Two different categories of data are used in a Weibull analysis which are time to failure (TTF) and censored or suspension data. In TTF data tells how long an item lasts before breaking. Censored data has failure data that has been saved over the operating period of an item and it has three different categories, right censored, left censored and interval data. Right censored data includes operating times for items that did not fail. If data is left censored, the exact time of failure is unknown. It is just known that the failure happened before it was found. Interval data includes all failures within a specific time interval but exact time to failure is not known.
Two parameter Weibull analysis is done by plotting the data manually or by software on Weibull probability paper. Failure times are ranked based on the amount of items that did and did not fail at that specific time. Most popular rankings resemble the mean and median of the data. A line that best fits the data points can be used to determine how well the Weibull analysis describes the data. This line gives item’s characteristic lifeηand the correlation coefficient of the line describes how well the line fits the data.
If the fit is not good, Weibull analysis can still give a direction to a more suitable distribution or suggest a better way to interpret the data. There are many different reasons why the fit might not be good. For example, unidentified failure modes, a change in the major cause of failures or different or changed environmental conditions all can influence the data and the fit. Plots that have the so called knees (corners) or S-shapes are the result of these problems.
When the Weibull parameters are calculated, reliability can be estimated using the reliability function R(t) or reading it from Weibull reliability plot’s y-axis. Reliability function can also be used for example to calculate when the population’s reliability is at a certain level. When the reliability is for example 90%, the calculation is called B 10 life of an item. (Mraz, 2013)