Xin Q. Diesel Engine System Design

Подождите немного. Документ загружается.

46 Diesel engine system design

to its specication as manufactured, while reliability denotes its ability to

continue to comply with its specication over its useful life. Reliability is

therefore an extension of quality into the time domain’. Such a denition

of quality is suitable for diesel engine system design. More details about

quality and reliability engineering are provided by Chandrupatla (2009).

The quality of an engine product, as designed and manufactured, refers

mainly to its consistency of conformance to the customer’s requirements

in performance, durability, and packaging attributes. As illustrated in Fig.

1.14, quality is an intermediate design goal of engine system design, and

it assembles and measures all three design attributes. Quality ultimately

extends or evolves to reliability in the time-in-service domain. During the

course of diesel engine system design, the measurement and assessment

of quality in performance, durability, packaging and their synthesis should

not be overlooked. The quality target can be used as an objective in design

optimization.

The demand for a product is usually affected by attributes and price

(Hazelrigg, 1998). The quality and reliability of engine performance,

durability and packaging features directly affect the brand image and product

demand of diesel engines. It should be noted that sometimes people think

a reliability problem is such a serious one that the product cannot function,

while a quality problem is often regarded as a less serious one that the

product can still function but with nuisance. That is a misunderstanding

and wrong perception of the concepts of reliability and quality from an

engineering sense. In fact, the severity of a problem can be characterized

by a cost function of quality loss. A small loss of quality may cause a small

cost to the customer without affecting the basic use of the product, while a

large loss of quality at a moment of time in service during the lifetime of

the product may cause a catastrophic failure of the function. Both scenarios

present a reliability problem of different severities.

Before Dr Genichi Taguchi developed the methodology of continuous

quality loss function, in the traditional discrete ‘pass–fail’ quality theory

where samples were viewed as either pass or fail, a design within the range

of specication was viewed as equally good, while any design outside the

specication was viewed as equally bad (Fig. 1.14). However, the quality

perception by the customer is not so simple. A product that is barely within

the specication is certainly not as good as a product that is perfectly on the

target. Therefore, quality loss, or functional performance loss, should be a

continuous and gradual event with respect to the functional performance rather

than a discrete or step event depicted in the traditional quality theory.

Quality engineering or robust design is a process to obtain the response that

is insensitive to noise factors. Robust design requires a quantitative denition

of quality. As mentioned earlier, the loss of quality is actually continuous

rather than discrete and sudden when the product quality deviates from the

Diesel-Xin-01.indd 46 5/5/11 11:41:26 AM



Probability

Design B

Not equally good

Speciﬁ cation limits

Design A

Very bad

Nominal

target

Nominal

target

Nominal

target

Equally good (for

Design B and the

Design A within

spec limits)

Functional

performance (Y)

Functional

performance (Y)

Functional

performance (Y)

Lower bound

Speciﬁ cation limits

Quality loss function:

: functional tolerance

: cost to functionality if tolerance is exceeded

Upper bound

uppe

()

LY()LY

uppe

–)

YY–)YY

–)

–)Y–)

Quality lossQuality loss

Speciﬁ cation limits

Equally

bad

(for

Design

outside

limits)

Equally

bad (for

Design

outside

limits)

Best

GoodAcceptable

Bad

1.14 Quality loss function.

Diesel-Xin-01.indd 47 5/5/11 11:41:26 AM



48 Diesel engine system design

nominal target due to various noise factors. The product quality is commonly

de ned by using a quality loss function (Fowlkes and Creveling, 1995). As

shown in Fig. 1.14, the two designs (A and B) have different probabilistic

distribution curves for a particular functional performance parameter. The

speci cation is de ned by a nominal target and a tolerance range for acceptable

limits from a lower speci cation limit to an upper limit. The design A has

a larger population of samples (i.e., higher probability) around the ‘good’

nominal target but widely spread samples outside the speci cation range due

to a larger standard deviation. On the other hand, the design B has a shifted

and narrower probability distribution curve which gives very few samples

meeting the ‘good’ nominal design target although all of its samples are within

the speci cation range. This example illustrates that ‘within speci cation’

and ‘on target’ do not have the same meaning. Therefore, it is important to

consider the probability distribution characteristics when evaluating different

designs.

Dr Genichi Taguchi promoted the idea of not just getting all the units

within the speci cation limits but getting all the units on target. He developed

the methodology of quality loss function to evaluate the  nancial impact of

the tolerance range. Taguchi de ned the quality loss of an off-target product

performance, due to deviation from the nominal target, as the life-cycle

monetary losses caused by functional variability and harmful side effects

related to customer functional tolerance and the cost to society. He de ned

that the losses represent a summation of rework, repair, warranty cost,

customer dissatisfaction, bad reputation, and eventual loss of market share

for the manufacturer. He de ned that the loss of quality as a cost increases

quadratically with the deviation from the target, and the quality reaches the

best at the nominal target. He used the quality loss function to quantify the

quality of a design and determine the tolerances.

Various models have been proposed to estimate the product quality loss by

using the mean value and the standard deviation of the response. Figure 1.14

and Table 1.5 show the quality loss function. Table A.1 also summarizes the

related formula used in statistics. The two basic sample statistics to quantify

variability are

and

, which describe the central location and the width

of the probability distribution. A histogram is another way to describe the

distribution of the frequency of occurrence. From the average quality loss

equation shown in Table 1.5,

target

(

LY(LY

(

LY(LY

)

· [

CY · [CY

+ (

CY + (CY

–

)]

+ (

CY + (CY

)]

1.1

where C

is a quality loss coef cient, it is observed that minimizing the

variance

(i.e., variability) or making the average response

on the

target Y

target

can reduce the quality loss. Such a method developed by Taguchi

lays out the theoretical ground of robust design.

Quality includes not only the conformance of geometric tolerances,

Diesel-Xin-01.indd 48 5/5/11 11:41:27 AM



49The analytical design process and diesel engine system design

but also the conformance of functional requirements. For example, a

television’s quality can be measured by its appearance and the sharpness of

the image. For diesel engine system design, engine product quality refers

to the conformance to the requirements of a combination of performance,

durability, and packaging. The quality loss function developed by Taguchi

is most useful to quantify the deviation of quality in the area of packaging

design and manufacturing tolerance. In general, it is dif cult to apply the

quality loss function directly to diesel engine system design because it is

dif cult to quantify the monetary cost during the system design for the

deviation of a design attribute. An easier way to utilize the concepts of

quality and quality loss in system design is to use an attribute parameter in

engineering units instead of the cost in monetary units. The quality loss of

an engine attribute can be simpli ed as a continuous function of another

dependent parameter. For instance, as shown in Fig. 1.15, if we want to

quantify the quality loss of EGR rate or coolant heat rejection, the quality

loss function can be constructed as a weighted function of engine outlet

coolant temperature and NO

emissions, which are dependent upon EGR rate

and heat rejection. The coolant temperature is an indicator of durability and

increases with coolant heat rejection. NO

emissions are affected by EGR

rate, which in turn directly affects the coolant heat rejection. The lower the

EGR rate or heat rejection, the higher the NO

emissions. Another example

of quality loss is the effect of air–fuel ratio. A deviation to higher air–fuel

ratio than the nominal target causes excessively high peak cylinder pressure,

while a deviation to lower air–fuel ratio results in high soot or high exhaust

manifold gas temperature. Essentially, the quality loss function in engine

system design can be simpli ed as any continuous composite function that

may be used as the objective function in system optimization in order to

determine the nominal design target and allowable tolerances.

Another key measure of robustness in Taguchi’s robust design for quality

improvement is the signal-to-noise ratio (or the S/N ratio). Good quality

requires the performance be on target with low variability. Taguchi focused

Table 1.5 Formulae of Taguchi’s quality loss in robust design

Parameter name Parameter

symbol

Formula

Taguchi’s quality

loss

(Y) L

(Y) = C

· (Y – Y

target

)

Taguchi’s mean

square deviation

MSD

(–

Y(–Y

(–

Y(–Y

YYSSYY

(–SS(–

YY(–YYSSYY(–YY

YY(–YY

YYSSYY

YY(–YY

)=SS)=

YYSSYY

(–

)

(–

)

)+YY)+

(–YY(–

)+YY)+

(–

(–YY(–

(–

Taguchi’s

average quality

loss

()

LY()LY

()

LY()LY

()

LY()LY

=·

CC=·CC

=·

CY=·CY

=·

CY=·CY

CY[+CY

(–

CY(–CY

)]

CY[+CY

(–

CY(–CY

)]

[+s[+

CY[+CYsCY[+CY

–)

]

–)

·[CY·[

+(CY+(

–)Y–)

–)

–)Y–)

–)

·[

·[CY·[

CYsCY

–)

–)Y–)

–)

Diesel-Xin-01.indd 49 5/5/11 11:41:28 AM



50 Diesel engine system design

on reducing variations in design and believed that it might be preferable

to have a product that gives consistently good results than one that gives

inconsistent results which are sometimes better but worse on average. The

details of Taguchi’s concepts of quality, quality loss function, signal-to-noise

ratio, and robust design were explained in detail by Fowlkes and Creveling

(1995).

1.3.6 The concept of reliability

The denition of reliability

Reliability has a great impact on the brand image of a product. Customers

expect a product of good quality that performs reliably over time. Issues

in durability, quality, and reliability lead to warranty. Manufacturers often

suffer high costs from failures under warranty. Good reliability leads to

reduced warranty costs and improved brand image. Automotive engines are

designed to possess longer warranty periods or mileage targets with lower

failure rates for better reliability.

Reliability emerged as a technical consideration after the First World

War in airplane engine applications. At that time, reliability was measured

as the number of accidents per hour of ight time (Rausand and Hoyland,

2004). Reliability means the dependability of a product related to its failure

and time to failure (Fig. 1.13). Reliability is usually concerned with failures

Quality loss function of coolant heat

rejection is a symmetric or asymmetric

weighted function

Z = w

+ w

emissions(Z

)

Performance or composite performance

value (quality loss)

Lower limit (for not

violating Y

limit)

Speciﬁcation range

EGR rate or coolant heat

rejection (Y)

Engine outlet coolant

temperature (Z

)

Upper limit (for not

violating Y

limit)

Nominal

design

target (for

optimum

of the

weighted)

1.15 Quality function in engine performance.

Diesel-Xin-01.indd 50 5/5/11 11:41:28 AM



51The analytical design process and diesel engine system design

in the time domain. In general, reliability is dened as the fulllment of

quality requirements over the life cycle under specic application conditions.

According to Chandrupatla (2009), ‘reliability is the probability that a system

or component can perform its intended function for a specied interval

under stated conditions’. According to O’Connor (2002), ‘reliability is the

probability that an item will perform a required function without failure under

stated conditions for a stated period of time’. As summarized by Rausand and

Hoyland (2004), until the 1960s reliability was dened as ‘the probability that

an item will perform a required function under stated conditions for a stated

period of time’. They pointed out that a more preferred and more general

denition of reliability given by ISO 8402 standard was that ‘reliability is the

ability of an item to perform a required function, under given environmental

and operational conditions and for a stated period of time.’ A required

function here may be a single function or a combination of functions that is

necessary to provide a specied service. A hardware system may pass the

initial quality specication test on the factory’s manufacturing assembly line,

but it may not perform reliably over a specied period of time in customer

usage. Rausand and Hoyland (2004) dened that ‘according to common

usage, quality denotes the conformity of the product to its specication as

manufactured, while reliability denotes its ability to continue to comply with

its specication over its useful life. Reliability is therefore an extension of

quality into the time domain’.

In diesel engine system design, reliability is the ability of how reliable

to maintain the quality of performance, durability, and packaging over time

after the engine is put into service. Reliability is not a design attribute.

Instead, it is a characteristic of the product quality extended into the time

domain. By denition, reliability contains two key factors: a probabilistic

impact by noises and time. Those factors are illustrated in the ‘bathtub’

curve of reliability in Fig. 1.16. When a product is produced as marginally

acceptable with respect to its specications, a little additional external

noise can cause a failure in service in the early life of the product. The

decreasing failure rate in Fig. 1.16 suggests infant mortality, which means

defective items fail early and the failure rate decreases over time as they

fall out of the population. The constant failure rate characterizes the bottom

of the ‘bathtub’ curve. It is caused by random failure events associated

with external or internal environment noise factors or customer usage noise

factors. Eventually, the cumulative effects of deterioration noise factors

(e.g., wear out) and environment/usage noise factors cause the end-of-life

failure, which is characterized by an increasing failure rate as time goes on.

The goal of design for reliability is to make the useful product life longer.

The goal of robust design is to make the design insensitive to all the noise

factors during the product life.

In general, reliability can be classied into three categories: hardware,

Diesel-Xin-01.indd 51 5/5/11 11:41:28 AM



52 Diesel engine system design

software, and human. The scope of diesel engine system design is primarily

concerned with hardware reliability of engine components and system.

Reliability can be deduced from a probability distribution function of the

time to failure (Rausand and Hoyland, 2004). Reliability is usually expressed

by the following means:

1. Failure rate and mean time to failure (Fig. 1.13, for example, 1% of failure

in total population by 250,000 miles; or 100,000 miles of passenger car

operation with no engine tune-ups; or one million miles of commercial

truck operation before the overhaul).

2. Number of failures over a period of time (e.g., R

/1000 or the number

of repairs per thousand units).

3. Probability of the event that the time to failure is longer than the specied

useful life (for example, if the specied useful life is three years and

the time to failure is four years, the reliability is 100%).

4. R = 1 – P

failure

, where P

failure

is the probability of failure or the failure

rate. The failure rate at a given mileage can be measured from the

warranty information of the engines in service.

To further illustrate the noise and time factors in reliability, Fig. 1.13 shows

an example of structural reliability in the random probability distribution

diagram of the stress–strength interference model. It should be noted that the

Early-life failure,

infant mortality

due to piece-to-

piece noise factors

Failures during

normal design life,

random failures due to

environment or customer

usage noise factors

End-of-life failure,

wear-out due

to deterioration

noise factors

Limit of

useful life

Failure rate

Running-in period

of new machine

Time in service

Wear-out failure

period

Normal running period

1.16 Reliability ‘bathtub’ curve.

Diesel-Xin-01.indd 52 5/5/11 11:41:29 AM



53The analytical design process and diesel engine system design

probability distribution curves of stress and strength shown in the top part

of Fig. 1.13 are constructed by considering noise factors, and they may or

may not include a time factor (recall that one of the noise factors discussed

earlier is changes over time). In the time-in-service domain, the stress or

load in usage varies with time in an uncontrolled manner. The strength of

the component also varies with time because the component deteriorates

in material properties over time due to failure mechanisms like creep and

corrosion. This time-dependent characteristic is shown in the bottom part

of Fig. 1.13. The reliability expressed by the time-to-failure is then the time

until the stress exceeds the strength. It should be noted that, since there is

a high level of uncertainty involved, it is very difcult to predict reliability

with statistical methods or precise mathematical calculations.

The role of reliability in engine system design

Reliability is affected by all the phases of a product life cycle (engineering

design, manufacturing, operation, and maintenance). Reliability engineering

is a process to ensure reliability is controlled in the design stage. Reliability

engineering efforts control the risks by addressing the causes of failures

in order to prevent or minimize their occurrence. Although reliability can

be improved by controlling production variability and human variation in

manufacturing and quality control or by implementing preventive maintenance

strategies, design plays a vital role to determine the reliability. The concept

of design for reliability was elaborated by Kececioglu (2003) and Kumar

et al. (2006). More in-depth discussions on reliability in engine system

design are provided in Chapter 2. The fundamentals on reliability can be

found in Ireson et al. (1996), Bignonnet and Thomas (2001), Kuehnel et al.

(2005), Dodson and Schwab (2006), Klyatis and Klyatis (2006), Rahman et

al. (2007), Zhou and Li (2009), and Klyatis (2010).

1.3.7 System design – from ‘design for target’ to ‘design

for variability’ and ‘design for reliability’

Overview of system design methodologies

As discussed earlier, variability and reliability are nondeterministic in nature.

Without considering the statistical probability distribution of the attribute the

design for a point target would be either an over-design or under-design. A

reliable and robust design of a diesel engine system requires the following

design methodologies:

∑ design for target (i.e., design for a specic point target, either nominal

or limit)

∑ design for variability (i.e., design for both mean value and tolerance

Diesel-Xin-01.indd 53 5/5/11 11:41:29 AM



54 Diesel engine system design

range of the nominal target to achieve a robust and sensitive design)

∑ design for reliability (i.e., design for time-dependent variability or

design-for-time-degradation).

Among the above three methods, design for target is the foundation and

the basic technique in system design. It is a deterministic design approach.

Many analytical techniques can be applied effectively in this category for

a precise design. The rest of the book gives a thorough introduction to this

basic technique to strengthen the foundation. Design for variability is a more

advanced technique and is nondeterministic. It expands the point design in

design for target to a more complex one-dimensional design by adding the

probability distribution of the attribute parameter. The probability distribution

involves uncertainty and the design loses its precise nature. Inadequate

sampling in measurement or test data may make the probability distribution

inaccurate. Design for reliability is more complicated because it further

expands the one-dimensional design to a two-dimensional design by adding

the random factor of time-in-service into the model. Design for reliability is

the least precise in nature due to more uncertainties involved. Owing to the

difculty of predicting reliability accurately, the time-dependent probability

distribution of the attribute parameter may not be accurate. However, once

modeled successfully (with the support of a large amount of lab/eld/service

data), design for reliability will give the best quality of engine system design.

Figure 1.17 illustrates the evolution of these design methodologies. Figure

1.17 shows that the ultimate design goal of diesel engine system design is

good reliability in service life.

Design for target

Design for target is the traditional deterministic design approach where a

point target is given. It can be further classied into two subsets: design for

nominal and design for limit. The target can be either a nominal value (design

for nominal) or limit value (design for limit) used in design and calibration.

For example, the engine is designed to have the structural strength to sustain

a maximum limit of peak cylinder pressure at 200 bar. In order to ensure the

pressure does not exceed the limit of 200 bar under any operating condition

with all the noise factors, the engine rated power needs to be calibrated at a

peak cylinder pressure of only 180 bar as a nominal target under standard lab

conditions. The difference of 20 bar serves as a design margin to cover any

deviations from the standard nominal (Fig. 1.18). Design for target is a very

useful and still prevalent approach in engine system and component designs

for both steady-state and transient topics. In the mathematical formulation of

design for target, the number of equations matches the number of unknowns

so that a deterministic solution can be found. Two examples of design for

target are provided below:

Diesel-Xin-01.indd 54 5/5/11 11:41:29 AM



1.17 Engine system design optimization – design for target, variability, and reliability.

Functional response

Target

Progress

Probability

Optimum

Design parameter

Design for target–

deterministic optimization

Optimum mean Initial mean

Design parameter

Optimum mean

Optimum speciﬁcation limits

Design for variability–probabilistic optimization

(without time-based effect)

Functional

response

Functional

response

At time T

Constraint

Optimum mean

Design parameters vary with

time due to wear, fatigue, etc.

Design for reliability –

reliability-based optimization

(with time-based effect)

Probability

Progress

Time in

service

Diesel-Xin-01.indd 55 5/5/11 11:41:29 AM

