This document provides an overview of several topics in industrial engineering including:
- Linear programming and how to formulate it as a minimization or maximization problem subject to constraints.
- Statistical process control methods like X-bar and R charts to monitor quality.
- Process capability analysis to determine if a process meets specifications.
- Queueing models and their fundamental relationships to model waiting times in systems.
- Simulation techniques like random number generation and the inverse transform method.
- Forecasting methods such as moving averages and exponentially weighted moving averages.
- Linear regression to model relationships between variables and determine coefficients.
- Experimental design topics including randomized block design and analysis of variance calculations.
1. INDUSTRIAL ENGINEERING
LINEAR PROGRAMMING n n
The general linear programming (LP) problem is: Minimize Z = ! ! cij xij
i=1j=1
Maximize Z = c1x1 + c2x2 + … + cnxn subject to
Subject to: n n
a11x1 + a12x2 + … + a1nxn ≤ b1 ! xij - ! x ji = bi for each node i ! N
j=1 j=1
a21x1 + a22x2 + … + a2nxn ≤ b2
and
h 0 ≤ xij ≤ uij for each arc (i, j)∈A
am1x1 + am2x2 + … + amnxn ≤ bm
x1, … , xn ≥ 0 The constraints on the nodes represent a conservation of flow
relationship. The first summation represents total flow out of
An LP problem is frequently reformulated by inserting non- node i, and the second summation represents total flow into
negative slack and surplus variables. Although these variables node i. The net difference generated at node i is equal to bi.
usually have zero costs (depending on the application), they
can have non-zero cost coefficients in the objective function. Many models, such as shortest-path, maximal-flow,
A slack variable is used with a "less than" inequality and assignment and transportation models can be reformulated as
transforms it into an equality. For example, the inequality minimal-cost network flow models.
5x1 + 3x2 + 2x3 ≤ 5 could be changed to 5x1 + 3x2 + 2x3 + s1 = 5
if s1 were chosen as a slack variable. The inequality STATISTICAL QUALITY CONTROL
3x1 + x2 – 4x3 ≥ 10 might be transformed into
3x1 + x2 – 4x3 – s2 = 10 by the addition of the surplus variable Average and Range Charts
s2. Computer printouts of the results of processing an LP n A2 D3 D4
usually include values for all slack and surplus variables, the 2 1.880 0 3.268
dual prices, and the reduced costs for each variable. 3 1.023 0 2.574
4 0.729 0 2.282
Dual Linear Program 5 0.577 0 2.114
Associated with the above linear programming problem is 6 0.483 0 2.004
another problem called the dual linear programming problem. 7 0.419 0.076 1.924
If we take the previous problem and call it the primal 8 0.373 0.136 1.864
9 0.337 0.184 1.816
problem, then in matrix form the primal and dual problems are
10 0.308 0.223 1.777
respectively:
Xi = an individual observation
Primal Dual
Maximize Z = cx Minimize W = yb n = the sample size of a group
Subject to: Ax ≤ b Subject to: yA ≥ c k = the number of groups
x≥0 y≥0 R = (range) the difference between the largest and smallest
It is assumed that if A is a matrix of size [m × n], then y is a observations in a sample of size n.
[1 × m] vector, c is a [1 × n] vector, b is an [m × 1] vector, and X1 + X2 + f + Xn
X= n
x is an [n × 1] vector.
Network Optimization
X1 + X 2 + f + X n
Assume we have a graph G(N, A) with a finite set of nodes N X=
k
and a finite set of arcs A. Furthermore, let
N = {1, 2, … , n}
R1 + R2 + f + Rk
xij = flow from node i to node j R=
k
cij = cost per unit flow from i to j
The R Chart formulas are:
uij = capacity of arc (i, j)
bi = net flow generated at node i CLR = R
UCLR = D4R
We wish to minimize the total cost of sending the available
supply through the network to satisfy the given demand. The LCLR = D3R
minimal cost flow model is formulated as follows:
INDUSTRIAL ENGINEERING 215
2. The X Chart formulas are: Tests for Out of Control
1. A single point falls outside the (three sigma) control limits.
CLX = X
2. Two out of three successive points fall on the same side of
UCLX = X + A2R and more than two sigma units from the center line.
LCLX = X - A2R 3. Four out of five successive points fall on the same side of
and more than one sigma unit from the center line.
Standard Deviation Charts 4. Eight successive points fall on the same side of the center
line.
n A3 B3 B4
2 2.659 0 3.267 PROCESS CAPABILITY
3 1.954 0 2.568
Actual Capability
4 1.628 0 2.266
PCRk = Cpk = min c -
5 1.427 0 2.089 n LSL USL - n m
,
6 1.287 0.030 1.970 3v 3v
7 1.182 0.119 1.882
Potential Capability (i.e., Centered Process)
8 1.099 0.185 1.815
9 1.032 0.239 1.761 PCR = Cp = USL - LSL , where
10 0.975 0.284 1.716 6v
μ and σ are the process mean and standard deviation,
UCLX = X + A3S respectively, and LSL and USL are the lower and upper
specification limits, respectively.
CLX = X
LCLX = X - A3S QUEUEING MODELS
UCLS = B4S Definitions
Pn = probability of n units in system,
CLS = S
L = expected number of units in the system,
LCLS = B3S Lq = expected number of units in the queue,
Approximations W = expected waiting time in system,
The following table and equations may be used to generate Wq = expected waiting time in queue,
initial approximations of the items indicated. λ = mean arrival rate (constant),
u
m = effective arrival rate,
n c4 d2 d3
2 0.7979 1.128 0.853 μ = mean service rate (constant),
3 0.8862 1.693 0.888 ρ = server utilization factor, and
4 0.9213 2.059 0.880 s = number of servers.
5 0.9400 2.326 0.864
6 0.9515 2.534 0.848 Kendall notation for describing a queueing system:
7 0.9594 2.704 0.833 A/B/s/M
8 0.9650 2.847 0.820
A = the arrival process,
9 0.9693 2.970 0.808
10 0.9727 3.078 0.797 B = the service time distribution,
s = the number of servers, and
v = R d2
t M = the total number of customers including those in service.
v = S c4
t
Fundamental Relationships
vR = d3 v
t
L = λW
2
vS = v 1 - c4 , where
t Lq = λWq
W = Wq + 1/μ
v = an estimate of σ,
t ρ = λ/(sμ)
σR = an estimate of the standard deviation of the ranges of the
samples, and
σS = an estimate of the standard deviation of the standard
deviations of the samples.
216 INDUSTRIAL ENGINEERING
3. Single Server Models (s = 1) Calculations for P0 and Lq can be time consuming; however,
Poisson Input—Exponential Service Time: M = ∞ the following table gives formulas for 1, 2, and 3 servers.
P0 = 1 – λ/μ = 1 – ρ
Pn = (1 – ρ)ρn = P0ρn s P0 Lq
2
L = ρ/(1 – ρ) = λ/(μ – λ) 1 1–ρ ρ /(1 – ρ)
Lq = λ2/[μ (μ– λ)] 2 (1 – ρ)/(1 + ρ) 2ρ 3/(1 – ρ 2)
W = 1/[μ (1 – ρ)] = 1/(μ – λ) 3 2(1 − ρ ) 9ρ 4
Wq = W – 1/μ = λ/[μ (μ – λ)] 2 + 4ρ + 3ρ 2 2 + 2ρ − ρ 2 − 3ρ 3
Finite queue: M < ∞
m = m _1 - Pni
u SIMULATION
P0 = (1 – ρ)/(1 – ρM+1) 1. Random Variate Generation
Pn = [(1 – ρ)/(1 – ρM+1)]ρn The linear congruential method of generating
L = ρ/(1 – ρ) – (M + 1)ρM+1/(1 – ρM+1) pseudorandom numbers Ui between 0 and 1 is obtained using
Lq = L – (1 – P0) Zn = (aZn–1+C) (mod m) where a, C, m, and Z0 are given
nonnegative integers and where Ui = Zi /m. Two integers are
Poisson Input—Arbitrary Service Time equal (mod m) if their remainders are the same when divided
by m.
Variance σ2 is known. For constant service time, σ2 = 0.
P0 = 1 – ρ 2. Inverse Transform Method
Lq = (λ2σ2 + ρ2)/[2 (1 – ρ)] If X is a continuous random variable with cumulative
L = ρ + Lq distribution function F(x), and Ui is a random number between
0 and 1, then the value of Xi corresponding to Ui can be
Wq = L q / λ
calculated by solving Ui = F(xi) for xi. The solution obtained is
W = Wq + 1/μ xi = F–1(Ui), where F–1 is the inverse function of F(x).
Poisson Input—Erlang Service Times, σ2 = 1/(kμ2)
Lq = [(1 + k)/(2k)][(λ2)/(μ (μ– λ))] F(x)
= [λ2/(kμ2) + ρ2]/[2(1 – ρ)] 1
Wq = [(1 + k)/(2k)]{λ /[μ (μ – λ)]} U1
W = Wq + 1/μ
Multiple Server Model (s > 1)
U2
Poisson Input—Exponential Service Times X
R V- 1 X2 0 X2
n s
Ss - 1 cn
S mm cn
mm W
1 W
n! + s ! f m pW
P0 = S ! Inverse Transform Method for Continuous Random Variables
Sn = 0 1 - sn W
T X FORECASTING
s - 1 ^ sth ^ sth
= 1 > ! n! + H
n s
Moving Average
n=0 s! ^1 - th n
s ! dt - i
P0 c n m t
m t
dt = i=1
n , where
Lq =
s! ^1 - th
2
t
dt = forecasted demand for period t,
s s+1
P0s t dt – i = actual demand for ith period preceding t, and
s! ^1 - th
= 2
n = number of time periods to include in the moving
Pn = P0 ^m nh /n!
n average.
0 # n # s
Pn = P0 ^m nh / _ s!s n - si n $ s
n
Exponentially Weighted Moving Average
Wq = Lq m dt = adt 1 + ^1 - ah dt 1, where
t
-
t
-
W = Wq + 1 n t
dt = forecasted demand for t, and
L = Lq + m n α = smoothing constant, 0 ≤ α ≤ 1
INDUSTRIAL ENGINEERING 217
4. LINEAR REGRESSION C = T2 N
Least Squares k
2
ni
SStotal = ! ! xij - C
t t
y = a + bx, where i=1j=1
SStreatments = ! `Ti2 ni j - C
k
t
y - intercept: a = y - bx,
t
i=1
t
and slope: b = Sxy Sxx,
SSerror = SStotal - SStreatments
Sxy = ! xi yi - _1 ni d ! xi n d ! yi n,
n n n
i=1 i=1 i=1 See One-Way ANOVA table later in this chapter.
2
Sxx = ! xi2 - _1 ni d ! xi n ,
n n
RANDOMIZED BLOCK DESIGN
i=1 i=1 The experimental material is divided into n randomized
n = sample size, blocks. One observation is taken at random for every
treatment within the same block. The total number of
y = _1 ni d ! yi n, and
n
i=1
observations is N = nk. The total value of these observations is
equal to T. The total value of observations for treatment i is Ti.
x = _1 ni d ! xi n .
n
The total value of observations in block j is Bj.
i=1
C = T2 N
Standard Error of Estimate k
2
n
2
SStotal = ! ! xij - C
2 SxxSyy - Sxy i=1j=1
Se = = MSE, where
Sxx ^n - 2h SSblocks = ! ` B 2 k j - C
n
j
2 j=1
Syy = ! yi2 - _1 ni d ! yi n
n n
SStreatments = ! `Ti2 nj - C
k
i=1 i=1
i=1
Confidence Interval for a SSerror = SStotal - SSblocks - SStreatments
e 1 + x o MSE
2
a ! ta
t See Two-Way ANOVA table later in this chapter.
2, n - 2 n Sxx
2n FACTORIAL EXPERIMENTS
Confidence Interval for b Factors: X1, X2, …, Xn
t MSE Levels of each factor: 1, 2 (sometimes these levels are
b ! ta 2, n - 2 represented by the symbols – and +, respectively)
Sxx
r = number of observations for each experimental condition
Sample Correlation Coefficient (treatment),
Sxy Ei = estimate of the effect of factor Xi, i = 1, 2, …, n,
r=
SxxSyy Eij = estimate of the effect of the interaction between factors
Xi and Xj,
ONE-WAY ANALYSIS OF VARIANCE (ANOVA)
Y ik = average response value for all r2n – 1 observations having
Given independent random samples of size ni from k
Xi set at level k, k = 1, 2, and
populations, then:
2
! ! _ xij - x i
k ni km
Y ij = average response value for all r2n – 2 observations having
i=1j=1
Xi set at level k, k = 1, 2, and Xj set at level m, m = 1, 2.
2
= ! ! _ xij - xi i + ! ni _ xi - x i or
k ni k 2
Ei = Y i2 - Y i1
i=1j=1 i=1
cY ij - Y ij m - cY ij - Y ij m
22 21 12 11
SStotal = SSerror + SStreatments
Eij =
2
Let T be the grand total of all N = Σini observations and Ti
be the total of the ni observations of the ith sample.
218 INDUSTRIAL ENGINEERING
5. ANALYSIS OF VARIANCE FOR 2n FACTORIAL where SSi is the sum of squares due to factor Xi, SSij is the sum
DESIGNS of squares due to the interaction of factors Xi and Xj, and so on.
Main Effects The total sum of squares is equal to
m r 2
Let E be the estimate of the effect of a given factor, let L be SStotal = ! ! Yck - T
2
the orthogonal contrast belonging to this effect. It can be c = 1k = 1 N
proved that where Yck is the kth observation taken for the cth experimental
E= L condition, m = 2n, T is the grand total of all observations, and
2n - 1 N = r2n.
m
L = ! a^chY ^ch RELIABILITY
c=1
2 If Pi is the probability that component i is functioning, a
SSL = rL , where
n reliability function R(P1, P2, …, Pn) represents the probability
2
that a system consisting of n components will work.
m = number of experimental conditions For n independent components connected in series,
(m = 2n for n factors),
R _ P , P2, fPni = % Pi
n
a(c) = –1 if the factor is set at its low level (level 1) in 1
i=1
experimental condition c,
a(c) = +1 if the factor is set at its high level (level 2) in For n independent components connected in parallel,
experimental condition c,
R _ P , P2, fPni = 1 - % _1 - Pi i
n
r = number of replications for each experimental condition 1
i=1
Y ^ch = average response value for experimental condition c,
and
LEARNING CURVES
SSL = sum of squares associated with the factor. The time to do the repetition N of a task is given by
Interaction Effects TN = KN s, where
Consider any group of two or more factors. K = constant, and
a(c) = +1 if there is an even number (or zero) of factors in the s = ln (learning rate, as a decimal)/ln 2; or, learning
group set at the low level (level 1) in experimental condition rate = 2s.
c = 1, 2, …, m If N units are to be produced, the average time per unit is
a(c) = –1 if there is an odd number of factors in the group set given by
8^ N + 0.5h^1 + sh - 0.5^1 + shB
at the low level (level 1) in experimental condition K
Tavg =
c = 1, 2, …, m N ^1 + sh
It can be proved that the interaction effect E for the factors in
the group and the corresponding sum of squares SSL can be
INVENTORY MODELS
determined as follows:
For instantaneous replenishment (with constant demand rate,
E= L known holding and ordering costs, and an infinite stockout
2n - 1 cost), the economic order quantity is given by
m
L = ! a^chY ^ch EOQ = 2AD , where
c=1 h
2
SSL = rLn A = cost to place one order,
2
D = number of units used per year, and
Sum of Squares of Random Error
The sum of the squares due to the random error can be h = holding cost per unit per year.
computed as Under the same conditions as above with a finite
SSerror = SStotal – ΣiSSi – ΣiΣjSSij – … – SS12 …n replenishment rate, the economic manufacturing quantity is
given by
EMQ = 2AD , where
h _1 - D Ri
R = the replenishment rate.
INDUSTRIAL ENGINEERING 219
6. ERGONOMICS
NIOSH Formula
Recommended Weight Limit (pounds)
= 51(10/H)(1 – 0.0075|V – 30|)(0.82 + 1.8/D)(1 – 0.0032A)(FM)(CM)
where
H = horizontal distance of the hand from the midpoint of the line joining the inner ankle bones to a point projected on the
floor directly below the load center, in inches
V = vertical distance of the hands from the floor, in inches
D = vertical travel distance of the hands between the origin and destination of the lift, in inches
A = asymmetry angle, in degrees
FM = frequency multiplier (see table)
CM = coupling multiplier (see table)
Frequency Multiplier Table
≤ 8 hr /day ≤ 2 hr /day ≤ 1 hr /day
–1
F, min V< 30 in. V ≥ 30 in. V < 30 in. V ≥ 30 in. V < 30 in. V ≥ 30 in.
0.2 0.85 0.95 1.00
0.5 0.81 0.92 0.97
1 0.75 0.88 0.94
2 0.65 0.84 0.91
3 0.55 0.79 0.88
4 0.45 0.72 0.84
5 0.35 0.60 0.80
6 0.27 0.50 0.75
7 0.22 0.42 0.70
8 0.18 0.35 0.60
9 0.15 0.30 0.52
10 0.13 0.26 0.45
11 0.23 0.41
12 0.21 0.37
13 0.00 0.34
14 0.31
15 0.28
220 INDUSTRIAL ENGINEERING
7. Coupling Multiplier (CM) Table
(Function of Coupling of Hands to Load)
Container Loose Part / Irreg. Object
Optimal Design Not Comfort Grip Not
Opt. Handles Not POOR GOOD
or Cut-outs
GOOD Flex Fingers 90 Degrees Not
FAIR POOR
Coupling V < 30 in. or 75 cm V ≥ 30 in. or 75 cm
GOOD 1.00
FAIR 0.95
POOR 0.90
Biomechanics of the Human Body
Basic Equations
Hx + Fx = 0
Hy + Fy = 0
Hz + W+ Fz = 0
THxz + TWxz + TFxz = 0 W
THyz + TWyz + TFyz = 0
THxy + TFxy = 0
The coefficient of friction μ and the angle α at which the floor is inclined determine the equations at the foot.
Fx = μFz
With the slope angle α
Fx = αFzcos α
Of course, when motion must be considered, dynamic conditions come into play according to Newton's Second Law. Force
transmitted with the hands is counteracted at the foot. Further, the body must also react with internal forces at all points between
the hand and the foot.
INDUSTRIAL ENGINEERING 221
8. PERMISSIBLE NOISE EXPOSURE (OSHA) Standard Time Determination
Noise dose (D) should not exceed 100%. ST = NT × AF
where
C NT = normal time, and
D = 100% # ! i
Ti
AF = allowance factor.
where Ci = time spent at specified sound pressure level,
SPL, (hours) Case 1: Allowances are based on the job time.
AFjob = 1 + Ajob
Ti = time permitted at SPL (hours)
Ajob = allowance fraction (percentage/100) based on
! Ci = 8 (hours) job time.
Case 2: Allowances are based on workday.
For 80 ≤ SPL ≤ 130 dBA, Ti = 2c
105 - SPL m
5 (hours)
AFtime = 1/(1 – Aday)
If D > 100%, noise abatement required. Aday = allowance fraction (percentage/100) based on
If 50% ≤ D ≤ 100%, hearing conservation program required. workday.
Note: D = 100% is equivalent to 90 dBA time-weighted
Plant Location
average (TWA). D = 50% equivalent to TWA of 85 dBA.
The following is one formulation of a discrete plant location
Hearing conservation program requires: (1) testing employee problem.
hearing, (2) providing hearing protection at employee’s
request, and (3) monitoring noise exposure. Minimize
m n n
Exposure to impulsive or impact noise should not exceed 140 z = ! ! cij yij + ! fjx j
dB sound pressure level (SPL). i=1j=1 j=1
subject to
FACILITY PLANNING m
! yij # mx j, j = 1, f, n
Equipment Requirements i=1
n PT n
ij ij ! yij = 1,
Mj = ! where i = 1, f, m
i = 1 Cij j=1
yij $ 0, for all i, j
Mj = number of machines of type j required per production
period, x j = ^0, 1h, for all j, where
Pij = desired production rate for product i on machine j,
m = number of customers,
measured in pieces per production period,
n = number of possible plant sites,
Tij = production time for product i on machine j, measured in
hours per piece, yij = fraction or proportion of the demand of customer i which
is satisfied by a plant located at site j; i = 1, …, m; j = 1,
Cij = number of hours in the production period available for
…, n,
the production of product i on machine j, and
xj = 1, if a plant is located at site j,
n = number of products.
xj = 0, otherwise,
People Requirements cij = cost of supplying the entire demand of customer i from a
n PT
ij ij plant located at site j, and
A j = ! ij , where
i=1 C fj = fixed cost resulting from locating a plant at site j.
Aj = number of crews required for assembly operation j,
Material Handling
Pij = desired production rate for product i and assembly Distances between two points (x1, y1) and (x2, y2) under
operation j (pieces per day), different metrics:
Tij = standard time to perform operation j on product i
Euclidean:
(minutes per piece),
D = ^ x1 - x2h + _ y1 - y2i
2 2
Cij = number of minutes available per day for assembly
operation j on product i, and
Rectilinear (or Manhattan):
n = number of products.
D = x1 - x2 + y1 - y2
222 INDUSTRIAL ENGINEERING
9. Chebyshev (simultaneous x and y movement): TAYLOR TOOL LIFE FORMULA
D = max _ x1 - x2 , y1 - y2 i VT n = C, where
Line Balancing V = speed in surface feet per minute,
Nmin = bOR # ! ti OT l T = tool life in minutes, and
i
C, n = constants that depend on the material and on the tool.
= Theoretical minimum number of stations
Idle Time/Station = CT - ST WORK SAMPLING FORMULAS
Idle Time/Cycle = ! ^CT - ST h p _1 - pi 1-p
D = Za 2 n and R = Za 2 pn , where
Idle Time/Cycle
Percent Idle Time = 100, where
Nactual # CT # p = proportion of observed time in an activity,
CT = cycle time (time between units), D = absolute error,
OT = operating time/period, R = relative error (R = D/p), and
OR = output rate/period, n = sample size.
ST = station time (time to complete task at each station),
ti = individual task times, and
N = number of stations.
Job Sequencing
Two Work Centers—Johnson’s Rule
1. Select the job with the shortest time, from the list of jobs,
and its time at each work center.
2. If the shortest job time is the time at the first work center,
schedule it first, otherwise schedule it last. Break ties
arbitrarily.
3. Eliminate that job from consideration.
4. Repeat 1, 2, and 3 until all jobs have been scheduled.
CRITICAL PATH METHOD (CPM)
dij = duration of activity (i, j),
CP = critical path (longest path),
T = duration of project, and
T = ! dij
_i, j i ! CP
PERT
(aij, bij, cij) = (optimistic, most likely, pessimistic) durations
for activity (i, j),
μij = mean duration of activity (i, j),
σij = standard deviation of the duration of activity (i, j),
μ = project mean duration, and
σ = standard deviation of project duration.
aij + 4bij + cij
nij =
6
cij - aij
vij =
6
n = ! nij
_i, j i ! CP
2 2
v = ! vij
_i, j i ! CP
INDUSTRIAL ENGINEERING 223
10. ONE-WAY ANOVA TABLE
Degrees of Sum of
Source of Variation Mean Square F
Freedom Squares
SStreatments MST
Between Treatments k–1 SStreatments MST =
k −1 MSE
SS error
Error N–k SSerror MSE =
N −k
Total N–1 SStotal
TWO-WAY ANOVA TABLE
Degrees of Sum of
Source of Variation Mean Square F
Freedom Squares
SStreatments MST
Between Treatments k–1 SStreatments MST =
k −1 MSE
SS blocks MSB
Between Blocks n–1 SSblocks MSB =
n −1 MSE
SS error
Error (k − 1 )(n − 1 ) SSerror MSE =
(k − 1 )(n − 1 )
Total N–1 SStotal
224 INDUSTRIAL ENGINEERING
11. PROBABILITY AND DENSITY FUNCTIONS: MEANS AND VARIANCES
Variable Equation Mean Variance
Binomial
Coefficient ( nx ) = x!(nn− x)!
!
b(x; n , p ) = ( ) p (1 − p )
n x n− x
Binomial np np(1 – p)
x
Hyper r ( ) N −r
n−x nr r (N − r )n(N − n )
h(x; n, r, N ) =
x () N 2 (N − 1)
Geometric
(N )
n
N
λ x e −λ
Poisson f (x; λ ) = λ λ
x!
Geometric g(x; p) = p (1 – p)x–1 1/p (1 – p)/p2
Negative
Binomial
f ( y ; r, p) = ( y + r −1 r
r −1 )
p (1 − p ) y r/p r (1 – p)/p2
n! x
Multinomial f ( x1, …, xk) = p1x1 … pk k npi npi (1 – pi)
x1! , …, xk !
Uniform f(x) = 1/(b – a) (a + b)/2 (b – a)2/12
x α − 1e − x β
Gamma f (x ) = ; α > 0, β > 0 αβ αβ 2
β α Γ (α )
1 −x β
Exponential f (x ) = e β β2
β
Weibull f (x ) =
α α − 1 − xα
β
x e β
β1 α Γ[(α + 1) α ] [ ( ) ( )]
β2 α Γ
α +1
α
− Γ2
α +1
α
2
Normal f (x ) =
1
e
−
2 σ ( )
1 x−μ
μ σ2
σ 2π
{
2(x − a )
if a ≤ x ≤ m
(b − a )(m − a ) a+b+m a 2 + b 2 + m 2 − ab − am − bm
Triangular f (x ) =
3 18
2(b − x )
if m < x ≤ b
(b − a )(b − m )
INDUSTRIAL ENGINEERING 225
12. HYPOTHESIS TESTING
Table A. Tests on means of normal distribution—variance known.
Hypothesis Test Statistic Criteria for Rejection
H0: μ = μ0
H1: μ ≠ μ0 |Z0| > Zα /2
H0: μ = μ0 X − μ0
Z0 ≡ Z0 < –Z α
H0: μ < μ0 σ n
H0: μ = μ0
H1: μ > μ0 Z0 > Z α
H0: μ1 – μ2 = γ
H1: μ1 – μ2 ≠ γ |Z0| > Zα /2
X1 − X 2 − γ
H0: μ1 – μ2 = γ Z0 ≡
2 2
σ1 σ 2 Z0 <– Zα
H1: μ1 – μ2 < γ +
n1 n2
H0: μ1 – μ2 = γ
H1: μ1 – μ2 > γ Z0 > Z α
Table B. Tests on means of normal distribution—variance unknown.
Hypothesis Test Statistic Criteria for Rejection
H0: μ = μ0
H1: μ ≠ μ0 |t0| > tα /2, n – 1
H0: μ = μ0 X − μ0
t0 = t0 < –t α , n – 1
H1: μ < μ0 S n
H0: μ = μ0
t0 > tα , n – 1
H1: μ > μ0
X1 − X 2 − γ
t0 =
H0: μ1 – μ2 = γ 1 1
Variances Sp + |t0| > t α /2, v
H1: μ1 – μ2 ≠ γ n1 n2
equal
v = n1 + n2 − 2
X1 − X 2 − γ
H0: μ1 – μ2 = γ t0 =
2 2 t0 < –t α , v
S1 S2
H1: μ1 – μ2 < γ Variances +
unequal n1 n2
2
)
2 2
H0: μ1 – μ2 = γ
ν=
( S1 S2
+
n1 n2
t0 > tα , v
2 2
H1: μ1 – μ2 > γ
(S n ) + (S
2
1 1
2
2 n2 )
n1 − 1 n2 − 1
In Table B, Sp2 = [(n1 – 1)S12 + (n2 – 1)S22]/v
226 INDUSTRIAL ENGINEERING
15. ERGONOMICS—HEARING
The average shifts with age of the threshold of hearing for pure tones of persons with “normal” hearing, using a 25-year-old
group as a reference group.
Equivalent sound-level contours used in determining the A-weighted sound level on the basis of an octave-band analysis. The
curve at the point of the highest penetration of the noise spectrum reflects the A-weighted sound level.
-
INDUSTRIAL ENGINEERING 229
16. Estimated average trend curves for net hearing loss at 1,000, 2,000, and 4,000 Hz after continuous exposure to steady noise.
Data are corrected for age, but not for temporary threshold shift. Dotted portions of curves represent extrapolation from
available data.
Tentative upper limit of effective temperature (ET) for unimpaired mental performance as related to exposure time; data are
based on an analysis of 15 studies. Comparative curves of tolerable and marginal physiological limits are also given.
Effective temperature (ET) is the dry bulb temperature at 50% relative humidity, which results in the same physiological effect
as the present conditions.
230 INDUSTRIAL ENGINEERING