Waste Not, Want Not

1

• Waste Not, Want Not
• Improving Manufacturing Process

2001-12017 ??? 2001-12030 ???
2
Index
1. Data mining in Manufacturing Process
2.1. Introduction to R.R. Donnelley
2. R.R. Donnelley
2.2. The Technical Problem
2.3. The Data
2.4. The Impact
3.1. Introduction to Time Inc.
3. Time Inc.
3.2. The Business Problem
3.3. The Data
3.4. Approach to the Problem
3.5. Types of Waste
3.6. Data Transformation
3.7. Decision Tree
1/19
3
Data mining in Manufacturing Process
1.DM in manufacturing
Data Minings another fruitful application
Cost reduction through industrial process
improvement
2.R.R.Donnelley
3. Time Inc.
Characteristic of Manufacturing process -
efficiency of process depends on hundreds or
thousands of variables - variables interaction
are not well understood - Small variations can
make the difference between profitable yields
and unprofitable waste
2/19
4
Introduction to R.R. Donnelley
Donnelley
1.DM in manufacturing

• Fortune-500 company founded in 1864
• The largest printing company in US
• Around 5 billion yearly from anywhere that can
  be printed
• Three principal strategies
• - continuous cost reduction
• - productivity improvement
• - optimizing manufacturing operations
• isolating conditions that lead to presses being
  stopped in the middle of a run

2.R.R.Donnelley
1) intro
2) problem
3) Data
4) mining
5) effects
3. Time Inc.
3/19
5
The Technical Problem
Donnelley
1.DM in manufacturing
Cylinder banding
2.R.R.Donnelley
A series of grooves that appears in the
chrome-plated, copper cylinder
1) intro
2) problem
Streak of ink running across the printed image
gt ruining the print job
3) Data
4) mining
5) effects
Cylinder bands became a problem in the 1950s when
process began operating at 1000 feet per minute
3. Time Inc.
Increases the overall cost of production
4/19
6
The Business Problem
Donnelley
When Cylinder band forms !!!
1.DM in manufacturing
- It makes run short - Technition remove the
cylinder from the press - Cylinder must be
transported to the plating station - Costly
printing cylinder must be entirely replaced
2.R.R.Donnelley
1) intro
2) problem
Technical problem is quickly translate into -
lost productivity - higher expense -
potentially missed deadlines - unhappy customers
3) Data
4) mining
5) effects
3. Time Inc.
Ruduce the incidence of cylinder banding for the
strategies
5/19
7
The Data
Donnelley
Collection of the Data
1.DM in manufacturing
Usefulness of data gt need the new data
collection

• People had made note of valuesof various
  variables that they believed to be important
• Not suitable data for data mining

2.R.R.Donnelley
1) intro
2) problem
3) Data
reporting data gt need the job records
4) mining
No Automatic reporting system exist gtData
collection was s laborious, manual process
5) effects
3. Time Inc.
Deciding on the right attributes
Paper type, solvent type, temperature, humidity,
viscosity, press type, press speed etc
6/19
8
The Data
Donnelley
Temperature Humidity Viscosity
Preparation for continuous inputs
1.DM in manufacturing
Categorical variavles
Continuous variables
Simple partitioning the continuous var. into
three arbitrary ranges
2.R.R.Donnelley
Decision tree algorithm only split
1) intro
2) problem
Not produce good result!!
3) Data
Median value for the runs with banding as one
boundary and the median value for the runs
without banding as the other
4) mining
5) effects
3. Time Inc.
Defining the target classes
- Final composition of the training set
All print runs that ran longer than the cut-off
and finished without banding
All print runs of any length resulted in banding
7/19
9
Induced Rules for Cylinder Bands
Donnelley

• Some examples of heuristics
• Lower values for anode distance increase
  likelihood of banding
• Lower values of chrome solution ratio increase
  likelihood of banding
• Lower values of humidity increase likelihood of
  banding
• Higher values of ink temperature increase
  likelihood of banding
• Higher values of blade pressure increase
  likelihood of banding

1.DM in manufacturing
2.R.R.Donnelley
1) intro
2) problem
3) Data
4) mining
5) effects

• Guidelines to try on the shop floor
• Keep the chrome solution ratio high
• Keep the ink temperature low
• Keep the ink viscosity high

3. Time Inc.
These rules do not in any way explain cylinder
banding!
8/19
10
The Impact
Donnelley
In 1989 538 banding incident caused more than 800
hours of downtime
1.DM in manufacturing
Data mining effort
2.R.R.Donnelley
1) intro
In 1995 21 banding incident resulting in 30 hours
of downtime
2) problem
3) Data
4) mining

• However, the guidelines developed in one
  location could not be transferred to another
• Continual effort to improve performance and
  productivity is needed
• This model also predict future as well ex)
  Banding will occur more in dry season

5) effects
3. Time Inc.
9/19
11
Introduction to Time Inc.
Time Inc.
1.DM in manufacturing

• The world largest magazine publisher
• Time Inc. publish magazines like Time, Life,
  People, Sports Illustrated, Teen People, In
  Style, Cooking Light, Fortune, Martha Stewart
  Living
• Buy huge quantity of paper from all around world
• Keep track of each 1-ton roll of paper from the
  time it is ordered to the time it is mounted on
  the press

2.R.R.Donnelley
3. Time Inc.
1) intro
2) problem
3) Data
4) mining
5) effects
10/19
12
The Business Problem
Time Inc.
1.DM in manufacturing
Two Ways of Increasing Profitability
Increase Revenue
Decrease Expense
2.R.R.Donnelley

• Increase circulation
• Increase selling advertisement pages

• Three main costs paper, printing, postage
• Paper cost is in the hundreds of millions
• One way to control paper is to buy when the
  price is right
• The other way to control paper cost is to use
  less paper

3. Time Inc.
1) intro
2) problem
3) Data
4) mining
5) effects
11/19
13
The Business Problem
Time Inc.
Plant 5

• Time Inc. does not own the printing plant
• Time Inc. contracts with more than twenty
  printing plants around the country
• Time Inc. purchases its own paper and has it
  shipped to the printing plants

1.DM in manufacturing
Plant 1
Time Inc.
Plant 2
Plant 4
2.R.R.Donnelley
Plant 3
3. Time Inc.
Performance of Printing Plants Varies Greatly
1) intro
2) problem

• Time Inc. wants to keep the printing plants in
  alignment by contractual limits on the amount of
  waste
• Time Inc. was sure that overall paper waste
  could be reduced substantially because of wide
  range of performance among the printing plant
• Each one tenth of one percent improvement is
  worth 200,000 per year.

3) Data
4) mining
5) effects
12/19
14
The Data
Time Inc.
1.DM in manufacturing
2.R.R.Donnelley
3. Time Inc.
1) intro
2) problem
3) Data
4) mining
5) effects
13/19
15
The Data
Time Inc.
1.DM in manufacturing
2.R.R.Donnelley
3. Time Inc.
1) intro
2) problem
Summary of Mining Data Set
3) Data

• Number of press runs 30,401
• Number of rolls 523,893
• Rolls/Press Run statistics
• Average 17 rolls/run Minimum 1 rolls/run
  Maximum 293 rolls/run
• Press run length statistics
• Average 6.68 hours Minimum 0.03 hours
  Maximum 759 hours
• Distribution of form types in press runs
• Advance 14,806 runs Cover 4,413 runs
  Current 11,092 runs

4) mining
5) effects
14/19
16
Approach to the Problem
Time Inc.
1.DM in manufacturing
Hypothesis Testing

• The printing experts at Time Inc. had many
  insights into likely causes of waste such as
  listed below
• Waste as a function of press type
• Waste as a function of paper age
• Waste as a function of basis weight
• Waste as a function of time of day
• Waste as a function of presses equipped with
  automatic blanket washers
• Waste as a function of number of rolls in press
  run

2.R.R.Donnelley
3. Time Inc.
1) intro
2) problem
3) Data
4) mining
5) effects
15/19
17
Types of Waste
Time Inc.
1.DM in manufacturing
2.R.R.Donnelley
3. Time Inc.
1) intro
2) problem
3) Data
4) mining
5) effects
Only Addressable Waste is the focus of this case
study!
16/19
18
Data Transformation
Time Inc.
Two kinds of transformation are required
1.DM in manufacturing
Gear to get more accessibility ex) date
Transform Format
2.R.R.Donnelley
Add several convenience fields ex) press_run_id
3. Time Inc.
Two types of derived fields are created
1) intro
2) problem
Creating derived field that contain information
from two or more fields
Convert continuous variables to categorical
variables
3) Data
4) mining
5) effects
Classified the Target
17/19
19
Decision Tree
Time Inc.
Classification versus Explanation
1.DM in manufacturing

• Decision Tree is a classifier
• The goal is not classification but explanation
• To understand the most important factors
  affecting waste
• As a tree become smaller, it becomes less
  discriminating but more informative

2.R.R.Donnelley
Extracting Rule for Addressable Waste
3. Time Inc.

• Increase the minimum number of records allowed
  in a node to prune decision tree
• Look only at nodes classified as high waste
• ex) If there is more than 899.5 lbs. of
  make-ready waste and the overrun percentage is
  greater than 2.95 then this is a wasteful run

1) intro
2) problem
3) Data
4) mining
Association Rules
5) effects

• They do not prove to be as effective as decision
  tree because many of the rules discovered by the
  software were trivial
• ex) Prints runs with a lot of wrapper waste also
  have al lot of core waste

18/19
20
Decision Tree - Inpacts
Time Inc.
1.DM in manufacturing
Putting It All Together

• There is correlation between paper age and
  running waste
• Save over 75,000 a year
• Print runs using paper from multiple mills have
  slightly higher running waste percentage
• A fifth color in addition to the usual four,
  something that happens fairly often with the
  cover of Time magazine, leads to increased
  running waste.
• Save over 10,000 a year

2.R.R.Donnelley
3. Time Inc.
1) intro
2) problem
3) Data
4) mining
5) effects
19/19