Statistical Process Control for ShortRuns

1
Statistical Process Control for Short-Runs

• Department of Industrial Manufacturing
• Engineering
• Tyler Mangin
• Canan Bilen
• 5-22-02

2
Background

• B.S. Industrial Engineering from North Dakota
  State University
• Emphasis on Statistical Quality Control
• Experience
• Quality control internship
• Consortium of contract manufacturers in North
  Dakota
• Center for Nanoscale Science and Engineering

3
Introduction

• Introduction to SPC
• Manufacturing environment in North Dakota
• Short-run manufacturing
• Short-run SPC techniques
• Strengths and weaknesses of these techniques
• Future work

4
Statistical Thinking

• All work occurs in a system of interconnected
  processes
• Variation exists in all processes
• Understanding reducing variation are keys to
  successes

5
Statistical Process Control

• Purpose
• Methodology for monitoring a process
• Proven technique for improving quality and
  productivity
• Identifies special causes of variation
• Signals the need to take corrective action
• Should be usable (with minimal or no math
  background)

6
Manufacturing in North Dakota

• Small to medium job shops and contract
  manufacturers are common
• Metal fabrication and electronics manufacturing
  facilities will be most accessible
• Operators have minimal mathematics and SPC
  training
• Limited resources available to implement SPC

7
Statistical Quality Needs in ND

• Should address short-run production
• The techniques should be kept as simple as
  possible
• Keep computation needs to a minimum
• SPC should demonstrate significant cost reduction
  (in short duration)

8
Short-Run Manufacturing
A production run that is not long enough to
provide adequate data to construct a control
chart.

• Standard for job shops
• Common in advanced manufacturing
• Driven by
• Demand for mass customization
• Availability of flexible production equipment
• Use of just in time techniques
• Short-runs result in
• Smaller lot sizes
• Shorter lead times
• Less available process data

9
Barriers to SPC in Short-Run Manufacturing

• Multiple part types
• Setups and changeovers
• Data scarcity
• Cost minimization
• Need for simplicity

10
Multiple Part Types

• Each part is likely to have a different average
  and standard deviation
• Unique control charts required for each chart
• Difficult to detect time-related changes
• Adds cost to the product
• Creates excessive paperwork
• Decreases operator efficiency

11
Setups and Changeovers

• Setup is a frequently occurring part of process
  operation
• Introduce special causes of variation into the
  process
• Importance of knowing whether the first part is
  on-target
• Two types of process capability
• Capability after process has been brought into
  control
• Capability across runs if the process were run
  without adjustment after initial setup
• Creates the need to monitor run-to-run
  variation
• Ensuring quick, consistent setups is critical

12
Data Scarcity

• Traditional charts require a large amount of data
  
• Recommended at least 25 subgroups of size 5
• Short-runs do not generate enough data
• If control limits are calculated, they will be
  unreliable
• Historical data may not be available
• The data for short-runs is likely to be
  auto-correlated

13
Minimizing Cost

• Maximize revenue by reducing quality-related
  costs
• Sampling and inspection costs
• Process repair costs
• Cost of false alarms
• Cost of poor quality
• Based on the lifetime of the production run
• Economic control chart design

14
Need for Simplicity

• Regional companies lack resources and experience
  with SPC
• Operator must be able to manage the control
  charts
• If it is not easy to use, it will not be used
• True benefits of SPC come from interaction with
  the process

15
Approaches to Short-Run SPC

• DNOM charts
• Standardized charts
• Q-charts
• Bayesian quality control
• Monitoring run-to-run variation

16
DNOM ChartsDeviation from Nominal

• Principles
• Different parts will have different target values
• Calculate the deviation from nominal value
• Plot deviation as the quality characteristic

17
Infinity Windows Sample Data

• Three part types
• Header
• Right jamb
• Left jamb
• Nominal length varies from part to part
• Continuous runs no batches

18
DNOM Chart
19
DNOM Charts

• Strengths
• Groups multiple parts and their data sets on a
  single chart
• Provides a continuous view of the process
• Fairly simple to construct and understand
• Shortcomings
• Assumes variation is equal for all parts
• Requires some historical data to calculate
  control limits
• Does not address quality costs
• Only tracks within-run variation

20
Standardized Control Charts

• Principles
• Multiple part-types flow through a single machine
• Different parts may have different target values
• Control limits and plot points are standardized
  to allow charting of multiple part-types

21
Standardized Control Charts

• Strengths
• Groups multiple parts and their data sets on a
  single chart
• Provides a continuous view of the process
• Fairly simple to construct and understand
• Does not assume all parts have equal variation
• Shortcomings
• Requires some historical data to calculate
  control limits
• Does not address quality costs
• Only tracks within-run variation

22
Sample Standardized Chart
23
Q-ChartsSelf-updating, standardized charts

• Principles
• Standardize the quality characteristic of
  interest
• The standardized statistic will be i.i.d. N(0,1)
• Plots multiple part types on a standardized chart
• Can begin charting with no historical data
• Uses all available information to estimate the
  parameters (updating control limits)

24
Q-Charts

• Strengths
• Charts can be made in real time beginning with
  the first production unit
• Does not assume process mean or variation are
  known in advance
• Does not assume all parts have the same variation
• Multiple part types can be plotted on a single
  chart
• Uses all available data to update control limits
• Shortcomings
• Does not address quality costs
• May not be clear to the operator
• Strictly monitors within-run variation
• Lacks simplicity? requires a PC

25
Bayesian Quality ControlEconomic charts

• Principles
• The system is modeled by partially observable
  Markov processes
• The system is generally assumed to have two
  states in-control out-of-control
• The operator is faced with certain
  action-decisions
• Do nothing
• Inspect output
• Inspect machine
• Repair machine
• The model is a decision-making tool for
  minimizing quality costs over the length of the
  production run

26
Bayesian Quality Control

• Strengths
• Addresses quality costs as a factor in process
  control
• Advises operators on which action to take based
  on probabilistic analysis
• Accounts for finite production horizon
• Shortcomings
• Models require accurate historical data
• Models must be individualized to the specific
  production process
• Not designed to handle multiple part types

27
Monitoring Run-to-Run VariationA new concept

• Setups are
• Time between last unit of one run and first good
  unit of the next run
• Integral part of process operation
• Occur frequently
• Reducing setup time implies reduction of
• Test runs
• Inspections
• Process adjustment
• Scrap rework

28
Monitoring Run-to-Run Variation

• Principles
• Plot the mean of the first sample taken after
  setup
• Each setup generates one plot point
• Plot each setup on one control chart
• Over time setup related variation is detected
• Attempts to detect run-to-run variation

29
Monitoring Run-to-Run Variation

• Strengths
• Addresses setup induced variation
• Becomes more effective as setups become more
  common
• Is a philosophy not a technique
• Shortcomings
• Long-term approach
• Does not address data scarcity
• Does not address quality costs
• Lacks a well-defined methodology

30
SPC Techniques Summary
31
Future Work

• Develop Run-to-Run Variation Charts
• as the focus of my thesis
• Further analysis of the shortcomings of the
  Monitoring Run-to-Run framework
• Determine needs of job-shops and other low-volume
  manufacturers
• Modify the Run-to-Run charts to fit the needs of
  regional companies
• Develop guidelines to maximize the potential for
  implementation

32
Review

• Introduction to SPC
• Manufacturing environment in North Dakota
• Short-run manufacturing
• Short-run SPC techniques
• Strengths and weaknesses of these techniques
• Future work

33
Thanks to

• Dr. Bilen
• Ritesh Saluja
• Faculty and staff of NDSUs Industrial and
  Manufacturing Engr. department
• QPR Conference

34
Discussion