Basic SPC Tools
Presented by Russell A. Boyles, PhD Six Sigma Master Black Belt [Link]@[Link]
SPC Statistical Process Control Statistical Process Monitoring
SPM?
Process Control Strategies
Reduce the chance that problems will occur Prevent problems from occurring Identify and remove causes of problems
Standardization Training Documentation Visual controls Periodic audits
Mistake proofing Kanban Warning systems
Visual controls Periodic audits Warning systems Monitor key variables
using statistical control charts and documented response plans
Key Concept in Statistical Monitoring
Common-cause variation:
Assignable-cause variation:
Two Kinds of Variation
Common causes
Random Inherent in the process Many small fluctuations Causes cannot be determined Outcomes are predictable
Assignable causes
Systematic Mistakes, malfunctions, external factors Occasional large fluctuations Causes can be determined Outcomes are not predictable
Two Kinds of Variation
Baseline phase
177
Common causes
Assignable cause
176 175 174 173 172 171 170
Monitoring phase
Two Kinds of Variation
Baseline phase
20 15 10 5
Assignable cause
Common causes
Monitoring phase
Two Kinds of Variation
No assignable causes!
Customer complaints
35 30 25 20 15 10 5 2 4 6 8 10 12 14 16 18 Month 20 22 24 26 28 30
New manager makes big improvement!
Manager gets bonus!
Manager is reassigned!
New manager has special meeting with CEO!
Statistical Monitoring Strategy
Establish expected limits of common-cause variation from baseline data
Statistical Monitoring Strategy
Do nothing unless an assignable cause is detected If and when one does occur...
9 Investigate to determine the cause 9 Take corrective action to eliminate the cause
Response Plan Example 1
Collect and enter data Assignable cause?
Y N
Continue
Verify the data Verify the gage Document problem and solution
Able to diagnose?
Y
Escalate
Able to fix?
Fix the problem
Response Plan Example 2
Take sample from current lot Assignable cause?
Y N
Start new lot
Take another sample
Assignable cause?
Y
Enter into process log
Do operator checklist
Call Technician Do technician checklist Problem Y solved?
N
Problem Y solved?
Call Engineer
Calculating Control Limits
Most often we use three-sigma limits to distinguish operationally between assignable causes and common causes
3 + 3
Assignable causes
Common causes
Assignable causes
Baseline distribution of quantity to be monitored
Calculating Control Limits
If the quantity to be monitored follows a Normal distribution, there is only a 0.3% chance of a false alarm
3 + 3
99.7%
Baseline distribution of quantity to be monitored
Calculating Control Limits
Dont need a Normal distribution
98.1%
Three-sigma limits are an economic
compromise between false alarms and missed signals
4 5 6
99.0%
0 2 4 6 8 10 12 14
99.4%
0 2 4 6 8 10 12 14 16 18 20 22
Control Chart
Evidence of assignable causes + 3 Upper Control Limit (UCL)
Baseline distribution of quantity to be monitored
Average
Lower Control Limit (LCL) Evidence of assignable causes Time
Regular sigma calculation
Based on deviations from the data average
40 35 30 25 20 15 10 5 0
Time
Control limits based on regular sigma
50 45 40 35 30
Average = 24.7 Standard deviation = 7.1
25 20 15 10 5 0
Time
Problem with using regular sigma
Often there are assignable causes in the baseline data (trends, outliers, . . . ) In this case, regular sigma is inflated by assignable causes, and is not an accurate estimate of commoncause variation Control limits based on regular sigma are too wide to detect assignable causes if and when they occur in the future
Calculating short-term sigma
Based on deviations from the previous data point
40 35 30 25
20 15 10 5 0
Time
Calculating short-term sigma
Data
Avg. Regular Moving moving Short-term Avg. sigma ranges range sigma 7.1 -0.80 2.40 6.40 2.40 2.50 5.70 1.20 4.00 1.80 3.20 4.50 3.70 1.20 8.60 7.50 3.73 3.30
15.10 24.7 14.30 16.70 23.10 25.50
=STDEV()
23.00 28.70 29.90 33.90 32.10 28.90 33.40 29.70 28.50 19.90 12.40
= Avg. moving range / 1.128
Control limits based on short-term sigma
50 45 40 35 30
Average = 24.7 Short - term sigma 3.3
Y 25
20 15 10 5 0
Time
Rationale for using short-term sigma
Often there are assignable causes in the baseline data (trends, outliers, . . . ) Short-term sigma is not inflated by assignable causes, so it is still an accurate estimate of common-cause variation Control limits based on short-term sigma will detect assignable causes if and when they occur in the future
What about specification limits?
Lower specification limit (LSL) Upper specification limit (USL)
Customers expectation is not met
Customers expectation is met
Customers expectation is not met
Out-of-specification event
What do we do?
USL
LSL
Well, that depends on
Process Capability
LSL USL
If our process has good capability, it will virtually never produce a defective outcome, except by assignable cause Therefore, any defective outcome should trigger the response plan Of course, we also need to disposition the affected material (scrap, rework, . . .)
Process Capability
LSL USL
If our process has bad capability, there will be defective outcomes that are not assignable causes Therefore, not all defect outcomes should trigger the response plan Of course, we still need to disposition the affected material (scrap, rework, . . .)
Exercise
LSL LCL
UCL USL
LCL
LSL
USL
UCL
Indicate in the table below which of the suggested actions are appropriate for process outcomes in each of the 4 zones shown above.
Initiate response plan Scrap, rework or other disposition of affected material
Zone 1 2 3 4
Do nothing
