Introduction to Statistical Process Control Engineering Experimental Design Valerie L. Young
Outline • Description of and justification for Statistical Process Control • Fundamental definitions and principles – Variability, specifications, capability – Process characterization – Why focus on variability first?
• Constructing control charts • Calculating process capability ratios
What is Statistical Process Control? • Strategy for process improvement that uses statistics-based techniques to evaluate the process and identify opportunities for improvement • Strategy that focuses on quantifying, classifying, and reducing variability in the process • Based on the philosophy that making the right product in the first place is better than trying to rework the wrong product
Quality Control vs. Process Control • Traditional quality control focuses on the product – Monitor product quality – Rework or scrap off-spec product
• Statistical process control focuses on the process – Monitor process behavior (including product quality) – Adjust the process to eliminate off-spec production
Quality Control vs. Process Control • Traditional quality control focuses on the values – A value outside specifications is a signal that the product must be reworked or scrapped
• Statistical process control focuses on the variability – Variation outside usual limits in ANY process variable is a signal that the process should be adjusted to prevent production of unacceptable product
Why Not Just Inspect & Reject? • Reality of escaping defects – Even the most careful inspection misses sometimes – Bad product means unhappy customers
• Inspection costs money • Rejection wastes resources • Reworking/scrapping wastes time, money, and resources
Why Use Statistics? • Intuition and gut feelings – Simple problems – Inexpensive solutions – Low risk in case of failure
• Statistical evaluation – Complex problems – Expensive solutions – High risk in case of failure
Is this theory, or is this relevant? • Major corporations all over the world have adopted a Statistical Process Control strategy called “Six Sigma”, and are applying it to ALL operations, including production, marketing, and customer service. • Many of the tools of Statistical Process Control (control charts, capability indices) can be used without any theoretical understanding of statistics.
Two Types of Variability • Common cause (Random) – Always present, even when process operation is consistent – Can be quantified with summary statistics that are consistent over time – CANNOT be reduced by adjusting the existing process, only by changing it
• Special cause (Assignable)
Two Types of Variability • Common cause (Random) • Special cause (Assignable) – Response to some inconsistency in process operation (purposefully adjusting that factor would give a predictable response) – Results in summary statistics that are not consistent over time – CAN be reduced by adjusting the existing process
Two Types of Variability •
How could you reduce the Common cause (Random) variability from – Precision limits of instrumentation each of these sources? – Changes in ambient conditions
• Special cause (Assignable) – Each operator has his own “style” – Raw materials purchased from different suppliers – Equipment wear
Two Types of Variability (This may hurt your brain at first)
• Common cause (Random) – Random, so its effect on the product is predictable. If only common cause variability is present, then product quality will only vary within a specified range. (99+ % of product will be within 3 standard deviations of the mean value.)
• Special cause (Assignable) – Non-random, so its effect on the product is UNpredictable until you identify the special cause. When special cause variability is present, but the cause has not been identified, product quality can change in any direction at any time.
Specifications • The range of acceptable values – May be given as Value ± Tolerance – May be given as USL (upper specification limit) and LSL (lower specification limit)
• Determined by the user, not by the process – Not calculated from process data
• Product that does not meet specifications is termed “off-spec”
Process Capability Ratios (Desired Performance) / (Actual Performance)
This curve is the distribution of data from the process
Process performance is not necessarily centered between The shaded areas the spec limits represent the percentage of off-spec production
Voice of Customer Voice of Process
Process Characterization • Ideal State – Process in control (all special causes of variability are eliminated, and only random variability remains) – 100 % acceptable product (mean value ± variability of product is inside the specification limits)
• Threshold State • Brink of Chaos • State of Chaos
Process Characterization • Ideal State • Threshold State – Process in control • all special cause variability eliminated • only random variability remains
– Some off-spec product • Mean value not centered between specification limits and/or • Random process variability exceeds specification limits
• Brink of Chaos • State of Chaos
Process Characterization • Ideal State • Threshold State • Brink of Chaos – Process out of control; product quality wanders due to • Uncontrolled special causes AND • Inherent random variability
– 100 % acceptable product
• State of Chaos
Process Characterization • • • •
Ideal State Threshold State Brink of Chaos State of Chaos
Which problem should you address first: an off-center mean, or special cause variability?
– Process out of control; product quality wanders due to • Uncontrolled special causes AND • Inherent random variability
– Some off-spec product • Mean value not centered between specification limits and/or • Process variability exceeds specification limits