Prof. Rob Leachman IEOR130 Fall, /11/2014 Rob Leachman Intro to Six Sigma

Transcription

1 Prof. Rob Leachman IEOR130 Fall,

2 Six Sigma is an engineering management paradigm originally developed at Motorola. Its application has spread from hi tech manufacturing to general business processes in many industries. Six Sigma seeks to improve the quality of process outputs by identifying and removing the causes of defects (errors) and variability. It involves statistical methods and usually relies on a special infrastructure of people within the organization who are experts in these methods. A Six Sigma project typically has specific financial goals for cost reduction or profit gain. 2

3 First written formulation by Bill Smith of Motorola in BUT, Six Sigma draws heavily on the previously published quality paradigms and methodologies (statistical quality control, TQM, Zero Defects, etc.) developed by Shewart, Deming, Juran, Ishikawa, Taguchi and others from the 1930s up to the 1980s. 3

4 A defect is anything that could lead to customer dissatisfaction. Defects are very costly. There must be continuous effort to achieve stable and predictable process results (i.e., to reduce process variation and hence defects). Processes have characteristics that can be measured, analyzed, improved and controlled. Achieving sustained quality improvement requires commitment from the entire organization. 4

5 Consider a process generating an on going stream of output. If on an on going basis we plot the averages of parameter measurements for groups of five or more output units of the process, then by the Central Limit Theorem of statistics, we should see a normal distribution. PROVIDED THAT the process is stable, i.e., provided that consecutive measurements are independent and identically distributed (IID) random variables. If not IID, then the process is not in statistical control and is said to be out of control (OOC). 5

6 For a normal distribution, 99.9% of the mass lies between [ 3, +3 ], where denotes the mean and denotes the standard deviation. Thus, a spread of 6 contains virtually all the output of the process: 6

7 We assume for each important process parameter that the engineers define specification limits, whereby if the parameter falls below the lower specification limit (LSL)or above the upper specification limit (USL), then the output unit is defective and no good for the customer. How many defects we experience can be characterized by comparing the spec limits to the 6 spread of the normal distribution for the process 7

8 A very well controlled process. The distribution is well centered between the spec limits, and the 6 spread is half the 12 spread of the spec limits. 8

9 Process capability concerns the ability of the process to generate output within the spec limits. Motorola developed some metrics for this purpose: C p = {USL LSL}/6 is termed the process capability index, where USL denotes the upper specification limit and LSL denotes the lower specification limit. C p >> 1 means good process capability. C p < 1 means bad process capability, i.e., lots of scrap is being generated. 9

10 C p = 1 might seem like there would be little scrap. But that would only be the case if the process was perfectly centered (i.e., the mean lies exactly halfway between LSL and USL). To allow for the fact that a process might not be well centered, a new metric was developed: C pk = Min { ( LSL)/3, (USL )/3 } is called the process performance index. C pk > 1 indicates the mean is more than 3 away from the nearest spec limit, i.e., there is little or no scrap. 10

11 It is found in practice that the process mean is often not stationary but tends to drift over time. Thus C pk = 1 does not imply quality is really great, because if the process drifts unfavorably we will start getting scrap. A common industry goal is to raise C pk for all important process parameters to at least 1.5, i.e., the nearest spec limit should be at least 4.5 away from the process mean. 11

12 Identify the occurrences of process variation and establish containment measures Add inspections or measurements (e.g., SPC charts) to detect out of control excursions Engineer fixes to eliminate root causes of excursions, OR: Re engineer the product so that the spec limits can be wider yet the customer will be just as happy 12

13 DMAIC (for existing processes) Define high level goals and define the existing process. Measure key aspects of the process and set up data collection systems. Analyze the data to verify cause and effect relationships. Determine those relationships. Improve or optimize the process based on application of techniques like design of experiments. Control to ensure that any deviations from targets are corrected before they result in defects. 13

14 DMADV (for new products or processes) Define design goals consistent with customer desires and enterprise strategy. Measure and identify CTQs (critical to quality characteristics), product capabilities, production process capability, and risks. Analyze to develop and design alternatives, create a highlevel design and evaluate design capability to select best design. Design details, optimize the design, and plan for design verification. (May require simulation.) Verify the design, set up pilot runs, implement the production process, hand over to process owners. 14

15 Statistical process control charts (AKA Shewart control charts, SPC charts) Ishikawa diagrams (AKA Fishbone diagrams) Design of experiments (DOE) Failure modes and effects analysis (FMEA) Fault detection and classification (FDC) Regression analysis, Analysis of variance Taguchi methods, Taguchi Loss Function Many others 15

16 Continuous process parameter X bar chart track mean of fixed size samples LCL = 3 / n, UCL = 3 / is process mean, is process standard deviation, n is sample size R chart track range of five unit samples LCL = d3r, UCL = d4r, R is average range of sample, d3 and d4 are constants from statistical tables = R/d2, d2 is constant from statistical table (can use this in X bar chart) n 16

17 Countable parameter (e.g., particles on wafer) C chart LCL = c 3 c, UCL = c 3 c c is the average count for a fixed size sample Binary parameter (e.g., good or not good) P chart LCL = p 3 p(1 p) / n, UCL = p 3 p(1 p) / n p is the average fraction bad, n is the sample size 17

18 1930s Shewhart invents control charts at Bell Labs and implements them in Western Electric (AT&T s manufacturing arm) 1940s Western Electric and a few others make good use of SPC, but most companies resist it 1950s Deming goes to Japan; SPC is embraced there 1970s & 1980s Many American industries are decimated by Japanese competition 1980s & 1990s TQM and 6 movements in USA 18

19 Introduce quality management professionals that cut across traditional dept. boundaries Executive Leadership define vision, empower role holders with freedom and resources Champions responsible for implementation, mentor the Black Belts Master Black Belt Full time, in house coaches, ensure consistent application across functions and departments Black Belt Full time, apply methodology to specific projects Green Belt handle implementation under guidance of Black Belts along with other job duties Yellow Belt trained in basic application of tools, work with Black Belts, closest to the work 19

20 For the period , Motorola claims its Six Sigma programs achieved $17 billion in savings. Starting in the 1990s, General Electric under Jack Welch became a strong disciple of Six Sigma, and GE claims major successes from it. Subsequently, Six Sigma became a management craze. Success at other companies has been mixed. Six Sigma is not a panacea. It is no substitute for inventing and marketing great products. It is an incremental improvement on previous quality theories. On the other hand, there is no substitute for great quality, and the methodologies embraced by Six Sigma embody the best knowledge we have on the subject. 20