STATISTICAL PROCESS CONTROL AND QUALITY MANAGEMENT

Quality Types of Quality Fitness for use, acceptable standard Based on needs, expectations and customer requests Types of Quality Quality of design Quality of conformance Quality of performance

Quality of Conformance Quality of Design Differences in quality due to design differences, intentional differences Quality of Conformance Degree to which product meets or exceeds standards Quality of Performance Long term consistent functioning of the product, reliability, safety, serviceability, maintainability

Quality and Productivity Improved quality leads to lower costs and increased profits

Statistics and Quality Management Statistical analysis is used to assist with product design, monitor the production process, and check quality of the finished product

Checking Finished Product Quality Random samples selected from batches of finished product can be used to check for product quality

Assisting With Production Design A variety of experimental design techniques are available for improving the production process

Monitoring the Process Control charts is used to monitor the process as it unfolds Sampling from the production line to see if variation in product quality is consistent with expectations

The Control Chart A special type of sequence plot which is used to monitor a process Throughout the process measurements are taken and plotted in a sequence plot Plot also contains upper and lower control limits indicating the expected range of the process when it is behaving properly

Examples

Variation There is no two natural items in any category are the same. Variation may be quite large or very small. If variation very small, it may appear that items are identical, but precision instruments will show differences.

3 Categories of variation Within-piece variation One portion of surface is rougher than another portion. Apiece-to-piece variation Variation among pieces produced at the same time. Time-to-time variation Service given early would be different from that given later in the day.

Source of variation Equipment Material Environment Operator Tool wear, machine vibration, … Material Raw material quality Environment Temperature, pressure, humadity Operator Operator performs- physical & emotional

Control Chart Viewpoint Variation due to Common or chance causes Assignable causes Control chart may be used to discover “assignable causes”

Control chart functions Control charts are powerful aids to understanding the performance of a process over time. PROCESS Output Input What’s causing variability?

Control charts identify variation Chance causes - “common cause” inherent to the process or random and not controllable if only common cause present, the process is considered stable or “in control” Assignable causes - “special cause” variation due to outside influences if present, the process is “out of control”

Control charts help us learn more about processes Separate common and special causes of variation Determine whether a process is in a state of statistical control or out-of-control Estimate the process parameters (mean, variation) and assess the performance of a process or its capability

Control charts to monitor processes To monitor output, we use a control chart we check things like the mean, range, standard deviation To monitor a process, we typically use two control charts mean (or some other central tendency measure) variation (typically using range or standard deviation)

Types of Data Variable data Product characteristic that can be measured Length, size, weight, height, time, velocity Attribute data Product characteristic evaluated with a discrete choice Good/bad, yes/no

Types of Control Charts Control Chart For The Mean(X chart) Control Chart For The Range(R chart) Control Chart For A Proportion(p chart) Control Chart For Attribute Measures(c chart)

Control Chart For The Sample Mean Assuming that the sample mean is approximately normal with mean  and standard deviation , a control chart for the mean usually consists of three horizontal lines The vertical axis is used to plot the magnitude of observed sample means while the horizontal axis represents time or the order of the sequence of observed means

Control Chart For The Sample Mean The center line is at the mean,  and upper and lower control limits are at +3 and -3 Since  (standard deviation)is usually unknown the term is usually replaced by an estimator based on the sample range

Control Chart For The Sample Mean The formula is given by where = average value of the range = k = number of samples The values of A2 depend on the sample size

Control Chart For The Sample Mean LCL = UCL =

Control Chart For The Range Designed to monitor variability in the product Range easier to determine than standard deviation Distribution of sample range assumes product measurement is normally distributed

Control Chart For The Range Upper and lower control limits and center line obtained from and the values of D3, D4 according to the formulae LCL = UCL = The values of D3, D4 are based on sample size

Control Chart For The Sample Proportion Population proportion  The sample mean is now a mean proportion given by Where = total number of objects in sample with characteristic divided by total sample size

Control Chart For The Sample Proportion Using the central limit theorem the control limits are given by:

Control Chart For The Sample Proportion Since the true proportion  is usually unknown we replace it by the average proportion If the sample size varies then the upper and lower limits will vary and the equations become: where ni = sample size in sample i

Control Chart For Attribute Measures Alternative method of counting good and bad items. Defects are measured by merely counting the no. of defects. Where c = total number of defects in a sample

Control Chart For Attribute Measures Process average or central line c= UCL: LCL:

Example: Control Charts for Variable Data 4/20/2017 Example: Control Charts for Variable Data Slip Ring Diameter (cm) Sample 1 2 3 4 5 X R 1 5.02 5.01 4.94 4.99 4.96 2 5.01 5.03 5.07 4.95 4.96 3 4.99 5.00 4.93 4.92 4.99 4 5.03 4.91 5.01 4.98 4.89 5 4.95 4.92 5.03 5.05 5.01 6 4.97 5.06 5.06 4.96 5.03 7 5.05 5.01 5.10 4.96 4.99 8 5.09 5.10 5.00 4.99 5.08 9 5.14 5.10 4.99 5.08 5.09 10 5.01 4.98 5.08 5.07 4.99 Example4.3 Statistics

Example: Control Charts for Variable Data 4/20/2017 Example: Control Charts for Variable Data Slip Ring Diameter (cm) Sample 1 2 3 4 5 X R 1 5.02 5.01 4.94 4.99 4.96 4.98 0.08 2 5.01 5.03 5.07 4.95 4.96 5.00 0.12 3 4.99 5.00 4.93 4.92 4.99 4.97 0.08 4 5.03 4.91 5.01 4.98 4.89 4.96 0.14 5 4.95 4.92 5.03 5.05 5.01 4.99 0.13 6 4.97 5.06 5.06 4.96 5.03 5.01 0.10 7 5.05 5.01 5.10 4.96 4.99 5.02 0.14 8 5.09 5.10 5.00 4.99 5.08 5.05 0.11 9 5.14 5.10 4.99 5.08 5.09 5.08 0.15 10 5.01 4.98 5.08 5.07 4.99 5.03 0.10 50.09 1.15 Example4.3 Statistics

Calculation From Table above: Sigma X-bar = 50.09 Sigma R = 1.15 Thus; X-Double bar = 50.09/10 = 5.009 cm R-bar = 1.15/10 = 0.115 cm Note: The control limits are only preliminary with 10 samples. It is desirable to have at least 25 samples.

Trial control limit UCLx-bar = X-D bar + A2 R-bar = 5.009 + (0.577)(0.115) = 5.075 cm LCLx-bar = X-D bar - A2 R-bar = 5.009 - (0.577)(0.115) = 4.943 cm UCLR = D4R-bar = (2.114)(0.115) = 0.243 cm LCLR = D3R-bar = (0)(0.115) = 0 cm For A2, D3, D4: see Table B, Appendix n = 5

3-Sigma Control Chart Factors Sample size X-chart R-chart n A2 D3 D4 2 1.88 0 3.27 3 1.02 0 2.57 4 0.73 0 2.28 5 0.58 0 2.11 6 0.48 0 2.00 7 0.42 0.08 1.92 8 0.37 0.14 1.86

X-bar Chart

R Chart

Subgroup 18 2 3 4 5 1 12.45 12.39 12.40 12.37 12.55 12.38 12.36 12.44 12.46 12.30 6 12.41 7 12.51 8 9 10 12.35 11 12 13 14 15

From a lot of 100 Sample No. Number of defects 1 10 2 9 3 8 4 11 5 7 6 12 13 14 From a lot of 100

From a lot of 100 Sample No. Number of defects Proportion 1 10 0.10 2 9 0.9 3 8 0.8 4 11 0.11 5 7 0.7 6 12 0.12 13 0.13 14 0.14 From a lot of 100

The Normal Distribution -3 -2 -1 +1 +2 +3 Mean 68.26% 95.44% 4/20/2017 -3 -2 -1 +1 +2 +3 Mean 68.26% 95.44% 99.74% The Normal Distribution  = Standard deviation LSL USL -3 +3 CL And at +/- 3 sigma, the most common choice of confidence/control limits in quality control application, the area is 99.97%. Statistics 20

34.13% of data lie between  and 1 above the mean (). 34.13% between  and 1 below the mean. Approximately two-thirds (68.28 %) within 1 of the mean. 13.59% of the data lie between one and two standard deviations Finally, almost all of the data (99.74%) are within 3 of the mean.

Normal Distribution Review Define the 3-sigma limits for sample means as follows: What is the probability that the sample means will lie outside 3-sigma limits? Note that the 3-sigma limits for sample means are different from natural tolerances which are at

4/20/2017 Common Causes The first set of slides presents the various aspects of common and assignable causes. This slide and the two following show a normal process distribution and how it allows for expected variability which is termed common causes. Statistics 2

Process Out of Control The term out of control is a change in the process due to an assignable cause. When a point (subgroup value) falls outside its control limits, the process is out of control.

Assignable Causes (a) Mean Average Grams 4/20/2017 The new distribution will look as shown in this slide. Grams Statistics 7

Assignable Causes (b) Spread Average Grams 4/20/2017 The new distribution has a much greater spread (higher standard deviation). Grams Statistics 9

Assignable Causes (c) Shape Average Grams 4/20/2017 A skewed (non-normal) distribution will result in a different pattern of variability. Grams Statistics 11

Control Charts Assignable causes likely UCL Nominal LCL 1 2 3 Samples 4/20/2017 Control Charts Assignable causes likely UCL Nominal However, the third sample plots outside the original distribution, indicating the likely presence of an assignable cause. LCL 1 2 3 Samples Statistics 24

Control Chart Examples 4/20/2017 Control Chart Examples UCL Nominal Variations LCL And finally, this chart shows a process with two points actually outside the control limits, an easy indicator to detect but not the only one. These rules, there are a few more, are commonly referred to as the Western Electric Rules and can be found in any advanced quality reference. Sample number Statistics 30

Control Limits and Errors 4/20/2017 Control Limits and Errors Type I error: Probability of searching for a cause when none exists (a) Three-sigma limits UCL Process average As long as the area encompassed by the control limits is less than 100% of the area under the distribution, there will be a probability of a Type I error. A Type I error occurs when it is concluded that a process is out of control when in fact pure randomness is present. Given +/- 3 sigma, the probability of a Type I error is 1 - 0.9997 = 0.0003, a very small probability. LCL Statistics 32

Control Limits and Errors 4/20/2017 Control Limits and Errors Type I error: Probability of searching for a cause when none exists (b) Two-sigma limits UCL Process average When the control limits are changed to +/- 2 sigma, the probability of a Type I error goes up considerably, from 0.0003 to 1 - 0.9544 = 0.0456. While this is still a small probability, it is a change that should be carefully considered. In practice, this would mean more samples would be identified inappropriately as out-of-control. Even if they were subsequently ‘Okd’, there would be increased costs due to many more cycles through the four step improvement process shown previously. LCL Statistics 33

Control Limits and Errors 4/20/2017 Control Limits and Errors Type II error: Probability of concluding that nothing has changed (a) Three-sigma limits UCL Shift in process average Process average Returning to the 3 sigma limits, we can see the probability of making a Type II error, in this case failing to detect a shift in the process mean. LCL Statistics 34

Control Limits and Errors 4/20/2017 Control Limits and Errors Type II error: Probability of concluding that nothing has changed (b) Two-sigma limits UCL Shift in process average Process average By reducing the control limits to +/- 2 sigma, we see the probability of failing to detect the shift has been reduced. LCL Statistics 35

Achieve the purpose Our goal is to decrease the variation inherent in a process over time. As we improve the process, the spread of the data will continue to decrease. Quality improves!!

Improvement

Examine the process A process is considered to be stable and in a state of control, or under control, when the performance of the process falls within the statistically calculated control limits and exhibits only chance, or common causes.

Consequences of misinterpreting the process Blaming people for problems that they cannot control Spending time and money looking for problems that do not exist Spending time and money on unnecessary process adjustments Taking action where no action is warranted Asking for worker-related improvements when process improvements are needed first

Process variation When a system is subject to only chance causes of variation, 99.74% of the measurements will fall within 6 standard deviations If 1000 subgroups are measured, 997 will fall within the six sigma limits. -3 -2 -1 +1 +2 +3 Mean 68.26% 95.44% 99.74%

Chart zones Based on our knowledge of the normal curve, a control chart exhibits a state of control when: Two thirds of all points are near the center value. The points appear to float back and forth across the centerline. The points are balanced on both sides of the centerline. No points beyond the control limits. No patterns or trends.

Acceptance Sampling

Acceptance sampling is a form of testing that involves taking random samples of “lots,” or batches, of finished products and measuring them against predetermined standards. A “lot,” or batch, of items can be inspected in several ways, including the use of single, double, or sequential sampling.

Single Sampling Two numbers specify a single sampling plan: They are the number of items to be sampled (n) and a pre specified acceptable number of defects (c). If there are fewer or equal defects in the lot than the acceptance number, c, then the whole batch will be accepted. If there are more than c defects, the whole lot will be rejected or subjected to 100% screening.

Double Sampling Often a lot of items is so good or so bad that we can reach a conclusion about its quality by taking a smaller sample than would have been used in a single sampling plan. If the number of defects in this smaller sample (of size n1) is less than or equal to some lower limit (c1), the lot can be accepted. If the number of defects exceeds an upper limit (c2), the whole lot can be rejected. But if the number of defects in the n1 sample is between c1 and c2, a second sample (of size n2) is drawn. The cumulative results determine whether to accept or reject the lot. The concept is called double sampling.

Sequential Sampling Multiple sampling is an extension of double sampling, with smaller samples used sequentially until a clear decision can be made. When units are randomly selected from a lot and tested one by one, with the cumulative number of inspected pieces and defects recorded, the process is called sequential sampling.

OPERATING CHARACTERISTIC (OC) CURVES The operating characteristic (OC) curve describes how well an acceptance plan discriminates between good and bad lots. A curve pertains to a specific plan, that is, a combination of n (sample size) and c (acceptance level). It is intended to show the probability that the plan will accept lots of various quality levels.

Figure shows a perfect discrimination plan for a company that wants to reject all lots with more than 2 ½ % defectives and accept all lots with less than 2 ½ % defectives.

OC Curves for Two Different Acceptable Levels of Defects (c = 1, c = 4) for the Same Sample Size (n = 100). So one way to increase the probability of accepting only good lots and rejecting only bad lots with random sampling is to set very tight acceptance levels.

OC Curves for Two Different Sample Sizes (n = 25, n = 100) but Same Acceptance Percentages (4%). Larger sample size shows better discrimination.

Sampling Terms Acceptance quality level (AQL): the percentage of defects at which consumers are willing to accept lots as “good” Lot tolerance percent defective (LTPD): the upper limit on the percentage of defects that a consumer is willing to accept Consumer’s risk: the probability that a lot contained defectives exceeding the LTPD will be accepted Producer’s risk: the probability that a lot containing the acceptable quality level will be rejected

THE OPERATING-CHARACTERISTIC (OC) CURVE For a given a sampling plan and a specified true fraction defective p, we can calculate Pa -- Probability of accepting lot If lot is truly good, 1 - Pa = a (Producers' Risk) If lot is truly bad, Pa = b (Consumer ‘s Risk) A plot of Pa as a function of p is called the OC curve for a given sampling plan

The OC curve shows the features of a particular sampling plan, including the risks of making a wrong decision.

Construction of OC curve In attribute sampling, where products are determined to be either good or bad, a binomial distribution is usually employed to build the OC curve. The binomial equation is where n = number of items sampled (called trials) p = probability that an x (defect) will occur on any one trial P(x) = probability of exactly x results in n trials In a Poisson approximation of the binomial distribution, the mean of the binomial, which is np, is used as the mean of the Poisson, which is λ; that is, λ = np

Example Probability of acceptance A shipment of 2,000 portable battery units for microcomputers is about to be inspected by a Malaysian importer. The Korean manufacturer and the importer have set up a sampling plan in which the risk is limited to 5% at an acceptable quality level (AQL) of 2% defective, and the risk is set to 10% at Lot Tolerance Percent Defective (LTPD) = 7% defective. We want to construct the OC curve for the plan of n = 120 sample size and an acceptance level of c ≤ 3 defectives. Both firms want to know if this plan will satisfy their quality and risk requirements.

Example N=1000 n = 100 AQL=1% LTPD=5% ß=10% α = 5% C<=2 Does the plan meet the producer’s and consumer’s requirement.

AVERAGE OUTGOING QUALITY In most sampling plans, when a lot is rejected, the entire lot is inspected and all of the defective items are replaced. Use of this replacement technique improves the average outgoing quality in terms of percent defective. In fact, given (1) any sampling plan that replaces all defective items encountered and (2) the true incoming percent defective for the lot, it is possible to determine the average outgoing quality (AOQ) in percent defective.

AVERAGE OUTGOING QUALITY The equation for AOQ is

Example Selected Values of % Defective Mean of Poisson, λ = np P (acceptance) .01 1 0.920 .02 2 0.677 .03 3 0.423 .04 4 0.238 .05 5 0.125 .06 6 0.062

Example The percent defective from an incoming lot in is 3%. An OC curve showed the probability of acceptance to be .515. Given a lot size of 2,000 and a sample of 120, what is the average outgoing quality in percent defective?