Sie sind auf Seite 1von 80

how-to guide

how-to guide
PROCESS CAPABILITY: How toCAPABILITY Measure PROCESS & Improve How to continuously improve the capability of a process Process Capability

v2
v2

2011 Rolls-Royce plc The information in this document is the property of Rolls-Royce plc and may not be copied or communicated to a third party, or used for any purpose other than that for which it is supplied without the express written consent of Rolls-Royce plc. This information is given in good faith based upon the latest information available to Rolls-Royce plc, no warranty or representation is given concerning such information, which must not be taken as establishing any contractual or other commitment binding upon Rolls-Royce plc or any of its subsidiary or associated companies.

How-To Guides Available Statistical Process Control Measurement System Analysis Process Capability

how-to guide
KEEPING PROCESSES IN CONTROL: How to Construct & How-To Guides Available Use SPC Charts

how-to guide
MEASUREMENT SYSTEM ANALYSIS:
How to ensure measurement systems provide good quality data

how-to guide
PROCESS CAPABILITY
How to continuously improve the capability of a process

v5

v2

v2

Produced by Smallpeice Enterprises Ltd www.smallpeice.com

Setting the Scene: The Improvement Context for Process Capability


Introduction to the Rolls-Royce Improvement Journey The focus on quality in Rolls-Royce is not new. It is one of the companys key values, part of the brand and what Rolls-Royce is known for by our customers. In todays market, having a reputation for quality is more important than ever. That said, quality is not static, but something we need to work to achieve every day, in everything we do. The Improvement Journey to Process Excellence is the framework which helps us meet our business and quality goals, and this How To Guide will help everyone to understand and apply a fundamental technique within the Rolls-Royce continuous improvement toolkit. To sustain progress and create a culture of continuous improvement, it is vital that these How To principles are ingrained in the way everything is done at Rolls-Royce. Whether you work in a manufacturing or transactional environment, these principles apply in all parts of the business and need to be understood and adopted by everyone. Overview of the Improvement Journey Steps The Improvement Journey is a proven, benchmarked description of the 4 stages to reaching and achieving Process Excellence: Process Basics, Process Control, Process Flow and Capable Processes.

Once the basics are in place and controlled, the use of all steps together will remove waste and reduce variation, ultimately helping to speed up company improvements.

To help company leaders and employees understand where their respective businesses are on the Improvement Journey a global assessment tool has been formulated. The overall results of the assessment (against a number of key building blocks) provide a percentage score for each element of the model. Central to everything are the people in the organisation and the way we behave. Journey to Process Excellence principles should become our company DNA the way we think, our way of life, if we are to sustain our progress and create a Continuous Improvement culture. Already millions of pounds are being saved across the business through the application of these principles. A Structured Approach to Improvement The Process Capability techniques which are introduced in this How To Guide should not be applied in isolation without the foundation elements of the Improvement Journey already being in place. Step 1: Checking the Process Basics The first part of the Improvement Journey is to ensure that we have our Process Basics in place. Process Basics provides the business with a suite of practical tools to simplify and standardise our workplace. The 8 Process Basics steps can be applied rapidly in all parts of our business and will create considerable improvements in quality and performance (and therefore bottom line improvements): Asset Care Visual Management Performance Mgt - Output Process Compliance Standard Processes HS&E Workplace Organisation Leadership & People Visit www.infocentre.rolls-royce.com/process_excellence for detailed explanations and further information on each of the Process Basics steps. Step 2: Process Control Process Control builds upon the foundation of Process Basics. The objective of process control is to allow us to take control of our processes. A process that is in control is one that is stable and predictable. We can predict within certain limits the outputs we are going to get. It is not possible to improve the flow or capability of a process until the process is under control. Therefore checking process control has been achieved is an important pre-requisite to measuring and improving Process Capability.

Step 3: Process Flow Process Flow builds on the foundations of the Process Basics and Process Control. Process Flow is concerned with identifying and eliminating waste in order to simultaneously improve on time delivery performance and make processes more efficient. On its own however Process Flow is not enough, the process must also be capable. Lack of Process Capability is sometimes the root cause of issues with the process flow. Process Capability must therefore be measured and understood at the same time as designing processes for flow. Step 4: Process Capability Process Capability is concerned with measuring and improving the quality performance of our processes. The ultimate vision for Process Capability is for all processes to achieve zero defects. The first step to Process Capability is to be able to measure and analyse current capability performance. That is the focus of this guide. Process Basics and Process Control are pre-requisites to measuring and improving Process Capability.

PRE-REQUISITES TO READING THIS GUIDE Before reading this guide you should first have read the How-To Guides for SPC and MSA. This guide assumes that you are familiar with the terminology and tools covered by these earlier guides in the series

how-to guide
KEEPING PROCESSES IN CONTROL: How to Construct & Use SPC Charts

how-to guide
MEASUREMENT SYSTEM ANALYSIS:
How to ensure measurement systems provide good quality data

v5

v2

CONTENTS INTRODUCTION 1.1: The need for Process Capability analysis 1.2: What is Process Capability? 1.3 Process Capability metrics 1.4: Understanding and Interpreting measures of potential Process Capability (Cp & Pp) 1.5: Understanding and Interpreting measures of actual Process Capability (Cpk and Ppk) 1.6: Interpretation of the metrics: when is the process capable? CONDUCTING A PROCESS CAPABILITY ANALYSIS Step 1: Pre-Study Requirements for Process Capability Ensuring the basics are in place Checking the process stability Setting specification limits Step 2: Collect and Structure the Data Step 3: Determining the Correct Capability Analysis to Use Determining the data type Determining the probability distribution Step 4: Carry out the Capability Analysis For Continuous Data Normal For Continuous Data Non Normal For Attribute Data Binomial Step 5: Improving the Process Compare actual capability to desired capability Make a decision concerning process changes Report results of study with recommendations Step 6: Maintaining Process Capability

CONTENTS

APPENDICES Appendix 1: Entering Data into Minitab For continuous data For binomial data Appendix 2: Checking the Normality Assumption Anderson-Darling in Minitab The normal distribution Appendix 3: Process Capability Analysis for Normal Data in Minitab Appendix 4: Distribution Fitting for Non-Normal Data in Minitab Appendix 5: Process Capability Analysis for Non Normal Data in Minitab Appendix 6: Process Capability Analysis for Binomial Data in Minitab Appendix 7: Basic Statistics for Process Capability Analysis Estimating process parameters Appendix 8: Formulae Explained Short term: Calculating Cp & Cpk Long term: Calculating Pp & Ppk

HOW TO USE THIS GUIDE This How To Guide is designed as a complete training package that you can work through individually at your own pace (or in small teams as part of a facilitated training exercise). By carefully reading the text, and practising the tools in the associated Workbook, you will become competent and confident in using these process capability tools in your work area. The How To Guide is designed to be applicable for use primarily by Manufacturing Engineers, Design Engineers and Lean Sigma Practitioners from any area of the business. For this reason, the technical explanations are based on general business application examples to ensure everyone can relate to them. Throughout the guide, there are case study examples which show how the theory is applied at the different stages of the process control sequence. Before you start, make sure you also have the Workbook available. It is essential that you work through this in parallel with the How To Guide, and that you complete the practise questions, plus case study exercise before you start to use Process Capability techniques in the business.

Icons are used throughout to highlight key elements, and to signpost supplementary information where appropriate. The technical explanation of the core terminology
SEEK GUIDANCE F.A.Q

The approved Rolls-Royce answers to main queries asked by users

Indicating where you must seek help from practitioner experts such as Black or Green Belts Where you can find additional information, and the next phase of the improvement journey Indicates a key learning point

Tips on the commonly observed pitfalls - & how to avoid them

SIGNPOST

WORKBOOK EXERCISE

KEY POINT

The separate Workbook which you must use in parallel with your learning

GUIDE STRUCTURE

The flowchart below illustrates the structure of this Guide. The Guide provides step by step instructions for conducting and interpreting the Process Capability Analysis. Steps 1, 2, 3, 5 & 6 below are common for all types of capability study. Step 4 differs depending on the type of the study (see step 2 for full guidelines on how to select the appropriate type of study). Supplementary information on how to carry out the analysis using Minitab statistical software is provided in the appendices as outlined below. Additional optional material is also provided in Appendix 7 & 8 explaining the underlying statistics and equations used for each analysis method. Step 11 Pre-Study Requirements

Step 22 Determining the correct metric/analysis to use

Appendix 1: Checking the normality assumption

Step 33 Collect the data

Appendix 2: Entering the data in Minitab

Step 44 Type of Study

Normal Capability Analysis Appendix 3

Non-Normal Capability Analysis Appendix 4 & 5

Binomial Capability Analysis Appendix 6

Step 55 Improving the process

Step 66 Maintaining the improvement

Section 1

Introduction
.................................................
In this section: 1.1: The need for Process Capability Analysis 1.2: What is Process Capability? 1.3 Process Capability Metrics 1.4: Understanding and Interpreting measures of Potential Capability and Performance (Cp & Pp) 1.5: Understanding and Interpreting measures of Process Capability (Cpk ) and Performance Capability (Ppk) 1.6: Interpretation of the metrics: when is the process capable?

.................................................
1.1: The Need for Process Capability Analysis
At Rolls-Royce we manage thousands of processes everyday. Each one of our processes has a customer or customers (either internal or external) who receive the outputs from each process. Our process outputs may be a physical part or assembly, a design, information or a service. In all cases we need to know how good each process is at delivering its output to our customers . This information enables us to effectively manage and improve process performance and to keep our customers happy. Without reliable measures of how good our processes are we do not have the information that we need to manage and improve our processes or understand our priorities! We must therefore develop methods to quantify and measure how good our processes are at satisfying our customers. We must also have a mechanism to allow us to compare the performance of different processes using common measures. Process Capability analysis provides us with a common set of comparable measures for measuring how good our processes are at satisfying our customer requirements. Capability analysis can help answer the following questions : Is my process meeting customer specifications? How will the process perform in the future? Are improvements needed in the process? Have we sustained these improvements, or has the process regressed to its previous unimproved state?

Introduction

1.2: What is Process Capability?


Process capability is broadly defined as the ability of a process to satisfy customer expectations. Some processes do a good job of meeting customer requirements and therefore are considered capable, whilst others do not and are designated not capable.

Process capability is the ability to produce products or provide services that meet specifications defined by the customer's needs. Capability analysis reveals how well each of our processes meets these specifications, and provides insight into how to improve the process and sustain your improvements. The concept of measuring and reporting how good our processes are is of course not new. Many processes already have measures in place to assess their ability to meet specification. The most common approach to measuring process capability is simply to measure the percentage pass or fail rate for the process. Percentage pass (or fail) is probably the most widely used and well understood method of summarising process performance. However what is often less well understood is that simply counting the number of units that pass or fail inspection or test is not the best or the most powerful method of calculating process capability.

3
1.2: What is Process Capability (continued)
To understand this consider the two data sets below:

Introduction

These two histograms each visualise the weight of a cake in grams (represented by the grey bars) in relation to the customer requirement that the weight must fall between (198g 202g represented by the two lines). In both cases we see that 100% of the cakes produced in these samples fall between the minimum and maximum weight requirements. Both processes can therefore claim, using % count as the measure, that their process capability is 100% good. However, as you can see this does not tell the whole story! Process 1 has much more variation in cake weight than process 2. We can see therefore (assuming both processes are in control) that there is a much higher likelihood of process 1 making a defective cake than process 2. Process 2 is visibly more capable than process 2 but how can we put this into numbers?

Introduction

1.3: Process Capability Metrics


When assessing process capability we need to use more than one metric to describe the process. A good way to understand the need for multiple metrics is to consider the different ways in which a process can fail to meet a customer requirements. There are three different scenarios which result in a process being not capable which we will illustrate here by returning to the cake weight example: 1) The process variation is too large

Here the amount of variation (the spread) in cake weight is wider than the acceptable tolerance limit of 198 202g which means that the process is not capable of delivering 100% of the cakes within customer requirements. 2) The process average is not properly centred

Here the amount of variation (the spread) of the cake weight is smaller than the acceptable tolerance limit however the average weight of the process is above the target weight of 200g. This means that the process is likely to produce a cake which is too heavy and is therefore not capable of delivering 100% of the cakes within customer requirements despite having a relatively small process spread.

5
1.3: Process Capability Metrics (continued)
3)

Introduction

The process average is not properly centred and the process variation is too large

Here you can see that the process spread is too wide and the process average is above the centre weight of 200g. This means that the process is not capable for more than one reason. Over time different capability metrics have been developed to cover each of the above scenarios. In this guide we will cover how to calculate and interpret metrics for two of the above scenarios: Metrics which consider only process variation (commonly known as potential capability metrics) Metrics which consider both process variation and centring of the average (commonly known as actual capability metrics) In addition, for both potential and actual capability metrics, we will also differentiate between what is known as short term and long term process capability. This will be fully explained later in this guide but in general terms this relates to the way that the process variation is estimated as explained on the next page.

Introduction

1.3: Process Capability Metrics (continued)


Understanding Short Term & Long Term Variation Process Capability can be calculated using either Short Term or Long Term metrics. The difference between the two relates to the way in which the process spread (variability) is calculated. Short term variation relates to variation that happens in the short term. For example within the period of one day, one shift or one machine batch run. We must be careful when using short term data as it will not reflect all of the variation that exists in a process as the process shifts and drifts over time. This is illustrated in the diagram below:

Day 1

Day 2

Day 3

Combined Variation

Short term

Long Term

Short term

Short term

In this illustration we can see how a process may change slightly from one time period to the next. These changes could be due to variation in the machines, methods, materials, operators or environmental conditions over time and are to be expected. If all of the data is combined then the longer term view gives the complete variation in the process. This is known as long term variation and takes into account the overall process variation. Short Term Variation is the variation seen in the short term only Short Term Process Capability is calculated using the spread seen solely from short term variation Long Term Variation is the variation of the overall process population Long Term Process Capability (commonly known as Process Performance) is calculated using the spread seen from long term variation

7
1.3: Process Capability Metrics (continued)

Introduction

In summary then, we will look at capability metrics for both potential capability and actual capability and for each will also consider where appropriate both short term and long term capability metrics. The metrics which Rolls-Royce uses (following common standards used within most manufacturing companies) are as follows:

KEY POINT

Capability (Short Term) Potential Actual Cp Cpk

Performance (Long Term) Pp Ppk

*Note: Cp & Cpk are only used for Normally Distributed Data

The correct choice of metric will depend on how the data has been collected, the data type and distribution and on which aspects of the process capability you are interested in. We will now provide an introduction to understanding and interpreting each of the above metrics. Full details of how to select and calculate each metric will be provided in section 2.

Introduction

1.4 Understanding and Interpreting Measures of Potential Capability


Defining Potential Capability In section 1.3 we learned that Cp and Pp are capability metrics which consider only process variation and that these measures are commonly known as potential capability metrics.

Potential capability is so called because, by not taking the location of the process average into account, it reveals what could happen if the process is centred. In other words it allows us to understand the potential performance of the process if the process is centred.
Potential capability metrics consider only process spread when calculating how good the process is Potential capability reveals how the process could perform if the process is centred To understand this better lets return to some of the cake weight scenarios we looked at earlier:

These two processes have an identical process spread. In the example to the left the process average is perfectly centred between the customer specification limits whilst in the example to the right the process average is off centre. Considering only process spread both processes have the same potential process capability. The process to the left is perfectly centred so already performs to its full potential. The process to the right is off-centre so does not perform to its full potential.

Introduction

1.4 Understanding and Interpreting Measures of Potential Capability continued


A common pitfall is for potential capability measures such as Cp or Pp to be used incorrectly to describe actual process performance. Remember that it is quite common that the potential and actual process performance differ. Therefore the use of potential process capability measures is best restricted to helping understand the improvement potential of our processes or for rare circumstances when the position of the process average is not important.
KEY POINT

Understanding Potential Capability The potential capability metrics we will use are Cp and Pp. Both are interpreted in essentially the same way. In general terms these potential capability metrics are a comparison of tolerance width against process spread. A simple ratio of Tolerance/Spread. To understand this, consider the example below. Here we can see that the process performance is clearly poor with 100% of the cake weights sitting beyond the maximum weight. The process potential however is quite good. You can see clearly that the process spread (approximately 2g) is small in comparison to the process tolerance (from 198g to 202g which equals a tolerance band of 4g). In this case the Tolerance/Spread = 2. For Cp & Pp metrics higher numbers represent better capability than lower numbers. This process has the potential, if centred, to perform well although in reality it has a big performance problem!

Tolerance Width

Spread

Introduction

10

1.4 Understanding and Interpreting Measures of Potential Capability continued


For this example, to understand the potential we can consider how many times the spread of this process could fit within the tolerance width. In this case the tolerance width is twice that of the process width giving a potential process capability metric of 2.

Tolerance Width

Spread

Spread

Quite often a process is considered to possess potential capability if its spread is equal to (or less than) the width of the tolerance. The narrower the process variation the greater the potential. If the process spread fits less than one times into the tolerance then the process potential would be considered not capable.

The precise methods used for estimating spread of the data and for calculating these metrics will be covered in section 2.

11

Introduction

1.5 Understanding and Interpreting Measures of Actual Capability


Defining Actual Capability In section 1.3 we learned that Cpk and Ppk are capability metrics which consider both process variation and centring of the average process variation and that these measures are commonly known as actual capability metrics.

Actual capability is so called because, by taking into account both process variation and the location of the process average into account, it reveals the actual expected performance of the process. To understand this better lets return to some of the cake weight scenarios:

These two processes have very similar process spreads and so have the same potential capability. However in the example to the left the process average is perfectly centred between the customer specification limits meaning it has a good actual capability. Whilst in the example to the right the process average is off centre meaning that is does not have a good actual capability. Moving the process average does not affect potential process capability but greatly affects actual capability. Actual capability metrics consider both process variation and centring of the average process variation Actual capability measures how well the process output actually conforms to specification

Introduction

12

1.5 Understanding and Interpreting Measures of Actual Capability continued


Understanding Actual Capability The actual capability metrics we will use are Cpk and Ppk. Both are interpreted in essentially the same way. Whilst the Cp and Pp solely considered the process spread in relationship to the tolerance width, the Cpk and Ppk metrics must also take into account the position of the process average. To understand this consider the cake weight scenarios below:

Scenario 1

Scenario 2

Scenario 3

In each of the above scenarios the process spread is identical and is approximately 2g wide. The process average however is different in each case. In scenario 1 the process average sits outside of the tolerance limits altogether at 203g, in scenario 2 the average sits on the upper tolerance limit at 202g and in scenario 3 the average sits at 201g. All have equal potential capability. Clearly out of the three the process that has the best actual capability is scenario 3 but how do we quantify this? For the process to be capable the process variation must be able to fit easily in the space between the process average and the closest tolerance limit. For example if we come back to scenario 3, the distance between the process average and the closest tolerance limit is 202g 201g = 1g. The process could only 100% satisfy the customer if it were able to squeeze half of its spread into this space which in this case you can see it is just able to do.

13

Introduction

1.5 Understanding and Interpreting Measures of Actual Capability continued


Calculating Actual Capability
KEY POINT

Distance from process average to closest customer limit = 202 201 = 1g Process spread = 200 202 = 2g

In our example (scenario 3) above we can see that half the process spread is 1g (estimated from the total spread seen in the histogram). This exactly fits once into the distance between the process average and the closest customer limit which therefore gives it an actual capability metric of 1.
Quite often a process is considered to be capable if half of its spread can fit one or more times into the distance between the process average and the closest specification limit. Therefore this process is capable. If you return to scenario 1 & 2 on the previous page you can see that neither of these is capable. Scenario 1 actually has a negative distance between its average the closest specification limit (202g 203g = -1g) which gives it a negative actual capability. This means that it is more probable that a cake will be out rather than in specification and this can be clearly seen from the graph. In Scenario 2 the process average (202g) sits directly on the customer upper tolerance limit. This means that there is zero distance between the process average and the customer specification limit giving the process an actual capability metric of 0. This means, as can be seen, there is an equal probability (50:50) of getting a cake in or out of specification. The precise methods used for estimating spread of the data and for calculating these metrics will be covered in section 2.

Introduction

14

1.5 Understanding and Interpreting Measures of Actual Capability continued


Calculating Actual Capability (continued) The difference between Short Term and Long Term Actual Capability The difference between short term actual capability and long term actual performance is the method used in estimating process variation. For actual process capability we are concerned with how many times we can fit half of our process spread into distance between the process average and its nearest customer limit. For Short Term Actual Capability (Cpk) the short term variation is used to calculate the process spread. For Long Term Actual Performance (Ppk) the long term variation is used. Full details of how to calculate these spreads is given in Appendices 7 & 8.

15

Introduction

1.6 Interpretation of the Metrics: When is my Process Capable?


To judge how good the process capability or performance is for an identified critical characteristic, the measure generated by a capability study must be compared to some desired goal. Obviously within any business different processes and product or service characteristics will have diverse process capability requirements. For example the capability requirement for a safety critical dimension is likely to be more stringent than the requirement for the time it takes to issue the invoice for that part. The capability goal must always be understood before assessing how capable a process is. However as a general guide, the following guidelines are often used for assessing process capability :
KEY POINT

Process

Actual Capabili ty

Total amount outside limits > 50%

Typical action to be taken

<0

Stop process. Process improvement required Stop process. Process improvement required Heavy process control, sorting, rework, etc. Heavy process control inspection

50%

0 - 1.0

> 5%

1.0

0.3%

1.33

64ppm

Reduced inspection, selected use of control charts Spot checking, selected use of control charts

2.0

0.001ppm

Section 2

16

Conducting a Process Capability Analysis


.................................................
In this section: A step by step guide for conducting a process capability analysis The key steps we will cover are as follows: Step 1

.................................................
Pre-study requirements for Process Capability Analysis

Step 2

Collecting and Organising the Data

Step 3

Determining the Correct Metric to Use

Step 4

Carrying Out and Interpreting the Capability Analysis

Step 5

Improving the Process

Step 6

Maintaining Process Capability

17

Conducting a Process Capability Analysis Conducting a Process Capability Analysis

18

Case Study Examples


We will introduce you to the application of these steps using three example scenarios which you can work through one each for normally distributed, non-normal and binomial data. You will then have the opportunity to turn to the workbook and practise applying the tools yourself using further case study examples. Case Study - Scenario 1 Anne works for a manufacturer of golf clubs and is the production engineer responsible for the production line that manufactures the steel shafts for their range of golf irons. A key characteristic of the steel shafts manufactured on her production line is their diameter at various points down the shaft. Anne is particularly interested in the manufactured diameter of the tip of the shaft. This is currently manufactured to be 9.400mm +/-0.2mm but due to negative feedback from end customers about the variable performance of the golf club range, the designers have made some design alteration to the club which requires the tolerance for the shaft tip diameter to be tightened to +/-0.1mm. Anne suspects that her current process should already be capable of manufacturing to these tightened customer specification limits but she needs data to fully understand the capability of the process and thus the likelihood of her process failing to deliver a club which meets the new specification. The company have budgeted that, to avoid future negative publicity, 99.9% of all the shafts must be made to this new specification level. Anne decides to carry out a process capability analysis to find out whether her process requires improvement to meet this new specification.

Conducting a Process Capability Analysis

18

Case Study Scenario 2 Jim works for the same golf club manufacturer as Anne but his job is in the customer service department. His team answer calls to their dedicated Customer Helpline and his team pride themselves on their high level of customer support and after-sales service.

Jims team is small and so have never adopted or measured themselves against service level agreements. Recently however he has noticed a growing number of customers complaining about the length of time they have had to wait on hold before their call is answered by one of his advisers. A recent article in a golfing magazine compared his companys service levels against some of their competitors and found them to come second last in terms of responsiveness. Jim has therefore been set the target to measure and improve call waiting times.
The first thing Jim does is to benchmark against his competitors and look at customer feedback to establish what an acceptable waiting time should be. From this he finds that the maximum waiting time that customers find acceptable is 30 seconds. He decides to collect data from his own process to establish his current process capability against the target of 30 seconds. As this type of data has not been collected before, Jim decides to collect it manually, with a simple data collection format to capture the time that the call was made and the duration of the call, measured in seconds.

19

Conducting a Process Capability Analysis

Case Study Scenario 3 Pat also works for the same golf club manufacturer. He works in the shipping department and is responsible for ensuring that all orders received are shipped right first time. His team believe that they are pretty good at shipping customers orders correctly. They do however occasionally ship an order that is not complete (usually where the customer has ordered additional accessories as well as clubs) and always quickly rectify any customer complaints. However the same magazine article that criticised the company for poor customer service also included customer feedback about receiving incorrect orders. Pat has been asked to review his process and make improvements where required. Since joining the department over two years ago Pat has kept careful records tracking for every month how many orders are shipped and out of those how many have to be reworked due to mistakes made in shipping. To date he has never done anything with this data other than to collect and review it. He now thinks there may be an opportunity to analyse the data more thoroughly to investigate the size of the problem and look for opportunities for improvement.

Conducting Pre-study a Process Requirements Capability Analysis for Process Capability Analysis

20

Step 1

1a) Ensure the basics are in place 1b) Check Process stability 1c) Set the specification limits Process Capability improvement builds on the earlier steps of the Process Excellence Journey. It is not advisable to measure process capability and not possible to improve it until the process basics are in place and the process is stable and in control.

As an initial step it is important to begin by checking that the process basics 1a) are in place. In particular the following should be checked: Ensuring Prerequisites: The basics Check the The work area where the process of interest is performed process is safe and well organised basics are Any equipment being used is in good working order and in place well maintained There is a standard operation in place for the process of interest There is evidence that the standard operation is being complied with For detailed guidance on how to assess and improve the process basics please go to RRPS Gain & Maintain Control How To Guide It is also vital to check the that the measurement system being used to collect the data is reliable. Where possible this should be assessed formally using Measurement Systems Analysis. As a minimum the following should be checked. Prerequisites: the measurement system Equipment used is calibrated Check

SIGNPOST

Equipment used has sufficient measurement resolution


That the measurement system is suitable for the full range of measurements Equipment used is stable That there are clear operational definitions in place The repeatability and reproducibility of the measurement system is known and is sufficient for the measurement of interest
SIGNPOST

For full details on how to check the Measurement System, see the How To guide available on Measurement System Analysis.

21

Pre-study requirements Pre-study for Requirements Process Capability for Process Analysis Capability Analysis

Process capability studies involve forecasting the future performance of the 1b) process output. This is an impossible task if the past process performance Check does not provide a sound basis for prediction [see the illustration below]. Process Thus, before any type of meaningful capability study can be undertaken, the Stability process being studied must be stable.

The illustration above shows an unstable process. The process performance changes and is unpredictable from one day to the next. Because of the instability there is no way of assessing its current or future ability to satisfy customer requirements. A process which is stable is in control. This means that the amount of variation in the process is consistent and predictable over time [see the illustration below].

When a process is stable it is repeatable, well defined and predictable. Process stability furnishes a high degree of assurance that the future will closely resemble the past and is thus an essential pre-requisite to conducting a process capability study.

Pre-study Requirements Be Prepared for Process Capability Analysis

22

Check Process An SPC Chart enables us to quickly see how a process performance is Stability changing over time is it getting better, getting worse, or staying the continued same? It helps us to quickly and easily identify any problems with the process It helps us to decide whether we need to take any action if the process appears to be getting worse (or better) It helps us see at a glance whether our process is in control

Step 1 1b)

One of the best ways to test if a process is stable is to use Statistical Process Control (SPC).

SPC Chart of Monthly Conumables Spend


2300 2200 Upper Control Limit

Consumables Spend ()

2100 2000 1900 1800 1700 1600 1 2 3 4 5 6 7 Month 8 9 10 11 12 Lower Control Limit _ Centre Line

An SPC chart plots data collected from a process in time order. As you can see in the example above Upper and Lower Control Limits are plotted to mark the values between which we would expect the majority of the data points to fall when the process is in control. If one of more points fall outside the control limits or if a trend, shift or pattern can be seen in the plot then this indicates that special cause variation may be present and thus that the process is not stable. Special Cause Variation is unpredictable variation resulting from one or more assignable causes acting on the process Special Cause variation can often be seen as a spike or a shift in the process performance over time A process which is Out Of Control has Special Cause Variation present

23 Step 1 1b)

Be Prepared Pre-study Requirements for Process Capability Analysis

If the SPC Chart shows that the process has special cause variation present then this must be fully investigated and, where possible eliminated, before Check proceeding with the Process Capability study. Process Stability continued

SIGNPOST

For full details on how to construct and interpret Statistical Process Control Charts see the How To guide available on Constructing and Using SPC Charts.

SEEK GUIDANCE

If you are unsure how to use SPC charts to check for stability or if you find that your process has special cause variation present, contact a local Black Belt for guidance.

Pre-study Requirements Be Prepared for Process Capability Analysis

24

Process capability is the ability to produce products or provide services that meet specifications defined by the customer's needs . In order to be able to Set the calculate process capability we therefore need to know what the specified Specification customer need is. Limits For example customer needs could be: - Delivery time of less than 24 hours - A dimension of between 24mm 25mm - A visual characteristic (such as colour) that passes inspection against a defined standard These specifications are usually found in documentation such as customer specification documents, engineering drawings or service level agreements. In all cases it is important to check that the specifications being used to assess process capability are both up to date and reflective of the actual customer needs. Specification Limits are values between which products or services should operate. Specification limits are usually set by customer requirements. Specification limits usually consist of both upper and lower Specification limits (for example the cake must weigh between 198g and 202g). These are sometimes referred to as bilateral specification limits. There may be situations where only one specification limit is appropriate (sometimes known as a unilateral specification limit), such as lead time. You may require an item to be delivered within 5 days this forms an upper specification limit but if you require the item the faster the better then there would be no lower limit. Specification limits are different from control limits. Specification limits are based on what is required for proper function or appropriate service. Control limits are calculated from process data. They represent how your process actually performs, while specification limits show the desired performance. Specification limits are commonly known interchangeably by other terms including customer acceptance limits and tolerance limits.

Step 1 1c)

25 Step 1 1c)
Set the Specification Limits continued

Be Prepared Pre-study Requirements for Process Capability Analysis

Make sure that you do not confuse the control limits on an SPC chart with the customer requirements or tolerance limits. These are NOT the same thing. Control limits are derived from variation in the process (the Voice of the Process) whilst specification limits are derived from the actual customer requirements (the Voice of the Customer).
A common pitfall, particularly in business areas where customer specifications are less formally captured (such as for response times to internal customers or service levels for transactional processes), specification limits can sometimes be set based on assumptions rather than facts. For example we may assume all requests must be responded to in 24 hours when in fact the customer would be happy to have a response within 5 days. Conversely we may set a specification limit for example to get things right first time 95% of the time which does not in fact satisfy the customer who perhaps needs things right 100% of the time. If in doubt make sure you understand the customers real requirements.

Pre-study Requirements for Process Capability Analysis

26

Case Study - Scenario 1 Annes production environment is highly controlled with standard processes in place which are complied to and a high degree of workplace organisation. Therefore, she is confident that the process basics are in place. Because of the criticality of the dimension of the golf clubs the measurement systems are also carefully controlled and the last measurement system analysis carried out 2 months ago confirmed that her measurement system was reliable.

Anne already has data available from her process as it is regularly sampled to monitor the process stability using an SPC Control Chart. Manufacturing is done on two shifts 5 days per week. Every shift (morning and afternoon) a consecutive sample of 5 shafts is taken and measured and this data is recorded and plotted on an SPC chart. Anne checks the SPC chart covering the last two weeks production:

There are no points out of control on the chart and no evidence of any trends, shifts in the process or patterns. Anne is confident that the process is in control and that all the pre-requisites are in place for her to proceed with the capability study. Anne calls the club designer who advised her of the need to tighten the tolerance of the shaft tips diameter to the club which requires the tolerance for the shaft tip diameter to be tightened to be 9.4+/-0.1mm to check that this tolerance really does reflect the customer requirements. The designer confirmed that trials involving their customers feedback had confirmed that this was the required tolerance to achieve the required consistency of performance for the golf club. Anne is therefore confident to move on to the data collection stage of her project.

27

Pre-study Requirements for Process Capability Analysis

Case Study Scenario 2 Jim is also confident that his area has the process basics and are in place. To ensure consistency of service his team are all trained to standard operating procedures for the answering, handling and logging of customer calls. He regularly audits the team to check for compliance and is sure there are no issues. As he doesnt currently have any data he is not sure whether the process is stable. He will need to check this once his data has been collected. He notes that it is a pre-requisite to have a reliable data collection system. He already has some ideas about how to collect the data but decides to ask a Black Belt from the Companys continuous improvement team to come over and help him design a fit for purpose measurement system. As he has already analysed both benchmarking data and a representative sample of his own customers feedback he is confident that setting a maximum acceptable time on hold of 30 seconds is representative of the customer requirements. He therefore, feels ready to move forward to the data collection planning phase of his study.

29

Pre-study Requirements for Process Capability Analysis

28

Case Study Scenario 3


Pat considers his process basics. His team are proud of their high level of workplace organisation. They also have standard processes for picking and packing the orders. Pat does wonder if these processes may need revision but is nevertheless sure that his team do comply with them as they stand so feels confident that the basics are sufficient. His data collection system has never been formally checked for reliability. The system is fairly simple though there are clear definitions documented for what counts as an order and what count as a defective order . Pat always completes the log personally so knows there cant be any agreement issues between data recorders. As he always follows the same operational definitions for counting the orders and the orders with complaints he feels confident that his data is reliable. He is not sure how to check whether the data is stable or not and so asks a local Black Belt to help him analyse his data. The Black Belt enters Pats data into Minitab and shows him how to run a P -chart to check the process is in statistical control with the following result:

The SPC chart shows no special cause variation in the process over the past 30 months. Pat is therefore confident that his process is stable over time.

Since Pats data is counting defective orders he is not quite sure what his specification limit is. He asks the Black Belt for advice on this too. She explains that Pat needs to check that his operational definition of a defective order matches what the customer thinks of as defective. Pat defines a defective order as one where the content shipped does not match the content on the customer order form. He knows from talking to customers that this matches the customers understanding of a defective order.
Pat is pleased to find that all the pre-requisites to running a capability analysis seem to be in place and asks his Black Belt to stay to help him with the capability analysis.

Collecting and Organising the Data

Step 2

Data Collection Guidelines To carry out the process capability analysis we must use a sample data set which is representative of the complete process. Data must be collected in time series order and at a frequency and sufficient length of time to fully represent all of the variation present within the process. For example if there are multiple shifts then data must be collected across all of these. If data already exists or has been collected for the construction of SPC control charts for the process then this same data can be used so long as the process is stable (as discussed in Step 1). Sample Size The principle concern must be to have sufficient data to be representative to capture the full variation which can be expected in the process. It is recommended that as a minimum at least 25 30 data points are used.

SEEK GUIDANCE

If you are unsure how much data to collect then please seek the advice of a local Black Belt. Sample Frequency and Subgroups In order to calculate performance (long term) we need to ensure that our data collection process captures all potential variation such as changes in operators, machines, materials or operating conditions. Determination of how long is long enough should be determined by the subject matter expert.

Collecting and Organising the Data

30

Case Study - Scenario 1 As discussed, Anne already has data available from her process as it is regularly sampled to monitor the process stability using an SPC Control Chart. Manufacturing is done on two shifts 5 days per week. Every shift (morning and afternoon) a consecutive sample of 5 shafts are taken and measured and this data is recorded in a spread sheet. Anne believes that most of the sources of variation that she would expect to influence the process such as changes in the materials, machine adjustments, staff and environmental factors should be seen within any two week period. She therefore decides to use the last two weeks SPC data for her study. She collects this data from the SPC spreadsheet and enters it into Minitab. She is now ready to analyse the data.

Case Study Scenario 2 Jim needs to design a data collection system with the help of his Black Belt. Ideally he would like to update his call handling system to allow for automatic data collection however this isnt realistic in the short term. He therefore decides to use a simple manual spreadsheet for each operator to start a stopwatch when the caller is placed on hold and stop the stopwatch when the call is picked up. This is less than ideal as a data collection system and for this reason Jim decides initially just to collect the data for one day. The local Black Belt helps Jim and his team to agree robust operational definitions for how the stopwatches should be used, when to start and stop the clock and how to record the data. They run a Measurement System Study to confirm the R&R of the system by each measuring 20 pre-recorded calls a total of 3 times each. This confirmed that the measurement system was reliable and so Jim and his team commenced with the data collection.
SIGNPOST

To review more guidelines on entering data into Minitab please turn to appendix 1

31 Step 2

Collecting and Organising the Data

Case Study Scenario 2 continued Jim and his team collect only a days worth of data but he has satisfied himself that it is representative and of sufficient quantity. He therefore decides that it will be suitable for the purposes of this study. He enters it into Minitab, remembering that an important pre-requisite for carrying out the study is that the process is stable. He therefore runs the appropriate SPC chart (an I MR Chart) to make sure that the process is stable over time with the following result:

All the data points are between the control limits and there is no evidence of any trends shifts or patterns in the data. Jim therefore concludes that the process is stable and that the data is suitable to use for his capability study.

Case Study Scenario 3 Pat, like Anne, already has his data. Having already organised it in Minitab to draw the SPC chart he is now ready to go with his process capability analysis.

Determining the Correct Capability Analysis to Use

32

Step Step 2 3

3a) Determining data type 3b) Determining the probability distribution

The type of process capability analysis required and the best capability Choosing metric to use depend on a number of factors as follows: the correct The type of data continuous data and attribute data are each metric: analysed using different capability methods Overview The probability distribution of the data for example, for continuous data it is important to know whether or not the data is Normally Distributed Whether we are interested in the Potential Capability or the Actual Capability of the process Whether we want to use Long Term or Short Term metrics In the following section we will cover each of these considerations and provide a summary flow chart which you can use to determine which metric or metrics you will to calculate. The first consideration is the type of data being used. There are two data 3a) types: continuous and attribute. Determining Data Type Where the data from the process is measured on a continuous scale such as time, weight, dimensions or pressure then we will be collecting and analysing numerical results (such as the dimension of a part in microns). This type of data is continuous data. Continuous data comes from measurements on a continuous scale such as: temperature, time, distance, weight, dimensions. Where the data from the process is measuring the count of items which fall into different categorises such as pass/fail or counting defects such as scratches in paintwork then the data is said to be Attribute data.

Attribute data is based on upon counting how many units fall into discrete distinctions such as: pass/fail or percentage defective.

33
Determining the probability distribution

Determining the Correct Capability Analysis to Use

Step we know the data type, the next step is to determine the probability 3 3b) Once distribution of the data.

The probability distribution of a data set is essentially a description of the datas shape. This shape once known can be described by a statistical equation . The equation in turn can then be used to calculate the likelihood of achieving results within any particular range of values from the process. Minitab uses these likelihoods to calculate the process capability. Therefore to get a reliable measure of process capability it is important that we ask Minitab to use the correct distribution. There are two families of probability distributions one for continuous data and one for attribute data. Continuous Data can follow a large number of different distributions. One of the most common and well known continuous data distributions is the Normal Distribution (often referred to as the bell shaped curve). Process capability for continuous data is calculated differently depending on whether the data is Normally Distributed or not. If the data does not follow the Normal distribution then we say it is Non-Normal.

SIGNPOST

To find out more about the Normal Distribution and how to test whether continuous data is Normally Distributed please turn to Appendix 2

SEEK GUIDANCE

If you are unsure about how to determine whether your data is Normally distributed or not then please refer to your local Black Belt

Determining the Correct Capability Analysis to Use

34

Step 2
Case Study - Scenario 1 Anne is measuring the diameter of her club shaft in mm. She therefore recognises that her data is continuous. Because of the nature of her manufacturing process Anne would expect her data to be normally distributed. She runs a Graphical Summary of the data in Minitab to check:

The graphical summary confirms Annes expectation that the data is normally distributed. The p-value for the Anderson-Darling test is > 0.05 and the histogram shows the data to be approximately bell shaped. Therefore Anne confirms to herself that she must use Normal Capability Analysis for her dataset.

35

Determining the Correct Capability Analysis to Use

Case Study Scenario 2 Jim is measuring hold time in seconds so also has continuous data. His next step is also to check for normality. First he carried out a graphical summary which is good practice as he could study the shape of the histogram and get some insight into whether or not the data was likely to be normal:

Jim noted that the data did not look bell shaped but rather seemed skewed. The p-value in the Anderson Darling test was <0.05 which confirmed Jims suspicion: The data was significantly non-normal. Jims conclusion? With this degree of non normality, he would need to carry out a non normal capability analysis.

Determining the Correct Capability Analysis to Use

36

Attribute Data has just two possible probability distributions Binomial 3b) and Poisson. Determining the probability distribution continued The Binomial distribution is used to describe a process where the outcomes can be labelled as an event or non-event. If, for example, an item passes or fails inspection. For process capability analysis we will use the binomial distribution where we are counting the number of fails out of a set number of parts inspected. Where we have attribute data that is Binomial we will use Binomial Capability Analysis. Case Study Scenario 3: Pats data collection system is counting forms. Pat realises this is attribute data. Since the assessment of the forms can lead to only two possible conclusions delivered to order or not delivered to order Pat recognises that he is counting the number of fails out of a set number of orders delivered each month. Therefore his data is Binomial.

Pat will use Binomial Capability Analysis to analyse his data.


The Poisson distribution describes the number of times an event occurs in a finite observation space. For example, a Poisson distribution can describe the number of defects in the mechanical system of an engine or the number of calls to a call centre within a specified period of time. Process Capability Analysis using the Poisson distribution is less frequently used and so is outside of the scope of this How To Guide.
SEEK GUIDANCE

If you think that your attribute data follows the Poisson Distribution then please contact your local Master Black Belt for advice. If you are unsure about how to determine the Probability Distribution of your data then contact your local Black Belt or Master Black Belt for advice.

37

Determining the Correct Capability Analysis to Use

Determining the correct analysis: Summary so far

So far we have established that the type of process capability study required depends on: 1) The type of data 2) The probability distribution of the data The flow chart below summarises the choices which can be made:

Type of data

Continuous

Attribute

Probability distribution

Probability distribution

Normal

NonNormal

Binomial

Poisson

Normal Capability Analysis

Non-Normal Capability Analysis

Binomial Capability Analysis

Talk to MBB

Determining Carrying the out Correct and interpreting Capability Analysis the capability to Useanalysis

38

Step 4

4a) Analysing the data in Minitab 4b) Interpretation of the graphical output

Now the data has been collected we are ready to start to analyse, interpret and communicate the results. As the methods for analysis are different for Normal, Non-Normal and Binomial Capability Analyses we will take each in turn using Anne, Jim and Pats scenarios as examples.

We will begin with Normal Capability Analysis (for continuous data) and then move on to Non-Normal Capability Analysis. Finally we will cover Binomial Capability Analysis (for Attribute data).
If you wish only to follow the procedure for Non-Normal data then please turn to page 49 or for Binomial data then please turn to page 57.
SEEK GUIDANCE

If you are unsure about how to determine the Probability Distribution of your data then contact your local Black Belt or Master Black Belt for advice.

39

Normally Distributed Data Step 4: Carrying out and interpreting Normal Process Capability Analysis

4a) Analysing the data in Minitab

Case Study 1 Data Analysis

This is the data on golf club shaft diameter collected by Anne and her team:

Normally Distributed Data As you can see the data is organised in columns with the subgroups marked by the Date Collected column. Anne runs her eye over the data. It looks complete and she cant see any mistyped or missing data. She therefore proceeds with the Normal Capability analysis.
SIGNPOST

For details on how to run the Normal Process Capability Analysis in Minitab please turn to Appendix 3. Further information on the statistics which underlies these calculations can be found in Appendix 7 & 8.

Normally Distributed Data Step 4: Carrying Analysing out and and Interpreting interpreting Gauge Normal R&R Process StudiesCapability Analysis

40

4b) Minitab presents the output of the Normal Process Capability Analysis Interpret using a concise graphical report. graphical output For Annes data the output is as follows:

As there are a number of different pieces of information summarised in this graph we will break it down and explain how to interpret the output section by section.
SEEK GUIDANCE

If you are unsure as to how to construct or interpret any of the graphs, ask for help from a local Black Belt.

41

Normally Distributed Data Step 4: Carrying Analysing out and and Interpreting interpreting Gauge Normal R&R Process Studies Capability Analysis

4b) Interpret graphical output

Interpreting the Process Data Process Data LSL 9.3 Target USL 9.5 Sample Mean 9.41538 Sample N 100 StDev(within) 0.0329631 St Dev(Overall) 0.0350541 The process data box simply summarises the data that is being used for the analysis. Always look at this first to check that it looks correct (for example that the sample size is what you expected). LSL & USL refer to the upper and lower specification limits you entered into the Process Analysis dialogue box. In this example you see these are the 9.3 9.5mm input for Annes data. Check that these have been input in the correct format (e.g. in mm not cm). Target will be blank unless you have entered the Target as optional information (not covered in this guide). Sample N is the total size of the sample of data used. In this example there were 10 days data each with 2 subgroups of 5 shafts each day so the total size of the sample is 100 data points. Always check that this number matches the sample size you collected in case you have made a data input error. Sample Mean is the mean (average) of all of the 100 data points. This is used in the calculation for each of the capability metrics (Cpk & Ppk). StDev (Within) is the calculated measure of short term variation. This is the standard deviation calculated from the variation within subgroups. This is used to calculate the short term capability metrics Cp & Cpk. If you are interested in understanding more about how this metric is calculated see appendix 7). StDev (Between) is the calculated measure of total long term variation. This is the standard deviation calculated using variation both within and between subgroups overall. This is used in the calculations for the long term capability metrics Pp & Ppk. If you are interested in understanding more about how this metric is calculated see appendix 7).

Normally Distributed Data Step 4: Carrying Analysing out and and Interpreting interpreting Gauge Normal R&R Process StudiesCapability Analysis Interpreting the Graph 4b) Interpret graphical output

42

The next thing to look at is the graph. This shows the complete data set represented as a histogram. The red dashed lines mark the location of the upper and lower customer specification limits (USL & LSL). From this we can visibly assess the capability of the process and get some clues as to any problems for example does the process look centred? How does the spread look in relation to the tolerance width? The diagram above looks slightly off centre closer to the upper than the lower limit. The process spread is also quite wide by eye it looks approximately the same width as the specification limits.

43

Normally Distributed Data Step 4: Carrying Analysing out and and Interpreting interpreting Gauge Normal R&R Process Studies Capability Analysis

4b) Interpret graphical output Cont.

Interpreting the Capability Metrics Potential (Within) Capability CP 1.01 CPL 1.17 CPU 0.86 Cpk 0.96 Overall Capability Pp 0.95 PPL 1.10 PPU 0.80 Ppk 0.80 Cpm The important Minitab outputs are the short term capability metrics: Cp and Cpk and the long term performance Pp and Ppk. Lets start with the short term capability. This is in the upper half of the box. Here you see the following: Cp: the short term potential capability of the process. The Cp only takes into account the spread of the data in relation to the tolerance. Here the spread of the data is calculated using the short term variation. Here the Cp is just over 1. Since we saw by eye that the spread of the data was roughly the same width as the tolerance band this makes sense. Cpk: the short term actual capability of the process. This takes both the spread and the position of the process average into account. If the process is perfectly centred the Cp and the Cpk will be the same. Here we see the Cpk is 0.86 which is a bit less than the Cp of 1.01. This tells us that the short term actual performance of the process is worse than its potential. That means that the process must be off centre. Again this does not surprise us as we saw by eye that the process was off centre when examining the histogram.

Normally Distributed Data Step 4: Carrying Analysing out and and Interpreting interpreting Gauge Normal R&R Process Studies Capability Analysis

44

4b) Interpret graphical output continued

Interpreting the Capability Metrics continued Overall Capability (Pp & Ppk) The overall capability calculations are summaries of the long term capability of the process. The metrics we can see here to understand the long term process capability are: Pp: Remember this is the long term potential capability of the process. The Pp only takes into account the spread of the data in relation to the tolerance. Here the spread of the data is calculated using the long term variation (StDev (Overall) on the data box). Because all the variation is considered the spread is seen as being slightly larger than for the short term metric. Therefore the Pp shows to be 0.95 indicating that the spread is slightly larger than the tolerance width. This is slightly worse than the Cp which is what we would expect since long term performance usually is worse than short term. Therefore we can see that the long term potential capability of the process is not particularly good. The process has too much variation to fit within the specification limits. Ppk: This is the long term performance capability of the process. This takes into account both the variation and the location of the process mean. The Ppk is 0.8 which is less than the Pp. Obviously we would expect this as we already saw with the short term capability metrics that the process was not performing to its full potential. Of all the metrics this is the one that gives Anne the best understanding of the actual expected overall performance of her process in the long term. But what does a Ppk of 0.8 actually mean? To understand this we look at the final part of the analysis output.

45

Normally Distributed Data Step 4: Carrying Analysing out and and Interpreting interpreting Gauge Normal R&R Process Studies Capability Analysis Interpreting the Performance Metrics 4b) Interpret graphical output continued

Observed Performance % < LSL 0.00 % > USL 0.00 % Total 0.00

Exp. Within Performance % < LSL 0.02 % > USL 0.51 % Total 0.54

Exp. Overall Performance

% < LSL % > USL % Total

0.05 0.79 0.84

Cpk and Ppk metrics can be difficult to understand and explain without tying them back to what they mean in terms of performance.
Here we see three different performances summarised. Lets look at each in turn: The observed performance is the actual Observed Performance performance seen within the sample of 100 % < LSL 0.00 shafts that were measured in the capability % > USL 0.00 study. Of those shafts all 100 were within the % Total 0.00 required specification. The % Total shows the % that are out of specification. It indicates zero percent. This is what Anne expected and why she was confident that her process was capable of meeting the tightened specifications without need for alteration. However the expected long term performance shows a different story. Remember that the power of this analysis is that it doesnt just count how many units in the sample hit or miss the specification. Rather it uses the probabilities of the Normal Distribution to assess the likelihood of the process hitting of missing the customer requirements in the longer term.

Normally Distributed Data Step 4: Carrying Analysing out and and Interpreting interpreting Gauge Normal R&R Process Studies Capability Analysis

46

4b) Interpret graphical output continued

Interpreting the Performance Metrics

In Annes case study for example she looked at a relatively small amount of club shafts just 10 per day for 10 days out of thousands that are manufactured. We can see from the histogram that her process spread just about squeezes into the specification limits so it is not totally unexpected to see that the observed performance of those 100 shafts sampled was 100% good. However given that the process is off centre and that its spread fills the whole of the tolerance range it should also be no surprise to expect that the process inevitably will produce a defect as some point. The expected performance quantify this likelihood as an expected defective percentage.
Exp. Within Performance % < LSL 0.02 % > USL 0.51 % Total 0.54

The Expected Within performance is the short term actual capability (Cpk) expressed as a % defective.

This means it is expected , based on the probabilities of the Normal distribution, that in the short term the process will produce 0.54% defective. To understand this further the % < LSL and % > USL show how this 0.54% defective would be expected to break down it is expected that 0.02% will be undersized and 0.51% will be oversized. This of course makes sense as we saw that the process sits closer to the upper than to the lower specification limit.
Exp. Overall Performance

% < LSL % > USL % Total

0.05 0.79 0.84

The Expected Overall performance is the long term actual performance (Ppk) expressed as a % defective.

This means it is expected, based on the probabilities of this normal distribution, that in the long term the process will produce 0.84% defective it is expected that 0.05% will be undersized and 0.79% will be oversized. This gives a worse picture than the expected short term capability. This is to be expected since the actual performance is long term and takes all the variation in the process into account. Process knowledge will aid in determining if your data represents long or short term variation. Note that if your observed performance is grossly different from your within and overall performance this may be an indication of a different distribution (see appendix 4).

47 49

Normally Distributed Data Step 4: Carrying out and interpreting Normal Process Capability Analysis

4b) Interpret graphical output continued

Interpreting the Performance Metrics

Case Study 1 Interpreting the output Anne is mindful of the business requirements that 99.9% of all the shafts must be manufactured within the these specification limits. Her initial feeling from reviewing her sample of 100 shafts was that her process was capable of hitting this target. Now however, looking at the expected performance she can see this isnt the case! She is most interested in the Ppk and associated Expected Overall Performance metric as this gives her an understanding of how she can expect the process to perform in the long term. An expected total % defective of 0.84 means she can only expect her process to deliver 99.16% right first time. This isnt good enough. However she can also see that the process is off centre so has a potential capability that is better than the actual performance. She asks her local Black Belt what Ppk she would need to have to achieve a right first time of 99.9% he refers to some capability tables and advised her she would need a Ppk of 1.03 to achieve a defect level of 0.1%. Since the Pp shows the long term potential capability to be only 0.95, Anne can see that even is she centred the process (something she thinks will be easy to do) this would not be enough in itself to meet her 99.9% target. She will also need to look at reducing the variation in the process.

Normally Distributed Data Step 4: Carrying out and interpreting Normal Process Capability Analysis Interpreting the Performance Metrics 4b) Interpret The table below summarises how the Minitab terminology aligns to the graphical standard terminology we used in Part 1. output continued

48

Standard Terminology Minitab Terminology

Capability (Short Term) Potential (Within) Capability Cp Cpk

Performance (Long Term) Overall Capability Pp Ppk

Potential

Actual

Also when interpreting the Expected Performances remember that:

Expected Within Performance Expected Overall Performance

Relates to Relates to

Cpk Ppk

49 Step 4

Carrying out and interpreting Non Normal Process Capability Analysis 4a) Analysing the data in Minitab 4b) Interpretation of the graphical output

We will now examine how to analyse and interpret the data for Non Normal Process Capability Analysis Case Study 2 Data Analysis This is the data on call hold times collected by Jim and his team

4a) Analysing the data in Minitab

Non Normally Distributed Data

As you can see the data is organised in columns with the time of each call given in the Time column and the call hold time in seconds recorded in the Waiting Time column. Jim checks the data. It looks complete and there are no obvious typos or missing data. He therefore decides to proceed with the data analysis. As his data is Non Normal his first step is to establish which probability distribution to use for the data analysis.

Non Normal Data Step 4: Carrying Out and Interpreting Non-Normal Capability Analysis The analysis of non-normal data in Minitab is a little more complicated than 4a) for data with a normal distribution and requires two stages of analysis. Analysing the data in The first stage is to identify which probability distribution best fits the data Minitab set. Fortunately Minitab has a function to do this easily for us. It checks the shape of the data set against a number of continuous probability distributions and analyses which is the best fit.

52 50

Full details of how to do this in Minitab are covered in Appendix 5.


Once the distribution is understood then the non normal capability analysis function in Minitab can be used to analyse the data.
SIGNPOST

To find out how to fit the best distribution for non-normal data turn to Appendix 4

SEEK GUIDANCE

If you are unsure about how to determine the best distribution for your data then please refer to your local Black Belt.

With the help of his local Black Belt Jim analyses his data and establishes that the best distribution to use is a 3-Parameter Weibull distribution. Jim now has all the information that he needs to be able to carry out the Non Normal capability analysis.

51

Non Normal Data Step 4: Carrying Out and Interpreting Non-Normal Capability Analysis

Minitab presents the output of the Non-Normal Process Capability Analysis 4b) using a concise graphical report. Interpret graphical For Jims data the output is as follows: output

As there are a number of different pieces of information summarised in this graph we will break it down and explain how to interpret the output section by section.
SIGNPOST

To find out how to do the Non-Normal Capability Analysis in Minitab turn to appendix 5

SEEK GUIDANCE

If you are unsure as to how to construct or interpret any of the graphs, ask for help from a local Black Belt.

Non Normal Data Step 4: Carrying Out and Interpreting Non-Normal Capability Analysis Interpreting the Process Data 4b) Interpret graphical output

52

The process data box simply summarises the data that is being used for the analysis. Always look at this first to check that it looks correct (for example that the sample size is what you expected). LSL & USL refer to the upper and lower specification limits you entered into the Process Analysis dialogue box. In this example you see these are 0 30 seconds which are the customer requirements for Jims data. Check that these have been input in the correct format (e.g. in seconds not minutes). Note, in the graphic above the LSL has been put in as a boundary as you cannot have negative time. Target will be blank unless you have entered the Target as optional information (not covered in this guide). Sample N is the total size of the sample of data used. In this example Jim collected one days worth of calls - a total of 75 data points. Always check that this number matches the sample size you collected in case you have made a data input error. The remaining two parameters Shape and Scale are parameters used to describe the Weibull distribution and can be ignored. Note however that these parameter will be different depending on the distribution used.

53

Non Normal Data Non Normal Data Step 4: Carrying Out and Interpreting Non-Normal Capability Analysis

4b) Interpret graphical output

Interpreting the Graph

The next thing to look at is the graph. This shows the complete data set represented as a histogram. The red dashed lines mark the location of the upper specification limit (USL) and lower boundary limit (LBL). From this we can visibly assess the capability of the process and get some clues as to any problems. Here straight away you can see a problem with Jims process capability the tail of the distribution goes beyond the upper specification limit of 30 seconds giving a clear visual indication of a capability problem. The dashed black line shows the theoretical Weibull distribution curve for the data. This curve is calculated by Minitab using the parameters we saw summarised in the Process Data box. Since we told Minitab to use the Weibull Distribution it assumes that the data would tend to look like this curve if a large enough sample was taken. It therefore uses the data from this theoretical curve rather than from the actual data to calculate the process capability. As we discussed in the introduction, this allows for much more powerful analysis of capability than if we simply counted how many of the sample of 75 calls fell outside of the required maximum call hold time.

Non Normal Data Step 4: Carrying Out and Interpreting Non-Normal Capability Analysis

54

4b) Interpret graphical output

Interpreting the Capability Metrics

Overall Capability Pp PPL PPU 0.43 Ppk 0.43

Here we see a summary of the capability metrics and the associated expected performance. You will note first of all that there are no metrics for Cp and Cpk. This is because these two metrics are only applicable to the Normal Distribution so will not be calculated for any non-normal capability analysis. The metrics of interest for all non-normal process capability analyses are therefore only Pp and Ppk . As with the Normal Capability Analysis the Pp for all Non-Normal Capability analyses relates to the size of the process spread in relation to the specification limits. If Pp < 1, then the specification spread is less than the process spread, indicating that the process is not capable of satisfying customer requirements.

In the example of Jims case study shown here you will note that no figure is given for the Pp. This is because in this case the lower specification limit (0 seconds) is bounded this means that it is not physically possible to have a waiting less than 0 seconds. In this circumstance and in circumstances where there is only one specification limit (so called Unilateral specifications) the Pp is not a valid metric to use.
The Ppk in this case is the only capability metric to consider. This gives us the Long Term Actual Performance. Remember this considers both process range and location when comparing the process with the specification. Here the Ppk is < 1 which indicates that the process is not capable. This is not a surprise to us as we already saw this from examining the graph. Where the Pp is given you can compare Pp and Ppk to assess if the process median is close to the specification midpoint. If Pp is greater than Ppk, the process median is off the specification midpoint and closer to one of the specification limits (the median rather than the mean is used for nonnormal data. This is a better measure of location for distributions that are not symmetrical).

55

Non Normal Data Non Normal Data Step 4: Carrying Out and Interpreting Non-Normal Capability Analysis

4b) Interpret graphical output

Interpreting the Performance Metrics

The observed performance is the actual performance seen within the sample of 75 calls that were measured in the capability study. The % Total tells that 6.67% of these measured calls had call hold times of more than the required 30 seconds. Remember however that the power of this analysis is that it doesnt just count how many units in the sample hit or miss the specification. Rather it uses the probabilities of the fitted distribution (in this case the Weibull distribution) to assess the likelihood of the process hitting or missing the customer requirements in the longer term. The Expected Overall Performance quantify this likelihood as an expected defective percentage. Therefore the 10.05% we see here is the long term actual performance (Ppk) expressed as a % defective. This means that it is expected, based on the probabilities of the Weibull distribution, that in the long term 10.05% of all the calls received will be on hold for more than 30 seconds. This gives a worse picture than the actual performance which is normally seen to be the case since the observed is based only on a relatively small sample of data whilst the probability distributions give a longer term view.

Non Normal Data Step 4: Carrying Out and Interpreting Non-Normal Capability Analysis

56

4b) Interpretation of the graphical output

Case Study 2 Interpreting the Output Jim is surprised by the results of his analysis. He didnt realise that so many callers were being put on hold for > 30 seconds. A likelihood of 10% of all customers calling his team getting this element of bad service is clearly unacceptable.

Jim and his team meet to discuss the findings of the study. They discuss what would be an acceptable capability for their process. After referring to some benchmarking data they agree that a Ppk = 1 would be a good improvement target to aim for. This would give an expected % defective for Long Term Actual Capability of 0.13% which the team felt was reasonable. The team also agreed that in order to monitor the control and capability of their process going forward that they should implement permanent long term measures to track this and other key characteristics for their process.

57 Step 4

Carrying out and interpreting Binomial Process Capability Analysis 4a) Analysing the data in Minitab 4b) Interpretation of the graphical output

We will now examine how to analyse and interpret the data for Binomial Process Capability Analysis 4a) Analysing the data in Minitab Case Study 3 Data Analysis This is the data on incorrectly shipped orders as collected by Pat.

Binomially Distributed Data

As you can see the data is organised in columns with the count of orders received each month in column 1 and the number of these recorded as incorrectly shipped in column 2 titled number of defective orders. The data has been entered in times series order. Pat checks the data. It looks complete and there are no obvious typographical errors or missing data. He therefore decides to proceed with the data analysis.
SIGNPOST

To learn how to use the Minitab Binomial Capability function to analyse this data turn to Appendix 6

Binomial Data Step 4: Carrying Out and Interpreting Binomial Capability Analysis

58

4b) Minitab presents the output of the Binomial Process Capability Analysis as a Interpret panel display which contains 4 graphical displays and a series of listed graphical statistics. output For Pats data the output is as follows:

The four graphs in the panel are used to check the assumptions for a binomial capability analysis. The table of results provides summary statistics and a capability index that can help you assess process capability. We will work through each of these one at a time.
SIGNPOST

To find out how to do the Binomial Capability Analysis in Minitab turn to appendix 6

SEEK GUIDANCE

If you are unsure as to how to construct or interpret any of the graphs, ask for help from a local Master Black Belt.

59

Binomial Data Step 4: Carrying Out and Interpreting Binomial Capability Analysis

4b) Interpret graphical output

Interpreting the P-Chart

The first assumption we check is that the process is stable over time. Minitab selects an appropriate SPC chart for the data (either a P-Chart or a NP-Chart depending on whether the sample size is varying or constant). We already generated this chart as part of the checks for pre-requisites in step 1. As you can see the process does not indicate any special cause variation, therefore we can consider it to be currently in control.
SIGNPOST

For more information on how to interpret SPC Control Charts please refer to the SPC How to Guide

Binomial Data Step 4: Carrying Out and Interpreting Binomial Capability Analysis

60

4b) Interpret graphical output

Interpreting the Cumulative % Defective

The next assumption we check is that you have collected data from enough samples to have a stable estimate of the %Defective. The %Defective chart analyses this by plotting for each sample the total %Defective to date. So the first data point is the defective rate for month 1, the next point is the defective rate for month 1 & month 2 combined and so on. Ideally, this graph should show that the %Defective stabilizes after several samples. This would appear as a flattening of the plotted line. Without this stability the capability analysis may be invalid and require more data to be gathered. In this case you can see that indeed the %Defective does level out after the first 7 data points or so. Therefore Pat can be confident that he has sufficient data for the analysis to be valid.

61

Binomial Data Step 4: Carrying Out and Interpreting Binomial Capability Analysis

4b) Interpret graphical output

Interpreting the Rate of Defectives

The next assumption we check is that the rate of defectives is not affected by sample size. This is checked by plotting the defective rate against the sample size (like a scatter plot). The solid line is a fitted line. This should be horizontal (as in this example) which means that the probability of a defective item is constant across all sample sizes. The two red dot-dashed lines represent the significance limits (also called confidence bounds). The points should all sit within these. Examine the plot to see if the %Defective is randomly distributed across sample sizes or if a pattern is present. In this example no pattern can be seen. The points appear to be randomly distributed across the different sample sizes. Note the plot here looks a little odd as in a number of months there were no defective orders which has led to a lot of the data points sitting along the bottom of the plot on the %Defectives line. If any pattern is found in the data then this indicates that the defective rate is not independent from the sample size which would mean that the data is not binomial and further investigation is required. Note when your sample sizes are equal, a binomial plot will be displayed instead of the defect rate plot. The binomial plot plots the expected and observed number of defectives. You should examine your plot to see if the plotted points are in a straight line. If not, then the assumption that the data were sampled from a binomial distribution may be false.

Binomial Data Step 4: Carrying Out and Interpreting Binomial Capability Analysis Interpreting the Histogram of % Defective

62

4b) Interpret graphical output

The final check that the data is binomial is to look at the shape of the distribution of the %Defectives. The shape can help determine the correct distribution has been used. This is quite a tricky graph to interpret however as the shape of the binomial distribution differs markedly depending on the underlying probability of defectives. If you are unsure of how to interpret this graph or think there may be a problem, then it is best to ask the advice of a Master Black Belt. In our example here a lot of recorded points are 0 (zero) defective with the distribution showing a tail stretching out to 1.2%.

63

Binomial Data Step 4: Carrying Out and Interpreting Binomial Capability Analysis Interpreting the Summary Stats 4b) Interpret After reviewing all of the four graphs to confirm that the capability analysis is graphical valid, the listed statistics can be read and interpreted for required actions. output %Defective gives the overall % of orders that were incorrect. This is calculated from the actual data collected. In this case 0.27% is quite small, however, the circumstances of defective levels should always be considered before reporting this as good or bad (i.e. 0.27% for an order entry process is amazingly good while 0.27% for a critical engine component may be unacceptable). The lower and upper confidence interval numbers for %Defectives are also shown. In practice this means the Defect rate could be as low as 0.12% or as high as 0.53%.

The Target value (as entered when completing the dialogue box) is also listed so that you can compare this to the actual defective level.

PPM Def (Parts per million defective) is a different way of reporting the proportion of orders that are incorrect. This simply expresses the proportion as defects per million rather than defects per hundred (as reported for %) the nearer the number is to zero (0) the better.

Binomial Data Step 4: Carrying Out and Interpreting Binomial Capability Analysis Process Z provides a capability index for the process. The Z value is often also known as the sigma level and here is a measure of Long Term Actual Performance. The larger the Z value the better the process capability. A process with a Process Z level of greater than 4.5 is often considered world class. The lower and upper confidence interval numbers for Process Z are also shown. Calculating an Equivalent Ppk from Z Values To be able to compare the capability level of a binomial process with the capability metrics from continuous data distributions the following conversion can be done: Equivalent Ppk = Process Z / 3 For our process this would give an equivalent Ppk of 2.78/3 = 0.93. This is less than the Rolls Royce goal Ppk of 1.33, which indicates that there is scope for improvement. However the context of the process needs to be understood. The process is 0.27% defective (or 99.72% correct) which for a complex order picking process such as in this example may be considered quite good. The actual process performance should always be discussed with the business owner and confirmed as good enough OR further action is required.

64

4b) Interpret graphical output

65

Binomial Data Step 4: Carrying Out and Interpreting Binomial Capability Analysis

4b) Interpret graphical output

Case Study Scenario 3 Interpreting the Capability Analysis Pat and his team are very interested in the output of the analysis. Although Pat always kept careful records of the orders delivered incorrectly this is the first time that they have analysed these to understand their long term actual performance. They have always prided themselves on their attention to detail and this analysis confirms what they have always believed that their % incorrect shipments is in fact relatively low.

Understanding the Ppk scale however has made them see that despite a relatively low % defectives there is still opportunity for improvement. They therefore resolve to try to improve their performance to 99.9% right first time (the equivalent of just 1 in 1000 orders delivered incorrectly).

Improving the Process

66

Step 5

Process Capability Improvement Guidelines Once the process capability study has been completed the next step is to compare the actual capability to the desired capability and make a decision regarding required improvements. There are two ways to improve process capability: 1) Centre the process location: as we have seen in Step 4 the potential capability metrics (Cp and Pp) can be used to give us an understanding of how much better our process capability could be if the process average was centred between the specification limits. Sometimes centring the process can be quite easy to achieve for example in Case Study Scenario 1 Anne can simply make some adjustments to her machine settings to reduce the average diameter of her golf club shafts. In other circumstances, for example cycle time, changing the process mean may be more difficult to achieve. 2) Reduce the variation in the process: as we have seen, the more spread there is in a process then the worse the process capability. Variation can come from many sources. We often think of the 6Ms Man, Machine, Methods, Materials, Measurement Systems, EnvironMent as the classic sources of variation. To reduce variation we need to understand through data analysis where the key sources of variation are in the process. For example, thinking again about the Case Study Scenario 1, Anne could carry out analysis to see whether there was a difference in performance between machines, shifts or batches of steel. Once the key sources of variation are understood they can then be eliminated or controlled. Often capability improvement will require a structured approach such as the DMAIC problem solving approach used for Lean Sigma projects.

SEEK GUIDANCE

For more guidance on how to approach Process Capability improvement please contact your local Black Belt

67

Step 5: Improving the Process

Case Study 1 Improving the Process Anne kicks off an improvement process straight away to improve her long term actual performance form 99.16% right first time to 99.9%.

to

The first thing she does is work with her Production Engineering team to centre her process. Until the process capability analysis was done they had not realised that the process average was sitting a little high. Adjustments are made to the machine settings which better centre the process average and improve the process capability to be 99.8% right first time. Anne now considers how to further improve the process to reduce the variation in the process. This is more difficult to achieve so Anne decides to train one of her team as a Lean Sigma Green Belt and sponsors a project to analyse the key sources of variation in the process. This identifies two possible sources of variation differences between batches of material and variation caused by the machine tooling set up process. Work begins to tackle the root causes of these key sources of variation and Anne feels confident of hitting her improvement target within 3 months.

Step 5: Improving the Process

68

Case Study 2 Improving the Process Jim and his team also immediately kick off an Improvement project to reduce the proportion of customer put on hold for > 30 seconds from 10% to 0.13%. The first step in this project was to implement ongoing process measurement systems so that the team could proactively monitor the control and capability of the process moving forward. At the same time the team decide to process map and waste walk their call handling process to understand why there is so much variation in the time for which customers are put on hold. This team activity identifies two key sources of variation one relating to Jims capacity planning process and another relating the ease with which customer files can be found. Although the team thought their process basics were in good shape, on closer analysis their workplace organisation particularly their customer account filing system is found to have room for improvement. Implementing some quick wins on this immediately results both in a reduction in average hold time and a reduction in the variation in call hold time which reduces the % of call on hold for too long from 10% to 5%. Jim and his team havent hit their improvement target yet but are confident that once their new automatic measuring system is in place they will be better able to monitor call volumes and improve their capacity planning process to make the final improvement required.

69

Step 5: Improving the Process

Step 6
Case Study Scenario 3 Improving the Process Pat and his team resolve to try to improve their order delivery performance from 99.73% to 99.9% right first time. To do this they need to understand the reasons for incorrectly despatched orders. Straight away they hit a barrier whilst Pat had carefully recorded the count of incorrect orders he had not noted the reasons for these. The team realise therefore that they will need to dig into the customer complaint records to find out the information they need. They decide to set aside a quiet week in winter to do a Single Week Improvement Project to get to the bottom of incorrect deliveries. This activity surfaces a number of problems with the ways that mixed orders (golf clubs + accessories) are recorded, picked and packed. Because the accessories order prints out on a separate sheet this is occasionally lost or missed when picking orders. The coding for some of the accessories is also confusing, particularly for golf balls, and occasionally leads to the wrong items being picked. The team work with the order processing and stores team to make some simple improvement to their process. They will need to wait a few months before they have enough data to validate their improvements but are confident that the capability should improve.

Maintaining Process Capability

70

Step 6

Maintaining the Gains Once the process capability is at an acceptable level it is of course imperative that it is sustained. The critical sources of variation identified for the process should be documented in a control plan and suitable process controls put in place to manage these. The best controls are mistake proofing but where this is not practical process control charts (SPC) are often effective at monitoring and controlling critical characteristics. Constant focus must also be given to maintaining the process basics since without processes that are standardised and complied to it will not be possible to maintain process control and capability. So long as the process remains unchanged and in control there should be no need to reassess the process capability. When you are collecting data, it is however good practice to check your capability periodically.

When ever the process is changed for example if a machine is replaced, a design change made or process improvement implemented then a new process capability study should be made.

Das könnte Ihnen auch gefallen