rm13006 process control methods

Revised September 7, 2021
RM13006 Process Control Methods
AESQRM006202109
SAE Industry Technologies Consortia provides that: “This AESQ Reference Manual is published by the AESQ Strategy Group/SAE ITC to advance the state of technical and engineering sciences. The use of this reference manual is entirely voluntary and its suitability for any particular use is the sole responsibility of the user.”
Copyright © 2021 AESQ Strategy Group, a Program of SAE ITC. All rights reserved.
No part of this publication may be reproduced, stored in a retrieval system, distributed, or transmitted, in any form or by any means, electronic, mechanical, photocopying, recording, or otherwise, without the prior written permission of AESQ Strategy Group/SAE ITC. For questions regarding licensing or to provide feedback, please contact [email protected].
Aerospace Engine Supplier Quality (AESQ) Strategy Group The origins of the AESQ can be traced back to 2012. The Aerospace Industry was, and still is, facing many challenges, including:
• Increasing demand for Aero Engines
• Customers expecting Zero Defects
• Increasing global footprint
The Aero Engine manufacturers Rolls-Royce, Pratt & Whitney, GE Aviation and Snecma (now Safran Aircraft Engines) began a collaboration project with the aim of driving rapid change throughout the aerospace engine supply chain, improving supply chain performance to meet the challenges faced by the industry and the need to improve the Quality Performance of the supply chain.
Suppliers to these Engine Manufacturers wanted to see greater harmonisation of requirements between the companies. Each Engine Manufacturer had Supplier Requirements that were similar in intent but quite different in terms of language and detail.
This collaboration was formalized as the SAE G-22 Aerospace Engine Supplier Quality (AESQ) Standards Committee formed under SAE International in 2013 to develop, specify, maintain and promote quality standards specific to the aerospace engine supply chain. The Engine Manufacturers were joined by six major Aero Engine suppliers including GKN, Honeywell, Howmet Aerospace, IHI, MTU and PCC Structurals. This collaboration would harmonise the aerospace engine OEM supplier requirements while also raising the bar for quality performance.
Subsequently, the Aerospace Engine Supplier Quality (AESQ) Strategy Group, a program of the SAE Industry Technologies Consortia (ITC), was formed in 2015 to pursue activities beyond standards writing including training, deployment, supply chain communication and value-add programs, products and services impacting the aerospace engine supply chain.
AESQ Vision To establish and maintain a common set of Quality Requirements
that enable the Global Aero Engine Supply Chain
to be truly competitive through lean, capable processes
and a culture of Continuous Improvement.
ii
The SAE G-22 AESQ Standards Committee published six standards between 2013 and 2019:
• AS13000 Problem Solving Requirements for Suppliers (8D) • AS13001 Delegated Product Release Verification Training Requirements (DPRV) • AS13002 Requirements for Developing and Qualifying Alternate Inspection Frequency Plans • AS13003 Measurement Systems Analysis Requirements for the Aero Engine Supply Chain • AS13004 Process Failure Mode & Effects Analysis and Control Plans • AS13006 Process Control
In 2021 the AESQ replaced these standards, except for AS13001, with a single standard, AS13100.
The AESQ continue to look for further opportunities to improve quality and create standards that will add value throughout the supply chain.
Suppliers to the Aero Engine Manufacturers can get involved through the regional supplier forums held each year or via the AESQ website http://aesq.saeitc.org/.
AESQ Reference Manuals AESQ Reference Manuals can be found on the AESQ website at the following link:
https://aesq.sae-itc.com/content/aesq-documents AESQ publishes several associated documents through the SAE G-22 AESQ Standards Committee supporting deployment of AS13100. Their relationship with APQP and PPAP is shown in Figure 1.
Figure 1: AESQ Standards and Guidance Documents and the link to AS9145 APQP / PPAP
This Reference Manual (RM) has been developed by the AESQ Process Control Methods Working Group, a group of Senior Industry Specialists from leading Aerospace companies, to promote the correct application of process control. Aerospace products are such that quality issues can be high profile and cause reputational damage to the producer, customer, and the industry. They also cause disruption to operations. Therefore, specialists from the leading Aerospace companies collaborate to improve the industry’s adoption and application of process control.
This Reference Manual includes both statistical and non-statistical tools for the application of control activities in the factory, and a range of statistical methods for process study of stability and capability leading to process improvement.
It also discusses process control from a principles level to help practitioners apply the techniques in the diverse array of manufacturing processes and environments. Common pitfalls and barriers are also discussed.
Many of the graphics in this guidance are produced using Minitab - a recognized statistical software application.
1
2
RM13006 - Process Control Methods
12. BENEFITS OF STATISTICAL PROCESS CONTROL (SPC) ................................................ 68 12.1 Background ......................................................................................................................... 68 12.2 Benefits ............................................................................................................................... 68 12.3 Resistance to SPC ............................................................................................................... 69 13. METHODS AND FORMULAE .............................................................................................. 71 APPENDIX A PROCESS CONTROL METHODS ASSESSMENT CHECKLIST ......................................... 76 APPENDIX B PROCESS CAPABILITY PLAN - EXAMPLE FORM ............................................................. 79 APPENDIX C TRAINING SYLLABUS ........................................................................................................ 80 APPENDIX D ACKNOWLEDGEMENTS .................................................................................................... 84 Figure 1 AESQ Standards and Guidance Documents and the Link for AS9145 APQP / PPAP ............ iii Figure 2 A Traditional View of Quality (Anything within Tolerance is Equally Good) ..............................6 Figure 3 Taguchi’s Loss Function (Any Deviation from Target Incurs Some Loss) ................................6 Figure 4 A Simple Control System .......................................................................................................7 Figure 5 Process Control Overview ......................................................................................................7 Figure 6 3 Step Process for Process Control........................................................................................9 Figure 7 The Deming (PDCA) Cycle .................................................................................................. 10 Figure 8 A Control Chart .................................................................................................................... 17 Figure 9 Variable Control Chart Selection .......................................................................................... 18 Figure 10 Process Showing No Signs of Special Cause Variation ........................................................ 19 Figure 11 Tests for Special Cause Variation ........................................................................................ 20 Figure 12 Run Chart with Non-Statistical Limits ................................................................................... 21 Figure 13 Pre-Control Chart for Bilateral Tolerance .............................................................................. 23 Figure 14 Pre-Control Chart for Unilateral Tolerance............................................................................ 23 Figure 15 Fuel Air Bracket Example ..................................................................................................... 24 Figure 16 Attribute Control Chart Selection .......................................................................................... 26 Figure 17 P Chart of Defectives ........................................................................................................... 27 Figure 18 P Chart with Varying Sample Sizes ...................................................................................... 27 Figure 19 C Chart ................................................................................................................................ 28 Figure 20 C Chart ................................................................................................................................ 29 Figure 21 Individuals Control Chart ...................................................................................................... 30 Figure 22 Process Checklist Format Example ...................................................................................... 31 Figure 23 Process Capability Index Cp/Pp ........................................................................................... 34 Figure 24 Elements of Process Capability Index (Cpk/Ppk) .................................................................. 35 Figure 25 High Capability - Practically Stable ....................................................................................... 38 Figure 26 Use of Ppk ........................................................................................................................... 39 Figure 27 Points Well Outside Control Limits ....................................................................................... 40 Figure 28 Binomial Capability Study .................................................................................................... 41 Figure 29 Poisson Capability Study ..................................................................................................... 41 Figure 30 A Non-Normal Distribution ................................................................................................... 42 Figures 31 and 32 A Bimodal Process Due to Oscillation ....................................................................... 43 Figures 33 and 34 A Bimodal Process Due to Step Changes ................................................................. 43 Figures 35 and 36 Normality Assessment (Process Approximately Normal) ........................................... 45 Figures 37 and 38 Normality Assessment (Non-Normal Process) ........................................................... 45 Figures 39 and 40 Normality Assessment (Bimodal Distribution) ............................................................ 46 Figure 41 Effect of Taking Averages on a Flat (Uniform) Distribution .................................................... 47 Figure 42 A Non-Normal (Skewed) Process Using an I-Mr Control Chart ............................................. 48 Figure 43 A Control Chart Using Transformed Data ............................................................................. 48 Figure 44 A Control Chart of Non-Normal Data with Appropriate Limits ................................................ 49 Figure 45 Distribution Identification Using Minitab Software ................................................................. 50 Figure 46 Process Capability Analysis Using a Weibull Distribution ...................................................... 50 Figure 47 Probability Plot of Original Data (Left) and Transformed Data (Right) ................................... 51 Figure 48 Capability Analysis of Transformed Data. The Capability Is Not Ideal. .................................. 51 Figure 49 Common Sources of Variation ............................................................................................. 52 Figure 50 Variation Within and Overall is Similar .................................................................................. 54 Figure 51 Xbar-R Chart Produced with Data from Figure 50................................................................. 54 Figure 52 Pattern of 20 Holes .............................................................................................................. 55 Figure 53 X Bar and R Chart of Pattern of 20 Holes ............................................................................. 56 Figure 54 I-MR Chart with Pattern of 20 Holes ..................................................................................... 56 Figure 55 A 3-Way Control Chart with Pattern of 20 Holes ................................................................... 57 Figure 56 Capability Analysis with Pattern of 20 Holes ......................................................................... 58
3
4
1. THE IMPORTANCE OF PROCESS CONTROL
If a process is in state of statistical control, it is likely to behave in a stable and predictable manner. This means that the process will perform at a reasonable performance level, provided the process’s capability is good, thus providing benefit for the producer. The process will produce less ‘surprises,’ and many aspects of operational planning can become more straightforward as a result.
Additionally, for product features that influence performance, a state of statistical control will offer the ability to maintain the process around the optimal design nominal. Thus, providing benefit for customers and users of the product.
Stability provides the potential for reliable planning. Instability causes un-predictable performance that is difficult to plan for.
But a state of statistical control is not necessarily a natural state. Processes that are not maintained and controlled will naturally decline over time. For this reason, methods of process control are needed.
2. KEY PRINCIPLES FOR PROCESS CONTROL
2.1 Key Principles
Process control tools can be used for a number of purposes such as performance calculations, root cause analysis, stability assessments, etc. The tools can be very useful. However, it is their application for the control of processes that maximizes their benefit, through being able to control quality proactively, thus avoiding quality issues.
The following principles underpin the use of the tools. All are important:
Principle 1 - On Target with Minimum Variation
A process with excessive variation will invariably lead to problems. The sources of variation should be managed proactively and in a systematic way. For all operations this will be through management of the process itself, but also foundational activities such as maintenance of equipment, training and competency, standardization of methods, correct measurement, etc. High quality tends to result from a well-managed and stable manufacturing environment.
Many product features have a design nominal that, if deviated from, causes a loss in the performance of the end product. For these features a process maintained ‘on target’ will perform better than one allowed to run ‘off target’ regardless of the conformance to specification. This concept is known as Taguchi’s Loss Function (see Figure 3).
Additionally, even processes without a performance related nominal will benefit from being ‘centralized’ between specifications due to the reduced likelihood of non-conformance.
5
Figure 2 - A Traditional View of Quality (Anything within Tolerance is Equally Good)
Figure 3 - Taguchi’s Loss Function (Any Deviation from Target Incurs Some Loss)
Principle 2 - Move from Inspecting in Quality to Controlling Quality
Reliance on inspection does not provide the optimal conditions for quality control. Even when inspection is introduced at the point of process, a ‘bad’ result is often detected too late to prevent further non-conformance due to buildup of work in process (WIP) or inability to spot trends if the data format is not appropriate. Inspection is rarely 100% effective due to gauging and process variations, and human factors.
To understand and control the process, it should be viewed using tools that offer the correct level of granularity to highlight trends and events and manage variation ‘on target’. Tools such as SPC charts (variable control charts) offer a far higher level of granularity than pass/fail inspection results.
Principle 3 - A Short Cycle Closed Loop Control System is Vital
A closed loop system (shown in Figure 4) involves the capture of information from the process, analysis of the information, a decision against some criteria (typically on whether a process anomaly is present), and a reaction to any such anomaly. The links between each of these activities need to be in place and be as short as possible in order to make decisions and actions timely.
Such control systems can be operated manually or built into the manufacturing process using automation. An example of an automated system is an in-cycle probing routine used in an NC machine tool.
6
Figure 4 - A Simple Control System
If the system is manually operated, ideally the measurer, decision maker and action taker will be the same person. If not then necessary communication channels, roles, responsibilities will need to be defined, agreed, and maintained.
Systems that are overly reliant on end of line inspection are compromised in all respects. They have severe delays, reliance on distant communication that is almost always too late to achieve anything constructive. At best, end of line inspection causes issues to be seen late, at the point where customer disruption is inevitable.
Principle 4 - The Operator Can Only Control if They Can See How the Process is Behaving
Often the process operator is in the possession of some process information. But if this information is not presented in an appropriate manner the operator will be unable to see any changes and trends. Then they will be unable to act on them. An example of information that is difficult to process is a Coordinate Measurement Machine (CMM) inspection report. The operator can recognize non-conformances easily enough, however. the amount of numbers and the discrete nature of each report means the data is not stitched together to show the process behavior. This will result in the operator only being able to detect non-conformance thus making control of the process reactive. On the other hand, a process control chart allows the operator to see the behavior of the process, and if it changes significantly the operator can take appropriate action to address the issue.
2.1.1 Overview of Process Control
Process Control has three main facets that are: Product Capability, Process Control Methods, and Foundational Activities (see Figure 5). High performance is not achievable without all three elements being in good order.
Figure 5 - Process Control Overview
Measure process
7
2.1.1.1 Level 1 - Importance of Product Capability
Process Capability (and thus Product Capability) is designed in during the selection and development of the manufacturing method. It is fundamental, because once designed in it can be very difficult to change. The capability should be high enough in the short-term that inevitable drifts and shifts over time do not result in non- conformance or deviations from the design nominal that result in a meaningful performance loss. Factors that may result in additional variation and process movements include multiple machine tools, batch to batch variations, operator to operator variations, tooling variations, raw material variations, etc.
The process designer should anticipate the potential effects of these factors when designing the process.
The better the short term capability the more tolerant the process will be to the sources of variation that affect the process in the medium and long term. A high capability such as a Cpk of 2.0 will allow the process to drift slightly without meaningful risk of non-conformance.
Process control will not fix an incapable process.
2.1.1.2 Level 2 - Importance of Process Control Methods
Once the manufacturing method is selected and the potential sources of variation have been determined, the process designer will develop process control systems that detect anomalies when they occur. The process and the product (process inputs and outputs) will be considered. A range of process control tools may be used (statistical and non-statistical). In many situations, control of process inputs will be preferable to monitoring of outputs, however this will be situation specific. These controls will ideally be closed loop systems.
2.1.1.3 Level 3 - Importance of Foundational Activities
The management of Foundational Activities provides the basis for stable operating conditions making process control achievable. These activities include, but are not restricted to: machine tool capability, condition and maintenance, standard methods, measurement systems, training and competence, factory environment, and raw material quality. It is expected that these be appropriately managed.
Regardless of the process capability and process control system, a process deployed into an environment which is unstable will cause significant problems. The result will likely be continual issues and frustration.
A stable environment will provide the conditions for anomalies to be the exception rather than the rule.
3. APPLYING PROCESS CONTROL
3.2 Process Control Activities
The Process Control Activities fall into three key steps (see Figure 6).
1. Process Control Method Selection - The selection of appropriate process control tools and methods for each item in the Control Plan.
2. Process Analysis and Improvement - Analytical study of the process to prove the effectiveness of the process controls described in the Control Plan. This involves the study of process stability, capability and any actions needed to address shortfalls.
The analytical study involves the following:
• The planning of the data that will be used to understand process control and capability, and any predefined acceptance criteria for control items. And the generation of a data collection plan.
• The execution of the data collection plan and application of visual tools to view initial data.
8
• The analysis of the process data using statistical techniques to describe process stability and capability, study the effect of the sources of variation, and understand the nature of any shortfalls.
• The actions to address any shortfalls in process stability, capability, or input variation.
3. Process Monitoring and Control - The application of the controls during continued production to detect issues and maintain process stability and capability.
Figure 6 - 3 Step Process for Process Control
3.3 Process Control in Process Design and Quality Planning
Process design and Quality Planning are concurrent activities. Process Control can be considered as much part of the process design as it is part of the Quality Planning activity. This point is often missed when one views tools such as PFMEA and Control Plans exclusively through a Quality Planning lens.
If the Quality Planning process is viewed without consideration for the process design activity, or one takes an overly document centric view, one could conclude that process control is only decided after the PFMEA activity. However, in reality one would begin to design the control system proactively as early as possible. This will often be done through applying pre-existing methods and considering past experience, often reapplying methods from similar products/processes.
Some controls will be based on part family standards and process best practices.
By the time the PFMEA is undertaken, the control system will mostly be decided upon. The controls provide the basis for scoring the detection in the PFMEA. The development of improvements is then based on the risk profile for the process. Additional controls and improvements may be developed based on this.
During process design and development, the capability of the process should be assessed to establish whether the process has sufficient capability to be adequately controlled within the specification limits or close to a target value. Ideally realistic tolerances will have been agreed during product development, based on customer needs and historic capability information.
In this early stage of development, the producer will likely be running the process on a limited run of product with fewer sources of variation present than would be expected in full manufacture. For example, a single machine with limited strip and reset of the process, and little raw material changes.
For this reason, the producer will need to estimate the likely effects on the process capability when the process goes into full manufacture, and judge the required capability for the initial proving run. Capability at the proving run is of no use if it cannot be translated into capability in volume manufacture.
Once the capability and stability are proven the process will be operated in serial manufacture using the adopted control system. The capability and stability may be assessed at various stages of product implementation and production.
Choice of Capability Metric
For initial capability the process may typically be run on a single machine tool and the product will likely be run on a continuous production run. For this reason, the capability may be reasonably well estimated by the Cp/Cpk metrics which are based on the analysis of variation in the short term, within subgroups from part to part variations of individuals.
9
For a full production run the data may have been derived from many subgroups (with sources of variation between each subgroup) or over a longer time period in which some natural process drifts and shifts occur. In this scenario the Cp/Cpk indices will be biased towards the short term (within subgroup) variation and may give overly optimistic results. For this reason, while the Cp/Cpk capability metrics can be informative they should be used in conjunction with the Pp/Ppk performance metrics which use an overall estimate of variation. In certain cases the difference between the short term (within subgroup) variation and the overall variation are such that a Between/Within capability study may be required.
For data involving multiple machine tools, the Cp/Cpk metrics may not make sense if the machines have systematic differences between them. In this case again the Pp/Ppk metrics may be more appropriate. However, there may be a case, where differences between machine tools are significant, to assess the capability of each machine separately.
3.4 Process Control in Continuous Improvement
The development of the control system in improvement may be done proactively (for example as an outcome of a PFMEA activity), or as a reaction to quality problems. When done proactively it follows a similar approach to that taken in process design and development, however the process control system will more likely be developed/refined following a process data study and PFMEA activity, to a pre-existing manufacturing process.
In problem solving it may be more or less regimented depending on the nature of the problem, the methodology used and whether the cause of the problem is obvious or not.
In continuous improvement activity usually some type of methodology will be used. Most methodologies follow a sequence of Plan, Do, Check, Act (known as the Deming cycle, see Figure 7). In the early phases, process data may be examined to understand the nature of the problem and decide on a course of action. The stability and capability of the process will be assessed. The work will be planned with an idea as to what the expected outcome will be. Sometimes this will involve modification of the control system. Once the work has been done, the result of the actions will be checked and compared against the expected outcome. Action will be taken based on this. Usually some form of Adopt, Adapt or Abandon decision for the change. This will be a data driven cycle.
Figure 7 - The Deming (PDCA) Cycle
3.4.1 Communication and Workforce Engagement
In certain situations, the closed loop system will involve multiple personnel. For instance, the person monitoring the process may not be the person responsible for making adjustments. In these situations, the responsibilities need to be made clear and particular attention will need to be given to the engagement of all personnel in the process.
A RACI analysis may be worthwhile to clarify who is Accountable for the control systems operation, Responsible for each activity within it and those Consulted and Informed periodically during its operation.
10
Process control and foundational activities are best sustained when the workforce is highly engaged in the operation of the controls, understand the importance of them, and are involved in the improvement of the systems.
4. PROCESS CONTROL METHODS OUTLINE
4.1 Nine Recognized Process Control Methods
The AESQ recognize the following Process Control Methods. Table 1 gives a simple summary. More comprehensive guidance follows in Section 5.
These are listed in a sequence that roughly aligns with the robustness or precision of each method. But their selection will depend on a number of factors. It is a case of selecting the right tool or tools for the job.
Table 1 - Overview of Recognised Process Control Methods
Method
Application
Example
See Also
Error/Mistake Proofing To avoid defects caused by inadvertent errors. The most robust and preferred method. Mistake Proofing devices build quality into a process in order to prevent and/or detect errors prior to defects being made.
Typical reaction:
Some error proofing devices prevent the possibility of entering an error state, so no Reaction Plan is required.
Some error proofing devices such as alarms and buzzers require the operator to stop and investigate the error cause. This reaction may involve following a prescribed recovery plan that eliminates the error condition or escalates the situation to an engineer or supervisor to determine next steps.
One-way fit of a die insert to prevent incorrect orientation during loading.
Use of a physical device to prevent installation of an oil- feed tube into the wrong port.
Use of electrical devices such as proximity switches and cameras to ensure proper alignment and orientation prior to the operation proceeding.
Section 5.1
See Also
Control Charts for Variable Data To monitor process inputs or process outputs that are continuous in nature for the purpose of establishing and maintaining a state of statistical control (also referred to as process stability).
Typical reaction:
Variable Control Charts alert the operator to “out of control” process behavior (special causes). If these occur, action is taken to identify the causes and bring the process back into statistical control. Recovery actions may be prescribed, or technical support may be provided depending on the situation.
Dimensional product features are plotted on Control Charts at the point of process and monitored by the operator. The operator takes action to investigate and remedy issues when special causes are detected.
The pressure drop in a vacuum furnace is monitored on a Control Chart to warn of developing issues. The operator responds to special causes by performing equipment diagnostic checks.
Section 5.2
Run Charts with Non-Statistical Limits To monitor process inputs that require adjustment within acceptable operating limits in response to natural drift. Likely to be used when statistical limits offer little practical benefit or lead to false signals of special cause.
To control conditions that follow a specific “profile” during the operation of the process.
Typical reaction:
Similar to Control Charts these Run Charts will have rules applied. Rules will typically be based on limits requiring some action (e.g., tool change). While these limits may not be statistically determined in the same way a Control Chart is, the Reaction Plan is similar to the ones used for Variable & Attribute Control Charts.
The viscosity of the slurry used in an investment casting process is monitored. When a limit is reached, the operator adds water to the mixture to correct for evaporation over time.
A highly capable characteristic of a machined part where tool wear is expected and can be tolerated to a point to maximize its effective use. The operator changes the tool at a predetermined dimension before the dimension becomes nonconforming.
Furnace Run Charts tracking thermocouple temperature levels throughout a cycle for heat treat and brazing processes. Each point in the cycle will have a normal operating window beyond which investigation occurs. Most likely to use software enabled system linked to the equipment.
Section 5.3
See Also
Pre-Control Charts To keep a capable process on target when the process has a tendency to move from the nominal value. Where processes are not sensitive to small changes, the use of a statistical Control Chart offers little additional value.
When simple operating rules are beneficial.
Typical reaction:
Pre-Control Charts have "warning limits". The action required is either one of further monitoring or action to investigate the reason for the process running off target. The reaction will depend on the ruleset being used.
Correct setup of a fuel control valve grinding process is confirmed by running the process and making adjustment until process is centered. Once centered, the process is monitored and only adjusted when Pre-Control rules are broken.
Monitoring of the outside diameter of an air cycle machine shaft where the operator controls adjustments using a machine offset in response to signals on the Pre-Control Chart.
Section 5.4
Life/Usage Control Processes that degrade over time where the useful life or usage is known. Limits to operation (time or number of cycles) will be set conservatively to avoid nonconformances.
Typical reaction:
The operator may be provided with a machine cycle counter. The reaction is to change the item that has reached its life limit at that point.
If cutting tool usage is monitored electronically, the machine may be programmed with control criteria, e.g., programmed not to allow further use of the tool after a certain number of cycles or hours use.
A forging die is run for a predetermined number of cycles before being removed for refurbishment/disposal. The life and die change are managed to coincide with batch changes.
Cutting tools with known wear characteristics are run for a specific cutting time. The tool life is electronically monitored by the Computer Numerical Control program to prevent overuse.
Section 5.5
See Also
Attribute Control Charts For monitoring quality levels of product or process attributes where the output is based on counts (typically defects) or classification (typically defectives). Used for recognizing changes in quality level due to special causes of variation.
Typical reaction:
Similar to Variable Control Charts. The action may be to stop the affected process or to investigate and resolve the problem.
Inspectors counting solder defects on a printed circuit board use a chart that monitors the number of defects per board. When a special cause is detected, the soldering process owner is informed and investigates the cause of the issue. The charts are reviewed by the operations management to identify opportunities for improvement, and to confirm results of improvement initiatives.
Section 5.6
Visual Process Check and Checklist Checking process attributes and recording them as meeting the requirements to run the process.
Typical reaction:
If the checklist cannot be completed, action will be taken to correct the gap. The process is not started. The execution of the process check should be audited for compliance.
A forging die is periodically examined by an operator for evidence of damage, wear, or scoring. The operator uses a checklist to record the result of the check.
An operator of a process with a lengthy setup operation uses a checklist to confirm each step of an operation is completed before running the machine. The checklist may also include safety items.
Section 5.7
First Piece Check To validate the setup and quality of a process prior to the production run.
Typical reaction:
If the criteria applied to the first-piece check are not met, the reason for the failure will be investigated. Once corrective action has been implemented the first-piece check will be repeated to validate the setup. Any activity of this kind should be documented for traceability.
A Coordinate Measuring Machine check of the first part in a batch of parts off a forming press is performed following change of press tooling. If the part meets the requirements, the process is allowed to run, and is then controlled using other Process Control Methods during the production run.
Section 5.8
See Also
Test Piece evaluation Commonly used along with process parameter control to provide validation of product quality. Typically, a destructive examination. It should be noted that a destructive examination processed with a batch of material is more inspection than control; so it needs to be used along with effective process input control.
Typical reaction:
For a test specimen that does not meet specifications upon the test conducted, the Reaction Plan will typically instruct the test operator to engage the appropriate engineer (e.g., Materials, Quality or Manufacturing Engineer) who will investigate the cause of the failure (process parameter inputs, furnace run schedule, etc.) as for clues to why the test specimen failed to meet the test. The product will be quarantined.
A piece of test material processed along with a batch of carburized gears in a heat treatment cycle is tested in a laboratory.
Tensile strength destructive examination of a test specimen used in a heat exchanger vacuum braze process.
Section 5.9
4.2 A Note on Automation
Process Control Methods can be incorporated using automation to add reliability and access to information at the earliest possible opportunity (e.g., in-cycle machine/part probing, automatic process compensations).
5. PROCESS CONTROL METHODS FURTHER EXPLANATION
5.1 Error/Mistake Proofing
Error proofing is the use of an automatic device or method that either makes error impossible or makes its occurrence immediately apparent. Error proofing should be chosen when the process is at risk of human error. The process risk analysis (PFMEA) should identify where human error is a potential cause of failure, where it has a high impact (severity) or may not be easily detected (detection). Safety related risks often require mistake proofed solutions.
Error proofing devices can take four forms. The hierarchy of these is:
1. Elimination - design the product or process hardware/software in such a way that an error is not possible.
2. Control - prevent an error being made by detecting it before it has an effect.
15
3. Signal - provide an immediate and obvious warning to prevent or highlight an error.
4. Facilitation - methods of guidance that make error less likely or will catch it.
NOTE: Error proofing methods are not industry specific. Some industrial sectors have a particularly well developed mistake-proofing culture often extending into product as well as process design. The automotive industry is very well known for its use of error proofing both from the manufacturing processes to the operation of the final product.
Examples:
• Guide Pins used to assure a one-way fit of a tool, fixture, or part to prevent incorrect orientation.
• An alarm used to alert an operator that a machine cycle has been attempted with a misaligned tool. The operator can take action to correct the problem.
• A limit switch used to detect correct placement of a work piece.
• Counters can be used to help an operator track the correct number of components needed in an assembly.
• A checklist used to assure all key steps are completed by the operator to prevent missing something that could cause an escape and/or defect. This approach is also described further in Section 5.7 - Visual Process Check & Checklist.
• Use of machine probing as either a control during manufacturing to check a size before final cut or as a signal after final cut to detect an anomaly or identify that an adjustment may be needed.
• Use of a Stopper Gate (physical barrier) affixed to a Fan Compressor assembly fixture to ensure an oil fill tube is installed in the correct port when there are multiple ports to choose from.
• Asymmetrical design of a nameplate that assures it is installed in only one possible orientation preventing backwards or upside down installation.
• A left/right two button hand operated system with foot switch operation to ensure hands are free prior to cycling a forging press.
• Automated weighing of a part or batch to ensure part is completely processed or batch is complete and present before moving to the next operation.
To ensure error proofing devices are robust, it is good practice to check that the failure of the device does not cause a problem (test to see what happens if the device fails to detect the error). Depending on the result (and the criticality of failure), revisit the design and maintenance requirements of the device and improve it.
If it is not possible to have an automated error proofing device, some of the other methods included in this standard may offer an adequate level of protection.
For further reading on the subject of Error/Mistake-Proofing the following may be referred to:
“Poka-Yoke,” by Productivity Press, ISBN 0-915299-31-3
“Mistake-Proofing for Operators: The ZQC System,” by Productivity Press, ISBN 1-56327-127-3
5.2 Control Charts for Variable Data
A control chart is a tool used to monitor and visually assess the behavior of a process over time. The control chart shows process data and ‘control limits’ which provide an approximation of the natural range of the process due to ‘common causes’ of variation. These limits (and other tests) are then used to detect abnormal events and trends (‘special causes’ of variation). The response to common cause issues and special cause issues are typically different, making the correct choice of approach important.
16
• Response to special causes of variation should be to immediately respond and investigate their cause. Provided the process was previously ‘in control’ the question is usually one of understanding what has changed.
• If common cause variation is behind the issue, then the fundamentals of the process should be understood, and the process changed to improve its capability.
The ‘control limits’ are derived using the process data and not product or process tolerances, thus minimizing the risk of responding in the wrong way (such as missing a signal to investigate or adjusting the process when it was not needed).
Assuming the process is capable the control chart will allow special causes of variation to be detected even if the data fall within the specification limits, meaning problems can be recognized earlier than if traditional inspection methods were used.
Figure 8 - A Control Chart
This section outlines four recognized control charts for variable data and provides guidance as to when they may be used. The list is not exhaustive. There are many more types of control charts not covered here that may be used for specific situations.
Figure 9 and Table 2 outline the basis for variable control chart selection.
71645750433629221 581 Observation
Table 2 - Variable Control Charts
Chart Its Use
Xbar and R
Xbar and S
Monitoring and control of characteristics on products being produced at a volume where typically a sample (subgroup) will be taken periodically to maintain quality.
Example: From a high volume process, five parts per hour are sampled from the line and measured. The average and range is plotted to understand if the process has changed (due to moving off target or through an increase in variation).
Can also be used for multiple similar products where it can used to plot ‘deviation from target’ thus avoiding the need for multiple charts.
The X bar chart displays the average of the subgroup. The R or S chart displays the variation within the subgroup (either the Range or Standard Deviation).
An X-Bar and R chart is used for subgroups of 3 to 8.
An X-Bar and S chart is used when subgroup size exceeds 8.
NOTE: The variation within the subgroups is assumed to be representative of the overall variation (no between batch effects expected). When this assumption is not met the process may appear out of control when in fact it is not. Consult an experienced practitioner if this appears to be the case.
Individual and Moving Range
Monitoring and control of characteristics on individual products being produced from continuous processes at a rate where subgrouping of data is not feasible.
Monitoring and control of process characteristics.
Can also be used for short run applications where there is product mix with similar characteristics (may be known as part families). In this situation the variability for all parts should be similar; used to monitor part families.
The Individuals chart displays the actual measured value (or deviation from target).
The Moving Range chart plots the difference between consecutive points (short-term variation).
NOTE: The variation from item to item is assumed to be representative of the overall process variation (no batching effects or systemic drifts/wear expected). When this assumption is not met the process may appear out of control when in fact it is stable. Consult a process control specialist if this appears to be the case.
18
Within control chart or ‘Three Way’ chart
Characteristics where the variation within the subgroup is not representative of the overall variation between them, usually the case when monitoring processes with ‘batching’ effects or multiple characteristics (a group of identical features) within a part are studied where the assumptions for an Xbar/R or S chart are not met.
The subgroup average is plotted on the Xbar chart.
The variation between consecutive subgroup averages is plotted on the Moving Range chart.
The Variation within the subgroup is plotted on the R or S chart.
NOTE: Higher subgroup sizes may lead to higher sensitivity to ‘special causes’ on R and S charts. Expected patterns within parts and batches can sometimes show signals that have no practical significance. Guidance may be sought from an experienced SPC practitioner if this appears to be the case.
There are eight industry standard tests for statistical control; to determine if the process data contains evidence of special causes of variation.
A process can be judged to be in statistical control (i.e., only common causes of variation present) when there is an absence of the patterns shown in Figure 11. An example of a stable process is shown in Figure 10. It should be noted when seeking to improve a process that the more tests used, the more signals will be detected. It may be worth using a selected few when starting out using control charts.
For process control purposes manufacturers often select the most appropriate tests for the process being operated, taking into account the actions that would be needed when they occur. Tests most frequently used by operators are Tests 1 and 5 (Figure 11), however, software applications make the use of all tests relatively simple.
Figure 10 - Process Showing No Signs of Special Cause Variation
2825221 91 61 31 0741 Observation
In di
vi du
al V
al ue
M ov
in g
Ra ng
19
5.3 Run Charts with Non-Statistical Limits
Some processes have characteristics that naturally drift in a certain direction as the process runs (i.e., the drift is a ‘common cause’ in the process). These processes when viewed on commonly used control charts tend to break tests for special cause long before the drift becomes a meaningful issue. Operation of traditional statistical control limits may then provide little benefit when compared to the characteristics’ ‘loss function’ and the cost and other implications of adjustment or reset. The more frequently the process is sampled the smaller the differences between measurements, which tends to exacerbate the issue.
Processes where this behavior may exist naturally are chemical etching (concentration changes), investment casting slurry control (through evaporation) and in some cases machining cutting tools (if they exhibit significant wear/drift with use).
An approach to manage this variation is to set limits on a time series chart. This limit will be set such that it detects drifts to avoid problems, but not so soon as it becomes uneconomic to adjust. This type of control is generally only useful when operated at the process rather than at an end of line inspection.
With appropriately set limits this method can be used effectively to control quality even using simpler measurement systems than downstream measurement equipment such as a CMM.
The following six step approach can be used:
1. Determine the variable to be monitored.
2. If the variable is an input or process variable, study, and quantify its relationship to the process outputs.
3. Establish the optimal process limits to be applied. In most cases this should be done using process data, to best ensure the limits are not too wide to allow a non-conformance.
20
4. Establish the adjustment to be made when the limit is reached. For example, this may be to adjust towards a lower limit, or an optimal setting, or in the case of a cutting tool, replace it. This reaction will be documented in the Control Plan and process instructions.
5. Operate the process and plot the measurements.
6. If the process limit is reached, adjust/set the process (see step 4). Confirm the adjustment has had the desired effect. If so continue. If not take action to understand why.
Figure 12 demonstrates how a chart of this type may be used. The process drifts upwards so a lower limit is not discussed within this example (for simplicity). It may, however, be wise to have one to mitigate other risks.
Figure 12 - Run Chart with Non-Statistical Limits
Process improvements can be made using the data from the run chart, for example in the following ways:
• Use process data and related process output to determine tighter reaction limits.
• Incorporation of automatic adjustments to the process to tighten the adjustment interval. This will decrease the spread between the limits.
• Make changes to the process or tools that decrease the rate of change of the process variable being controlled.
• Optimize the initial location for the process to increase the time between adjustments.
Features controlled in the way described should typically have a relatively flat ‘loss function’ when compared to the cost of reset or adjustment. The designer should be consulted where implications of process drift is not understood.
201 81 61 41 21 08642
1 0.075
1 0.050
1 0.025
1 0.000
set-point' is reset to 'process reached process When limit is
Diameter
21
Processes with systematic drift and infrequent ‘large adjustments’ may produce distorted capability analysis. There are two reasons for this.
1. The within subgroup range is typically small relative to the overall variation, resulting in Cp metrics being overly optimistic and not representative of the spread of the process.
2. The distribution of the data may not fit a distribution well enough to make accurate capability predictions. Both Cp and Pp derived capability may be inaccurate and alternative methods (e.g., non-normal methods such as Johnson Transformation, Box-Cox Transformation (see Section 7. Guidance for Non-Normal Data) may be required. If these methods do not help, then the process performance may need to be characterized by other means.
5.4 Pre-Control Charts
5.4.1 Background
Pre-Control is a method for monitoring and controlling the process within specification limits. It may be particularly useful when applied to process outputs or parameters that have a tendency to drift but for which the process is not overly sensitive to small changes. For example, a measurement taken on a ground feature where the grinding wheel wears over time.
Pre-Control may also be useful where it is important to maintain a capable process centered or ‘on target’, when detection of process ‘special causes’ are less important.
NOTE: The use of Pre-Control dates back to the 1950s. The merits of its use are often debated, with some favoring and some opposing its use. There are definitely valid arguments for and against which should be considered.
Pre-Control uses a chart that monitors items by classifying the measurements into colored zones (Red, Yellow, or Green). Decisions are made whether to adjust or stop the process based on where in these zones the measurements lie.
The advantages of Pre-Control are its simplicity and that it drives behavior towards on-target thinking.
NOTE: It is commonplace for the bands to be set as follows (see Figure 13):
• Green - the central 50% of the tolerance band (or 50% tolerance around a specific target).
• Yellow - outer quartiles (or remainder) of the tolerance band.
• Red - outside the tolerance.
Where tolerance is unilateral, the chart will have a single green, yellow, and red zone (see Figure 14).
5.4.2 Method
Following setup, a qualification phase runs according to a predefined ruleset to ensure the process is ‘on target’. Typically, qualification is passed after five consecutive units are produced in the green zone.
Three styles of Pre-Control exist:
1. Classical Pre-Control: Rules based around sampling two consecutive items periodically from a production run:
• Single item in Yellow - continue to run (but check subsequent item).
• Both items in Yellow - stop and investigate. Correct the process.
• Single item in Red - stop and investigate. Correct the process.
22
2. Two Stage Pre-Control: Based on a single item being sampled periodically.
• A single measurement in the yellow zone triggers measurement of additional items.
• A single Red will trigger process to be stopped and corrected.
3. Modified Pre-Control:
• A standard control chart with colored zones applied as described for Classical Precontrol (but to control limits, not tolerances).
With the exception of modified Pre-control, the limits and rules are not statistically derived. Opponents argue there is a risk of process tampering (over-control), if applying Pre-Control to an incapable process; or missing special causes that would be detected by statistical control charts. It is therefore not advisable to use Pre- Control on processes with poor capability or in situations where small changes in process need to be recognized.
Figure 13 - Pre-Control Chart for Bilateral Tolerance
Figure 14 - Pre-Control Chart for Unilateral Tolerance
NOTE: If analyzing the capability of a process that uses Pre-Control methods, a statistical control chart should be constructed to ensure the process is stable prior to analysis of capability and communication of capability indices such as Cp/Cpk.
Despite the concern of an unstable process on capability, a measure of goodness such as extended period in green zone on a Pre-Control Chart may serve as satisfactory evidence of capability to meet customer requirements if the customer permits this. This is more likely for minor characteristics than for KCs or special characteristics such as those categorized as Major or Critical.
For further reading on the subject of Pre-Control refer to Implementing Six Sigma (2nd Edition) - Breyfogle 2003. ISBN 0-471-26572-1).
5.4.3 Pre-Control Example
An aerospace manufacturer produces a Fuel Air Bracket (see Figure 15) with a key feature having an engineering tolerance of 0.386 ± 0.005 inches. The central 50% of the total tolerance (±0.0025 inches) defines the green zone.
23
Figure 15 - Fuel Air Bracket Example
The engineer defines the zones on the Pre-Control chart. The edges of the green zone are known as Upper and Lower Pre-Control limits (UPC and LPC).
UPC limit = 0.386 + 0.0025 = 0.3885 inches.
LPC limit = 0.386 - 0.0025 = 0.3835 inches.
The control method selected is two stage Pre-Control.
Set-Up Procedure
Following successful setup, the process operator runs five parts and records the dimensions of the features being controlled. If all five parts fall within the green zone on the Pre-Control chart (UPC = 0.3885 inches and LPC = 0.3835 inches) the setup is judged to be targeted properly and sample measurements are taken at a frequency of 20% (check every 5th part). This measurement frequency is for the purpose of maintaining process control and does not relate to product inspection frequency.
Executing the Pre-Control Monitoring Technique
The 10th piece comes up for inspection. It has a measured value of 0.387 inches. This is within the Pre-Control (UPC and LPC) limits, and the operator continues with production. The next piece to be inspected is the 15th. Its measurement is 0.3854 inches, well within the Pre-Control limits so the operator continues. The 20th part measures 0.3892 inches. This value is outside the UPC limit. The reaction plan referenced in the Control Plan determines that the operator now measures the next part produced, in this case the 21st. This part measures 0.3867 inches, again outside the UPC limit. The operator stops the process and investigates according to the prescribed reaction plan.
Pre-Control Rule 1: If the measured value is within the green zone (Pre-Control limits UPC and LPC) the operator may continue to check every 5th part (apply a 20% monitoring frequency).
Pre-Control Rule 2: When two consecutive measured values fall outside the same Pre-Control limit (UPC and LPC), the operator should react making an appropriate process adjustment. The reaction plan reference in the Control Plan (refer to AS13004) should describe the actions required.
Pre-Control Rule 3: When one measurement violates one Pre-Control limit and the following part violates the opposite Pre-Control limit, the variability may have increased. The operator should investigate the cause engaging support if needed (e.g., Quality/Manufacturing Engineer). The reaction plan referenced in the Control Plan (refer to RM13004) should describe the actions required.
5.5 Life/Usage Control
Processes may have factors that are dynamic in nature and change through use or over time. Such processes may require control methods that prevent the process (or its factors) reaching a condition that will adversely affect the product of the process. Such controls can be placed on, e.g., chemicals, wearable items such as cutting tools, and other consumables.
KEY
24
The control criteria for life/usage controls may be defined in many ways. Control is often not simply a question of ‘how old’. Examples of control criteria are: number of parts processed, total running time, number of cycles, once opened use by date, weight of parts processed and surface area processed.
Examples of control application include:
• A cutting tool has a maximum operating time. The tool life is recorded on a machine readable chip. The machine program includes code that checks the life of the tool prior to use. When cutting tips are replaced and the tool is set a pre-setting operation resets the readable chip to zero.
• A peening operation has media that is controlled based on the total equipment running time. A timer is installed on the equipment to indicate how close the process is to a media change. In addition to this method of control, the process also has assessment for media quality and uses test pieces to qualify the process for correct operation.
• The concentration of a chemical etch bath is routinely maintained with an auto-dosing system. However, once a month the entire system is emptied, cleaned out, and refilled. To keep the planning of this control simple this is done at a defined time regardless of use - for example the morning of the first Monday in every month.
A life/usage limit may also incorporate a check and reset. For example, a wearable item may be tested after a number of cycles and found to have not reached a point where change is required. The tool may be returned for use for a defined number of cycles. It should be noted that this does not imply the tool will be run to the point of failure.
The life/usage limits should ideally be determined to maximize the process quality. Statistical studies and experiments will allow the life to be optimized for other factors such as cost. These studies may be performed on test pieces and scaled to the production process. The life/usage limits should be validated however usually at process qualification.
NOTE: These guidelines and examples do not replace specific process standards or customer requirements that may exist to govern the life/usage controls.
5.6 Control Charts for Attribute Data
Attributes are characteristics, or conditions characterized as present or not-present or counted, typically through some form of inspection or check. A number of charts may be used depending on the attribute being studied.
NOTE: Process control via attributes is less effective than variable methods. Some checking methods may provide attribute data despite being variable in their nature. An example is a hole size, that may be checked via variable methods or attribute (e.g., plug gauge). If an attribute method were selected based on its speed and simplicity, it should be on the basis that the process is proven capable, because an attribute go/no-go gauge will not give early warning of emerging issues, the way a variable gauge does. A robust control strategy in the case of hole size may be to use a variable tool measurement device such as a presetter to assure the quality of the tool, and an attribute style plug gauge as a quick conformance check but with a periodic sample taken from production for variable measurement.
Figure 16 and Table 3 outline the basis for attribute control chart selection.
25
Table 3 - Attribute Control Charts
Scenario
A process that observes discrete values, such as pass/fail, go/no-go, present/absent, or conforming/nonconforming.
For example, a circuit card could consist of a number of solder joints that either conform or do not conform to a set standard.
Appropriate:
When it is important to control the number or % of defects over a given time period, lot to lot, or unit to unit such as measuring improvement over time, when go/no-go gauges are employed or when visual inspections are used.
Not Appropriate:
Cannot be used for establishing process control or process capability in the same way as variables data due to the scale not being continuous. Measures of performance and stability can be undertaken with a view to directing improvement activities, but true process control needs to be done through process variables, inputs, and foundational activities.
Not appropriate for rare events.
P Chart
Plot the percent defective - classifying product as good or bad with changing or constant subgroup size.
Plot the monthly percent defective rate of a critical supplier; plot the On Time Delivery performance of a critical supplier.
NP Chart
Plot the number defective - classifying parts as good or bad with constant subgroup size.
A machining cell produces fuel control valves in standard lot sizes of 50. Final Inspection performs a 100% inspection of the product and plots the number of valves that are determined to be nonconforming.
C Chart
Plot the count of defects based where the same area of opportunity (constant subgroup size) exists.
An aerospace manufacturer produces one type of heat exchanger for a customer. After vacuum braze a leak check is performed. A C chart is used to plot the number of leaks requiring weld repair.
U Chart
Plot Defects Per Unit (DPU) based on counts and varying or constant area of opportunity (changing or constant subgroup size) the defects come from.
An aerospace manufacturer operating Production Part Approval Process (PPAP) tracks the DPU on a monthly basis for all the inspected PPAP packages. An accompanying Pareto Diagram suggests the categories driving the DPU rate are poor PFMEAs, part marking errors and poorly written Control Plans. Projects are established to address these issues in order to reduce the overall DPU rate shown on the Uchart.
26
Figure 17 - P Chart of Defectives
Example: The non-conformities from a series of batches of 50 parts are monitored by the manufacturer on a P- Chart (Figure 17). The manufacturer observes an overall defective rate of 2.2%. The manufacturer concludes from the control chart that - despite the variability from batch to batch - the rate of defectives is statistically stable over time.
Figure 18 - P Chart with Varying Sample Sizes
2825221 91 61 31 0741
0.09
0.08
0.07
0.06
0.05
0.04
0.03
0.02
0.01
0.00
1 .00
P Chart of % Yield
Example: A manufacturer monitors the yield % in their goods produced per week on a P Chart (Figure 18). The weekly output varies. The manufacturer concludes that the process yield is not stable over time and seeks to understand the cause of the ‘bad’ weeks.
C Chart Example:
Figure 19 - C Chart
Example: A manufacturer produces a similar quantity of product each day. The number of defects noted from a visual inspection area is plotted on a C Chart (Figure 19) in order to understand the process performance and behavior over time. In this case the supplier notes a run of improved performance between days 12 and 22, and an increase in defects on day 30. In reaction to the defect rate on day 30 the manufacturer launches a problem solving activity.
NOTE: The use of NP charts and U charts are not illustrated in this document. Implementing Six Sigma - Breyfogle 2003. ISBN 0-471-26572-1 may be referred to for explanation and examples of their use.
The tests for special causes of variation for attribute control charts are as follows:
• One or more points beyond a control limit.
• A run of eight or more points on the same side of the center line.
• Six points in a row increasing or decreasing.
• Fourteen points in a row alternating up and down.
It is considered good practice to use a Pareto chart to support attribute methods to allow further prioritization and insight on the defects/defectives within the attributes plotted.
Assumptions for Attribute Charts
Attribute control charts have assumptions that need to be met in order for the chart to function correctly. If these assumptions are not met, then the control limits for the chart may be incorrect.
A C chart works best with a minimum average defect rate per subgroup of approximately 4, and a minimal number of zero values. Where this is not met the chart’s usefulness will be compromised.
2825221 91 61 31 0741
20
28
The calculations for the control limits for a C chart are based on the Poisson distribution.
A P Chart and NP Chart assume that the defectives are randomly dispersed and independent. In situations where defectives are generated in clusters the limits generated may be too narrow to reliably represent the common cause variation of the process.
The calculations for the control limits for the P Chart are based on the binomial distribution.
Use of Variable Methods for Attributes
In some scenarios, attribute data may be monitored quite adequately using variables control charts. For example, the Right First Time measure of a manufacturing operation whilst based on an attribute (good/bad), may be expressed as a ratio and plotted on a simple individual’s control chart. In many cases an Individuals chart is simpler to interpret and construct than attributes charts. Also of consideration is the sample sizes used, that when large may result in tighter control limits that result in the majority of data showing as ‘out of control especially when defective items occur naturally in clusters. The individual’s chart may help put the process in a better perspective.
A Note on Rare Events
For rare/infrequent events, attribute control charts can give less definitive results. The absence of events/defects/failures for example will have an adverse effect on the control limits and averages. In these cases, a time between failures may be a more useful measure to track. Mean Time Between Failure (MTBF) is a commonly used measure of equipment reliability for example.
Figure 20 - C Chart
29
Example - A manufacturer plots the failures of a machine tool, counting how many failures were experienced over a 100 day period (Figure 20). The chart is not very informative.
Figure 21 - Individuals Control Chart
Example: The manufacturer plots the time between failures for the data on an Individuals chart (Figure 21). The chart is much more informative. The average days between failures of 7.7 days and the control limits can help guide the manufacturer on equipment reliability and maintenance activity planning.
5.7 Visual Process Check and Checklist
A visual process check provides positive confirmation of goodness either prior to allowing a process to run, or during its operation.
The process checks need to become part of routine operation. The personnel conducting the check will ideally understand the importance of the check and also understand the reaction if the check fails against the criteria. In many cases the check will confirm that a particular step of the sequence has been done correctly.
The checks may be conducted by a single person, however on important items or high consequence failure items the method may use two persons who jointly confirm that the correct condition is achieved. An example of this approach is the standard pre-flight checks that are undertaken by pilot and co-pilot when preparing for a flight. One pilot calls out the check, the other performs the check and confirms as correct, and then the first records the check on a checklist before proceeding.
An example is shown in Figure 22.
To increase robustness, a “double scrutiny” and/or “buddy check” may involve two personnel to positively confirm an action or result of a check; or the check may be performed by someone independent of the operation.
A single person check may have some inherent risks of error. A preferred approach is automation or error proofing devices, (see Section 5.1 - Error/Mistake Proofing). Prior to finalizing the check, it is advisable to confirm the PFMEA risk level - as the method of control relates to the detection score in the PFMEA (refer to RM13004 for guidance).
1 0987654321
1 8
1 6
1 4
1 2
1 0
30
Pre-Operation Process Checklist Note to operator: Use this checklist prior to execution of the process operation and sign off each item below. Part No: 123456-78 Process operation number: 110 Run date: 08/12/2016 Process step name: Machine air holes
in Fuel/Air bracket
Check item number
Reaction (if Fail) Sign off (initial and date)
1 Health/Safety check Stop and isolate equipment. Contact cell leader
2 Work instructions are latest version
Contact Manufacturing Engineer - obtain instructions
3 Machine asset care checks complete and correct
Raise issue with cell leader
4 Gages in calibration Contact Quality engineer
5 Fixture damage check Contact Manufacturing Engineer
6 CNC programme correct (as per instruction)
Contact Manufacturing Engineer
8 Etc.
5.8 First Piece Check
The objective of a first piece check is to validate the set-up and quality of a process prior to the full production run. Alongside other controls it serves to verify and confirm the integrity of the production system (man, machine, fixture, tool, NC program, etc.) at a point in time, and hence to avoid economic damage of non-conformance (through timely action to ensure process conformance).
Prerequisite to a first piece check should be the adherence and confirmation that all other foundational control requirements are met (e.g., calibration, machine tool diagnostics, tooling within prescribed life limits, acceptable parameter settings, consumables level, etc.) typically approved through positive confirmation (see Section 5.7).
As a general rule, all manufacturing processes can be subject to first piece inspection.
It may be called out in a control strategy:
• Whenever a new production lot is started.
• Following maintenance/repairs of measurement systems and production equipment, as well as after software updates of production equipment control systems.
• At a defined interval (e.g., at the start of each shift).
• When tools used to produce the component contour are replaced (e.g., diamond rolls, profiled grinding/cutting wheels, etc.).
31
First-piece checking/inspection may be independent from the production method in a number of ways:
• Inspection by an operator other than the person having performed the operation (two person rule); thus, avoiding risks due to bias and other human factors.
• Inspection using another inspection tool or inspection method (where possible); thus avoiding/highlighting measurement discrepancies.
If independent inspection is to be used the method should be at least as good as the production method, free from bias and have adequate resolution to make the decisions valid. Tighter limits may apply to first piece checks, and this should be considered when evaluating such measurement equipment.
In order that the process is correctly judged as sufficiently good to continue additional criteria may be applied. Such criteria should have a rational and/or scientific basis for its application. For instance, a process capability study or designed experiments.
Example 1: A machined dimension with a known adequate level of capability, achieved at first part check may be deemed sufficient if within 50% of process tolerance; a measurement close to normal limits of operation may result in adjustment and further measurement to bring the process on target.
Example 2: A process with a tendency towards upward drift may have a zone in the lower region of the specification band that provides a standard for process acceptance of the first item. Continued conformity as the process drifts naturally through use is provided by a tool life/usage control. The zone has been determined through a previous tool wear study. If the measurement is outside this zone, the operator refers to a process guidance document (referenced in the Control Plan) to determine appropriate action (e.g., tool replacement, or adjustment to the tool life/usage standard).
A first piece check strategy may extend to multiple parts - depending on process risk and behavior. For example, a very large batch of parts, a rapidly cycling process or high cost parts may require inspection of the first five parts (Pre-Control may be beneficial (see Section 5.4)).
It is good practice to require formal record keeping for approval of first piece checks (e.g., a signature, and/or countersignature/inspection report).
NOTE: The method should be used in conjunction with other methods to make the control strategy robust to variations that may occur as production continues.
NOTE: First Piece Check should not be confused with First Article Inspection (FAI). For further information in FAI, refer to AS9102.
5.9 Test Piece Evaluation
Some characteristics and properties that are created or changed through processing may not be directly measurable other than through destructive or damaging testing. Use of test pieces processed alongside the product may help to determine the result of the process and also its stability. These test pieces are tested following processing to validate the products of the process and/or confirm the effectiveness of the other process controls.
Such processes should be highly controlled through process parameter controls and monitoring and may be categorized as ‘fixed processes’ or ‘special processes’ often with regulatory control requirements.
A test piece/coupon should be to a defined standard (thus minimizing the variation in the test material itself).
In some instances a test piece may be operated within a first piece check to qualify the process setup prior to the full production run (see Section 5.8).
Examples of processes that use representative test pieces include the following:
• Heat treatment operations
32
• Mechanical property testing using test bars
• Surface contamination coupons in heat treat or thermal processes
• Coupons determining material removal rates in etch and electro-polish processes
• Cast coupons determining chemical analysis of parts from melts
• A forging that has extra material outside the finished part envelope that will be removed for testing
Once a result has been obtained from a test piece the result can be analyzed with a variety of process control tools such as control charts (variable and attribute) and run charts.
Acceptance of process results by the use of test specimens or coupons is typically approved and agreed to by the customer.
NOTE: There may be regulatory, customer, product specifications, and other requirements that address the extent to which test piece evaluation, or requirements are permissible and established as part of process qualification. Equivalence between test piece and physical product should be understood.
6. PROCESS CAPABILITY INDICES
Process Capability is the ability of a process/product to consistently meet a specification or customer requirement.
Various indices are computed to assess the Process Capability of a given product characteristic.
The definition and calculation of these is often misunderstood and thus misinterpreted. The methods described within this section are based on recognized industry methods. Software tools such as Minitab calculate capability in line with these methods and additionally cater for some specific scenarios that exist such as batch processing where information may be sought about the capability both within and between batches of production.
Process Capability can be assessed for Variable and Attribute data.
6.1 Fundamentals for Variable data
At the heart of capability for variable data, is the need to manage process variation and location to align with customer specification to ensure that requirements can be continually met.
Variability of the process is calculated through statistical methods; these methods aim to anticipate the total process variation rather than just the range seen in the data collected for the capability study. A process spread of six standard deviations is used to represent this spread. This six standard deviation range theoretically covers 99.73% of the area under a normal distribution curve. Data is assumed to be normally distributed (symmetrical, bell shaped).
Many processes have a tendency - even naturally - to periodic drift or shift. Therefore, borderline capability is not desirable for either supplier or customer. A capability of 1.33 is often seen as a minimum to assure continued conformance while allowing for minor process drift. However, depending on the process, a higher level of capability may be required. Products with large numbers of characteristics that cannot be controlled independently may require some additional margin for small drifts that may occur through production.
For any capability calculation to be reliable, it is important that the process be in a state of statistical control thus behaving in a predictable manner - otherwise any perceived goodness may be short-lived. It is possible for a process with a ‘good’ capability index to be producing non-conforming product if a state of control is not reached. Process stability is therefore a prerequisite to capability calculation.
33
Capability Indices Cp and Pp
Cp and Pp indices are simply a ratio of specification width to process variation thus calculating the ‘potential of the process if centered’. The indices increase if variation is reduced. A Cp or Pp of exactly 1.0 indicates that six standard-deviations of process variation match the width of the specification. Such a process if centralized within the specification would be intolerant to even minor drift over time. Not an ideal situation.
Figure 23 - Process Capability Index Cp/Pp
The process shown in Figure 23 has a Cp or Pp>1. The process is less variable than allowed by the specification.
Cp and Pp use different methods for estimating process variability. Cp uses ranges of the data within subgroups (or difference between individual values) to estimate the process variation. A statistical constant d2 is used to adjust for the subgroup size. This method estimates the standard deviation of the process rather than calculating by the more involved ‘root sum of squares’ method (which is used to calculate Pp).
The average range over d2 method generates the estimate denoted by sigma hat (Equation 1).
(Eq. 1)
The root sum of squares method generates the standard deviation denoted by s (see Equation 2).
= ∑ (−)2 =1 −1
(Eq. 2)
Cp is typically used to assess short term (within subgroup) capability whereas Pp is used to assess longer term (overall) capability.
LSL USL
These are incorporated into the formulae (Equations 3 and 4) as follows:
(Eq. 3)
(Eq. 4)
For a stable continuous process behaving in a random manner, Cp, and Pp calculations can be expected to deliver similar values.
Capability Indices Cpk and Ppk
In order to estimate the likely performance - against a specification - of the process Cpk and Ppk indices are used. These indices are similar ratios to Cp and Pp but additionally take into account the process location.
Cpl and Cpu, and Ppl and Ppu measure capability against each of the specification limits. The ‘l’ and ‘u’ indices will be equal only if the process is centered. The Cpk or Ppk is the smaller of the upper and lower values.
The ‘l’ and ‘u’ indices can be used to determine how the process is located relative to specifications, however, a visual assessment of the capability histogram is usually preferred to understand this situation.
The formulae for these indices is shown (Equations 5 to 10).
(Eq. 5) (Eq. 8)
(Eq. 6) (Eq. 9)
(Eq. 7) (Eq. 10)
LSL USL
3σ 3σ
The process shown in Figure 24 has a Cp of approximately 1.0 but due to being too close to the upper specification limit (with the tail of the distribution outside it) the Cpk is <1 If the process average is outside the specification, the Cpk will be negative.
NOTE: It will not be possible to calculate Cp or Pp indices for processes with unilateral (single sided) tolerances as the tolerance width cannot be defined. However, Cpk and Ppk can be calculated from the Cpl/Ppl or Cpu/Ppu (whichever can be calculated).
Table 4 provides guidance on approximate expected performance levels at various levels of Process Capability. The performance rates assume a process perfectly centered between two specification limits. This table assumes a normal distribution.
Table 4 - Expected Performance for Cpk
CPK Sigma Level
(assumes “centered” process)
%YIELD (assumes “non-
REJECT RATE Parts Per Million
(assumes “non-centered” process with 1.5 sigma
shift) 0.50 1.5 86.64 133614 49.87 501350 0.67 2.0 95.45 45500 69.16 308417 0.80 2.4 98.36 16395 81.59 184108 1.00 3.0 99.73 2700 93.32 66811 1.20 3.6 99.97 318 98.21 17865 1.33 4.0 99.994 63 99.377 6227 1.50 4.5 99.9993 6.8 99.865 1350 1.67 5.0 99.99994 0.57 99.977 232 1.80 5.4 99.999993 0.067 99.9952 48 2.00 6.0 99.9999998 0.002 99.99966 3.4
For the Cpk and Ppk calculations in this section, the process is assumed normally distributed. If the data are non-normal (skewed for example) alternative methods can be used (see Section 7 - Guidance for Non-Normal Data).
NOTE: The descriptions in this section are fundamentals. Some additional methods for specific situations are described in Section 9 - Scenarios requiring specific analysis methods.
Some characteristics may benefit from being ‘targeted’ to a particular nominal value. These are usually characteristics that influence performance of the product, that have a loss associated with deviation from target even within the specification. These characteristics may have additional requirements communicated by the customer. For these types of characteristics, it is important to examine the location of the process relative to this target. It should be noted that due to the calculation methods, high Cpk/Ppk indices do not necessarily imply the process is on target as their calculations use the distance of the process mean to the specification limits. The nominal location is not considered in the calculation.
A target-based process capability index (Cpm) may be used in these situations. Cpm is not covered in this RM but is described in statistical texts and provided in statistical software applications.
Data Collection and Sample Size Considerations (Added)
A process capability index can mathematically be produced on any dataset with two samples or more. However, the confidence one would have in the capability metric will depend on the amount of data that has been gathered and how representative it is of the study in question.
The studies that are more likely are the initial process study and the ongoing performance study.
36
Consideration should be given to the following
• The items and the period over which the data will be collected
• The method of collection (either manual or automatic)
• The inspection method
• Interim review to act on obvious signals and trends (prior to full st