22.1 STATISTICAL PROCESS CONTROLSta

22.1 STATISTICAL PROCESS CONTROL
Statistical Process Control was briefly introduced in part two especially in chapter
6 (W. Edwards Deming). Its use relies on three key points:
• a pre-defined process;
• an established measurement system;
• operational (practical) definitions of the quality characteristics of the product or
service must have been agreed and understood by all parties.
The method for process charting was outlined in chapter 20.
STATISTICAL METHODS 245
A measurement system is required to monitor the performance of the defined
process with particular emphasis on quality achievement, that is, the proportion of
defective parts or other outputs (such as services) in relation to the total produced.
The measure of defective parts relies in turn on the operation definition of quality
characteristics. Construction of a measurement system consists in deciding what
should be measured and where, developing a recording and reporting system and,
to be effective, taking action on the results. Decisions about the measurement
system should be relatively straightforward once the process chart is constructed
and enables the construction of control charts which are much favoured by Deming,
Juran, Taguchi and others.
The platform for measurement is regular calculation and recording of standards
achieved to establish a pattern of system behaviour over time. Measurements must
be taken at time intervals appropriate to the process such as hourly, daily or weekly
and should capture a suitable sample, for example, an individual operator, a group,
shift or department. The plotted results provide the basis for corrective action and
performance improvement.
Plotting of the performance measurements enables the calculation of control
limits based on the actual performance of the process. While variations fall within
the limits, the process is deemed to be under control, that is in a stable state.
Variations outside the limits require immediate corrective action to restore stability.
Over time, the objective is to eradicate causes of variation outside the limits,
what Deming calls ‘special causes’. These may be removed through a variety of
approaches, for example improved equipment, training or improvement in quality of
inputs to the process.
Once special causes have been eradicated, the random variations within the
control limits are considered to have ‘common causes’; that is, the sources of
variation lie in the process itself although it is considered to be under control. The
reduction or eradication of these causes requires action on the process itself. This
is seen as the responsibility of management since only they have the authority to
make this level of change. Consistent and continuous effort is required to maintain
the system in a stable state and to continuously reduce variation.
It should be noted that the control limits relate to the degree of stability of the
process itself and are driven by process performance. They are not determined by
the product or service specifications. Operational definitions are specifications for
a product or service which include the acceptable limits of variation and the criteria
against which they can be measured. These must be expressed in terms which can
be understood by the supplier and the customer and must be useful in practice.
Specifications are important, since regardless of whether the process is ‘in
control’ or ‘out of control’ they provide the targets towards which the process
should be oriented. To have a process which is apparently in statistical control –
that is, the process is in a stable state – yet producing parts which do not meet the
specifications is useless. It is important therefore that the process is not only
246 METHODS, TOOLS AND TECHNIQUES
statistically in control but also capable of creating products or services which meet
the specifications. The specification limits will often be narrower, at least at the
outset, than the control limits of the process.
22.2 CONSTRUCTING CONTROL CHARTS
A control chart is used to record each occurrence of a particular event. This may be
either measurement of a continuous variable, for example, temperature, humidity,
thickness, weight or an attribute, that is, conformance to a requirement – specified
as conforming/non-conforming, or acceptable/ not-acceptable. Statistical analysis
is used to determine upper and lower control limits for the process in its current
state. The vertical axis of the control chart records the number of occurrences of a
particular event such as a product failure or defective parts produced. The horizontal
axis is normally based on time periods. An appropriate time period must be used for
the particular process, that is, it must relate to the production time cycle. It is
unhelpful to record defective parts produced in a week if the number of units
produced is measured in minutes!
Figure 22.1 provides a sample control chart. A chart may also be constructed
where the number of defective parts is measured against individuals or teams of
workers. This may help to identify where problems are occurring within a set of
common processes, say, where ten teams are working on identical processes, and
can highlight where investment in training or other problem resolution techniques
will bring benefit.
The upper and lower control limits for a ‘variables’ control chart are calculated
using the properties of normal distributions. For a normal distribution, approximately
99.8 per cent of values fall within a band of 6 standard deviations, that is, plus or
minus 3 standard deviations from the norm, and 68 per cent (approximately twoFigure
22.1 Sample control chart
STATISTICAL METHODS 247
thirds) fall within one standard deviation. On this basis, the likelihood of an event
falling outside the control limits is roughly 3 in every 1,000. It is considered therefore
that any event falling outside is caused by a ‘special’ rather than ‘common’ cause.
The upper and lower control limits (± 3 standard deviations) are treated as
triggers for action – any occurrence outside the limits must provoke an intervention
in the process. Limits may also be set at ± 2 standard deviations from the mean to
act as warning pointers to emergent deviation in the system.
In addition to using the basic information provided, the control charts may be
used for trend analysis and forecasting purposes. These enable feed-forward control
of processes which appear to be going out of control, and allow a degree of
predictability to be generated.
Control limits for ‘attribute’ based charts are calculated differently. There are
four versions of attribute chart (figure 22.2) chosen according to the sample size to
be used and the quality characteristic to be measured. The upper and lower limits
continue to be set at ± 3 standard deviations from the mean.
• p chart: varying batch size, varying sample size
• np chart: varying batch size, constant sample size
• u chart: component failure, varying sample size
• c chart: component failure, constant sample size
Figure 22.2 Attribute control charts
A p chart is used for defective items when it is not possible to take samples of a
constant size, for example if the batch sizes or flow volumes vary. An np chart deals
with the same problem but in situations where a constant sample is possible. The u
chart is used for monitoring component failure where it is not possible to take
constant sample sizes, the c chart when constant sample size is possible.
While this section has introduced control charts and provided the basis of their
operation it is not intended to turn readers into experts in SPC. It is intended to
provide only an introductory insight. All readers are recommended to employ the
services of a qualified statistician if seeking to employ these techniques in any
substantial manner. This is of particular importance in the choice of the appropriate
tool (s) and in calculating such matters as sample sizes, validity of results and
confidence limits. It is no use making changes to a process based on inaccurate or
misleading data.
248 METHODS, TOOLS AND TECHNIQUES
22.3 INTERPRETING CONTROL CHARTS
As well as informing the user whether or not a process is in statistical control,
control charts can provide clues to help determine and eliminate special causes of
variation.
A process which experiences only common causes of failure (that is those
inherent in the process) will exhibit the following characteristics:
• all points will fall within the control limits;
• points will be distributed evenly either side of the mean;
• the pattern will appear random;
• most points will be near the mean, that is, less than one standard deviation.
In general, an ‘in control’ process will require modification or redesign in order for
sustainable improvement to be achieved. For ‘out of control’ processes there are a
number of cues for investigation/action.
A process which continuously or regularly produces results falling below the
established lower control limit calls for two actions. First, is validation of the limit –
have the calculations been performed correctly? The second action is to investigate
the driver of the change and implement changes which reinforce the desirable
improvement.
Second, any point falling outside the control limits should be investigated since
it is likely to have a particular cause and the point of the system is to help identify
and eradicate such causes.
Third, a continuing run of points to one side or the other of the mean, or a trend
in one direction should be investigated. Again, a special cause may well be evident.
Fourth, any apparently non-random, cyclical or repeated patterns should be
investigated. These may relate to a particular operator or work group, a parts or
materials supplier or may coincide with particular events (Monday mornings! Friday
afternoons!). Again, desirable and undesirable patterns should be investigated to
enable eradication or repeatabilit

Kết quả (Việt) 1:

22.1 THỐNG KÊ QUY TRÌNH KIỂM SOÁTĐiều khiển quá trình thống kê đã được một thời gian ngắn giới thiệu trong phần hai đặc biệt là trong chương6 (W. Edwards Deming). Sử dụng của nó dựa trên ba điểm chính:• một quá trình được xác định trước;• một hệ thống đo lường được thành lập;• hoạt động (thực tế) định nghĩa về đặc điểm chất lượng của sản phẩm hoặcDịch vụ phải đã được đồng ý và hiểu bởi tất cả các bên.Phương pháp cho quá trình biểu đồ được nêu trong chương 20.PHƯƠNG PHÁP THỐNG KÊ 245Một hệ thống đo lường là cần thiết để giám sát hiệu suất của các định nghĩaxử lý với sự nhấn mạnh đặc biệt vào thành tích chất lượng, có nghĩa là, tỷ lệphần khiếm khuyết hoặc các kết quả đầu ra (ví dụ như dịch vụ) liên quan đến tổng số sản xuất.Các biện pháp của phần khiếm khuyết dựa lần lượt vào định nghĩa hoạt động chất lượngđặc điểm. Xây dựng một hệ thống đo lường bao gồm trong việc quyết định những gìnên được đo và ở đâu, phát triển một ghi âm và hệ thống báo cáo và,để có hiệu quả, thực hiện hành động trên các kết quả. Quyết định về đo lườngHệ thống nên tương đối đơn giản một khi biểu đồ quá trình được xây dựngvà cho phép xây dựng biểu đồ kiểm soát mà nhiều được ưa thích bởi Deming,Juran, Taguchi và những người khác.Nền tảng cho đo lường là tính toán thường xuyên và ghi âm của tiêu chuẩnđạt được để thiết lập một mô hình của hành vi hệ thống theo thời gian. Số đo phảiđược thực hiện tại khoảng thời gian thích hợp để trình chẳng hạn như hàng giờ, hàng ngày hoặc hàng tuầnand should capture a suitable sample, for example, an individual operator, a group,shift or department. The plotted results provide the basis for corrective action andperformance improvement.Plotting of the performance measurements enables the calculation of controllimits based on the actual performance of the process. While variations fall withinthe limits, the process is deemed to be under control, that is in a stable state.Variations outside the limits require immediate corrective action to restore stability.Over time, the objective is to eradicate causes of variation outside the limits,what Deming calls ‘special causes’. These may be removed through a variety ofapproaches, for example improved equipment, training or improvement in quality ofinputs to the process.Once special causes have been eradicated, the random variations within thecontrol limits are considered to have ‘common causes’; that is, the sources ofvariation lie in the process itself although it is considered to be under control. Thereduction or eradication of these causes requires action on the process itself. Thisis seen as the responsibility of management since only they have the authority tomake this level of change. Consistent and continuous effort is required to maintainthe system in a stable state and to continuously reduce variation.It should be noted that the control limits relate to the degree of stability of theprocess itself and are driven by process performance. They are not determined bythe product or service specifications. Operational definitions are specifications fora product or service which include the acceptable limits of variation and the criteriaagainst which they can be measured. These must be expressed in terms which canbe understood by the supplier and the customer and must be useful in practice.Specifications are important, since regardless of whether the process is ‘incontrol’ or ‘out of control’ they provide the targets towards which the processshould be oriented. To have a process which is apparently in statistical control –that is, the process is in a stable state – yet producing parts which do not meet thespecifications is useless. It is important therefore that the process is not only246 METHODS, TOOLS AND TECHNIQUESstatistically in control but also capable of creating products or services which meetthe specifications. The specification limits will often be narrower, at least at theoutset, than the control limits of the process.22.2 CONSTRUCTING CONTROL CHARTSA control chart is used to record each occurrence of a particular event. This may beeither measurement of a continuous variable, for example, temperature, humidity,thickness, weight or an attribute, that is, conformance to a requirement – specifiedas conforming/non-conforming, or acceptable/ not-acceptable. Statistical analysisis used to determine upper and lower control limits for the process in its current
state. The vertical axis of the control chart records the number of occurrences of a
particular event such as a product failure or defective parts produced. The horizontal
axis is normally based on time periods. An appropriate time period must be used for
the particular process, that is, it must relate to the production time cycle. It is
unhelpful to record defective parts produced in a week if the number of units
produced is measured in minutes!
Figure 22.1 provides a sample control chart. A chart may also be constructed
where the number of defective parts is measured against individuals or teams of
workers. This may help to identify where problems are occurring within a set of
common processes, say, where ten teams are working on identical processes, and
can highlight where investment in training or other problem resolution techniques
will bring benefit.
The upper and lower control limits for a ‘variables’ control chart are calculated
using the properties of normal distributions. For a normal distribution, approximately
99.8 per cent of values fall within a band of 6 standard deviations, that is, plus or
minus 3 standard deviations from the norm, and 68 per cent (approximately twoFigure
22.1 Sample control chart
STATISTICAL METHODS 247
thirds) fall within one standard deviation. On this basis, the likelihood of an event
falling outside the control limits is roughly 3 in every 1,000. It is considered therefore
that any event falling outside is caused by a ‘special’ rather than ‘common’ cause.
The upper and lower control limits (± 3 standard deviations) are treated as
triggers for action – any occurrence outside the limits must provoke an intervention
in the process. Limits may also be set at ± 2 standard deviations from the mean to
act as warning pointers to emergent deviation in the system.
In addition to using the basic information provided, the control charts may be
used for trend analysis and forecasting purposes. These enable feed-forward control
of processes which appear to be going out of control, and allow a degree of
predictability to be generated.
Control limits for ‘attribute’ based charts are calculated differently. There are
four versions of attribute chart (figure 22.2) chosen according to the sample size to
be used and the quality characteristic to be measured. The upper and lower limits
continue to be set at ± 3 standard deviations from the mean.
• p chart: varying batch size, varying sample size
• np chart: varying batch size, constant sample size
• u chart: component failure, varying sample size
• c chart: component failure, constant sample size
Figure 22.2 Attribute control charts
A p chart is used for defective items when it is not possible to take samples of a
constant size, for example if the batch sizes or flow volumes vary. An np chart deals
with the same problem but in situations where a constant sample is possible. The u
chart is used for monitoring component failure where it is not possible to take
constant sample sizes, the c chart when constant sample size is possible.
While this section has introduced control charts and provided the basis of their
operation it is not intended to turn readers into experts in SPC. It is intended to
provide only an introductory insight. All readers are recommended to employ the
services of a qualified statistician if seeking to employ these techniques in any
substantial manner. This is of particular importance in the choice of the appropriate
tool (s) and in calculating such matters as sample sizes, validity of results and
confidence limits. It is no use making changes to a process based on inaccurate or
misleading data.
248 METHODS, TOOLS AND TECHNIQUES
22.3 INTERPRETING CONTROL CHARTS
As well as informing the user whether or not a process is in statistical control,
control charts can provide clues to help determine and eliminate special causes of
variation.
A process which experiences only common causes of failure (that is those
inherent in the process) will exhibit the following characteristics:
• all points will fall within the control limits;
• points will be distributed evenly either side of the mean;
• the pattern will appear random;
• most points will be near the mean, that is, less than one standard deviation.
In general, an ‘in control’ process will require modification or redesign in order for
sustainable improvement to be achieved. For ‘out of control’ processes there are a
number of cues for investigation/action.
A process which continuously or regularly produces results falling below the
established lower control limit calls for two actions. First, is validation of the limit –
have the calculations been performed correctly? The second action is to investigate
the driver of the change and implement changes which reinforce the desirable
improvement.
Second, any point falling outside the control limits should be investigated since
it is likely to have a particular cause and the point of the system is to help identify
and eradicate such causes.
Third, a continuing run of points to one side or the other of the mean, or a trend
in one direction should be investigated. Again, a special cause may well be evident.
Fourth, any apparently non-random, cyclical or repeated patterns should be
investigated. These may relate to a particular operator or work group, a parts or
materials supplier or may coincide with particular events (Monday mornings! Friday
afternoons!). Again, desirable and undesirable patterns should be investigated to
enable eradication or repeatabilit