1
00:00:03,376 --> 00:00:06,876
We have learned in this course that
data can be used for several objectives.

2
00:00:07,336 --> 00:00:11,366
The first is simply just to learn about your
process, and understand what is going on.

3
00:00:11,866 --> 00:00:16,026
For example, in the section on t-tests
we learned how to verify whether changes

4
00:00:16,026 --> 00:00:18,426
to our process are actually significant or not.

5
00:00:18,696 --> 00:00:21,996
We can also learn by observing
trends in the data visualizations,

6
00:00:22,296 --> 00:00:24,236
to troubleshoot problems, and so forth.

7
00:00:24,776 --> 00:00:27,696
In this video we are going to
look at process monitoring,

8
00:00:28,126 --> 00:00:30,996
where we build on that topic
and take it step further.

9
00:00:31,536 --> 00:00:35,216
Other interesting things that we can do with
our data are to build predictive models,

10
00:00:35,216 --> 00:00:37,586
such as a least squares model
to make a prediction

11
00:00:37,586 --> 00:00:39,966
of a property that's really hard to measure.

12
00:00:40,676 --> 00:00:44,746
We've also just recently seen how we can
optimize our processes using response

13
00:00:44,746 --> 00:00:45,556
surface methods.

14
00:00:46,146 --> 00:00:50,476
Climbing that mountain and maximizing some
value on our process, such as the profit.

15
00:00:51,066 --> 00:00:56,186
The last section here considers the fifth
major objective, that of process monitoring.

16
00:00:56,716 --> 00:01:01,206
Process monitoring is a way that we can
track how our system behaves in real time,

17
00:01:01,516 --> 00:01:03,766
to ensure that it remains on target.

18
00:01:04,346 --> 00:01:08,776
Any patterns that we observe in these monitoring
plots are then used for troubleshooting.

19
00:01:09,466 --> 00:01:12,636
It nicely ties the second and
the first objectives together.

20
00:01:13,116 --> 00:01:15,926
Let's start with a few monitoring
examples that you've seen before.

21
00:01:16,526 --> 00:01:18,196
The first is that of a hospital.

22
00:01:18,386 --> 00:01:22,856
We've all seen this on TV, where a patient
is being monitored for various vital signs

23
00:01:23,116 --> 00:01:28,096
such as heartbeat, blood pressure, oxygen
level, blood glucose level, body temperature.

24
00:01:28,206 --> 00:01:29,936
These are all parameters that now go

25
00:01:29,936 --> 00:01:33,126
into hospital databases to
track patients in real time.

26
00:01:33,636 --> 00:01:37,126
Perhaps you have seen these monitoring
charts in a control room in a chemical plant,

27
00:01:37,416 --> 00:01:41,386
or perhaps you've observed stock market
charts and people who do intra-day trading.

28
00:01:41,936 --> 00:01:46,156
In an engineering context, we can monitor
our processes for their vital signs

29
00:01:46,446 --> 00:01:50,536
to ensure they're behaving on-target
and away from unsafe operation.

30
00:01:51,076 --> 00:01:53,026
When problems are observed in these charts,

31
00:01:53,136 --> 00:01:55,406
engineers and operators quickly
react quickly to them.

32
00:01:55,906 --> 00:02:00,066
Notice that process monitoring is a
reactive step, it is not proactive.

33
00:02:00,766 --> 00:02:04,536
Optimization, which we saw earlier in
this course, is a proactive activity.

34
00:02:05,006 --> 00:02:09,416
There we proactively moved the process
to a better location or operating point.

35
00:02:09,986 --> 00:02:14,476
More than any other aspect that we've learned
about in this course, this area of monitoring

36
00:02:14,476 --> 00:02:17,496
and data acquisition is quickly
growing in importance.

37
00:02:17,846 --> 00:02:21,946
You may have heard of the terms big
data, lean manufacturing, "six sigma".

38
00:02:22,656 --> 00:02:25,196
All of these come out of this
topic of process monitoring.

39
00:02:25,696 --> 00:02:27,636
If you want to learn more about this area,

40
00:02:27,816 --> 00:02:30,516
I strongly suggest reading
these good books shown here

41
00:02:30,516 --> 00:02:32,706
on the screen for an engineering context.

42
00:02:33,126 --> 00:02:37,246
However, I can guarantee this topic will be
quite different, even five years from now.

43
00:02:37,726 --> 00:02:41,036
Let's start off though and look at
the workhorse of process monitoring.

44
00:02:41,276 --> 00:02:45,696
This is a chart that has been around for
almost a century, it's not going to go away.

45
00:02:46,186 --> 00:02:50,466
The first feature that you notice is the fact
that it is a time series or sequence plot,

46
00:02:50,936 --> 00:02:53,266
where new points are added
on the right-hand side,

47
00:02:53,526 --> 00:02:56,816
and previous points get removed,
or disappear, on the left.

48
00:02:57,286 --> 00:03:00,766
It is displayed in real time, or
as close to real time as possible.

49
00:03:01,236 --> 00:03:05,066
The units on the vertical axis are the
units of the variable being measured.

50
00:03:05,596 --> 00:03:10,226
There might be several horizontal lines
also drawn, one of which is the target line.

51
00:03:10,616 --> 00:03:13,626
There might also be upper control
limits and lower control limits.

52
00:03:13,816 --> 00:03:16,216
We'll see how these are derived
in the coming videos.

53
00:03:16,826 --> 00:03:21,326
I'm going to start with a demonstration though
that will quickly illustrate how these are used.

54
00:03:21,986 --> 00:03:25,056
Here is an example of an actual
system I had a chance to work on.

55
00:03:25,516 --> 00:03:29,896
The company was monitoring the appearance
of bubbles on the top of a flotation froth.

56
00:03:30,876 --> 00:03:34,636
Flotation is a process whereby
minerals attach themselves to a bubble;

57
00:03:34,806 --> 00:03:37,206
float to the top of the tank and are removed.

58
00:03:37,686 --> 00:03:41,616
Air is bubbled into the system and
mixed in to assist this process.

59
00:03:41,966 --> 00:03:45,616
The company places a digital video
camera over the top of the tank

60
00:03:45,686 --> 00:03:47,556
and observes the appearance of the bubble.

61
00:03:47,946 --> 00:03:50,976
Some examples of these images
are now shown here on the screen.

62
00:03:51,546 --> 00:03:55,126
One of the parameters the operators
are interested in, is the bubble size.

63
00:03:55,386 --> 00:03:59,106
They're also interested in the bubble's
colour, and other textural features.

64
00:03:59,526 --> 00:04:03,926
I'm showing an accelerated version of the two
monitoring charts developed for this process.

65
00:04:04,206 --> 00:04:08,136
The first is the bubble diameter,
measured in millimetres, and the lower

66
00:04:08,136 --> 00:04:11,076
and upper control limits are
shown, as well as the target value.

67
00:04:11,436 --> 00:04:17,476
The second chart monitors the grey-scale
colour, which is a scale from 0 to 255,

68
00:04:17,786 --> 00:04:19,616
and there are also limits for this value.

69
00:04:20,066 --> 00:04:22,876
Now at some point in time the
following might take place.

70
00:04:23,116 --> 00:04:25,246
We see the bubble diameter has shifted down.

71
00:04:25,656 --> 00:04:28,766
The grey-scale colour has
also changed, and shifted up.

72
00:04:29,556 --> 00:04:33,366
The operator notices this signature,
of a very particular problem

73
00:04:33,366 --> 00:04:35,406
that occurs periodically in the process.

74
00:04:35,826 --> 00:04:39,386
When they are alerted to this,
through an automated alarm process,

75
00:04:39,606 --> 00:04:42,466
they know exactly what to do
to counteract the problem.

76
00:04:42,876 --> 00:04:47,216
Notice however that the monitoring chart will
never conclusively tell you exactly what is

77
00:04:47,216 --> 00:04:48,186
wrong with the process.

78
00:04:48,556 --> 00:04:51,316
It will simply alert you to the
fact that something is wrong.

79
00:04:51,566 --> 00:04:53,596
You will have to use your
judgement and knowledge

80
00:04:53,596 --> 00:04:55,906
of the physical system to the troubleshooting.

81
00:04:56,446 --> 00:05:00,206
This is no different to a nurse or a
doctor in a medical facility who will have

82
00:05:00,206 --> 00:05:04,876
to observe the signals being shown on the
medical devices to determine the problem is,

83
00:05:05,336 --> 00:05:07,946
and then make the subsequent
diagnosis to fix it.

84
00:05:08,336 --> 00:05:13,246
Notice however that a variety of different
problems can have the same signature

85
00:05:13,246 --> 00:05:14,046
on the chart.

86
00:05:14,556 --> 00:05:19,216
There is never a one-to-one relationship between
a problem, and its signature on the plot.

87
00:05:19,556 --> 00:05:22,276
This is why there is always
human intervention required

88
00:05:22,276 --> 00:05:24,586
to diagnose, and then to fix the problem.

89
00:05:25,056 --> 00:05:29,646
This chart that we've just used is called the
Shewhart chart, and named after Walter Shewhart

90
00:05:29,646 --> 00:05:33,076
from the Bell Telephone Company,
who developed in the 1920s

91
00:05:33,076 --> 00:05:35,706
to monitor the production of
parts at the phone company.

92
00:05:36,156 --> 00:05:40,946
It is a chart for monitoring the location of a
variable; where it lies on that vertical axis.

93
00:05:41,496 --> 00:05:44,716
Shewhart charts often have a
lower and an upper control limit,

94
00:05:45,056 --> 00:05:47,006
as well as a target line drawn on them.

95
00:05:47,336 --> 00:05:51,766
A process is considered to be "in
control" if it lies within those limits.

96
00:05:52,236 --> 00:05:54,606
The opposite is a process
being "out of control",

97
00:05:55,006 --> 00:05:59,146
when it lies beyond the upper control
limit or below the lower control limit.

98
00:05:59,466 --> 00:06:03,036
We say that a process is in
control when there is variation,

99
00:06:03,446 --> 00:06:06,766
but we call that common cause
variability, within the limits.

100
00:06:07,186 --> 00:06:13,446
Regular operation is stable, the product being
produced has variability, but it is still sold

101
00:06:13,446 --> 00:06:15,866
to the customer as good quality product.

102
00:06:16,476 --> 00:06:21,746
When we are out of control, we say that a
special cause, or special causes, have occurred.

103
00:06:22,276 --> 00:06:27,536
Some destabilizing event has happened, we
are out of control, we are "off-target".

104
00:06:27,866 --> 00:06:30,816
This is product that we typically
will not sell to our customer.

105
00:06:31,166 --> 00:06:35,576
We may have to modify it, or sell it
at a lower price, or even destroy it.

106
00:06:36,146 --> 00:06:39,866
Now one of the toughest problems that
engineers often face is to figure out which

107
00:06:39,866 --> 00:06:42,896
of the many variables available
to us, should be monitored.

108
00:06:43,326 --> 00:06:46,906
Companies have hundreds, if not
thousands, of variables available to them.

109
00:06:46,906 --> 00:06:49,726
Especially on newer plants
with multiple sensors.

110
00:06:50,076 --> 00:06:54,386
Consider the following situations before
we continue on: What would you monitor

111
00:06:54,386 --> 00:06:56,856
if you were running a waste-water
treatment process?

112
00:06:57,426 --> 00:07:00,426
Which variables would you monitor
in an oil and gas facility?

113
00:07:01,066 --> 00:07:05,756
What might be of interest to track in a food
processing unit, or a mineral processing plant?

114
00:07:06,126 --> 00:07:07,906
Or what if you were producing plastics?

115
00:07:08,366 --> 00:07:12,976
Which variables monitor the key quality
properties in each of those situations?

116
00:07:13,356 --> 00:07:15,136
What about a pharmaceutical facility,

117
00:07:15,136 --> 00:07:18,736
how do we know we are producing good
quality product in that location?

118
00:07:19,266 --> 00:07:23,426
Once you have identified which variable
you would like to monitor, we can then go

119
00:07:23,426 --> 00:07:25,426
and start constructing a monitoring chart.

120
00:07:25,836 --> 00:07:28,616
Figuring out what the upper
control limit is that we should use,

121
00:07:28,616 --> 00:07:32,136
what the lower control limit should be,
and what should the target value be?

122
00:07:32,546 --> 00:07:36,216
In the process monitoring literature
that step of building the chart,

123
00:07:36,416 --> 00:07:41,786
of figuring out those limits and testing the
chart on prior operating data is called phase 1.

124
00:07:42,146 --> 00:07:44,656
This is where you will spend
most of your time as an engineer.

125
00:07:45,226 --> 00:07:50,166
Phase 2 is the phase where we go and use this
chart on new data that we've never seen before.

126
00:07:50,626 --> 00:07:54,726
This is where the operators and the final
end-users of your chart will spend their time.

127
00:07:55,216 --> 00:07:59,696
In the next video we will look at the
phase 1 construction of a Shewhart chart.