All language subtitles for Internet of Things for Everything Equipment Maintenance Assistant.en.transcribed
Afrikaans
Albanian
Amharic
Arabic
Armenian
Azerbaijani
Basque
Belarusian
Bengali
Bosnian
Bulgarian
Catalan
Cebuano
Chichewa
Chinese (Simplified)
Chinese (Traditional)
Corsican
Croatian
Czech
Danish
Dutch
English
Esperanto
Estonian
Filipino
Finnish
French
Frisian
Galician
Georgian
German
Greek
Gujarati
Haitian Creole
Hausa
Hawaiian
Hebrew
Hindi
Hmong
Hungarian
Icelandic
Igbo
Indonesian
Irish
Italian
Japanese
Javanese
Kannada
Kazakh
Khmer
Korean
Kurdish (Kurmanji)
Kyrgyz
Lao
Latin
Latvian
Lithuanian
Luxembourgish
Macedonian
Malagasy
Malay
Malayalam
Maltese
Maori
Marathi
Mongolian
Myanmar (Burmese)
Nepali
Norwegian
Pashto
Persian
Polish
Portuguese
Punjabi
Romanian
Russian
Samoan
Scots Gaelic
Serbian
Sesotho
Shona
Sindhi
Sinhala
Slovak
Slovenian
Somali
Spanish
Sundanese
Swahili
Swedish
Tajik
Tamil
Telugu
Thai
Turkish
Ukrainian
Urdu
Uzbek
Vietnamese
Welsh
Xhosa
Yiddish
Yoruba
Zulu
Odia (Oriya)
Kinyarwanda
Turkmen
Tatar
Uyghur
Would you like to inspect the original subtitles? These are the user uploaded subtitles that are being translated:
1
00:00:04,029 --> 00:00:08,629
hi my name is David Boyle and I'm a
2
00:00:06,649 --> 00:00:11,480
technical specialist with IBM Watson IOT
3
00:00:08,629 --> 00:00:12,889
I work across all the industries with
4
00:00:11,480 --> 00:00:14,600
the focus of helping clients better
5
00:00:12,889 --> 00:00:17,509
manage the performance of their assets
6
00:00:14,600 --> 00:00:19,160
in this video I'll be discussing how the
7
00:00:17,510 --> 00:00:21,140
equipment maintenance assistance
8
00:00:19,160 --> 00:00:23,300
solution is used to capture the
9
00:00:21,140 --> 00:00:25,460
knowledge of an ageing workforce and how
10
00:00:23,300 --> 00:00:27,490
its underlying AI is used to help field
11
00:00:25,460 --> 00:00:29,390
technicians increase their productivity
12
00:00:27,490 --> 00:00:30,948
please note that throughout this video
13
00:00:29,390 --> 00:00:34,600
I'll be referring to equipment
14
00:00:30,949 --> 00:00:37,130
maintenance assistant as EMA for short
15
00:00:34,600 --> 00:00:38,810
now before I dive into this solution
16
00:00:37,130 --> 00:00:41,120
itself I think it's important to
17
00:00:38,810 --> 00:00:42,980
understand the motivations behind it at
18
00:00:41,120 --> 00:00:44,660
a high level companies with capital
19
00:00:42,980 --> 00:00:46,400
equipment have always been challenged
20
00:00:44,660 --> 00:00:49,429
with lowering operating cost and
21
00:00:46,400 --> 00:00:51,019
increasing production it used to be that
22
00:00:49,430 --> 00:00:52,880
the preventative maintenance approach of
23
00:00:51,020 --> 00:00:54,890
replacing parts every X number of run
24
00:00:52,880 --> 00:00:55,750
hours was the only way to address these
25
00:00:54,890 --> 00:00:58,880
challenges
26
00:00:55,750 --> 00:01:01,400
eventually SCADA systems PLC's and IOT
27
00:00:58,880 --> 00:01:03,170
analytics allowed operators to become
28
00:01:01,400 --> 00:01:04,938
more proactive by monitoring the
29
00:01:03,170 --> 00:01:08,210
condition of their assets in real time
30
00:01:04,938 --> 00:01:11,178
or by predicting failures and extending
31
00:01:08,210 --> 00:01:12,469
maintenance intervals these insights
32
00:01:11,179 --> 00:01:13,820
have proven to be valuable in
33
00:01:12,469 --> 00:01:16,939
understanding the performance of their
34
00:01:13,820 --> 00:01:18,288
assets but they are not resolutions no
35
00:01:16,939 --> 00:01:19,999
matter the approach a field technician
36
00:01:18,289 --> 00:01:22,729
still needs to go out and service that
37
00:01:19,999 --> 00:01:25,100
equipment that being said the key to
38
00:01:22,729 --> 00:01:27,770
actually lowering operating costs and
39
00:01:25,100 --> 00:01:29,749
actually increasing production there's a
40
00:01:27,770 --> 00:01:32,240
combination of insights and effective
41
00:01:29,749 --> 00:01:34,969
field technicians and so in order for
42
00:01:32,240 --> 00:01:36,408
IBM to have developed EMA and provide
43
00:01:34,969 --> 00:01:38,630
complete end-to-end maintenance
44
00:01:36,409 --> 00:01:42,499
solutions we had to first understand the
45
00:01:38,630 --> 00:01:43,969
challenges of the field technician from
46
00:01:42,499 --> 00:01:45,798
that perspective we see that there's a
47
00:01:43,969 --> 00:01:47,770
continued increase in the complexity of
48
00:01:45,799 --> 00:01:50,509
machinery and the systems they make up
49
00:01:47,770 --> 00:01:52,490
at the same time the availability of
50
00:01:50,509 --> 00:01:55,429
equipment expertise is shrinking as a
51
00:01:52,490 --> 00:01:57,499
result of an aging workforce so when you
52
00:01:55,429 --> 00:01:58,939
couple these two challenges it causes
53
00:01:57,499 --> 00:02:01,249
technicians to have to spend more time
54
00:01:58,939 --> 00:02:03,678
in the field and revisit equipment more
55
00:02:01,249 --> 00:02:06,259
than once which in other words means an
56
00:02:03,679 --> 00:02:09,289
increase in mean time to repair and a
57
00:02:06,259 --> 00:02:11,269
decrease in first-time fix rate these
58
00:02:09,288 --> 00:02:13,548
metrics are associated with operating
59
00:02:11,269 --> 00:02:15,349
cost and production levels and the way
60
00:02:13,549 --> 00:02:17,450
we're addressing them directly is with
61
00:02:15,349 --> 00:02:20,629
EMA
62
00:02:17,450 --> 00:02:22,349
so what is EMA and how does one use it
63
00:02:20,629 --> 00:02:26,340
simply put ei
64
00:02:22,349 --> 00:02:28,410
EMA is an AI digital assistant that
65
00:02:26,340 --> 00:02:31,280
field technicians can use at the point
66
00:02:28,410 --> 00:02:33,870
of work to surface diagnosis and repair
67
00:02:31,280 --> 00:02:37,260
recommendations from a corpus of tribal
68
00:02:33,870 --> 00:02:39,180
knowledge EMA's ability to transfer this
69
00:02:37,260 --> 00:02:41,670
knowledge from subject matter experts to
70
00:02:39,180 --> 00:02:42,930
less experienced field technicians is
71
00:02:41,670 --> 00:02:45,329
what allows them to make the best
72
00:02:42,930 --> 00:02:47,549
repairs the first time and in a timely
73
00:02:45,330 --> 00:02:50,549
manner thus addressing first-time
74
00:02:47,549 --> 00:02:52,920
fixed-rate and mean time to prepare now
75
00:02:50,549 --> 00:02:55,410
in order to get started with EMA the
76
00:02:52,920 --> 00:02:57,450
knowledge base that bill refer to must
77
00:02:55,410 --> 00:02:59,880
be prepared this is typically done by
78
00:02:57,450 --> 00:03:01,488
subject matter expert and can include
79
00:02:59,880 --> 00:03:04,890
both structured and unstructured
80
00:03:01,489 --> 00:03:07,170
documentation such as OAM manuals repair
81
00:03:04,890 --> 00:03:09,630
guides historical work orders online
82
00:03:07,170 --> 00:03:12,809
forums and diagnosis models to name a
83
00:03:09,630 --> 00:03:14,160
few from there the underlying watson
84
00:03:12,810 --> 00:03:16,799
algorithms will automatically identify
85
00:03:14,160 --> 00:03:18,930
and enrich the content found within the
86
00:03:16,799 --> 00:03:22,650
uploaded documentation to help us search
87
00:03:18,930 --> 00:03:25,350
be in accuracy once this process is
88
00:03:22,650 --> 00:03:27,630
complete the AI will need to be trained
89
00:03:25,350 --> 00:03:29,400
on relevancy that is what are the
90
00:03:27,630 --> 00:03:30,840
questions being asked what are the
91
00:03:29,400 --> 00:03:34,350
responses and how should they be
92
00:03:30,840 --> 00:03:36,060
prioritized this second step in the
93
00:03:34,350 --> 00:03:38,069
preparation phase is typically performed
94
00:03:36,060 --> 00:03:40,500
by the same subject matter expert and
95
00:03:38,069 --> 00:03:42,600
involves a simple iterative process of
96
00:03:40,500 --> 00:03:44,670
asking questions and rating responses
97
00:03:42,600 --> 00:03:47,850
until a high degree of confidence is
98
00:03:44,670 --> 00:03:50,220
achieved once the SME is comfortable
99
00:03:47,850 --> 00:03:52,590
with the results EMA is ready to be
100
00:03:50,220 --> 00:03:53,970
deployed so now when work orders are
101
00:03:52,590 --> 00:03:56,310
issued from an enterprise asset
102
00:03:53,970 --> 00:03:58,709
management system such as Maximo or sa
103
00:03:56,310 --> 00:04:01,109
ppm the less experienced field
104
00:03:58,709 --> 00:04:03,630
technician can use that context to
105
00:04:01,109 --> 00:04:05,250
prepare recommended parts tools and
106
00:04:03,630 --> 00:04:08,579
safety equipment before heading out to
107
00:04:05,250 --> 00:04:11,069
the site once at the site the same field
108
00:04:08,579 --> 00:04:13,019
technician can leverage a diagnosis
109
00:04:11,069 --> 00:04:15,450
model to help determine root cause or
110
00:04:13,019 --> 00:04:17,760
use EMA's query capabilities to gather
111
00:04:15,450 --> 00:04:20,459
step-by-step procedures you need to make
112
00:04:17,760 --> 00:04:22,289
a quick and accurate fix then when they
113
00:04:20,459 --> 00:04:24,300
go to close work orders the field
114
00:04:22,289 --> 00:04:26,430
technician can also provide feedback to
115
00:04:24,300 --> 00:04:29,039
help the underlying AI continuously
116
00:04:26,430 --> 00:04:29,550
improve and then add that to the growing
117
00:04:29,039 --> 00:04:31,620
knowledge
118
00:04:29,550 --> 00:04:33,930
future technicians can take advantage of
119
00:04:31,620 --> 00:04:35,520
it and the demonstration I'm about to
120
00:04:33,930 --> 00:04:38,220
give will look at how an SME would
121
00:04:35,520 --> 00:04:40,440
prepare EMA as well as how an automotive
122
00:04:38,220 --> 00:04:42,540
technician would use EMA to determine
123
00:04:40,440 --> 00:04:47,700
why a car won't start and how it should
124
00:04:42,540 --> 00:04:50,190
be repaired now as I just stated the
125
00:04:47,700 --> 00:04:51,810
desired outcome of using EMA is to
126
00:04:50,190 --> 00:04:53,610
transfer knowledge from subject matter
127
00:04:51,810 --> 00:04:56,490
experts to less experienced field
128
00:04:53,610 --> 00:04:58,830
technicians I also mentioned that the
129
00:04:56,490 --> 00:05:00,390
first step in doing so is to build out
130
00:04:58,830 --> 00:05:03,050
the knowledge base from which they will
131
00:05:00,390 --> 00:05:06,120
extract those repair recommendations
132
00:05:03,050 --> 00:05:08,430
that being said we'll first take on the
133
00:05:06,120 --> 00:05:11,790
role of a subject matter expert or an
134
00:05:08,430 --> 00:05:14,490
SME and look at how this process is done
135
00:05:11,790 --> 00:05:19,200
using EMA's document manager and the
136
00:05:14,490 --> 00:05:21,930
diagnosis model manager in the document
137
00:05:19,200 --> 00:05:25,229
manager the SME is able to create
138
00:05:21,930 --> 00:05:28,200
various collections of documents as they
139
00:05:25,230 --> 00:05:29,670
relate to specific assets so as you see
140
00:05:28,200 --> 00:05:31,680
in the shared environment there's a
141
00:05:29,670 --> 00:05:34,920
number of different assets with their
142
00:05:31,680 --> 00:05:36,750
own distinctive collections but in our
143
00:05:34,920 --> 00:05:39,840
case we want to create a collection
144
00:05:36,750 --> 00:05:41,730
that's related to a car so to do that we
145
00:05:39,840 --> 00:05:43,710
would hit the create button we would
146
00:05:41,730 --> 00:05:47,340
give a name to the collection in this
147
00:05:43,710 --> 00:05:49,140
case my car we would define the
148
00:05:47,340 --> 00:05:52,500
collection type which in this case we'll
149
00:05:49,140 --> 00:05:55,229
keep as documents and then we'll need to
150
00:05:52,500 --> 00:05:58,770
associate this collection with one in
151
00:05:55,230 --> 00:06:00,930
Watson discovery and this is what the
152
00:05:58,770 --> 00:06:03,000
Watson discovery interface looks like
153
00:06:00,930 --> 00:06:05,670
and as you can see there are collections
154
00:06:03,000 --> 00:06:09,060
similar to what we saw in the document
155
00:06:05,670 --> 00:06:11,550
manager so what you would do here is hit
156
00:06:09,060 --> 00:06:14,120
the upload and essentially replicate
157
00:06:11,550 --> 00:06:17,460
this collection in Watson discovery and
158
00:06:14,120 --> 00:06:20,970
then you would come back here select it
159
00:06:17,460 --> 00:06:22,890
from the drop-down and continue and this
160
00:06:20,970 --> 00:06:24,900
is arguably the most important step
161
00:06:22,890 --> 00:06:27,719
because without this the SME would not
162
00:06:24,900 --> 00:06:30,210
be able to understand their
163
00:06:27,720 --> 00:06:32,130
documentation and also they would not be
164
00:06:30,210 --> 00:06:34,950
able to perform relevancy training on
165
00:06:32,130 --> 00:06:37,020
that documentation now normally I would
166
00:06:34,950 --> 00:06:38,640
hit create at this point but I've
167
00:06:37,020 --> 00:06:42,849
actually gone ahead and created two
168
00:06:38,640 --> 00:06:44,438
collections for the car so scroll down
169
00:06:42,849 --> 00:06:49,959
take a look at the first one called my
170
00:06:44,439 --> 00:06:54,389
car this collection contains various PDF
171
00:06:49,959 --> 00:06:56,349
documents containing either OAM manuals
172
00:06:54,389 --> 00:06:59,199
repair guides in any other
173
00:06:56,349 --> 00:07:01,389
troubleshooting procedures if we come
174
00:06:59,199 --> 00:07:04,089
back out of the document manager we can
175
00:07:01,389 --> 00:07:08,069
look at the second collection titled my
176
00:07:04,089 --> 00:07:11,319
car work orders which as you can imagine
177
00:07:08,069 --> 00:07:14,289
contains historical work orders that the
178
00:07:11,319 --> 00:07:15,909
field technicians can use as a reference
179
00:07:14,289 --> 00:07:19,389
point to see what other technicians have
180
00:07:15,909 --> 00:07:21,849
done on these repairs in the past now
181
00:07:19,389 --> 00:07:24,580
both of these collections are very
182
00:07:21,849 --> 00:07:27,628
useful in situations where the
183
00:07:24,580 --> 00:07:29,800
technician has some sort of context
184
00:07:27,629 --> 00:07:32,800
related to the repair that they have to
185
00:07:29,800 --> 00:07:35,080
make and that may come from work orders
186
00:07:32,800 --> 00:07:36,999
or from predictive maintenance systems
187
00:07:35,080 --> 00:07:39,609
or even condition based maintenance
188
00:07:36,999 --> 00:07:41,679
systems but there are a lot of times
189
00:07:39,610 --> 00:07:44,740
where the field technician must also
190
00:07:41,679 --> 00:07:47,859
perform a diagnosis to determine the
191
00:07:44,740 --> 00:07:50,249
root cause before they understand the
192
00:07:47,860 --> 00:07:53,139
types of questions that they need to ask
193
00:07:50,249 --> 00:07:55,389
so another way in which the SME can
194
00:07:53,139 --> 00:07:58,269
build out the knowledge base is by
195
00:07:55,389 --> 00:08:00,490
creating a diagnosis model and as you
196
00:07:58,269 --> 00:08:02,769
can see I've already created one for the
197
00:08:00,490 --> 00:08:07,360
car so we'll go ahead and open that one
198
00:08:02,769 --> 00:08:11,679
up here the SME has a canvas where they
199
00:08:07,360 --> 00:08:16,419
can drop caused nodes and symptom nodes
200
00:08:11,679 --> 00:08:19,779
and then define them and link them
201
00:08:16,419 --> 00:08:22,508
together using probabilities so from the
202
00:08:19,779 --> 00:08:24,159
technician perspective the technician is
203
00:08:22,509 --> 00:08:26,559
able to go through a list of symptoms
204
00:08:24,159 --> 00:08:30,490
click on the ones that they hear see
205
00:08:26,559 --> 00:08:35,169
smell and so on and then run a diagnosis
206
00:08:30,490 --> 00:08:37,389
to determine the most likely cause now
207
00:08:35,169 --> 00:08:40,838
as far as preparing the knowledgebase
208
00:08:37,389 --> 00:08:42,909
goes the SME has completed their job so
209
00:08:40,839 --> 00:08:45,220
the next step in this phase as I had
210
00:08:42,909 --> 00:08:47,829
shown earlier in that diagram was that
211
00:08:45,220 --> 00:08:50,439
we need to now train the knowledgebase
212
00:08:47,829 --> 00:08:52,660
and since we're already in the diagnosis
213
00:08:50,439 --> 00:08:55,120
model manager we'll go ahead and look at
214
00:08:52,660 --> 00:08:56,620
how we can train the diagnosis model
215
00:08:55,120 --> 00:08:58,360
then after that
216
00:08:56,620 --> 00:09:00,580
we'll go into Watson discovery to see
217
00:08:58,360 --> 00:09:04,000
how we would do the same with the
218
00:09:00,580 --> 00:09:06,670
documents we uploaded earlier so if the
219
00:09:04,000 --> 00:09:08,380
SME were to open up the data tab they
220
00:09:06,670 --> 00:09:11,140
would see three different methods for
221
00:09:08,380 --> 00:09:15,390
training the diagnosis model the first
222
00:09:11,140 --> 00:09:19,300
one is to map historical work orders to
223
00:09:15,390 --> 00:09:22,870
instances where symptoms occurred and
224
00:09:19,300 --> 00:09:24,760
causes were found another way in which
225
00:09:22,870 --> 00:09:28,240
they could train this is through
226
00:09:24,760 --> 00:09:30,490
feedback and right now we don't have any
227
00:09:28,240 --> 00:09:31,900
feedback in the system but when we look
228
00:09:30,490 --> 00:09:33,970
at this from the field technician
229
00:09:31,900 --> 00:09:37,720
perspective we'll see how the field
230
00:09:33,970 --> 00:09:40,650
technician uses EMA's interface to send
231
00:09:37,720 --> 00:09:45,779
feedback through to the diagnosis model
232
00:09:40,650 --> 00:09:48,279
now the third way of training a
233
00:09:45,779 --> 00:09:50,860
diagnosis model is to create predefined
234
00:09:48,279 --> 00:09:53,080
examples so this is arguably the
235
00:09:50,860 --> 00:09:55,570
quickest way in which we can train the
236
00:09:53,080 --> 00:09:59,110
diagnosis model to understand real-life
237
00:09:55,570 --> 00:10:02,800
situations where symptoms were found and
238
00:09:59,110 --> 00:10:05,230
causes occurred and so what happens is
239
00:10:02,800 --> 00:10:07,660
as you start to use these various
240
00:10:05,230 --> 00:10:10,480
methods of training the diagnosis model
241
00:10:07,660 --> 00:10:12,870
the AI will start to dwarf the
242
00:10:10,480 --> 00:10:16,240
probabilistic nature of the solution and
243
00:10:12,870 --> 00:10:19,209
then override it with more realistic
244
00:10:16,240 --> 00:10:22,630
examples of again where symptoms
245
00:10:19,209 --> 00:10:24,219
occurred and causes were found so this
246
00:10:22,630 --> 00:10:26,950
includes the training portion for the
247
00:10:24,220 --> 00:10:30,100
diagnosis model manager so to get into
248
00:10:26,950 --> 00:10:32,020
the process of training the documents
249
00:10:30,100 --> 00:10:35,709
that we uploaded earlier we'll have to
250
00:10:32,020 --> 00:10:37,810
go into Watson discovery now the first
251
00:10:35,709 --> 00:10:40,060
thing I want to show you is how we
252
00:10:37,810 --> 00:10:43,209
improve the search speed and efficiency
253
00:10:40,060 --> 00:10:46,029
of Watson discovery by using what is
254
00:10:43,209 --> 00:10:48,689
known as smart document understanding so
255
00:10:46,029 --> 00:10:51,580
to do that I will use a sample
256
00:10:48,690 --> 00:10:53,320
collection called SDU test to
257
00:10:51,580 --> 00:10:57,750
demonstrate this as I've already done
258
00:10:53,320 --> 00:10:57,750
some training to the my card collection
259
00:11:05,830 --> 00:11:12,120
so once we open this up we will see
260
00:11:12,660 --> 00:11:18,610
details around this collection we will
261
00:11:16,570 --> 00:11:23,230
see some of the enrichments that were
262
00:11:18,610 --> 00:11:25,810
added in the enrichments are is
263
00:11:23,230 --> 00:11:28,600
basically data that has been added to
264
00:11:25,810 --> 00:11:32,018
the existing content to help with the
265
00:11:28,600 --> 00:11:34,510
search process so we see some
266
00:11:32,019 --> 00:11:38,170
enrichments that have been added already
267
00:11:34,510 --> 00:11:40,540
as far as entities go concepts sentiment
268
00:11:38,170 --> 00:11:42,459
analysis and categories so this was done
269
00:11:40,540 --> 00:11:45,219
automatically but what we can do through
270
00:11:42,459 --> 00:11:48,790
smart document understanding is we can
271
00:11:45,220 --> 00:11:57,430
configure this data to and to include
272
00:11:48,790 --> 00:12:01,390
new and better enrichments so as you can
273
00:11:57,430 --> 00:12:04,599
see through Watson discovery we get a
274
00:12:01,390 --> 00:12:08,199
view of the document and what Watson has
275
00:12:04,600 --> 00:12:10,750
come back as or come back with as
276
00:12:08,200 --> 00:12:12,760
identified fields so we see initially
277
00:12:10,750 --> 00:12:15,700
that everything is labeled as it is
278
00:12:12,760 --> 00:12:20,470
what's known as a text entity and so the
279
00:12:15,700 --> 00:12:25,810
text entity alone is not very efficient
280
00:12:20,470 --> 00:12:27,940
in terms of search speed in accuracy so
281
00:12:25,810 --> 00:12:30,899
one thing that the SME can do to improve
282
00:12:27,940 --> 00:12:34,510
this is to use field labels to identify
283
00:12:30,899 --> 00:12:37,029
different objects within their document
284
00:12:34,510 --> 00:12:41,680
so if they drag over things such as the
285
00:12:37,029 --> 00:12:46,029
title and drag over things such as these
286
00:12:41,680 --> 00:12:48,329
subtitles they can begin to teach Watson
287
00:12:46,029 --> 00:12:51,220
to identify these objects on their own
288
00:12:48,329 --> 00:12:53,979
so as they upload new documents to a
289
00:12:51,220 --> 00:12:57,640
collection these documents will be
290
00:12:53,980 --> 00:12:59,380
annotated automatically and the subject
291
00:12:57,640 --> 00:13:02,010
matter expert will no longer need to
292
00:12:59,380 --> 00:13:04,570
provide these sort of annotations and
293
00:13:02,010 --> 00:13:06,430
here in a second we'll talk about why
294
00:13:04,570 --> 00:13:09,010
it's important to do these annotations
295
00:13:06,430 --> 00:13:11,140
but I want to continue annotating for
296
00:13:09,010 --> 00:13:13,630
right now because I want to show you the
297
00:13:11,140 --> 00:13:17,410
real-life machine war machine learning
298
00:13:13,630 --> 00:13:19,060
happening right here in front of us so
299
00:13:17,410 --> 00:13:20,160
I'll go ahead and finish up this page
300
00:13:19,060 --> 00:13:24,969
here
301
00:13:20,160 --> 00:13:27,310
with some annotations and then as you
302
00:13:24,970 --> 00:13:29,260
see here Watson has started to
303
00:13:27,310 --> 00:13:32,050
incorporate machine learning and
304
00:13:29,260 --> 00:13:34,390
understand these objects within within
305
00:13:32,050 --> 00:13:37,930
the documents without me even having to
306
00:13:34,390 --> 00:13:39,280
tell it to do so and to continue
307
00:13:37,930 --> 00:13:40,750
training it a little further we'll go
308
00:13:39,280 --> 00:13:44,770
ahead and highlight this footer here
309
00:13:40,750 --> 00:13:47,650
submit this page and again we can see
310
00:13:44,770 --> 00:13:51,670
that Watson has now picked up the footer
311
00:13:47,650 --> 00:13:53,470
of this document and has correctly
312
00:13:51,670 --> 00:13:56,500
identified that there are no subtitles
313
00:13:53,470 --> 00:13:58,570
found within here now the reason it's
314
00:13:56,500 --> 00:14:00,460
important to do this smart document
315
00:13:58,570 --> 00:14:03,400
understanding is that we can then manage
316
00:14:00,460 --> 00:14:05,860
these fields afterwards so for the
317
00:14:03,400 --> 00:14:09,579
subject matter expert and their
318
00:14:05,860 --> 00:14:11,830
documents they know that there is not no
319
00:14:09,580 --> 00:14:13,750
relevant information or no information
320
00:14:11,830 --> 00:14:16,750
of value to the technicians and things
321
00:14:13,750 --> 00:14:20,770
like the answer the author the footer
322
00:14:16,750 --> 00:14:23,050
the header may be the question and table
323
00:14:20,770 --> 00:14:26,560
of contents for instance so now we're
324
00:14:23,050 --> 00:14:28,449
filtering out what gets returned to the
325
00:14:26,560 --> 00:14:29,680
field technician so if it's not a value
326
00:14:28,450 --> 00:14:34,210
we don't want to give it to the
327
00:14:29,680 --> 00:14:35,920
technician also what we can do is split
328
00:14:34,210 --> 00:14:38,920
up the document based on one of those
329
00:14:35,920 --> 00:14:41,589
fields so if I choose subtitle for
330
00:14:38,920 --> 00:14:43,750
instance this original document will get
331
00:14:41,590 --> 00:14:46,270
split into as many new documents as
332
00:14:43,750 --> 00:14:47,860
there are subtitles with the subtitle
333
00:14:46,270 --> 00:14:50,829
being the subject line of each new
334
00:14:47,860 --> 00:14:53,110
document and what this does is this
335
00:14:50,830 --> 00:14:56,440
provides granularity in the search
336
00:14:53,110 --> 00:14:58,810
process so we have new documents that
337
00:14:56,440 --> 00:15:01,090
Watson can then identify very quickly in
338
00:14:58,810 --> 00:15:03,430
return answers in what I call answer
339
00:15:01,090 --> 00:15:07,330
units to the field technician in a
340
00:15:03,430 --> 00:15:09,459
shorter period of time and to extend the
341
00:15:07,330 --> 00:15:13,530
enrichments even further we can add in
342
00:15:09,460 --> 00:15:15,880
the entities such as the subtitle and
343
00:15:13,530 --> 00:15:21,480
choose which enrichments we want to
344
00:15:15,880 --> 00:15:21,480
include on each of those fields
345
00:15:23,540 --> 00:15:29,870
now this process of the smart document
346
00:15:26,570 --> 00:15:31,820
understanding is very useful in terms of
347
00:15:29,870 --> 00:15:33,980
search speed and efficiency but it
348
00:15:31,820 --> 00:15:37,430
doesn't necessarily address the concern
349
00:15:33,980 --> 00:15:39,590
of accuracy and to do that we need to
350
00:15:37,430 --> 00:15:42,530
look at how we perform relevancy
351
00:15:39,590 --> 00:15:44,720
training on a collection so we'll first
352
00:15:42,530 --> 00:15:45,740
open up the my car collection and
353
00:15:44,720 --> 00:15:48,800
there's one thing that I want to point
354
00:15:45,740 --> 00:15:51,530
out so I already did training on this
355
00:15:48,800 --> 00:15:54,020
collection and what we saw earlier was
356
00:15:51,530 --> 00:15:55,730
that I uploaded six documents but
357
00:15:54,020 --> 00:15:57,760
because I've annotated these they have
358
00:15:55,730 --> 00:16:00,530
now been broken into thirty four
359
00:15:57,760 --> 00:16:04,910
distinct documents that Watson can then
360
00:16:00,530 --> 00:16:08,480
take advantage of to search with with
361
00:16:04,910 --> 00:16:09,949
better granularity but to do the
362
00:16:08,480 --> 00:16:13,610
relevancy training we would click on the
363
00:16:09,950 --> 00:16:15,230
performance tab here select view all and
364
00:16:13,610 --> 00:16:16,640
perform relevancy training and then
365
00:16:15,230 --> 00:16:19,730
choose the collection that we want to
366
00:16:16,640 --> 00:16:23,210
train and I talked about this briefly
367
00:16:19,730 --> 00:16:26,000
earlier the process of training a
368
00:16:23,210 --> 00:16:28,100
collection on relevancy is to ask it
369
00:16:26,000 --> 00:16:31,550
questions that a common field technician
370
00:16:28,100 --> 00:16:32,870
would ask so for instance we can look at
371
00:16:31,550 --> 00:16:35,569
one of these questions I asked earlier
372
00:16:32,870 --> 00:16:38,750
where I asked what tools do I need to
373
00:16:35,570 --> 00:16:40,760
remove a battery and what the SME can
374
00:16:38,750 --> 00:16:43,730
then do is validate these results by
375
00:16:40,760 --> 00:16:45,470
clicking rate results to then see the
376
00:16:43,730 --> 00:16:48,380
smaller documents we talked about
377
00:16:45,470 --> 00:16:51,980
earlier and rate them as relevant or not
378
00:16:48,380 --> 00:16:55,880
relevant and what that does is it
379
00:16:51,980 --> 00:16:57,740
teaches Watson what the question is how
380
00:16:55,880 --> 00:17:00,290
it's supposed to be responded to and
381
00:16:57,740 --> 00:17:02,540
then how to prioritize these documents
382
00:17:00,290 --> 00:17:05,420
and we'll look at this again in the end
383
00:17:02,540 --> 00:17:07,359
user interface to see what it looks like
384
00:17:05,420 --> 00:17:09,860
from the field technician perspective
385
00:17:07,359 --> 00:17:12,050
and we'll do that here shortly because
386
00:17:09,859 --> 00:17:17,000
we are now finished with the process of
387
00:17:12,050 --> 00:17:19,310
training the knowledgebase so to look at
388
00:17:17,000 --> 00:17:23,359
this from the field technician
389
00:17:19,310 --> 00:17:29,149
perspective we will jump to the sample
390
00:17:23,359 --> 00:17:31,790
application now for this let's imagine
391
00:17:29,150 --> 00:17:34,430
that an automotive technician just had a
392
00:17:31,790 --> 00:17:36,110
car towed to the shop because the
393
00:17:34,430 --> 00:17:37,420
customer could not get it to start on
394
00:17:36,110 --> 00:17:40,250
their way to work that
395
00:17:37,420 --> 00:17:42,830
this particular customer is of value to
396
00:17:40,250 --> 00:17:44,210
the shop and has requested to pick it up
397
00:17:42,830 --> 00:17:47,510
on their way back from the office that
398
00:17:44,210 --> 00:17:49,730
same day that being said the technician
399
00:17:47,510 --> 00:17:52,280
now has to repair the vehicle in a
400
00:17:49,730 --> 00:17:54,260
timely manner analysts also get it done
401
00:17:52,280 --> 00:17:57,230
correctly as to avoid having it back in
402
00:17:54,260 --> 00:17:58,640
the shop later that week so the first
403
00:17:57,230 --> 00:17:59,810
thing that the technician is going to
404
00:17:58,640 --> 00:18:01,750
want to do is they're going to want to
405
00:17:59,810 --> 00:18:04,940
determine the root cause of this failure
406
00:18:01,750 --> 00:18:07,250
so here they can select the model that
407
00:18:04,940 --> 00:18:10,640
we looked at earlier related to the car
408
00:18:07,250 --> 00:18:13,340
and then what the technician can then do
409
00:18:10,640 --> 00:18:16,850
is assess what are the things that I
410
00:18:13,340 --> 00:18:19,760
hear smell see hopefully not taste and
411
00:18:16,850 --> 00:18:22,399
so on and what they see is that the
412
00:18:19,760 --> 00:18:25,460
lights are turning on but there is no
413
00:18:22,400 --> 00:18:27,530
sound coming from the car they could
414
00:18:25,460 --> 00:18:28,970
then submit these symptoms and the
415
00:18:27,530 --> 00:18:31,040
diagnosis model will run in the
416
00:18:28,970 --> 00:18:33,920
background to determine the most likely
417
00:18:31,040 --> 00:18:36,889
cause and if you click on the most
418
00:18:33,920 --> 00:18:38,870
likely cause the technician can then use
419
00:18:36,890 --> 00:18:42,770
the support documentation we uploaded
420
00:18:38,870 --> 00:18:47,000
earlier to find the best recommendation
421
00:18:42,770 --> 00:18:48,710
for making this repair and because we're
422
00:18:47,000 --> 00:18:51,530
working in a shared environment I have
423
00:18:48,710 --> 00:18:54,500
to filter by the asset that I want to
424
00:18:51,530 --> 00:18:56,000
focus on but what comes back to me is
425
00:18:54,500 --> 00:18:58,130
one of those answer units I had
426
00:18:56,000 --> 00:19:01,070
mentioned earlier which is a snippet of
427
00:18:58,130 --> 00:19:03,560
the larger documents we uploaded in the
428
00:19:01,070 --> 00:19:05,899
first step and I also get a confidence
429
00:19:03,560 --> 00:19:08,360
level that's relatively low because of
430
00:19:05,900 --> 00:19:10,820
the amount of training that's been
431
00:19:08,360 --> 00:19:13,699
incorporated into this but another way
432
00:19:10,820 --> 00:19:15,139
the technician can add feedback is
433
00:19:13,700 --> 00:19:18,170
through the simple thumbs-up or
434
00:19:15,140 --> 00:19:20,630
thumbs-down and that's one way to train
435
00:19:18,170 --> 00:19:22,550
it before the technician moves on with
436
00:19:20,630 --> 00:19:25,340
actually making the repair it is
437
00:19:22,550 --> 00:19:30,889
suggested to also provide feedback in a
438
00:19:25,340 --> 00:19:33,169
more detailed format and this is very
439
00:19:30,890 --> 00:19:34,850
important because this is something that
440
00:19:33,170 --> 00:19:36,650
we can incorporate into the
441
00:19:34,850 --> 00:19:41,110
knowledgebase to let future technicians
442
00:19:36,650 --> 00:19:41,110
use if they encounter similar problems
443
00:19:42,130 --> 00:19:48,380
now we're gonna go look at this from the
444
00:19:46,040 --> 00:19:50,180
perspective of a field technician who
445
00:19:48,380 --> 00:19:50,640
now has the context they need from a
446
00:19:50,180 --> 00:19:52,560
diagnose
447
00:19:50,640 --> 00:19:55,530
is to then ask some questions along the
448
00:19:52,560 --> 00:19:58,740
way so from the query a sample
449
00:19:55,530 --> 00:20:00,780
application within EMA our technician
450
00:19:58,740 --> 00:20:04,080
might find himself stuck on some parts
451
00:20:00,780 --> 00:20:06,210
or have concerns about certain steps
452
00:20:04,080 --> 00:20:09,060
within the process of repairing a car
453
00:20:06,210 --> 00:20:13,680
that won't start so for instance we can
454
00:20:09,060 --> 00:20:20,940
say or ask the question of how do I
455
00:20:13,680 --> 00:20:26,490
safely disconnect the jumper cables hit
456
00:20:20,940 --> 00:20:29,370
entered and EMA will use that that
457
00:20:26,490 --> 00:20:31,620
context to then go out in search for the
458
00:20:29,370 --> 00:20:34,229
best recommendation so if we again
459
00:20:31,620 --> 00:20:37,290
filter by the collection of interest by
460
00:20:34,230 --> 00:20:39,480
the asset of interest we can see various
461
00:20:37,290 --> 00:20:42,030
recommendations based off of our
462
00:20:39,480 --> 00:20:44,940
training based off of the SMEs training
463
00:20:42,030 --> 00:20:47,520
and the technician can then use these
464
00:20:44,940 --> 00:20:50,910
answer units to help troubleshoot the
465
00:20:47,520 --> 00:20:53,310
car so what we're doing overall with the
466
00:20:50,910 --> 00:20:56,220
EMA solution is we're streamlining the
467
00:20:53,310 --> 00:20:58,679
process that the technician goes through
468
00:20:56,220 --> 00:21:02,190
in searching for information that they
469
00:20:58,680 --> 00:21:07,080
can use to make a repair and in doing so
470
00:21:02,190 --> 00:21:13,380
we're reducing the mean time to repair
471
00:21:07,080 --> 00:21:15,210
or rather reducing the sorry yeah
472
00:21:13,380 --> 00:21:16,740
reducing the mean time to repair an
473
00:21:15,210 --> 00:21:20,840
increase in the first time fix rate on
474
00:21:16,740 --> 00:21:24,180
this alone so I want to come back to the
475
00:21:20,840 --> 00:21:27,030
graph we looked at earlier because we
476
00:21:24,180 --> 00:21:30,060
see that machine complexity has remained
477
00:21:27,030 --> 00:21:34,830
the same and it likely always will if
478
00:21:30,060 --> 00:21:37,620
not increased but what has changed was
479
00:21:34,830 --> 00:21:39,899
the equipment expertise so we've closed
480
00:21:37,620 --> 00:21:43,770
the gap between expertise and the
481
00:21:39,900 --> 00:21:45,180
complexity of machinery and as you saw
482
00:21:43,770 --> 00:21:47,520
in this demonstration the automotive
483
00:21:45,180 --> 00:21:50,520
technician was given the required
484
00:21:47,520 --> 00:21:52,350
knowledge to make the right fix the
485
00:21:50,520 --> 00:21:56,220
first time in a shorter period of time
486
00:21:52,350 --> 00:21:58,409
so this means that the potential for a
487
00:21:56,220 --> 00:22:00,180
decrease in the operating cost and an
488
00:21:58,410 --> 00:22:02,370
increase in earnings has just been made
489
00:22:00,180 --> 00:22:04,970
available simply by focusing on
490
00:22:02,370 --> 00:22:07,469
improving the productivity of fields
491
00:22:04,970 --> 00:22:09,120
if you think this solution is something
492
00:22:07,470 --> 00:22:11,669
of interest to you I recommend reaching
493
00:22:09,120 --> 00:22:13,770
out to either myself or Hina purohit who
494
00:22:11,669 --> 00:22:16,049
is our offering manager to explore the
495
00:22:13,770 --> 00:22:17,580
solution further for client references
496
00:22:16,049 --> 00:22:20,100
please see the links in the comments
497
00:22:17,580 --> 00:22:22,939
below thank you for watching this video
498
00:22:20,100 --> 00:22:22,939
I hope you enjoyed it
499
00:22:23,240 --> 00:22:33,819
[Music]
38044
Can't find what you're looking for?
Get subtitles in any language from opensubtitles.com, and translate them here.