AWS re:Invent 2022 - Improve uptime with predictive maintenance for industrial equipment (AIM211)

Aug 16, 2023

AWS re:Invent 2022 - Improve uptime with predictive maintenance for industrial equipment (AIM211)

Predictive maintenance is an effective way to avoid industrial machinery failures and expensive downtime by proactively monitoring the condition of your equipment so that you can be alerted to any anomalies before equipment failures occur. However, this technology has historically been difficult to implement. In this session, learn how industrial customers like Baxter International are activating predictive maintenance and avoiding downtime with AI services from AWS such as Amazon Monitron, with no machine learning experience required.

Learn more about AWS re:Invent at https://go.aws/3ikK4dD.

Subscribe:
More AWS videos http://bit.ly/2O3zS75
More AWS events videos http://bit.ly/316g9t4

ABOUT AWS
Amazon Web Services (AWS) hosts events, both online and in-person, bringing the cloud computing community together to connect, collaborate, and learn from AWS experts.

AWS is the world’s most comprehensive and broadly adopted cloud platform, offering over 200 fully featured services from data centers globally. Millions of customers—including the fastest-growing startups, largest enterprises, and leading government agencies—are using AWS to lower costs, become more agile, and innovate faster.

#reInvent2022 #AWSreInvent2022 #AWSEvents

Content

0.339 -> (upbeat music)

4.35 -> - Good afternoon.

7.389 -> I'm so excited to be here.

10.11 -> Since our launch of our industrial AI services,

12.66 -> we've heard from hundreds of industrial

14.85 -> and manufacturing customers just like yourselves.

17.76 -> I'm so excited to present our latest evolution

20.995 -> on our industrial AI services

23.34 -> to help improve your digital transformation journey.

27.24 -> Today, we'll be hearing from guest speakers

28.98 -> such as Shalika Pargal,

30.06 -> who's the senior product manager

31.738 -> of our industrial AI services,

33.36 -> as well as guest speakers from Baxter International,

35.88 -> A.K. Karan, senior director of Baxter International,

38.61 -> as well as Adam Aldridge.

41.13 -> Since I started in this industry

42.78 -> about two and a half decades ago.

44.85 -> Hmm, that seems a little long, huh?

48 -> You know, we used to monitor only

50.13 -> the most critical assets and the most critical applications.

54.15 -> Today you were able to apply industrial AI

56.7 -> to hundreds or thousands of pieces of equipment

58.83 -> that you could never even dream of monitoring In the past.

62.88 -> we'll be able to show you, you know,

64.29 -> how customers today that previously couldn't monitor

67.17 -> their assets and apply additional industrial AI services,

71.34 -> whether it's computer vision,

72.96 -> time series data using machine learning

74.85 -> and anomaly detection across their plants,

77.52 -> production lines, as well as their assets.

84.39 -> We have a packed agenda day today.

86.34 -> And so what we'll do is give an overview

88.23 -> of our industrial AI services,

90.54 -> the evolution based on your feedback of our specific

93.57 -> service, Monitron,

94.62 -> which is an end to end vibration, temperature sensing

97.11 -> to machine learning application.

98.79 -> And the bulk of the time we'll spend going through

101.73 -> a real life case study

103.11 -> of how industrial AI can impact your operation,

106.32 -> as well as how you can scale

107.64 -> it across niche applications

110.31 -> to across your overall production facility.

118.353 -> So just show of hands,

120.39 -> who here has used the AWS app for a re event?

125.1 -> Yeah, yeah.

127.71 -> Anybody use the queue lines?

130.59 -> Anybody use that to decide whether or not, no?

132.946 -> You just drew a drew a straw to see if the line was long.

136.41 -> You just, you know, figured out

137.64 -> whether or not to get a badge.

139.38 -> Well, those are the types of insights

141.87 -> that we wanna bring to you, right?

143.49 -> So on apps that you're using today,

145.65 -> be able to get real time insights in terms of queue lines,

149.7 -> in terms of how your pieces

151.05 -> of equipment are performing, right?

153 -> To people who can make the decisions, right?

155.16 -> So if you're waiting in line and you look at, hey, look,

158.37 -> if I go next door, that line

159.93 -> is half the time that I could

162.206 -> be standing there to get my badge, why not do that?

165.42 -> Right, so same thing with our industrial AI services.

168.15 -> We wanna bring those insights to the operators,

171.09 -> the maintenance team,

172.14 -> quality managers who are making decisions

174.84 -> for your operational facility, right?

177.39 -> In the past it was really set to data scientists,

180.296 -> IT experts, to be able to manage your data

183.45 -> as well as the data science models.

185.55 -> No more we'd like to take that step away.

187.8 -> That those machine learning steps and the thought process

190.92 -> is now more around what type of data

193.23 -> do I need to drive my problems?

195.36 -> So some areas that we're very much focused

197.4 -> on are condition monitoring,

199.14 -> being able to predict the health of your specific asset,

202.32 -> as well as quality management,

204.09 -> leveraging computer vision at the edge

206.64 -> to be able to understand

208.89 -> the quality of your production lines.

215.7 -> So our focus in AWS, as I mentioned,

217.98 -> is that we're trying to simplify

220.23 -> the machine learning workflow.

221.52 -> In the past, you need to have a data scientist,

224.46 -> you know, generally somebody

225.51 -> of a PhD degree to be able to select

228.45 -> and process the data that you have.

230.19 -> Whether that's time series data,

232.05 -> computer vision data,

233.34 -> we know it comes in all sorts of forms.

235.62 -> And that pre-processing, from what we've heard from you,

238.864 -> can be between 40 to 60% of the time.

242.012 -> And across our industrial AI services,

244.8 -> what we do is we look for commonalities across this data,

247.92 -> commonalities across these challenges of the data

250.17 -> that you see, and to be able to automate away those.

253.47 -> So things like timestamp alignment, imputation,

258.15 -> traditional steps that you need

259.53 -> just to get started in machine learning,

261.99 -> as well as algorithms, right?

263.94 -> You have hundreds or thousands of different algorithms

266.37 -> that you can- to select from.

268.92 -> We pre-select these different algorithms so

271.02 -> that you don't need a PhD to determine what type

273.66 -> of algorithms to select,

275.22 -> and also the hyper parameters

276.87 -> to tune them to specific applications.

279.39 -> What we know is that every asset is different.

282.33 -> Every line that you operate is different.

284.695 -> You know, you might say, okay,

285.78 -> automotive and healthcare are very different applications,

289.32 -> but what we found is even in

290.61 -> the same manufacturing facility,

292.47 -> you might have different lines

294.21 -> that operate a little bit differently,

295.44 -> whether that's how you operate it,

297.66 -> however it's maintained, or the different products

299.85 -> that you send through those different lines.

302.67 -> Based on that,

303.57 -> we're able to select the algorithms that are pre-built based

306.3 -> on those operational data that you send in.

310.32 -> So our focus is to completely automate those steps.

312.63 -> So in the past it might have taken you months

314.61 -> to be able to fully develop

316.74 -> and manage those machine learning models.

318.78 -> Today you can get started in a matter of hours,

321.3 -> literally going to amazon.com.

323.16 -> You can offer, you can purchase our Monitron service,

325.8 -> glue it on anything in your system.

327.63 -> We've heard everything from water pumps, to HVAC pumps,

330.87 -> to anything in your operating facility

333.18 -> where you can automatically glue your sensor

334.89 -> and be able to monitor your assets on day one.

341.218 -> So specifically for this presentation,

343.68 -> we're focused on condition monitoring and specifically

346.41 -> around how condition monitoring

348.491 -> can be applied to predictive maintenance.

350.25 -> As I mentioned, across facilities we've visited,

352.98 -> we've seen the most advanced facilities, right?

355.26 -> So Amazon fulfillment centers where we have have robotics,

358.38 -> where we have AGVs, where we have

360.39 -> the most advanced automation systems

362.894 -> that we've seen, to facilities where

365.73 -> they built in the eighties, not the 1980s,

368.46 -> but the 1880s where monitoring was just a dream, right?

372.18 -> And we've seen that whole gamut between

373.92 -> the most advanced to these 1880 facilities,

377.16 -> where some pieces of equipment are monitored,

379.59 -> others partially monitored, and in other cases,

382.68 -> there's fully developed machine learning models.

385.2 -> And so because of that we developed

387.75 -> an array of different services

389.37 -> and solutions based on those specific assets

391.77 -> that you're monitoring.

393.03 -> For assets that don't have any monitoring like vibration,

396.27 -> temperature sensing today or continuously monitored today,

399.66 -> we have a solution called Amazon Monitron

401.97 -> which we'll be discussing later,

403.17 -> which is an end to end offering

404.91 -> from hardware all the way to the software

407.55 -> and application solution.

409.53 -> If you already have condition monitoring

411.45 -> where you have vibration, pressure,

413.19 -> temperature already monitored today,

415.53 -> you can begin to build machine learning models using

418.95 -> our service called Amazon Lookout for Equipment.

421.26 -> As I mentioned, you bring in the data

423.717 -> and the system automatically decides what type

426.03 -> of algorithms to be able to use

427.74 -> for that specific machine learning model.

430.32 -> And lastly, for the smaller subset of pieces

432.99 -> of equipment where you've gone down the journey,

434.67 -> you've built your machine learning model,

436.56 -> we have Amazon SageMaker where you're able to optimize

440.07 -> that machine learning workflow or your ML ops

442.59 -> to be able to optimize how you deploy those specific models.

450.06 -> So what I'll leave you

450.893 -> with before I hand it over to my colleague Shalika,

453.3 -> is that since the launch of our industrial AI services,

456.54 -> we've worked with hundreds of customers

458.76 -> and our focus is not only the machine learning is,

461.88 -> but also how do we integrate machine learning

464.19 -> into your operations.

465.72 -> For Koch AG and Energy Solutions,

467.4 -> they've been monitoring their system through

469.02 -> a remote monitoring diagnostic center for years now.

471.96 -> They've been using a variety

473.22 -> of different machine learning methods.

475.14 -> However, they've been applying both Monitron

477.18 -> and Lookout for Equipment to one,

479.01 -> start to monitor pieces of equipment

482.16 -> that they haven't been able to continuously monitor,

484.68 -> as well as billing- being able

486.3 -> to build machine learning models

487.86 -> on pieces of equipment they haven't been able to monitor.

490.47 -> And based on this new data

492.24 -> and based on this new machine learning,

493.74 -> they're able to detect issues hours,

495.72 -> if not weeks in advance of their previous method

498.81 -> of detecting issues.

500.94 -> In addition, one of- one other customer

502.92 -> that we're working with,

505.002 -> they've embedded our Lookout for Vision service

506.97 -> into their quality management system.

509.79 -> That's where we see the most impact for our machine learning

512.61 -> is where we're able to fully embed machine learning

515.13 -> into your operation.

516.32 -> In this case, it was fully adopted

518.25 -> into their quality management system,

519.75 -> saving them tens of thousands of hours of manual labor

523.29 -> and instead that that data could be used to optimize

526.32 -> their facility rather than monitoring

528.09 -> that specific assembly line.

530.34 -> And lastly, what I'll leave you with

531.78 -> is that one of our largest customers

534.3 -> and one of the largest asks is that,

536.4 -> hey look, we've seen that machine learning

538.41 -> can be used by the few,

539.88 -> and the few being data scientists and IT personnel,

542.88 -> but how can I transform my organization

545.25 -> where data is at the forefront?

546.87 -> And the way that we see that

548.34 -> is by automating the machine learning

549.99 -> and bringing that closer to operators

552.93 -> and maintenance staff.

553.77 -> So that's the feedback from our customer,

555.72 -> GS EPS, is that they're able to transform

557.786 -> their organization and be able to use machine learning

560.76 -> across their organization.

562.53 -> So without further ado, I'd like to introduce Shalika again,

565.26 -> the senior product manager for industrial AI service.

568.02 -> - Thanks, Alicia.

569.4 -> I'm now going to go to the details of the buy

571.83 -> or the out the box solution that Alicia mentioned.

574.68 -> Before I get started,

575.73 -> I just wanted to understand,

577.08 -> any Monitron users in the crowd?

578.73 -> Can I see a show of hands?

581.52 -> You see a few. (laughs)

583.74 -> All right.

584.573 -> So for others,

586.526 -> Monitron is an end-to-end system

588.75 -> that monitors industrial assets

591.06 -> and detects abnormal conditions

593.04 -> in these assets using machine learning.

595.65 -> In doing so, it enables predictive maintenance

598.35 -> in an easy and cost effective way,

600.54 -> thus helping reduce unplanned downtime.

603.81 -> You can get started

604.83 -> in monitoring your rotating equipments like pumps, motors,

607.92 -> gear boxes, and bearings using Amazon Monitron

611.97 -> within a matter of minutes.

614.85 -> Now let's look at how Monitron works.

618.173 -> Monitron has a purpose built wireless sensors

621.99 -> that capture the temperature and vibration patterns

624.39 -> of the asset.

625.95 -> Temperature and vibration

627.03 -> are good indicators of the health of the asset.

629.88 -> These sensors can be enabled using near field communication

633.12 -> from the Monitron mobile app.

635.82 -> These sensors have been optimized

637.59 -> and built to optimize for both cost and simplicity.

642.18 -> Next is the fully managed Monitron gateway.

645.66 -> This gateway securely and automatically transfers

648.45 -> the sensor data to the cloud.

651.03 -> In the cloud we have the Monitron service

653.46 -> that analyzes the sensor data using ISO vibration standards

657.96 -> and machine learning algorithms.

660.48 -> And finally we have the Monitron app.

663.06 -> We have a mobile app and a web app

665.76 -> which are used to set up the hardware,

667.53 -> which is the sensor and gateway,

669.42 -> and to receive notifications

671.37 -> in case of abnormal conditions on the industrial assets.

675.69 -> Monitron is quick to install.

677.73 -> Within a matter of minutes,

679.26 -> a reliability engineer can set up

680.94 -> the sensor and gateway and start monitoring their assets.

685.206 -> The service uses machine learning

687.78 -> to analyze the sensor data.

689.64 -> So no prior machine learning knowledge

691.95 -> or expertise is required to use Amazon Monitron.

695.85 -> Since the launch of Monitron,

697.2 -> our focus has been to try and add additional value

699.87 -> for our customers around three core themes.

703.2 -> The devices, notifications,

705.63 -> and making Monitron data available to customers

708.6 -> in a form they can add more value to.

713.07 -> Starting with the devices theme,

715.35 -> when we launched Monitron,

716.58 -> our devices were certified for the US, Canada, EU, and UK.

721.89 -> However, over time we have seen

723.63 -> that our customers who operate globally

726.03 -> want to install or deploy Monitron in their overseas plants

729.48 -> in Asia Pacific.

731.28 -> This has now led us to certify Monitron in Australia,

734.37 -> New Zealand, Japan, Turkey, and Singapore.

738.63 -> To reduce the operational overhead for customers in

741.42 -> replacing sensor battery,

743.25 -> And to reduce the overall cost of ownership,

747.06 -> We have increased the Monitron sensor battery life

749.55 -> from three to five years.

751.83 -> This update was done via the firmware,

753.96 -> so no action is required from the customer.

757.74 -> The updated firmware has been deployed on all the sensors in

761.1 -> the field using Monitrons over the air capability.

765.3 -> So now if a customer has bought

767.22 -> the sensors and installed it,

768.63 -> then they can avail this benefit

770.25 -> at no additional cost to them.

773.58 -> Some of our customers had network protocol requirements,

777.12 -> which required them to use wired gateways.

779.7 -> To address this need, we launched the ethernet gateway,

783.21 -> which facilitates the transfer of sensor data to the cloud

786.81 -> using the customer's ethernet network.

789.99 -> Now the ethernet gateway along with the WiFi gateway,

793.14 -> which was available at the time of launch,

795.54 -> provides customers with multiple options

797.97 -> for their Monitron connectivity needs.

802.32 -> Moving to our next theme,

803.79 -> which has been focused around enhancing our notifications

806.67 -> to provide additional value for our end user personal,

810.3 -> the reliability manager, and the shop floor technician.

812.994 -> Our focus has been on adding more to the notifications,

817.32 -> more commentary and details around severity,

820.05 -> which is an indicator of the health of

821.58 -> the asset type of notification,

824.73 -> which is determined if the abnormal condition is seen either

828.24 -> in the vibration data or in the temperature data.

831.63 -> The source of the alert,

833.1 -> which can be based on the ISO 20816 vibration standards,

837.81 -> or it could be due to ML enabled models.

842.13 -> ISO based notifications are a result of-

844.8 -> are determined by the ISO thresholds.

847.56 -> We have ISO thresholds,

849.42 -> which are associated with four different equipment classes.

852.93 -> These equipment classes are very broad,

854.76 -> so sometimes we have customers who run their equipments

857.91 -> in different operating envelopes.

859.86 -> This results in getting notifications

861.6 -> which might not apply to that particular asset.

864.84 -> To address this, we now allow customers

867.24 -> to mute their ISO based notifications if this-

870.18 -> if they want to do so.

873.93 -> Monitron now provides indications

875.61 -> when a sensor or a gateway goes offline.

878.13 -> This tells customers that their devices have gone offline

881.22 -> and not sending information to the cloud.

884.16 -> The customers can troubleshoot and bring

885.87 -> the devices back online and resume monitoring their assets.

892.2 -> Our last theme has been focused

893.91 -> on making Monitron data available

896.31 -> and accessible to customers within their own ecosystem.

900.48 -> Amazon Monitron can now stream live sensor data

903.96 -> to a Kinesis data stream.

906.24 -> So customers can use this data, analyze the data,

908.94 -> and create work orders in their asset management systems

912 -> like IBM Maximo, SAP PM or Infor and others.

916.89 -> So a technician can be sent to look at assets with issues.

921.36 -> We have seen many

922.193 -> of our customers build operational dashboards

924.54 -> to monitor the health of their assets.

927.24 -> Given the data in the Kinesis data stream,

929.55 -> these customers can now augment these dashboards

932.34 -> with the real time data

933.75 -> and insights from Monitron.

937.2 -> The Monitron data and inferences

939.09 -> can also be used and leverage along with telemetry data

942.9 -> and operational data to build advanced analytics

945.42 -> for other use cases.

948.57 -> Now changing gears,

949.77 -> we wanna show you Monitron in action

951.63 -> at Baxter's manufacturing facility.

960.99 -> - Of healthcare and life saving products.

963.57 -> Our vision is to save and sustain lives.

966.84 -> We have 70 plus manufacturing sites located globally

970.38 -> and we run 24/7, 365.

973.86 -> Every minute of production is very critical to us.

976.732 -> And every instance of downtime that we can avoid

980.023 -> is crucial and very valuable.

982.02 -> - Most of our systems were manual inspections,

985.8 -> which requires a lot of resources and labor.

988.26 -> In some cases we have to enter confined spaces,

991.05 -> so there's a safety aspect.

992.4 -> You have to shut down critical operations

994.53 -> to do the maintenance activities.

996.09 -> - Equipment reliability is very, very critical to us.

999.45 -> We have over thousands of manufacturing assets

1001.82 -> that we need to monitor.

1003.02 -> Monitoring these assets using time-based methods

1005.51 -> or periodic alerting wasn't good enough.

1008.51 -> On the other hand,

1009.83 -> Amazon Monitron gives us machine learning

1012.14 -> and AI capabilities that helps us

1014.36 -> to generate custom vibration

1015.977 -> and temperature profile for every asset.

1018.56 -> And this was truly a game changer.

1022.67 -> - Without further ado,

1023.63 -> we'd like to welcome A.K. Karan and Adam Aldridge

1027.35 -> to walk us through Baxter's journey with Monitron.

1030.68 -> - Hello and good afternoon.

1032.874 -> (crowd applauds) - Hello.

1035.78 -> - Yeah, what a great video, right?

1037.22 -> Just because it is not, I'm in that video.

1039.38 -> But truly, It's a great video.

1042.14 -> So let's start with wha- who we are, right? Baxter.

1044.69 -> So Baxter is a global manufacturer of healthcare

1047.99 -> and lifesaving products.

1049.61 -> We have a pretty broad portfolio,

1051.11 -> around 50,000 SKUs.

1053 -> But the big thing is,

1053.833 -> I mean we impact the life of 350 million patients in a year,

1058.16 -> in a given year.

1059.36 -> That is the reach of our products.

1061.43 -> And from a manufacturing footprint,

1063.553 -> we have over 70 manufacturing sites.

1066.35 -> Look at it globally,

1067.61 -> and we have our own challenges in terms of

1069.92 -> how to keep our operations running 24/7, 365.

1075.05 -> So talk about why we start- embarked on this journey, right?

1079.97 -> So if you look at the, one of the words that was used,

1083.12 -> I would say in the last few years you heard resiliency.

1087.62 -> So when we look into operations, you knew the supply chain.

1089.75 -> That mean it's a close in to link of all

1091.22 -> those different groups functions.

1093.62 -> But we said from an operations standpoint,

1097.19 -> we want to be resilient.

1098.24 -> We don't want to be the weakest link.

1100.31 -> And especially we are facing labor shortages.

1102.89 -> How can we find, we have jobs but we can't find people,

1105.304 -> we have inflation, we have cost pressures.

1107.99 -> So how do we keep our operations running 24/7, 365?

1111.464 -> How do we build resiliency?

1113.9 -> And that was going to be the determining factor

1116.12 -> for us to start on with,

1117.98 -> I would say a predictive maintenance tool.

1121.82 -> So what were we doing before, right?

1124.31 -> I mean it was not like we didn't have any systems.

1126.41 -> We did have systems.

1127.37 -> I mean we used condition based monitoring.

1128.96 -> We did have our technicians who would go on our rounds.

1132.05 -> I mean every some,

1133.13 -> every day they just go all check all the assets.

1135.56 -> I mean they have thousands of assets.

1137.3 -> So going on a periodic frequency, checking the assets.

1140.39 -> But things might break down during those checks.

1143.93 -> So when all those (indistinct) systems fail,

1146.36 -> they can create a very serious disruption

1149.03 -> to our supply chain.

1150.5 -> So how do we kind of have a system

1152.42 -> that's going to monitor these assets,

1154.899 -> 24/7, 365 continuous monitoring.

1157.37 -> Just think like a Fitbit on you, right?

1159.38 -> Or on a person. It tells you all your vitals,

1162.44 -> making sure the equipment is operational,

1164.18 -> it's in a good healthy state of operation.

1167.93 -> And that's one of the reasons we started this journey.

1170.6 -> But then when this kind of goes down,

1172.82 -> I mean we are always waiting for this element of surprise.

1177.23 -> This goes down, so what do we do?

1178.88 -> Call technicians, dispatch people

1180.53 -> and bring emergency maintenance crew.

1182.66 -> It's- it creates havoc.

1184.19 -> So how do we kind of go from the existing tools

1187.07 -> to more of a preventive based approach?

1193.07 -> But before I go,

1193.903 -> and I would say why or what are we looking for

1197.39 -> in terms of the value proposition.

1199.885 -> First and foremost, right,

1201.23 -> we said speed is gonna be key.

1203.57 -> When you talk about speed, it's not about,

1205.19 -> we look for as a company, we look at weeks,

1211.04 -> to kind of get this thing, days and weeks,

1212.54 -> not months and years.

1214.25 -> And it has to be cost effective because how can we generate

1217.19 -> quick value, right?

1218.63 -> And also the the big thing is, I mean we have 70 sites.

1221.45 -> How do we scale this? So it's just,

1223.703 -> it's not like you have to go reinvent when you go from plant

1227.06 -> one to a plant two across global sites.

1230.21 -> So Monitron gave us all

1231.29 -> those features that we were looking for.

1236.84 -> Like some Shalika and Alicia mentioned,

1240.59 -> solutions end to end.

1242.12 -> Like you just slap the sensor on an asset,

1244.76 -> flip the button on and say stream the data.

1247.19 -> And that's as a consumer, that's what we did.

1249.95 -> So the system had data streaming services to the cloud,

1254.51 -> analytics built in,

1256.31 -> especially all this embedded machine learning

1258.5 -> that came out give us a custom vibration profile,

1260.72 -> a custom temperature profile for every single asset.

1263.57 -> Asset could be 10 years old, could be 20 years old,

1266.6 -> could be five years old, right?

1268.13 -> But every single asset had a unique signature

1271.85 -> that we can now use.

1272.81 -> And machine learning helped us to take it to the next step.

1275.99 -> So this was truly, I would say a game changer,

1278.39 -> because it helped us to generate value very, very rapidly.

1285.8 -> So I'm gonna pass it down to Adam

1287.21 -> he's gonna talk about operational challenges.

1290.03 -> - So, yep.

1291.74 -> Hello everyone. So Adam Aldridge,

1293.03 -> a reliability engineering manager at our flagship facility,

1296.48 -> helped lead and develop this program and deploy

1299.42 -> it throughout our facility.

1300.8 -> And I'm gonna be talking to you about

1302.21 -> our first case scenarios

1304.85 -> and everything that we ran into and

1306.05 -> the learning curve that we did over

1307.97 -> the one year journey

1308.81 -> that we've had since we started installing Monitron sensors.

1312.29 -> So some of the operational challenges that we were trying

1314.54 -> to be able to successfully mitigate with

1318.65 -> the way we were doing it compared

1320.03 -> to the way we're able to do it now with Monitron

1322.22 -> is realtime data collection.

1324.47 -> As you can see in the top picture in the photo,

1326.45 -> we've been doing manual data collection

1328.7 -> for vibration analysis, where a technician will go out

1331.37 -> and physically sample each site, each motor every month.

1335 -> We'll get one reading for a whole month's window,

1337.82 -> we'll send it to a third party for a data analysis

1340.85 -> and then we'll receive that information back.

1343.19 -> So that comes at a cost to the site

1345.47 -> for the third party analysis

1346.76 -> as well as a lot of labor for our technicians

1349.28 -> to collect this data.

1350.9 -> Now we're reading one reading per hour,

1353.84 -> and being able to have the machine learning

1355.82 -> to notify our technicians when

1357.29 -> there's a potential failure being presented

1360.29 -> and then we can go out and investigate.

1362.39 -> It also has allowed us to do rapid decision making

1364.7 -> and readily available data to make decisions.

1367.43 -> So we have the trend lines from the Monitron app

1370.49 -> as you can see in figure one and figure two,

1372.44 -> we're able to monitor those and determine

1374.63 -> is this issue developing something that we need

1377.12 -> to go in and investigate or do we want

1378.59 -> to continue watching it to see how it develops over time?

1381.915 -> We've had issues where it's took almost three to four months

1386.54 -> for it to double in value for its RMS velocity.

1390.5 -> So it's not something that

1391.37 -> would immediately recognize we need to go out

1393.83 -> and respond immediately, but after you look

1395.36 -> at that data over a three to four month period,

1397.82 -> there's absolutely a trend developing that we need

1400.28 -> to investigate that potential cause of failure.

1403.49 -> It's increased productivity.

1405.11 -> So we have a fully deployed maintenance

1406.97 -> and predictive maintenance program at our facility,

1409.37 -> but you can imagine how much time it takes in

1412.04 -> a 1.4 million square foot facility

1414.295 -> to do those manual data collections.

1416.898 -> Now we're able to use the app, monitor the app,

1420.53 -> and we're able

1421.363 -> to use our technicians time pursuing

1422.63 -> other predictive maintenance technologies

1424.97 -> or further scaling our Monitron program.

1428.69 -> And it's a disruptive and complimentary technology.

1431.732 -> The disruptive part should be pretty clear

1434.18 -> where we talked about,

1435.013 -> we've been able to get away from our manual route running

1437.96 -> and we're bringing it in house,

1439.43 -> but it's also been complimentary,

1441.29 -> as you can see in figure one and figure two,

1443.9 -> when our technicians respond to it,

1445.82 -> We don't just go out and listen to the vibration data

1449.48 -> and immediately say let's do a repair.

1451.49 -> We want to further investigate and get to the root cause

1453.98 -> of what that problem is before we go through

1455.66 -> a costly repair process.

1457.61 -> And figure one is one where we received

1459.513 -> a machine learning temperature alarm.

1462.17 -> We go out with our thermal imaging cameras,

1464.33 -> we take pictures of the motor gear box on

1466.34 -> the one that we're in question

1467.57 -> and then the one next to it where we can see

1469.49 -> a like for like comparison.

1471.14 -> And we're able to determine

1472.22 -> that it's 30 to 40 degrees fahrenheit more hot than

1476.149 -> the neighboring one under the same load and same conditions.

1479 -> So that's a determination, let's start initiating a repair.

1482.57 -> We've also had a lot of benefit with utilizing

1484.94 -> and pairing this with ultrasonic technologies.

1487.1 -> As you can see in figure two and three.

1489.35 -> Figure two comes right outta the Monitron app

1491.51 -> for where we would respond as our technicians

1493.85 -> and then we take our ultrasonic listening,

1496.37 -> and actually listen to that bearing on the motor

1498.71 -> that we were alerted to with Monitron,

1500.96 -> confirm and set a second opinion

1503.45 -> that yes absolutely there is

1505.16 -> an additional wear in that bearing.

1507.53 -> And then we initiate our repair work after that,

1509.96 -> as opposed to just listening and putting in

1513.26 -> a repair work without doing some kind

1514.97 -> of second confirmation.

1516.35 -> And then also a robust visual based inspection.

1520.88 -> So our first win came during our proof of concept

1523.298 -> and it was, it's actually our single biggest win as well,

1526.46 -> ironically enough.

1527.753 -> But it was our cooling tower that provides chilled water

1530.96 -> to two production areas, 33 machines in that area.

1534.677 -> And they're very sensitive to the room ambient temperature

1538.04 -> to be able to produce their product correctly.

1540.92 -> So we received a high vibration ISO alarm

1543.53 -> that our technician was respondin'

1545 -> and was able to identify

1546.23 -> the gearbox leaking oil and in a early stage of failure.

1549.86 -> But because we were so early in that failure curve,

1552.68 -> if you're familiar with a p to f curve

1554.36 -> and the probability of failure,

1555.92 -> we were able to catch it much sooner.

1557.87 -> And as you can see in the pictures,

1559.73 -> this is very large cooling tower,

1561.08 -> So you can imagine the size of the gearbox as well,

1563 -> larger than the podium here.

1564.77 -> So we had to have a contractor crew

1566.33 -> and a crane come in to do that repair work.

1569.06 -> So the real value is we chose when

1571.058 -> that cooling tower went down,

1573.32 -> we did our repair work at minimal cost instead of having

1576.56 -> an emergency deployment for our contractor.

1579.32 -> And we had no downtime to that production area

1582.2 -> that we served 'cause we was able to rearrange

1584.84 -> our chilled water distribution

1586.82 -> and pick on a day where it was very cool,

1588.92 -> So we wouldn't have as much temperature problems.

1591.762 -> Our estimated value creation for that one single instance

1595.76 -> was roughly 400 machine hours of downtime avoidance when you

1598.94 -> added it all together.

1600.11 -> So as a major impact and a very big first win.

1605.69 -> So some other things that we've been able to learn

1607.7 -> and establish as we've been going through this journey,

1610.46 -> normal operating baselines in a matter

1612.8 -> of weeks instead of years.

1614.09 -> So we're able to determine how our equipment is performing

1616.97 -> and what is and isn't normal

1619.31 -> In figure one, You can see where we had normal operation

1622.46 -> for a large window of time,

1624.35 -> and some equipment vibrates simply at

1626.75 -> a larger RMS velocity than other equipment.

1629.21 -> But it still can be considered normal,

1631.01 -> especially in a facility that is older like ours,

1634.04 -> some equipment just is already

1635.63 -> at an early, early stage of failure.

1637.79 -> But we're not gonna respond to it at that time.

1640.4 -> We're not getting a pure baseline

1641.84 -> from a beginning installation.

1645.5 -> We're able to improve our asset maintenance plans.

1648.35 -> So one of the issues that we found

1650.84 -> in some of our rotating spares

1653 -> where we'll have redundant systems.

1655.25 -> Which basically will mean every, you know,

1656.72 -> roughly six months we'll switch from one pump

1658.67 -> and motor to another one.

1660.637 -> And then we learned that our maintenance practices

1662.87 -> were not sufficient for the motor

1664.7 -> that was not being ran at that time.

1667.01 -> 'Cause we continually found when we switched,

1670.52 -> within a couple of weeks,

1671.81 -> a lot of times we were getting early stage coupling failure,

1674.48 -> or finding out other issues from where that motor

1677 -> had been setting dormant for about six months.

1679.28 -> So we did a root cause analysis

1681.02 -> and we was able to identify some

1682.818 -> of our maintenance practices that we were not doing

1685.43 -> when that motor was not operational.

1687.44 -> We've changed those to try to eliminate that,

1689.45 -> Cause granted it is a redundant system,

1691.61 -> but if you lose one of the motors and pumps,

1693.86 -> now it's not a redundant system

1695.61 -> and it is for absolutely critical equipment in our facility.

1699.32 -> So we've also been able to identify the differences between

1701.63 -> the ISO and machine learning alarms,

1703.91 -> we have responded to

1706.216 -> and got tangible wins from all varieties

1708.26 -> of alarms provided by the Monitron system.

1710.84 -> So we get the ISO alarms and we're seeing high volume,

1713.93 -> high RMS velocity.

1715.94 -> We're not capturing the trend over time patterns,

1718.13 -> like we get in our machine learning program,

1720.17 -> but we're able to see this should not be as high as it is

1723.89 -> and we're able to go out and investigate.

1725.84 -> And of course the way this functions after six hours,

1728.57 -> you're able to start bringing in those ISO alarms

1730.73 -> and you have to wait a 21 day period

1732.41 -> for your machine learning to take in effect.

1734.15 -> So you're able to start getting alarms

1736.43 -> within six hours of installation,

1738.8 -> and then within a couple of days,

1740.51 -> you have enough trend data

1741.92 -> to see how that's been progressing

1743.78 -> and start responding to it.

1745.82 -> It's also been something that we've been able to transform.

1749.66 -> In our plant, we already had

1751.01 -> a predictive maintenance program and culture established,

1753.8 -> but as we've been working on other facilities

1755.9 -> that do not have that program established,

1758.09 -> Monitron is able to do as a foundational

1760.67 -> for all of the predictive maintenance in the areas.

1763.49 -> We're able to eliminate our manual route running not only

1766.19 -> from vibration analysis,

1768.26 -> but ultrasonic analysis and thermal imaging routes,

1771.41 -> utilizing the capabilities of the Monitron program

1774.47 -> to monitor a large amount of assets with minimal amount

1777.71 -> of technicians, and then use those technician time

1780.71 -> to do supplemental technologies when we get

1784.075 -> those Monitron alarms, as I showed in the previous slide.

1786.83 -> So in figure two is probably one of our best examples

1790.61 -> of how we've been learning our Monotron alarm

1793.13 -> and the machine learning program.

1794.78 -> It was one of those redundant systems I made

1797.24 -> an example of a second ago,

1798.877 -> you could see where it turned on,

1800.72 -> and that's where we switched from one pump to the other.

1803.06 -> We actually got our first machine learning alarm in an area

1805.82 -> that we would've very much considered normal operation

1808.28 -> based off of not having

1809.48 -> a very trend over time change in pattern.

1812.81 -> However, as we continued to watch it,

1814.61 -> then we started seeing a exponential climb towards

1817.61 -> the end of that graph.

1818.93 -> And as you can see, it continued getting worse

1820.94 -> and worse until we pulled the trigger,

1822.74 -> did the work order, was still able to get it,

1824.72 -> catch it before any kind of failure.

1827.21 -> And that hearkens back to, like A.K, was talking about

1828.976 -> with the supply chain woes and everything like that.

1832.4 -> We aren't able to get motors and pumps and gear boxes

1836.135 -> in a week like we used to.

1838.34 -> Sometimes it's a month lead time, two month,

1840.32 -> three month lead times.

1841.76 -> So the more early in advance we're able to determine

1845.15 -> there is a failure pending on this equipment.

1847.76 -> We're able to order our supplies and not have to depend off

1851.09 -> of our stockroom supplies or have extended amount

1854 -> of downtime because we cannot get the part in time.

1857.57 -> And that's the real value of

1858.92 -> the predictive maintenance in the Monitron.

1862.55 -> So our journey continues,

1864.08 -> so we're only one site out of

1865.46 -> the 72 facilities that Baxter has.

1868.094 -> We started with the 400 sensor proof of concept

1871.13 -> and that was in November, December of last year.

1874.52 -> We have since scaled the site to 2,600 sensors

1877.49 -> with the last few going in during

1879.26 -> our plan maintenance outage in this upcoming Christmas

1882.08 -> and you know, areas that we're waiting

1883.55 -> to shut down equipment for confined space entry.

1886.79 -> But that's just one year, 2,600 sensors and a team of five,

1891.65 -> four predictive maintenance technicians

1893.39 -> and myself have been helping to lead this.

1896.81 -> And now we're scaling to other facilities.

1899.57 -> We've actually got one facility that had

1901.25 -> to shut down earlier than ours

1902.81 -> and they're actually at a hundred percent already.

1904.64 -> So they've, they've surpassed us with our deployment

1907.07 -> even though we're the original site.

1908.57 -> But that just shows you the speed and scalability of

1910.576 -> the deployment solutions that we're able to do.

1913.28 -> A team of two to three individuals

1915.08 -> can install several hundred sensors

1917.15 -> with machine downtime in a matter of a couple of days,

1920.75 -> and you start getting data within six hours after that.

1925.07 -> So over the next of our real five year plan

1927.52 -> as we're targeting to finish our global deployment

1929.66 -> at the other 70 production facilities,

1934.52 -> and I think we'll turn it over

1936.08 -> to a question and answer section.

Source: https://www.youtube.com/watch?v=hvIjkOEe9C4