predictive maintenance on azure iot -...
TRANSCRIPT
Predictive Maintenanceon Azure IoTIoT와 Big Data 를 활용한 예지 정비 시스템 구축
In Kee Paek, Cloud Solution Architect
IoT 기반 predictive maintenance
를 통해 장비의 다운타임을 50%1
까지 절감할 수 있음
Predictive maintenance 적용으로
설비 종합 효율(OEE)을 30%3
까지 증대 시킬 수 있음
다운타임 최소화를 통해 제조사는
연간 약 $630B 의 잠재적 수익
효과를 볼 것으로 예측됨 (2025년기준)
특정 비즈니스에서는, 계획된 유지보수 작업이 계획되지 않은 수리
작업에 비해 10배이상 비용이
절약될 수 있음
1 McKinsey, The Internet of Things: Mapping the Value Beyond the Hype, 2015
2 McKinsey, The Internet of Things: Mapping the Value Beyond the Hype, 2015
3 GE, The Impact of No Unplanned Downtime, 2014
4 GE, The Impact of No Unplanned Downtime, 2014
“In contrast to a traditional preventive maintenance system, predictive maintenance solutions enables customers to strategically plan their maintenance tasks and group them in a way that allows them to perform the required maintenance
more efficiently…This leads to even more dramatic savings in terms of labor and maintenance costs” – Melissa Topp, Director of Global Marketing, ICONICS
Address business needs across a range of scenarios…
가동시간(uptime) 최적화 및장비 기대 수명의 연장을통해 제품의 품질 개선
Gain visibility into product performance and enable workflows to respond to changing conditions.
43.1° | 35.2°
91.6° | 87.3°
MAINTENANCE INITIATED
고객 맞춤의 선도적인 서비스대응을 통해 고객 관계 및충성도 강화
Offer predictive and proactive services to address customer and product needs.
데이터 기반의 혁신적이고차별화된비즈니스 모델 개발및 기존 모델의 개선
Identify trends and potential growth opportunities using customer sentiment and product usage data.
ADD BESTPRACTICE
절감된 비용을 제품디자인이나서비스의 개선에투자
Utilize service and performance data to proactively detect product failure.
실시간 제조 공정에 예지분석 기술을효과적으로 적용한 사례
Jabil
“We’ll be able to improve our efficiencies, cut costs, and decrease our lead times,
which tie directly to our customers’ requirement to increase flexibility.”
Matt Behringer
CIO, Enterprise Operations and Quality Systems, Jabil
구현 목적Jabil wanted to better predict
errors or failures on the
assembly floor before they
occur, saving customers’ time
and money.
실행 전략Jabil was able to
transform their
manufacturing production
lines with advanced
analytics solutions built
on Microsoft technology.
적용 성과• Predicted machine processes
that will slow down or fail with
an 80% accuracy
• Reduced costs of scrap and re-
work by 17%
• Delivered energy savings of 10%
As-Is 이슈및문제
• Reduced manufacturing cycle time
• Higher cost of wasted materials, time and resources
• Inability to address customers’ critical requirement for speed to market
Product
quality not
acceptable
Jabil 이당면한비즈니스과제
• Continuous requirement to increase yield, reduce amount of scrap and re-work
• Traditional inspection techniques for ensuring quality quickly becoming
outdated
with more one-off production runs
• Adding more equipment and people to existing manufacturing processes would
not have significant impact on increasing throughput
Inspection steps along the SMT line cannot always detect the quality issues
Source of failure can be introduced at multiple stages but cannot be detected until it ispowered-up for testing at the end
디지털 기반 예지정비 역량 강화로 제품 품질 개선 실현
0
0
1
1
00
0
1
1
1
0
0
0
0
1
1
00
0
1
1
1
0
0
0
0
10
1
0
0
0
1
1
1
0
0
0
0
1
1
00
0
1
1
1
0
0
0
0
1
1
00
0
1
1
1
0
0
0
0
1
1
00
0
1
1
1
0
0
00
1
1
0
0
0
1
1
0
0
0
10
1
0
0
0
1
1
1
0
0
0
0
1
0
0
1
0
0
10
1
0
0
0
1
1
적용결과
• Predicted machine processes that will slow down
or fail with an 80% accuracy
• Reduced costs of scrap and re-work by 17%
• Delivered energy savings of 10%
Variation: 11% (tolerance ≤ 11%)
Vibration frequency: Too high
Bit wear: High
Result in plant 2: Failed
PREDICTED
FAILURERecommended maintenance
in next 48 hours
PREVENTATIVE
MAINTENANCE
FOR TOMORROW
10 TuesdayOctober
7:00 AM
실행전략
Jabil was able to transform their manufacturing
production lines with advanced analytics solutions
built on Microsoft technology
001
0
100
1
1
001
100
001
100 0
01
100
001
1
1
100
001
00
100
CRM ERP MES SPC
001
100
01
00
1
1
010
11
구현목적
Jabil wanted to better predict errors or failures on
the assembly floor before they occur, saving
customers’ time and money
001
100
001
0
100
1
1
001
100
Other systems
00
1
100
0
11
10
0
00
1
100
0
11
10
0
00
1
100
0
11
1
0
0
00
1
100
0
11
10
0
00
1
100
0
11
10
0
00
1
100
0
11
10
0
00
1
100
0
11
1
0
0
00
1
100
0
11
11
00
1
100
0
11
10
0
00
1
100
0
11
10
0
00
1
100
0
11
1
0
0
00
1
100
0
11
10
0
00
1
100
0
11
10
0
00
1
100
0
11
10
0
00
1
100
0
11
1
0
0
00
1
100
100
100
00
0
11
0
11
0
11 1
0 0
0
00
1
100
0
11
10
0
00
1
100
0
11
10
0
00
1
100
0
11
1
0
0
00
1
100
0
11
10
0
0
110
0
00
1
100
0
11
10
0
00
1
100
0
11
1
0
0
00
1
100
100
100
100
0
11
0
11
0
11
0
1
00
1
0100
10
11
0
10
0
1
00
1
1
00
11
0
0
00
1
0 0
100
0
11
1
0
0
00
1
0100
10
11
0
10
0
1
1 0
00
1
100
0
11
10
0
11
1
00
11
1
0
0 1 0
00
1
1
100
0
0
11
1
10
0
0
00
10
0
110
00
1
1
100
0
110
0
10
0
11
10
0
0
1 00
1
100
0
11
10
0
00
10
0
11
1
0
0
00
1
PREDICTED FAILURE
IN NEXT 48 HOURS
Supplier data
Customer
data
Production
data
Historical data
Azure
Services Our rate of rejection has decreased dramatically now
that we can predict these failures early in the process
디지털 기반 예지정비 역량 강화로 제품 품질 개선 실현
예지 정비 도입으로 글로벌 현장의엘리베이터 운영 역량 극대화
ThyssenKrupp Elevator
“We wanted to go beyond the industry standard of preventative maintenance,
to offer predictive and even preemptive maintenance.”
Andreas Schierenbeck
CEO, ThyssenKrupp Elevator
구현 목적ThyssenKrupp wanted to
better compete in their
industry by offering
dramatically increased
uptime, taking preventative
maintenance a step further to
predictive and even
preemptive service.
실행 전략Microsoft technology
enabled ThyssenKrupp to
monitor products via a
real-time dashboard and
instruct technicians on
optimal maintenance
activities through
dynamic predictive
models.
적용 성과• Increased elevator uptime
• Reduced costs for ThyssenKrupp
and its customers
• Developed real-time data
visualization and awareness of
issues
현장 기술자에 예지분석 기반의 정비 가이드를 적시에 제공
구현목적Thyssenkrupp wanted to better monitor their more
than 1.1 million elevators worldwide. Lack of insight
led to downtime and unpredicted failures in some
of the world’s most famous buildings
수행전략They connected thousands of sensors embedded in
their elevators to the cloud to monitor real-time
performance and proactively address issues with
Microsoft technology
적용성과
• Reduced costs for thyssenkrupp and its customers
• Increased reliability through predictive maintenance
and rapid, remote diagnostic capabilities
Maintenance in progress
Scheduled Maintenance
• Hydraulic/Passenger Elevator
• Replace brake shoes
• Check AC motor for health
!
!NEXT STOP
Microsoft Offers Two Approaches to IoT Solutions
PaaS – Azure IoT Suite
SaaS – Microsoft IoT Central
PaaS
SaaS
Microsoft Offers Two Approaches to IoT Solutions
Azure IoT Suite Microsoft IoT Central
Primary usageTo accelerate development of a custom IoT solution
that needs maximum flexibility
To accelerate time to market for straightforward IoT solutions
that don’t require deep service customization
Access to underlying
PaaS Services
Access to the underlying Azure services to
manage them, or replace them as needed.
SaaS. Fully managed solution,
underlying services aren't exposed.
FlexibilityHigh. The code for the microservices is open
source to be modified.
Medium. leverage built-in browser based user experience to cust
omize the solution model and aspects of the UI.
Skill level
Medium-High
Java or .NET skills are required to customize the solution back end.
JavaScript are required to customize the visualization.
Low
Modeling skills are required to customize the solution.
No coding skills are required.
Get started experiencePreconfigured solutions implement common
IoT scenarios. Can be deployed in minutes.Templates provide pre-built models.
Can be deployed in minutes.
Pricing You can fine-tune the services to control the cost. Simple, predictable pricing structure.
Azure IoT Suite - Accelerate Time to Value!
Modify existing rules and alerts
Fine-tuned to specific assets and processes
Integrate with back-end systems
Highly visual for your real-time operational data
Get started in minutes
Add your devices and begin tailor to your needs
Azure IoT Suite Solutions
What you get with the preconfigured solution?
Azure IoT Suite Predictive Maintenance, Remote Monitoring
Existing
Business
Process
ERP/CRM
Devices
Azure IoT SDK
(OSS)
Linux, RTOS, mBed,
Windows,
Android, iOS
Event Hub
Storage blobs DocumentDB
Web/
Mobile App
Stream
Analytics
Logic Apps
Azure
Active Directory
IoT Hub Web Jobs
Power BI
Machine Learning
Azure IoT Suite Reference Architecture
Cloud Gateway
Web App
Strea Analytics
(Rules/Insights)
IoT Edge(s)
IoT Device(s)
User
Management
Machine
LearningCold Path
StoreWarm Data
Config. &
Rules
Definition
Data Xformation
Microservices
Notifications /
Actions
Orchestration
Stream Analytics
(Rules/Insights)
AAD
OAuth2 Providers
Logic Apps
Event Grid
Azure Storage
Cosmos DB
Azure MLTSI
IoTHub
App Service
Kubernetes
Service Fabric
Spark
AF
ASA
Optional
Tech Options
Predictive Maintenance project 수행 절차
잠재적 이슈와 고장이 발생하기 전 자동 감지를 효과적으로 수행하기 위한 가이드
Azure IoT Suite solutions come with pre-built sample scenarios that include:
• Background information on the business need and objectives
• Simulated devices and sample data
• Pre-set rules and alerts, pre-defined dashboards and more
Identify the
target outcome
Inventory data
sources
Capture &
combine data
Model, test and
integrate
Integrate into
operations
1 2 3 4 5
Approach: Checklist for Machine Learning
Question is sharp.
Data measures what you care about.
Data is connected.
Data is accurate.
A lot of data.
The better the raw materials, the better the product.
E.g. Predict whether component X will fail in the next Y days
E.g. Identifiers at the level they are predicting
E.g. Will be difficult to predict failure accurately with few examples
E.g. Failures are really failures, human labels on root causes
E.g. Machine information linkable to usage information
Approach: Data Science Process
Define Objective
Data Sources
Explore Data
Machine Learning
Analysis Dataset Publish
Approach: Data Sources for Predictive Modeling
The failure history of a machine or
component within the machine.
The repair history of a machine, e.g.
previous maintenance records,
components replaced, maintenance
activities performed.
The operating characteristics of a
machine, e.g. data collected from
sensors.
FAILURE HISTORY REPAIR HISTORY MACHINE CONDITIONS
The features of machine or
components, e.g. production date,
technical specifications.
Environmental features that may
influence a machine’s performance,
e.g. location, temperature, other
interactions.
The attributes of the operator who
uses the machine, e.g. driver.
MACHINE FEATURES OPERATING CONDITIONS OPERATOR ATTRIBUTES
Approach: Predictive Modeling Techniques
Predict failures within a future period of time
BINARY CLASSIFICATION
Predict failures with their causes within a future
time period.
Predict remaining useful life within ranges of future
periods
MULTICLASS CLASSIFICATION
Predict remaining useful life, the amount of time
before the next failure
REGRESSION or SURVIVAL ANALYSIS
Identify change in normal trends to find anomalies
ANOMALY DETECTION
Predictive Maintenance framework
1Identify the target outcomeDetermine what outcome you ultimately want to achieve. In this case, we want to know how long until any given AC unit fails.
I want to understand how much time each
AC unit has left before it needs maintenance
so we can prevent unplanned failures.
Last time an AC unit failed, it cost
thousands of dollars and operations
were down for days.
AC unit
out of order
Predictive Maintenance framework
2Inventory data sourcesIdentify all potential sources of data, including the types of data and the amounts available. The outcome you are seeking will influence what data is essential and what’s optional.
100101011000
101000101101
010011001110
101000110011
Performance
data
Maintenance logs
Weather data Failure logs
Sensor data
Include data from a variety
of sources – you may be
surprised about the places
where key information can
come from.
3Capture and combine dataConnect all of your data to a single place and prepare it for analysis.
Lay the groundwork for
a robust predictive model
by pulling in data that
includes both expected
behavior and failure logs.
Predictive Maintenance framework
4Model, test, and iterate
Identify unexpected patterns by developing statistical models using advanced analytics techniques. Stank-rank models to determine which model is best at forecasting the timing of AC unit failures.
Make your model
actionable by understanding
how much advance notice
the maintenance team needs
in order to respond.
▪ Model A
▪ Model B
▪ Model C
Predictive Maintenance framework
5Validate model using your latest data
Apply your model to live, streaming data and observe how it works in real-world conditions.
Use machine learning to refine your model and ready it for full implementation.
Be willing to refine your
approach based on the
data you gather during
the real-world pilot.
100101011000
101000101101
010011001110
101000110011
Predictive Maintenance framework
6Integrate into operationsOperationalize the model by adjusting maintenance processes, systems and resources to act on near-real-time data. Make ongoing improvements by gaining insights from machine learning and advanced analytics.
Strengthen your processes
and procedures to take
advantage of what you learn.
3 weeks until failure: Order
replacement part
2 weeks until failure: Send
repair team
Predictive Maintenance framework